Computational Modeling of Drugs Against Alzheimer’s Disease [2 ed.] 1071633104, 9781071633106

Citation preview

Neuromethods 203

Kunal Roy  Editor

Computational Modeling of Drugs Against Alzheimer’s Disease Second Edition

NEUROMETHODS

Series Editor Wolfgang Walz University of Saskatchewan Saskatoon, SK, Canada

For further volumes: http://www.springer.com/series/7657

Neuromethods publishes cutting-edge methods and protocols in all areas of neuroscience as well as translational neurological and mental research. Each volume in the series offers tested laboratory protocols, step-by-step methods for reproducible lab experiments and addresses methodological controversies and pitfalls in order to aid neuroscientists in experimentation. Neuromethods focuses on traditional and emerging topics with wide-ranging implications to brain function, such as electrophysiology, neuroimaging, behavioral analysis, genomics, neurodegeneration, translational research and clinical trials. Neuromethods provides investigators and trainees with highly useful compendiums of key strategies and approaches for successful research in animal and human brain function including translational “bench to bedside” approaches to mental and neurological diseases.

Computational Modeling of Drugs Against Alzheimer’s Disease Second Edition

Edited by

Kunal Roy Department of Pharmaceutical Technology, Jadavpur University, Kolkata, West Bengal, India

Editor Kunal Roy Department of Pharmaceutical Technology Jadavpur University Kolkata, West Bengal, India

ISSN 0893-2336 ISSN 1940-6045 (electronic) Neuromethods ISBN 978-1-0716-3310-6 ISBN 978-1-0716-3311-3 (eBook) https://doi.org/10.1007/978-1-0716-3311-3 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Humana imprint is published by the registered company Springer Science+Business Media, LLC, part of Springer Nature. The registered company address is: 1 New York Plaza, New York, NY 10004, U.S.A.

Dedication For Aatreyi, Arpit and Chaitali

v

About the Editor KUNAL ROY is Professor and Ex-Head at the Department of Pharmaceutical Technology, Jadavpur University, Kolkata, India (https://sites.google.com/site/kunalroyindia). He has been a recipient of Commonwealth Academic Staff Fellowship (University of Manchester, 2007) and Marie Curie International Incoming Fellowship (University of Manchester, 2013) and a former visiting scientist of Istituto di Ricerche Farmacologiche “Mario Negri” IRCCS, Milano, Italy. The field of his research interest is Quantitative Structure-Activity Relationship (QSAR) and Molecular Modeling with application in Drug Design, Property Modeling and Predictive Ecotoxicology. Dr. Roy has published more than 350 research articles (ORCID: http://orcid. org/0000-0003-4486-8074) in refereed journals (current SCOPUS h index 51; total citations till date > 13,000). He has also coauthored two QSAR related books (Academic Press and Springer), edited seven QSAR books (Springer, Academic Press, Wiley and IGI Global) and published more than ten book chapters. Dr. Roy is the Co-Editor-in-Chief of Molecular Diversity (Springer Nature).

vii

Preface to the Series Experimental life sciences have two basic foundations: concepts and tools. The Neuromethods series focuses on the tools and techniques unique to the investigation of the nervous system and excitable cells. It will not, however, shortchange the concept side of things as care has been taken to integrate these tools within the context of the concepts and questions under investigation. In this way, the series is unique in that it not only collects protocols but also includes theoretical background information and critiques which led to the methods and their development. Thus, it gives the reader a better understanding of the origin of the techniques and their potential future development. The Neuromethods publishing program strikes a balance between recent and exciting developments like those concerning new animal models of disease, imaging, in vivo methods, and more established techniques, including, for example, immunocytochemistry and electrophysiological technologies. New trainees in neurosciences still need a sound footing in these older methods in order to apply a critical approach to their results. Under the guidance of its founders, Alan Boulton and Glen Baker, the Neuromethods series has been a success since its first volume published through Humana Press in 1985. The series continues to flourish through many changes over the years. It is now published under the umbrella of Springer Protocols. While methods involving brain research have changed a lot since the series started, the publishing environment and technology have changed even more radically. Neuromethods has the distinct layout and style of the Springer Protocols program, designed specifically for readability and ease of reference in a laboratory setting. The careful application of methods is potentially the most important step in the process of scientific inquiry. In the past, new methodologies led the way in developing new disciplines in the biological and medical sciences. For example, Physiology emerged out of Anatomy in the nineteenth century by harnessing new methods based on the newly discovered phenomenon of electricity. Nowadays, the relationships between disciplines and methods are more complex. Methods are now widely shared between disciplines and research areas. New developments in electronic publishing make it possible for scientists that encounter new methods to quickly find sources of information electronically. The design of individual volumes and chapters in this series takes this new access technology into account. Springer Protocols makes it possible to download single protocols separately. In addition, Springer makes its print-on-demand technology available globally. A print copy can therefore be acquired quickly and for a competitive price anywhere in the world. Saskatoon, SK, Canada

Wolfgang Walz

ix

Preface Dementia is a syndrome in which a deterioration in cognitive function mainly affects older people, and is currently the seventh leading cause of death among all diseases. It has significant physical, psychological, social, and economic impacts. According to the World Health Organization [1], more than 55 million people are living with dementia worldwide, and there are nearly 10 million new cases every year. Dementia results from a variety of diseases and injuries primarily or secondarily affecting the brain. Alzheimer’s disease (AD) is the most common form of dementia contributing to 60–70% of cases. Alzheimer’s disease (AD) is a heterogeneous disease with a complex pathobiology with the presence of extracellular β-amyloid deposition as neuritic plaques and intracellular accumulation of hyperphosphorylated tau as neurofibrillary tangles, which are the primary neuropathologic criteria for the diagnosis of the disease. Although recent studies highlight important pathological roles for other critical cellular and molecular processes, no diseasemodifying treatment currently exists, and several phase III clinical trials have failed to demonstrate benefits [2]. Although drug development is a time-consuming and expensive process [3], computational biology and bioinformatics methodologies have simplified the initial identification of potential therapeutic targets. For decades, chemometrics, chemoinformatics, and computeraided molecular design approaches have been utilized to identify and optimize novel compounds with improved therapeutic potential and selectivity in various fields. In this perspective, in silico approaches such as quantitative structure-activity relationship (QSAR), chemical read-across, pharmacophore modeling, molecular docking, molecular dynamics (MD) simulations, etc. are playing significant roles in the design and discovery of new compounds with enhanced therapeutic efficacy. Computational modeling techniques are also efficient for multi-target ligand design against multifactorial Alzheimer’s disease [4]. Computational Modeling of Drugs Against Alzheimer’s Disease (2018) presented the applications of different computational methods encompassing ligand- and structure-based approaches for anti-Alzheimer drug design along with different background topics like molecular etiologies of Alzheimer’s disease, targets for new drug development, and different cheminformatic modeling strategies [5]. The current volume presents a second edition supplementing the information on the recent developments mainly from 2018 onward. This volume has four parts. In the first part, Kumar and Roy summarize and highlight recent advancements in research on the development of novel therapies and their implications in the treatment of AD. The second part deals with the recent progress in computational modeling of anti-Alzheimer’s drugs. In this part, Rao et al. describe a systematic in silico study of Aβ42 self-assemblies and their interactions with the capped VQIVYK hexapeptide, to understand the mechanisms of Aβ42 fibrillogenesis. Papavasileiou et al. discuss the recent developments in the computational modeling search for novel BACE1 inhibitors in the next chapter. Froes et al. in the next chapter discuss that the synergy of ligand- and structure-based molecular modeling tools helps to gain a deeper understanding of BACE-1 conformational flexibility, its allosteric regulation, and identify novel molecular scaffolds that explore cryptic and allosteric sites in BACE-1 or improve BACE-1/BACE-2 selectivity profile. De and Roy have extensively reviewed different tau kinases, their systemic roles, and

xi

xii

Preface

the mechanism of tau phosphorylation relevant to cause AD. Jain et al. focus on the computer-assisted design of some novel Tau kinase inhibitors. Kumar and Roy discuss different structure- and ligand-based strategies applied to the computational modeling of anti-Alzheimer drugs and present some case studies of the development of phosphodiesterase inhibitors. Thai et al. highlight various computational modeling methods utilized in designing multi-targeting inhibitors as anti-Alzheimer agents. De and Roy have reviewed different computational modeling studies recently performed on the PET and SPECT diagnostic imaging agents for AD. Part III presents computational studies against some newer targets for anti-Alzheimer drug development. Serrano-Candelas et al. explored different ligand- and structure-based computational approaches to develop a workflow for the selection of potential candidates as inhibitors of DYRK1A. Kaur et al. compile recent computational studies on the natural and synthetic inhibitors of monoamine oxidases. Tsopka and Hadjipavlou-Litina discuss the results of computational modeling studies on phosphodiesterase inhibitors as anti-Alzheimer agents. Bagri et al. review the design and development of glutaminyl cyclase inhibitors for the use in Alzheimer’s disease. Part IV presents some special topics related to anti-Alzheimer drug development. Lushington et al. discuss how non-neurological disorders may lay foundations of neurodegeneration long before clinical neuropathology manifests and offer information-based suggestions for how to pursue novel target perception, outlining simple recipes and examples of how literature query tools and techniques may illuminate insight from emerging research trends. Dash et al. discuss the importance and success of drug repositioning in anti-Alzheimer drug development and the prospects and methodologies of network pharmacology in understanding various aspects of drug repositioning. Perkin et al. provide an overview of 13 online tools that can be used to predict the ADMET profile of a compound, with a special focus on the properties such as lipophilicity, aqueous solubility, hERG inhibition, blood–brain barrier permeability, Caco-2 permeability, human intestinal absorption, and Ames mutagenicity in the context of anti-Alzheimer drug development. With the above 16 chapters, the current volume supplements the previous edition of Computational Modeling of Drugs Against Alzheimer’s Disease [5]. We hope that the new additions will be useful to the researchers working in the field. Kunal Roy Kolkata, West Bengal, India References 1. https://www.who.int/news-room/fact-sheets/detail/dementia 2. Long JM, Holtzman DM (2019) Alzheimer disease: an update on pathobiology and treatment strategies. Cell 179(2):312–339. https://doi.org/10.1016/j.cell.2019.09.001 3. Roy K (ed) (2017) Advances in QSAR modeling. Applications in pharmaceutical, chemical, food, agricultural and environmental sciences. Springer, New York. http:// www.springer.com/in/book/9783319568492 4. Kumar V, Saha A, Roy K (2023) Multi-target QSAR modeling for the identification of novel inhibitors against Alzheimer’s disease. Chemom Intell Lab Syst 233:104734. https:// doi.org/10.1016/j.chemolab.2022.104734 5. Roy K (ed) (2018) Computational modeling of drugs against Alzheimer’s disease. Springer, New York. http://www.springer.com/in/book/9781493974030

Contents Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . About the Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preface to the Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contributors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

PART I

RECENT PROGRESS IN UNDERSTANDING MOLECULAR ETIOLOGY OF ALZHEIMER’S DISEASE

1 Recent Progress in the Treatment Strategies for Alzheimer’s Disease . . . . . . . . . . Vinay Kumar and Kunal Roy

PART II

v vii ix xi xv

3

RECENT ADVANCES IN COMPUTATIONAL MODELING OF ANTI-ALZHEIMER DRUGS

2 Understanding the Mechanisms of Amyloid Beta (Aβ) Aggregation by Computational Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Praveen P. N. Rao, Yusheng Zhao, and Rui Huang 3 Recent Advances in Computational Modeling of BACE1 Inhibitors as Anti-Alzheimer Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Konstantinos D. Papavasileiou, Francesco Dondero, Georgia Melagraki, and Antreas Afantitis 4 Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents . . . . . . . . . . . . . . . . . Thamires Quadros Froes, Deyse Brito Barbosa, Mayra Ramos do Bomfim, Franco Henrique Andrade Leite, and Marcelo Santos Castilho 5 Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents . . . . . . Priyanka De and Kunal Roy 6 Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors and Its Implications in Alzheimer’s Disease . . . . . . . . . . . . . . . . . . . . . . . Arvind Kumar Jain, C. Karthikeyan, Piyush Trivedi, and Anita Dutt Konar 7 Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease Agents: Case Studies of Phosphodiesterase Inhibitors . . . . . . . . . . . . . . . . Vinay Kumar and Kunal Roy 8 Recent Advances in Computational Modeling of Multi-targeting Inhibitors as Anti-Alzheimer Agents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Khac-Minh Thai, Thai-Son Tran, The-Huan Tran, Thi-Cam-Nhung Cao, Hoang-Nhan Ho, Phuong Nguyen Hoai Huynh, Tan Thanh Mai, Thanh-Dao Tran, Minh-Tri Le, and Van-Thanh Tran 9 Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for Alzheimer’s Disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Priyanka De and Kunal Roy

xiii

51

73

99

127

165

187

231

279

xiv

Contents

PART III

COMPUTATIONAL MODELING OF ANTI-ALZHEIMER DRUGS AGAINST NEWER TARGETS

10

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Eva Serrano-Candelas, Laureano E. Carpio, and Rafael Gozalbes 11 Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents . . . . . . . Gurmeet Kaur, Deepti Goyal, and Bhupesh Goyal 12 Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina 13 Computational Methods for the Design and Development of Glutaminyl Cyclase Inhibitors in Alzheimer’s Disease . . . . . . . . . . . . . . . . . . . . . Kiran Bagri, Ashwani Kumar, Parvin Kumar, Archana Kapoor, and Vikas Verma

PART IV

295 325

355

383

SPECIAL TOPICS

14

Basic Information Science Methods for Insight into Neurodegenerative Pathogenesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407 Thomas H. W. Lushington, Mary I. Zgurzynski, and Gerald H. Lushington 15 Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433 Raju Dash, Yeasmin Akter Munni, Sarmistha Mitra, Nayan Dash, and Il Soo Moon 16 Web Services for the Prediction of ADMET Parameters Relevant to the Design of Neuroprotective Drugs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465 Valentin O. Perkin, Grigory V. Antonyan, Eugene V. Radchenko, and Vladimir A. Palyulin Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

487

Contributors ANTREAS AFANTITIS • NovaMechanics Ltd, Nicosia, Cyprus; NovaMechanics MIKE, Piraeus, Greece KIRAN BAGRI • Department of Pharmaceutical Sciences, Guru Jambheshwar University of Science & Technology, Hisar, India DEYSE BRITO BARBOSA • Departamento de Sau´de, Universidade Estadual de Feira de Santana, Feira de Santana, BA, Brazil MAYRA RAMOS DO BOMFIM • Departamento de Sau´de, Universidade Estadual de Feira de Santana, Feira de Santana, BA, Brazil THI-CAM-NHUNG CAO • Faculty of Pharmacy, Hue University of Medicine and Pharmacy, Hue University, Hue City, Vietnam LAUREANO E. CARPIO • ProtoQSAR SL, Centro Europeo de Empresas Innovadoras, Parque Tecnologico de Valencia, Valencia, Spain MARCELO SANTOS CASTILHO • Faculdade de Farma´cia da Universidade Federal da Bahia, Salvador, BA, Brazil NAYAN DASH • Department of Computer Science and Engineering, BGC Trust University Bangladesh, Chittagong, Bangladesh RAJU DASH • Department of Anatomy, Dongguk University College of Medicine, Gyeongju, Korea; Department of New Biology, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Daegu, Republic of Korea PRIYANKA DE • Drug Theoretics and Cheminformatics Laboratory, Department of Pharmaceutical Technology, Jadavpur University, Kolkata, India FRANCESCO DONDERO • Department of Science and Technological Innovation (DISIT), ` del Piemonte Orientale “Amedeo Avogadro” – Alessandria, Novara, Vercelli, Italy Universita THAMIRES QUADROS FROES • Faculdade de Cieˆncias Farmaceˆuticas de Ribeira˜o Preto – Universidade de Sa˜o Paulo, Sa˜o Paulo, SP, Brazil BHUPESH GOYAL • School of Chemistry & Biochemistry, Thapar Institute of Engineering & Technology, Patiala, Punjab, India DEEPTI GOYAL • Department of Chemistry, DAV College, Sector 10, Chandigarh, India RAFAEL GOZALBES • ProtoQSAR SL, Centro Europeo de Empresas Innovadoras, Parque Tecnologico de Valencia, Valencia, Spain; Moldrug AI Systems SL, Valencia, Spain DIMITRA HADJIPAVLOU-LITINA • Department of Pharmaceutical Chemistry, School of Pharmacy, Faculty of Health Sciences, Aristotle University of Thessaloniki, Thessaloniki, Greece HOANG-NHAN HO • Faculty of Pharmacy, Hue University of Medicine and Pharmacy, Hue University, Hue City, Vietnam RUI HUANG • Department of Chemistry, University of Guelph, Guelph, ON, Canada PHUONG NGUYEN HOAI HUYNH • Faculty of Pharmacy, University of Medicine and Pharmacy at Ho Chi Minh City, Ho Chi Minh City, Vietnam ARVIND KUMAR JAIN • School of Pharmaceutical Sciences, Rajiv Gandhi Technological University, Gandhinagar, Bhopal, India ARCHANA KAPOOR • Department of Pharmaceutical Sciences, Guru Jambheshwar University of Science & Technology, Hisar, India C. KARTHIKEYAN • Department of Pharmacy, Indira Gandhi National Tribal University, Amarkantak, MP, India

xv

xvi

Contributors

GURMEET KAUR • School of Chemistry & Biochemistry, Thapar Institute of Engineering & Technology, Patiala, Punjab, India ANITA DUTT KONAR • School of Pharmaceutical Sciences, Rajiv Gandhi Technological University, Gandhinagar, Bhopal, India; Department of Applied Chemistry, Rajiv Gandhi Technological University, Bhopal, India ASHWANI KUMAR • Department of Pharmaceutical Sciences, Guru Jambheshwar University of Science & Technology, Hisar, India PARVIN KUMAR • Department of Chemistry, Kurukshetra University, Kurukshetra, India VINAY KUMAR • Drug Theoretics and Cheminformatics Laboratory, Department of Pharmaceutical Technology, Jadavpur University, Kolkata, India MINH-TRI LE • Faculty of Pharmacy, University of Medicine and Pharmacy at Ho Chi Minh City, Ho Chi Minh City, Vietnam; School of Medicine, Vietnam National University Ho Chi Minh City, Ho Chi Minh City, Vietnam FRANCO HENRIQUE ANDRADE LEITE • Departamento de Sau´de, Universidade Estadual de Feira de Santana, Feira de Santana, BA, Brazil GERALD H. LUSHINGTON • LiS Consulting, Lawrence, KS, USA; Qnapsyn Biosciences, Inc., Oklahoma City, OK, USA THOMAS H. W. LUSHINGTON • LiS Consulting, Lawrence, KS, USA TAN THANH MAI • Faculty of Pharmacy, University of Medicine and Pharmacy at Ho Chi Minh City, Ho Chi Minh City, Vietnam GEORGIA MELAGRAKI • Division of Physical Sciences and Applications, Hellenic Military Academy, Vari, Greece SARMISTHA MITRA • Department of Anatomy, Dongguk University College of Medicine, Gyeongju, Korea IL SOO MOON • Department of Anatomy, Dongguk University College of Medicine, Gyeongju, Korea YEASMIN AKTER MUNNI • Department of Anatomy, Dongguk University College of Medicine, Gyeongju, Korea VLADIMIR A. PALYULIN • Department of Chemistry, Lomonosov Moscow State University, Moscow, Russia KONSTANTINOS D. PAPAVASILEIOU • NovaMechanics Ltd, Nicosia, Cyprus; NovaMechanics MIKE, Piraeus, Greece PRAVEEN P. N. RAO • School of Pharmacy, Health Sciences Campus, University of Waterloo, Waterloo, ON, Canada KUNAL ROY • Drug Theoretics and Cheminformatics Laboratory, Department of Pharmaceutical Technology, Jadavpur University, Kolkata, India; Drug Theoretics and Cheminformatics Laboratory, Department of Pharmaceutical Technology, Jadavpur University, Kolkata, India EVA SERRANO-CANDELAS • ProtoQSAR SL, Centro Europeo de Empresas Innovadoras, Parque Tecnologico de Valencia, Valencia, Spain KHAC-MINH THAI • Faculty of Pharmacy, University of Medicine and Pharmacy at Ho Chi Minh City, Ho Chi Minh City, Vietnam THAI-SON TRAN • Faculty of Pharmacy, Hue University of Medicine and Pharmacy, Hue University, Hue City, Vietnam THANH-DAO TRAN • Faculty of Pharmacy, University of Medicine and Pharmacy at Ho Chi Minh City, Ho Chi Minh City, Vietnam THE-HUAN TRAN • Faculty of Pharmacy, Hue University of Medicine and Pharmacy, Hue University, Hue City, Vietnam

Contributors

xvii

VAN-THANH TRAN • Faculty of Pharmacy, University of Medicine and Pharmacy at Ho Chi Minh City, Ho Chi Minh City, Vietnam PIYUSH TRIVEDI • Centre of Innovation and Translational Research, Poona College of Pharmacy, Bharati Vidyapeeth Deemed University, Pune, India IOANNA-CHRYSOULA TSOPKA • Department of Pharmaceutical Chemistry, School of Pharmacy, Faculty of Health Sciences, Aristotle University of Thessaloniki, Thessaloniki, Greece VIKAS VERMA • Department of Chemistry, Guru Jambheshwar University of Science & Technology, Hisar, India MARY I. ZGURZYNSKI • Massachusetts General Hospital, Boston Dermatology & Laser Center, Boston, MA, USA YUSHENG ZHAO • School of Pharmacy, Health Sciences Campus, University of Waterloo, Waterloo, ON, Canada

Part I Recent Progress in Understanding Molecular Etiology of Alzheimer’s Disease

Chapter 1 Recent Progress in the Treatment Strategies for Alzheimer’s Disease Vinay Kumar and Kunal Roy Abstract Alzheimer’s disease (AD) is a neurological ailment that affects older people and causes a steady decline in their cognitive function. The cognitive impairments found are presumed to be the result of synapse disruption and neurochemical deficits. Several neurochemical abnormalities have been found throughout progressive aging, and these have been connected to cognitive dysfunction seen in the sporadic stage of AD. There are various hypotheses explaining AD, such as aberrant deposits of amyloid β (Aβ) protein in the extracellular spaces of neurons, production of twisted fibers of tau proteins inside neurons, cholinergic neuron damage, inflammation, oxidative stress, and so on, and many anti-AD therapeutics have been developed based on these hypotheses. While current pharmacological treatments assist in relieving the symptoms of AD and enhance a patient’s quality of life, they do not halt or cure the disease. Presently, targeted drug delivery to the central nervous system (CNS) for AD therapy is hampered by the difficulties posed by blood-brain interfaces surrounding the CNS, reducing therapeutic bioavailability. Among innovative ways to overcome these restrictions and successfully deliver pharmaceuticals to the CNS, nanoparticles (NPs) can overcome these barriers, offering new therapeutic strategies in terms of dealing drugs to cross the blood-brain barrier (BBB) and enter the brain more adequately. Various innovative therapeutic options for the treatment of AD have shown promising results in preclinical research and are currently being tested in clinical trials throughout the last decade. In addition to generating chemical entities, various natural compounds such as alkaloids, terpenoids, flavonoids, and curcumin have been isolated and evaluated for AD, and all demonstrated promising actions against a range of targets. Moreover, computational techniques have also proven to be quite useful in reducing time and money when developing new therapies. Molecular modeling, virtual screening, and docking have been widely used by researchers worldwide in recent years. These techniques have already aided in the development of several promising compounds. The purpose of this chapter is to summarize and highlight recent advancements in research on the development of novel therapies and their implications in the treatment of AD. Key words Alzheimer’s disease, CNS, Amyloid β, Cholinergic neuron, Clinical trials, Nanoparticles

1

Introduction Alzheimer’s disease (AD) is a complex, progressive neurological disease, which contributes to about 60–80% of all dementia cases [1]. Most people are unsure about the distinction between

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_1, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

3

4

Vinay Kumar and Kunal Roy

Alzheimer’s and dementia. Dementia is a broad term that refers to a set of symptoms [2, 3]. Dementia is characterized by concerns with memory, language, problem-solving, and other intellectual abilities that limit a person’s ability to carry out daily activities [2]. Dementia has various causes such as AD, cerebrovascular disease, Lewy body disease, frontotemporal lobar degeneration (FTLD), Parkinson’s disease (PD), hippocampal sclerosis (HS), and mixed pathologies [2]. AD is the most prevalent type of dementia. AD is estimated to start 20 years or more before it manifests symptoms [2, 3]. It begins with undetectable alterations in the brain of the individual who is affected [3]. Individuals only begin to notice symptoms such as memory loss and language difficulties after years of brain changes. Symptoms occur as a result of nerve cell damage or destruction in areas of the brain involved in intellectual functions, learning, and memory (cognitive abilities) [2, 4]. Other sections of the brain’s neurons are also damaged as the disease progresses [4]. Neurons in areas of the brain that allow a person to walk and swallow become impaired with time. Individuals become bedridden and demand 24-h care. Alzheimer’s disease is fatal in the end. AD is a developing healthcare concern, with increased life expectancy as the primary risk cause. Disease prevalence is expected to more than double over the next several decades in the absence of adequate prevention and treatment strategies [3, 4]. According to the World Alzheimer Report 2021 [5], there were around 55 million people worldwide living with dementia, and the figure is expected to exceed 80 million in 2030 and 152 million in 2050 worldwide. Aside from the immediate impact on human health and well-being, long-term care for affected individuals is a significant financial burden. Efforts to develop disease-modifying treatments for AD, which have included around 200 clinical studies to date, have generally failed, with many failures due to lack of effectiveness or extreme toxicity [5]. The current research reveals that the major histopathologic lesions of AD are extracellular amyloid plaques and intracellular Tau neurofibrillary tangles (NFTs), based on the basic pathophysiology and neuropathology of the disease (NFTs) [5, 6]. The amyloid or senile plaques are primarily composed of quite insoluble and proteolysis-resistant peptide fibrils formed by β-amyloid (Aβ) cleavage [6]. After the sequential breakdown of the large precursor protein amyloid precursor protein (APP) by the two enzymes, γ-secretase and β-secretase (BACE1), Aβ peptides with Aβ38, Aβ40, and Aβ42 being the most prevalent forms are formed [7]. However, Aβ is not formed if APP is initially acted on and cleaved by the enzyme α-secretase rather than by β-secretase [6, 8]. As per the “amyloid hypothesis,” Aβ formation in the brain initiates a cascade of events that leads to the clinical syndrome of AD [8]. The formation of amyloid oligomers is primarily responsible for neurotoxicity and the commencement of the amyloid

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

5

cascade [6–8]. The elements of the cascade comprise local inflammation, oxidation, excitotoxicity (excessive glutamate), and tau hyperphosphorylation [6–8]. Tau protein is a microtubuleassociated protein that attaches to microtubules in cells to aid neuronal transport [8]. Microtubules also help to support the developing axons, which are essential for neuronal development and function [8]. Intraneuronic tangles are formed when abnormally hyperphosphorylated tau produces insoluble fibrils and folds [8]. It was formerly believed that tau hyperphosphorylation followed the deposition of amyloid, but it is equally likely that tau and amyloid function in separate routes, generating AD and amplifying each other’s detrimental impacts [8]. Progressive neuronal loss results in a deficiency and imbalance of numerous neurotransmitters (e.g., acetylcholine, dopamine, and serotonin), as well as the cognitive deficits found in AD [5–8]. Subsequently, many possibilities have since been presented to explain AD, particularly the lateonset sporadic type (LO-SAD), including aging aggravation, gender, genetic variables, aluminum exposure, head injury, nutrition, mitochondrial dysfunction, vascular illness, immune system dysfunction, and infectious disease [5–8]. However, no single explanation has achieved widespread acceptance, and it is now widely accepted that numerous risk factors are likely to be implicated. Given recent improvements, it seems appropriate to update previous reviews, assess the present level of knowledge about risk factors for AD, and answer two questions: (1) to what degree does the existing knowledge identify credible risk factors for AD, and (2) how can this information be integrated into a hypothesis to explain the pathogenesis of AD [4, 5]. Figure 1 shows the multifactorial AD pathogenesis and progression which is influenced by various pathophysiological events occurring sequentially in the brain. A classification of therapy treatments for Alzheimer’s disease based on distinct mechanisms is given in Table 1. Sex-specific differences contribute to its complexity, according to evidence from basic and clinical investigations [9]. Around two-thirds of Alzheimer’s patients are female. Females have a higher prevalence of Alzheimer’s disease than males, which could be due to their longer life expectancy or higher dementia rate [4, 5, 9]. There are conflicting claims in the literature about the incidence of AD in men and women, with some claiming that there are no differences [4, 5]. However, sex differences in Alzheimer’s disease have been reported in clinical, neuroimaging, and pathology studies [4, 5, 9]. Surprisingly, sex-specific analyses were only undertaken on rare occasions during the AD clinical trial phases, and the significance of addressing sex as a significant modulator of patient responses to therapies was grossly underestimated [5]. While it is crucial to address sex differences in clinical trial design, the present obstacles stem from a lack of knowledge of the mechanisms that govern sex-biased variances in AD [4, 5].

6

Vinay Kumar and Kunal Roy

Fig. 1 Complex AD pathogenesis and progression which influenced by various pathophysiological events occurring sequentially in the brain

In recent years, significant research has been devoted to developing medications that slow neurodegeneration in AD, but we are quite far from finding specific treatment strategies. The early diagnosis and treatment of AD is now a fast-developing field of both scientific and clinical research because current treatments only help with the symptoms of the disease (symptomatic treatment) [10]. There are now only five approved drugs (see Fig. 2) for the treatment of cognitive symptoms of Alzheimer’s disease [10]. Among them, four drugs are acetylcholinesterase enzyme (AChE) inhibitors (tacrine, rivastigmine, galantamine, and donepezil), and the remaining one drug is a noncompetitive glutamate (NMDA) receptor antagonist (memantine “FDA-approved”)

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

7

Table 1 Different therapeutic strategies for the treatment of AD and their mode of action Hypothesis

Mode of action

Strategies

Neurotransmitter hypothesis

Improvement of the acetylcholine response Inhibition of glutamate cytotoxicity

AChE enzyme inhibitor, α7nAChR agonists NMDA receptor antagonists

Aβ hypothesis

Inhibition of Aβ production

BACE-1 inhibitors Gamma-secretase inhibitors Metal protein attenuating compounds Anti-Aβ vaccine Anti-Aβ antibody

Inhibition of Aβ aggregation Anti-Aβ immunotherapy Tau hypothesis

Inflammation hypothesis

Inhibition of tau phosphorylation Improvement of microtubule stabilization Inhibition of tau aggregates Enhancement of tau clearance Inhibition of neuroinflammation

GSK-3β inhibitors Microtubule stabilizer Tau aggregation inhibitors Vaccination therapy NSAIDs Microglial activation inhibitors

Fig. 2 List of approved drugs and molecules in clinical trials

[10, 11]. The benefit of their use is symptomatic, and no medicine has been clearly shown to delay or halt the progress of the disease [10, 11]. Recently, it has been proposed that the long-lasting effects of AChE inhibitors are brought about by these drugs’

8

Vinay Kumar and Kunal Roy

interference with APP metabolism [12]. The cause and progression of Alzheimer’s disease are still not well understood [12]. In particular, the search for drugs in the field of neurodegenerative disorders is very active, though regrettably, no cure exists nowadays for AD [12]. Although drug research is a time-consuming and expensive process, computational biology and bioinformatics methodologies have simplified the initial identification of potential therapeutic compounds. Such computational biology strategies have become extremely relevant, from lead identification to optimization. In this regard, computational approaches such as quantitative structureactivity relationship (QSAR), chemical read-across, pharmacophore modeling, molecular docking, molecular dynamics (MD) Simulations, etc. are playing imperative roles in the design and discovery of new compounds with enhanced therapeutic activity [13–18]. For decades, chemoinformatics and molecular modeling approaches have been utilized to identify and optimize novel compounds with improved therapeutic potential in various fields [13–18]. Currently, in silico modeling is a part of the conventional drug discovery process, and such methods are usually employed in the search for novel drugs or the optimization of the therapeutic activity of a chemical series at the initial phases of drug development [13–18]. Computational methodologies have given several potential drug molecules against the predictable and promising targets against AD [19]. Many molecules have been discovered to be excellent lead compounds (including flavonoids, carbamates, pyridonepezil, and coumarin derivatives) that can be propagated against AD [19]. Numerous molecules have entered various stages of clinical trials, including MK-8931 (against β-secretase, Merck), TAK-070 (against β-secretase, Takeda Pharmaceuticals), and LMTX (against tau hyperphosphorylation, TauRx Inc.) (See Fig. 2) [19].

2

Estimated Morbidity and Mortality Rate of AD Country-Wise It is challenging to predict how many deaths are caused by AD each year due to the way the causes of death are recorded. According to statistics, AD is the sixth leading cause of death in the country and the fifth leading cause of death for persons 65 and older [4, 5]. However, it may be responsible for considerably more deaths than official sources acknowledge. Alzheimer’s disease is also a prominent cause of disability and poor health (morbidity) among older persons. Before dying from Alzheimer’s, a person has years of morbidity as the disease advances [4, 5]. Table 2 shows the estimated numbers of people (in millions) living with Alzheimer’s by region, and Table 3 shows the overall estimated mortality rate of AD country-wise.

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

9

Table 2 Estimated numbers of people (in millions) living with Alzheimer’s by regiona Years Region

2020

2025

2030

2035

2040

2045

2050

Australasia

0.35

0.4

0.48

0.56

0.67

0.77

0.87

0.96

Asia Pacific high-income

3.57

4.23

4.86

5.48

6.1

6.67

7.04

7.4

Oceania

0.02

0.03

0.03

0.04

0.05

0.06

0.08

0.09

Asia Central

0.33

0.36

0.41

0.47

0.57

0.71

0.84

0.95

Asia East

9.49

14.14

17.68

21.85

26.35

31.21

36.12

Asia South

6.96

8.41

10.12

12.15

14.56

17.47

20.94

24.86

Asia Southeast

3.55

4.31

5.23

6.33

7.56

8.85

10.18

11.37

60.9

71.16

81.75

Asia

11.5

24.28 29.23 35.28 42.71 51.37

Europe Central

1.51

1.69

1.85

2.04

2.25

2.5

2.7

2.82

Europe Eastern

2.03

2.1

2.2

2.23

2.45

2.68

2.77

2.77

Europe Western

8.01

8.92

9.92

11.06

12.28

13.63

14.88

16.05

11.55 12.71 13.97 15.33 16.98

18.82

20.35

21.64 11.1

Europe North America high-income

4.4

5.01

5.79

6.77

8

9.25

10.29

Caribbean

0.41

0.48

0.56

0.66

0.77

0.88

1.01

1.14

Latin America Andean

0.42

0.52

0.63

0.77

0.94

1.15

1.4

1.68

Latin America Central

1.8

2.22

2.73

3.37

4.19

5.19

6.39

7.71

Latin America Southern

0.9

1.04

1.2

1.39

1.62

1.89

2.19

2.51

Latin America Tropical

1.69

2.14

2.7

3.38

4.19

5.18

6.3

7.48

The Americas

9.62 11.42 13.6

23.54

27.59

31.63

North Africa/Middle East

2.49

2.99

3.62

4.44

5.52

6.85

8.4

10.1

Sub-Saharan Africa Central

0.21

0.25

0.3

0.36

0.44

0.53

0.65

0.79

Sub-Saharan Africa East

0.85

1.02

1.21

1.46

1.77

2.17

2.7

3.38

Sub-Saharan Africa Southern

0.25

0.28

0.32

0.36

0.42

0.49

0.58

0.68

Sub-Saharan Africa West

0.66

0.76

0.89

1.05

1.26

1.53

1.86

2.26

Africa

4.47

5.3

6.34

7.67

9.4

11.57

14.18

17.22

World a

2015

49.92 58.66 69.2

16.33 19.7

82.05 97.45 114.83 133.28 152.24

Date of data collection: June 6, 2022, data source: World Health Organization (WHO); 2020, WHO excludes the following countries: Andorra, Cook Islands, Dominica, Marshall Islands, Monaco, Nauru, Niue, Palau, Saint Kitts, San Marino, and Tuvalu

Table 3 The estimated mortality rate of Alzheimer’s disease country-wise Rank Country

Rate Rank Country

1

Finland

54.6 5

47

Tunisia

19.9 2

93

Jordan

17.5 3

139

Barbados

14.3 8

2

The UK

42.7

48

Malawi

19.9

94

Sudan

17.5 2

140

Ecuador

14.3 4

3

Slovakia

38.1 5

49

Gambia

19.8 9

95

Hungary

17.5

141

Turkmenistan

14.1 9

4

Albania

36.9 2

50

Mali

19.8 8

96

Kuwait

17.4 2

142

Bangladesh

13.8 9

5

Iceland

35.5 9

51

Bolivia

19.8 6

97

China

17.3 6

143

Czech Republic

13.7 1

6

Brunei

33.8 7

52

DR Congo

19.8 3

98

Azerbaijan

17.3 6

144

Ukraine

13.6 7

The Netherlands 33.7 8

53

Mauritania

19.7 6

99

Lesotho

17.2 4

145

Nepal

13.6

7

Rate Rank Country

Rate Rank Country

Rate

8

The USA

33.2 6

54

Togo

19.7 5

100

Seychelles

17.2 2

146

Serbia

13.5 8

9

Ireland

32.2 3

55

Rwanda

19.4 9

101

Portugal

17.2 1

147

Austria

13.3 3

10

Sweden

30.9 6

56

Uruguay

19.4 9

102

Syria

17.1 4

148

Egypt

13.1 4

11

Denmark

29.4 1

57

Chad

19.4 9

103

Saint Lucia

17.1 2

149

Solomon Isl.

12.6 8

12

Norway

28.9 4

58

Cambodia

19.4 5

104

Myanmar

17.1 2

150

Trinidad/To B.

12.2 7

13

Canada

27.8 7

59

Nigeria

19.4 5

105

Bosnia/Herze G.

17.1

151

Brazil

11.9 3

14

Sri lanka

27.6 2

60

Uganda

19.3 6

106

Indonesia

17.0 1

152

Chile

11.6 4

15

Kiribati

25.2 6

61

Thailand

19.3 3

107

The UAE

16.9 6

153

Costa rica

11.5 9

16

New Zealand

24.8 4

62

Sierra Leone

19.2 9

108

Malaysia

16.9 5

154

South Korea

11.0 4

17

Bahrain

24.3 2

63

Cote d’ Ivoire

19.2 7

109

Cape Verde

16.8 9

155

Latvia

9.94

18

Mozambique

23.8 8

64

France

19.1 9

110

Bahamas

16.8 3

156

Greece

9.74

19

Switzerland

23.7 8

65

Yemen

19.1 8

111

Israel

16.7 5

157

Uzbekistan

8.36

20

Vietnam

23.7 2

66

Iraq

19.1 6

112

Bhutan

16.7 3

158

Croatia

7.99

21

Suriname

23.5 3

67

Lebanon

19.1 4

113

Swaziland

16.7 3

159

Japan

7.87

22

Qatar

22.6 9

68

Samoa

19.1 3

114

Guatemala

16.6 6

160

Grenada

6.91

23

Australia

22.6 3

69

Turkey

19.0 7

115

Belize

16.6 6

161

Lithuania

6.73

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

11

26

Afghanistan

22.3 4

72

Zambia

18.8 7

118

South Africa

16.4 4

164

Slovenia

5.8

27

Belgium

22.3

73

Burundi

18.7 2

119

Laos

16.4 2

165

Romania

5.41

28

Cuba

21.9

74

Iran

18.6 3

120

South Sudan

16.3 7

166

Panama

5.37

29

Djibouti

21.7 6

75

Zimbabwe

18.4 1

121

Pakistan

16.3

167

Argentina

4.95

30

Ethiopia

21.7 4

76

Kenya

18.3 5

122

North Korea

16.2 6

168

El Salvador

4.47

31

Georgia

21.6 4

77

Luxembourg

18.3 3

123

Peru

16.2 3

169

Poland

3.73

32

Spain

21.5 2

78

Maldives

18.3

124

Algeria

16.0 8

170

Colombia

3.6

33

Honduras

21.3 2

79

Namibia

18.2 5

125

Belarus

15.8 6

171

Guyana

3.28

34

Burkina Faso

20.9 8

80

Morocco

18.1 1

126

Dominican Rep.

15.8 4

172

The Philippines

3.1

35

Equ. Guinea

20.9 4

81

Tanzania

18.1 1

127

Fiji

15.7

173

Nicaragua

3.05

36

Senegal

20.9 2

82

Somalia

18.0 2

128

Germany

15.5 4

174

Moldova

2.92

37

Cameroon

20.7 1

83

GuineaBissau

17.9 9

129

Malta

15.5 1

175

Mexico

2.7

38

Comoros

20.7

84

Tonga

17.9 7

130

Kazakhstan

15.3 9

176

Paraguay

2.56

39

Eritrea

20.7

85

Ghana

17.7 9

131

Kyrgyzstan

15.2 2

177

Mauritius

2.41

40

Gabon

20.5 5

86

Sao Tome

17.7 3

132

Haiti

15.1 4

178

Venezuela

2.18

41

Saudi Arabia

20.5 2

87

Montenegro

17.7 1

133

Jamaica

15.0 9

179

North Macedonia

2.15

42

Benin

20.2 3

88

Tajikistan

17.7

134

Russia

15.0 7

180

Saint Vincent

1.18

43

Congo

20.2 2

89

Mongolia

17.6 3

135

Italy

14.8 6

181

Armenia

0.87

44

Niger

20.1 8

90

Vanuatu

17.6 2

136

Madagascar

14.6 5

182

Bulgaria

0.79

45

Angola

20.1 1

91

Central Africa

17.6

137

India

14.6

183

Singapore

0.43

46

Guinea

20.0 4

92

Liberia

17.5 6

138

New Guinea

14.4 7

High

Low

Note: Death Rate per 100,000 age Standardized, Date of data collection: 06/06/2022, Data Source: WORLD HEALTH ORGANIZATION (WHO), 2020, WHO Excludes Cause of Death Data For the following countries: Andorra, Cook Islands, Dominica, Marshall Islands, Monaco, Nauru, Niue, Palau, Saint Kitts, San Marino, Tuvalu.

12

3

Vinay Kumar and Kunal Roy

The Effect of the COVID-19 Epidemic on the Mortality from AD The data presented in the Mortality and Morbidity section (Tables 2 and 3) are from the most recent data sources accessible, all of which preceded the global COVID-19 pandemic that erupted in 2020 [4, 5]. According to preliminary data from the Centers for Disease Control and Prevention (CDC) [4, 5], excess mortality (the difference between the observed and projected number of deaths for a specific period) was extremely high in 2020 compared to prior years, particularly among older persons. Most of these increased deaths happened among vulnerable aged persons suffering from AD and other dementias [4, 5]. Preliminary CDC records show that in 2020, there were at least 42,000 more deaths from AD and other dementias than there were on average in the 5 years before 2020 [4, 5]. This is about 16% higher than expected [4]. Furthermore, the number of COVID-19 deaths for which the death certificate stated more than one condition as a cause of death was also counted by the CDC: in 4% of death certificates stating COVID-19 as the primary cause of death, AD was also indicated as one of the various reasons of death, and in 11% of death certificates identifying COVID-19 as the primary cause of death, there was also a record of an unspecified form of dementia [4, 5]. AD was identified as one of several reasons for death for 8% of patients over the age of 85 who died from COVID-19, whereas vascular dementia was indicated for 20% [4, 5]. As a result, we anticipate that the substantial impact of this epidemic on Alzheimer’s death patterns will be evident in the coming years [4, 5].

4

Neurobiology of AD and Promising Drug Targets AD is a complex disease with numerous contributing causes. The specific pathophysiology of AD is still unknown due to the complexity of human brains, and a lack of suitable animal models, and research tools. In this chapter, we address treatment techniques based on a variety of known hypotheses aiming at explaining the genesis of AD. The primary objective of this chapter is to give a critical overview of the clinical signs and synaptic abnormalities, such as neurotransmitter deficiencies, and how they relate to cognitive difficulties, as well as newly developed therapeutic approaches for the treatment of AD. There have been numerous hypotheses established for AD listed below.

4.1 Amyloid Hypothesis

AD is distinguished by the formation of aberrant neuritic plaques and neurofibrillary tangles in the brain [4, 5]. Neuritic plaques are hemispheric minuscule plaques with an extracellular amyloid betapeptide core surrounded by increased axonal terminals. A transmembrane protein known as an amyloid precursor protein (APP)

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

13

produces beta-amyloid peptide (β-AP) as a byproduct [20]. In the brain, generally, the β-AP is cleaved from APP by the action of three proteases christened alpha (α), beta (β), and gamma (γ)-secretase [20–22]. APP is normally cleaved by alpha or beta-secretase, and the resulting microscopic fragments are not toxic to neurons. [20–22]. Through successive cleavage by beta and then gammasecretase, it results in 42 amino acid peptides called beta-amyloid42 peptide. Raise in the levels of beta-amyloid 42 leads to the accumulation of amyloid which causes neuronal toxicity by the deposition at the synaptic site in the brain [20–22]. Over normal APP breakdown, beta-amyloid 42 promotes the formation of aggregated fibrillary amyloid protein [21]. The APP gene is positioned on chromosome 21, which is associated with familial AD [22]. In AD, amyloid deposition happens around meningeal and neural vessels, as well as a gray matter [20]. Multifocal gray matter deposits aggregate to become milliary structures known as plaques [20–22]. Though some people without dementia had amyloid plaques detected during brain scans, some people with dementia did not. 4.2

Tau Hypothesis

In neurons, a protein called tau forms fibrillary intracytoplasmic formations known as neurofibrillary tangles. The primary role of the tau protein is to maintain the stability of axonal microtubules, which are necessary for intracellular transport [23]. Microtubule assembly is detained and organized by tau protein [23–25]. In AD, tau is hyperphosphorylated as a consequence of extracellular betaamyloid accumulation, which leads to the formation of tau clumps [25]. Tau clumps form twisted paired helical threads known as neurofibrillary tangles [23–25]. They form initially in the hippocampus and thereafter spread throughout the cerebral cortex [24]. Neurofibrillary tangles are deposited within the neurons. The National Institute of Aging and Reagan Institute’s neuropathological standards for the diagnosis of AD include the Braak staging, which was devised by Braak and Braak and is based on the topographical staging of neurofibrillary tangles into six different phases [23]. In comparison to plaques, tangles have a stronger correlation with AD [23]. Another hallmark of AD is granulovacuolar degeneration of hippocampus pyramidal cells caused by amyloid angiopathy [23–25]. According to some research, cognitive decline is more closely associated with a decrease in the density of presynaptic boutons formed by pyramidal neurons in laminae III and IV than with an increase in plaque density [24]. It has also been shown that the Meynert nucleus basalis has lost neurons, resulting in low acetylcholine levels [25]. Uncertainty surrounds the role of vasculature in the neurodegenerative process of Alzheimer’s disease [25]. Subcortical infarcts quadruple the risk of developing dementia [23]. The severity and rate of progression of dementia are further exacerbated by cerebrovascular pathology [23–25].

14

Vinay Kumar and Kunal Roy

4.3 Oxidative Stress and Mitochondrial Hypothesis

Byproducts of the electron transport chain such as hydrogen peroxide, hydroxyl, and superoxide radicals that accumulate excessively in the mitochondria cause oxidative stress, which can lead to cell damage and death in AD [26–28]. There is growing evidence that Aβ plaques directly disrupt the electron transport chain and generate free radicals, which enhances the level of oxidative stress [26–28]. The peroxynitrite radicals, another extremely reactive oxidative substance, are formed when some of these superoxide radicals react with nitric oxide due to their high level of reactivity [26–28]. Additionally, the generation of free radicals during AD can lead to the oxidation and destruction of DNA (deoxyribonucleic acid) strands, the cross-linking of DNA proteins, and also changes in DNA base pairs [26–28]. On the other hand, oxidative stress has been demonstrated to cause the oxidation and glycation of certain proteins and lipids, as well as the formation of advanced glycation end products, both of which worsen oxidative stress and neuroinflammation [26–28]. The progression of AD and oxidativenitrosative stress are more likely to affect the neurons in the entorhinal cortex, hippocampus, frontal cortex, and amygdala [26–28].

4.4 Inflammatory Hypothesis

The first description of a link between microglia cells and inflammation as being connected to AD was established in the middle of the 1980s [29, 30]. According to the report, microglia and astrocyte cells are activated by amyloid plaques and NFT during the progression of AD [29, 30]. Interleukin-1, interleukin-6, tumor necrosis factor (TNF), and interleukin-8 are only a few examples of the pro-inflammatory cytokines and chemokines that are expressed more frequently when microglia are activated [29, 30]. The neuroinflammatory response is further exacerbated by the interaction of activated astrocytes with Aβ sites and the release of various pro-inflammatory mediators such as interleukins, prostaglandins, leukotrienes, and thromboxanes [29, 30]. Moreover, it has been also reported that Aβ arouses nitric oxide activity, subsequently leading to inflammation and neuronal destruction [29, 30]. In contrast to conventional neuro-inflammatory diseases like multiple sclerosis and encephalitides, however, recent investigations have found a connection between AD inflammation and the innate immune system [29, 30]. Hence, more research is required to completely develop the inflammatory cascade theory and to support the idea of inflammation in AD.

4.5 Epigenetics Hypothesis

Recent studies have demonstrated that the pathologies of AD are significantly influenced by epigenetic alterations, particularly DNA methylation and histone acetylation, which control gene expression at the transcriptional level [26, 31]. DNA methylation takes place on cytosine residues known as C (cytosine) bases followed immediately by G (guanine) base (CpG) regions, which are followed by guanine residues [26, 31, 32]. DNA methyl transferases (DNMTs),

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

15

which move the methyl group from S-adenosyl methionine to cytosine residues, catalyze the methylation process [26, 31, 32]. De novo or maintenance DNMTs are the two broad categories for DNMTs [26, 31]. The de novo DNMTs, DNMT3a and DNMT3b, generate initial methylation patterns on formerly unmethylated DNA, whereas DNMT1 establishes methylation patterns on hemimethylated replicating DNA [26, 31, 32]. The role of DNA methylation in neuropathological pathways leading to AD has been thoroughly documented [31, 32]. When compared to healthy persons, Mastroeni et al. [33] found that DNMT1 expression was lower in AD patients, which was associated with lower levels of DNA methylation in the cortex, particularly in NFT-containing neurons. Klein et al. [34] demonstrated that DNMT1 gene mutations cause abnormal DNA methylation and cause dementia, hearing loss, and neurodegeneration in individuals. Many more studies have also confirmed that AD is related to abnormal methylation in the brain [31, 32]. In AD pathogenesis, DNA methylation is altered not just at the global level but also at the gene-specific promoter level [31, 32]. The APP gene promoter contains a GC-rich region that can be controlled by methylation [26]. A hyper-methylation-dependent change in the expression of APP has been seen in the AD brain. A synaptic plasticity marker known as a brain-derived neurotrophic factor (BDNF) is involved in the development of long-term memory [31–34]. Rao et al. [35] revealed that lower BDNF expression in AD brains is caused by hyper-methylation in the promoter region. Additionally, cAMP response element binding protein (CREB), a different synaptic plasticity measure, revealed promoter hyper-methylation in AD brains. Folate, vitamin B12, and their end product S-adenosylmethionine (SAM) all play important roles in carbon metabolism in Alzheimer’s disease, and SAM is the primary methyl donor for DNA methylation [31–35]. According to studies, patients with AD have lower levels of folate, vitamin B12, and SAM, which leads to altered DNA methylation [31–35]. Histone acetyltransferases (HATs) catalyze the acetylation of the N-terminal tail of histones, resulting in a relaxed chromatin state that is easily accessible for the binding of transcription factors and active transcription; histone deacetylase (HDAC) removes an acetyl group from the N-terminal tail of histones, resulting in a compacted chromatin state that inhibits gene transcription [31–35]. Of the HDACs, HDAC2 is a negative regulator of memory-related genes involved in synaptic plasticity [31–35]. Graff et al. [36] demonstrated that the expression of HDAC2 was upregulated in the hippocampus of a neurodegenerative mouse model along with lower levels of histone acetylation and expression of synaptic plasticity genes such as BDNF, activity-regulated cytoskeleton-associated protein, and early growth response protein. They also discovered that amyloid beta protein or oxidative stress caused by

16

Vinay Kumar and Kunal Roy

H2O2 increased HDAC2 expression via glucocorticoid receptor activation, which is a transcription factor for HDAC2 [31– 36]. Consequently, oxidative stress stimulates the expression of HDAC2 in AD-affected brains. HDAC6 expression is increased to 52% above baseline in the cortex and 91% in the hippocampus of brains with AD pathology, similar to HDAC2 [31–36]. Sirtuins, which are class III HDACs and primarily deacetylate cytoplasmic proteins, have been reported to have a neuroprotective effect because they prevent the aggregation of tau by deacetylating its acetylation [36]. SIRT1 also regulates α-secretase production, which cleaves APP and inhibits Aβ deposition. However, SIRT1 expression is reduced in the cortex, which is linked to the accumulation of tau pathology and deterioration in memory as observed in AD [36]. 4.6 Genetic Causes of AD

5

AD can be genetic as an autosomal dominant disease with nearly perfect penetrance. The disease’s autosomal dominant type is caused by mutations in three genes: the AAP gene on chromosome 21, the presenilin 1 (PSEN1) gene on chromosome 14, and the presenilin 2 (PSEN2) gene on chromosome 1 [37, 38]. The formation and aggregation of the beta-amyloid peptide may increase as a result of APP mutations. PSEN1 and PSEN2 mutations cause beta-amyloid aggregation by interfering with gamma-secretase processing [37, 38]. Most cases of early-onset AD and 5–10% of all cases are caused by mutations in these three genes [37, 38]. Apolipoprotein E is an additional genetic marker that raises the risk of AD [37, 38]. It is a lipid metabolism regulator with an affinity for beta-amyloid protein [37, 38]. The APOE isoform e4 gene, which is located on chromosome 19, has been linked to more sporadic and familial types of AD that manifest after age 65 [37, 38]. The existence of one APOEe4 allele does not always result in AD, but roughly 50% of people with one APOE-e4 allele have AD, and 90% of people with two alleles develop AD [37, 38]. The age of disease onset decreases with each APOEe4 allele. The APOEe4 allele is a significant risk factor for developing AD [37, 38]. Both familial and sporadic types of AD have been linked to variations in the sortilin receptor SORT1 gene, which is crucial for moving APP from the cell surface to the Golgi-endoplasmic reticulum complex [37, 38].

Role of Neurotransmitters and Their Receptors in AD Almost every neurotransmitter system is important in controlling brain function. Specific neurological illnesses have been linked to either hyper- or hypofunction of any of the neurotransmitters [26, 39]. Growing data reveals that AD causes pathological changes to the receptor networks for several neurotransmitters as well as neurochemical changes to those same neurotransmitters

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

17

[26, 39]. More precisely, aberrant signaling of different neurotransmitting systems including cholinergic, serotonergic, noradrenergic, GABAergic, and probably dopaminergic neurotransmission has been described in the course of AD [26, 39]. 5.1 Cholinergic Signaling in AD

The severity of AD pathology is connected with the cholinergic system, which is thought to be most negatively impacted [39– 41]. Degeneration of basal forebrain cholinergic neurons and hypofunction, in particular, have been linked to the course of AD [42]. A neurotransmitter named acetylcholine interacts with cholinergic receptors and activates them [43]. Cholinergic receptors are divided into two types: nicotinic receptors and muscarinic receptors. The brain and parasympathetic effector organs express muscarinic receptors, whereas nicotinic receptors are established in the neuromuscular junctions, autonomic ganglia, and CNS [40, 41, 44]. Nicotinic receptors are found primarily in the striatum, cortex, superior colliculus, lateral geniculate nucleus, and cerebellum, whereas muscarinic receptors are found mostly in the hippocampus, cortex, and thalamus [40, 41, 44]. A rising body of research points to the involvement of muscarinic and nicotinic receptors in AD [40, 41, 44]. The muscarinic M1 receptor has been implicated in several neurological diseases. Remarkably, the density of M1 receptors does not drop dramatically, but the coupling between M1 and G-protein in the frontal cortex decreases and is directly related to the severity of AD [40, 41, 44]. According to previous research, M1 receptors are specifically implicated in cognitive activities. Furthermore, M2 receptor knockout mice exhibited impaired cognitive behavior, indicating that these receptors may play a role in AD [40, 41, 44]. Additionally, a study discovered that mice lacking the M3 receptor are less sensitive to the learning and memory of conditioned fear [40, 41, 44]. This decline is revealed to be dependent on the amount of muscarinic receptor phosphorylation, confirming the significance of these receptors in AD [40, 41, 44]. However, with the progression of AD, nicotinic receptor density and ACh binding efficiency both significantly decreased [40, 41, 44]. The pathogenesis of AD is said to be facilitated by Aβ interaction with nicotinic receptors [40, 41, 44].

5.2 Serotonergic Signaling in AD

Serotonin (5-hydroxytryptamine, 5-HT), a monoamine neurotransmitter, has been found to have an important function in the CNS in regulating cognitive behavior, sensory and affective processes, autonomic responses, and motor activity [44, 45]. The behavioral elements of AD development are influenced by serotonergic neuron degeneration and 5-HT neurotransmitter hypofunction [44, 45]. In particular, 5-HT concentration was shown to be significantly lower in the hippocampus area of affected brains. 5-HT receptors have been classified into seven families based on transducer processes, spanning from 5-HT1 to 5-HT7

18

Vinay Kumar and Kunal Roy

[44, 45]. The pathophysiology of AD has been associated with these kinds of serotonergic receptors, including 5-HT6 receptors, which are implicated in learning and memory. This is corroborated by the finding that the 5-HT6 receptor gene’s single nucleotide polymorphism C267T is a risk factor in the genesis of AD. Notably, apolipoprotein E epsilon 4, a crucial factor in the development of AD, is not dependent on the genetic polymorphism C267T, which is involved in the late onset of AD [44, 45]. The function of several neurotransmitters, including glutamate and acetylcholine, which are crucial for memory and learning, is regulated (more precisely, downregulated) by 5-HT6 receptors [44, 45]. 5-HT3 receptors have been discovered in the hippocampus [44, 45]. Serotonin reduces cholinergic tone via interacting with 5-HT3 receptors located in the hippocampus [44, 45]. The reduced cholinergic tone during AD may also contribute to the course of the disease [44, 45]. Thus, 5-HT3 receptor antagonism may be advantageous in Alzheimer’s patients. Furthermore, stimulation of 5-HT4 receptors has been shown to enhance acetylcholine release simultaneously decreasing Aβ toxicity [44, 45]. However, 5-HT1, 5-HT2, and 5-HT7 receptor dysfunction may potentially be involved in the etiology of AD [44, 45]. 5.3 Glutamatergic Signaling in AD

Glutamate, a key excitatory neurotransmitter in the CNS, has been proven to be crucial for several neuronal functions under physiologic conditions, including synaptic transmission, neuronal differentiation and growth, synaptic plasticity, and learning and memory [46–48]. Ionotropic and metabotropic are the two types of glutamate receptors [46–48]. Based on their agonists, ionotropic receptors are further divided into three classes: kainic acid, amino3-hydroxy-5-methyl-4-isoxazole propionic acid, and NMDA [46– 48]. However, in disease states, NMDA receptor-mediated excitotoxicity plays an important part in the development of AD [46– 48]. Excess glutamate-mediated NMDA receptor over-activation appears to result in the formation of amyloid plaques, which leads to neuronal death [46–48]. NMDA receptors become overactivated after Aβ aggregation and NFT formation in AD, resulting in Ca2+ overflow into the cytoplasm [46–48]. The vital enzyme CREB (cyclic AMP response element binding protein) is activated by Ca2+ influx, which leads to mitochondrial dysfunction and signal suppression that lowers phospho-CREB levels [46–48]. The synthesis of pro-survival molecules like brain-derived neurotrophic factor (BDNF) is decreased when phospho-CREB is downregulated, making cells more susceptible to oxidative stress-induced cellular malfunction and neuronal death [46–48]. All of the aforementioned activities are synergistically mediated by nitric oxide, which stimulates the Aβ protein to produce more glutamate [46–48]. Overexpression of Aβ reduces glutamate absorption by glial cells, which increases glutamatergic excitotoxicity [46–48].

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

19

5.4 Dopaminergic Signaling in AD

The nigrostriatal pathway’s neurons in particular exhibited several pathological alterations, including NFT, Aβ plaques, neuropil threads, neuronal loss, and a drop in dopamine content, which is consistent with the function of dopaminergic neuronal degeneration in AD [49, 50]. These variables point to the potential involvement of dopaminergic signaling in the pathophysiology of AD [49, 50]. Learning, memory, and motivation are all influenced by the agonism of dopaminergic receptors, which are members of the family of metabotropic G-protein-coupled receptors [49, 50]. Dopamine receptors have been divided into five groups, designated as D1–D5 receptors, based on how sensitive they are to dopamine receptors [49, 50]. Many AD patients experienced Parkinson’s disease symptoms, often known as extrapyramidal symptoms of AD, indicating that postsynaptic D2 receptors in nigrostriatal pathways were downregulated [49, 50]. However, similar symptoms are missing in patients who do not have Parkinson’s disease. Furthermore, Kemppainen et al. [51] proposed a function for extra-striatal D2 receptor depletion in the course of AD. Interestingly, there has been a reduction of D2 receptors in both temporal and hippocampal regions, which are related to memory consolidation [49–51]. D2 receptors specifically influence caudate and hippocampal areas and their interactions to contribute to hippocampus-based cognitive activities [50].

5.5 Adrenergic Signaling in AD

Another class of metabotropic receptor is the adrenergic receptor, which is divided into two categories, α and β, and its subtypes α1, α2, and β1–β3 [52–54]. The brain expresses both α and β adrenergic receptors, which regulate cognitive processes [52–54]. It has been observed that noradrenergic locus coeruleus neurons, which are the main source of norepinephrine to the cortical region of the brain, are destroyed [52–54]. Though in the brain, the β1 adrenergic receptor was diminished, β2 receptors were significantly increased [52–54]. In contrast, the hippocampus area of the affected brain exhibited an increase in both β1 and β2 adrenergic receptors [52–54]. Additionally, AD patients have much more beta-2 adrenergic receptors than healthy individuals, particularly in the cerebral microvasculature, which is indicative of functional abnormalities in the blood-brain barrier [52–54]. Pascual et al. proposed that the density of α2 receptors in the cortex and hippocampal areas of AD brains is significantly reduced [52–54]. Observations on the changes in norepinephrine levels in AD-affected brains remain controversial despite the loss of locus coeruleus neurons and diverse variations in the density of adrenergic receptors in different parts of the brain [52–54]. It has been proposed that as AD advances, the loss of noradrenergic locus coeruleus neurons increases amyloid plaques, neurofibrillary tangles, and dementia severity [52–54]. Remarkably, Kong et al. revealed that locus coeruleus projections play a critical role in Aβ uptake and clearance by

20

Vinay Kumar and Kunal Roy

microglia cells, lending credence to the notion that these projections act as a link between neuron and microglia to design the host response to Aβ in AD [52–54]. Additionally, it has been noted that soluble Aβ, which is dependent on pro-inflammatory cytokines and causes changes in locus coeruleus neurons, establishes a connection between inflammation and noradrenergic signaling in AD [52–54].

6

Novel Therapeutic Strategies for the Treatment of AD Undeniably, worldwide researchers have made significant efforts to the identification of novel strategies for the treatment of AD. AChE enzyme inhibitors (donepezil, galantamine, tacrine, and rivastigmine) and NMDA receptor blockers (memantine) are the only two kinds of drugs available for the treatment of AD [55]. Very recently, aducanumab (brand name: Aduhelm) was approved by FDA for the treatment of AD [56]. It is an amyloid beta-directed monoclonal antibody that targets aggregated forms of amyloid beta found in the brains of people with AD to reduce its buildup [56]. It was developed by Biogen and Eisai [56]. Additionally, the availability of more recent atypical neuroleptics and serotonin-modulating antidepressants has increased hope for effective therapy of AD [56]. The therapeutic efficacy of these substances is still somewhat limited, despite their usage in medicine. As a result, throughout the last decade, the emphasis has switched to the development of a new therapeutic strategy for the treatment of AD. New hope for the treatment of AD has been provided by the development of a 7-mehtoxytacrine (7-MEOTA) derivative based on the multitarget-directed ligand [56].

6.1

Antioxidants

Since there is growing evidence that oxidative stress may play a role in the etiology of AD, the therapeutic emphasis has turned to the use of antioxidants as a target for both curative and preventative management of AD [57, 58]. Clinical studies have shown that vitamin E, which may have antioxidant properties, can lessen caregiver stress by slowing the deterioration in mild to moderate AD [57, 58]. It has also been demonstrated that long-term use of Ginkgo biloba can prevent the age-dependent decline in spatial cognition in a transgenic mice model of AD without changing Aβ levels [57, 58]. An endogenous antioxidant called coenzyme Q10 analog known as ubiquinone has demonstrated modest enhancements in cognitive abilities in AD patients [57, 58]. However, scientists have not discovered any therapeutic benefits of omega-3 fatty acids in mild to moderate cases of AD [57, 58]. The development of effective therapeutic interventions for the treatment of AD still requires additional clinical research, even though antioxidants have demonstrated positive benefits in several clinical studies involving AD patients.

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

21

6.2 Antiinflammatory Agents

The inflammatory hypothesis has been widely established as a fundamental factor in the pathophysiology of AD, and as a result, nonsteroidal anti-inflammatory medicines (NSAIDs) have been examined but failed in clinical trials for the treatment of AD [59, 60]. Flurbiprofen, an NSAID, has demonstrated encouraging outcomes in reducing cognitive, behavioral, and mental dysfunctions in AD [59, 60]. This is currently undergoing phase 3 clinical studies for the treatment of AD and has also been stated to modulate γ-secretase [59, 60]. However, the current data suggest that this NSAID may fail because of very limited pharmacological activity against γ-secretase [59, 60]. In its phase 3 clinical studies, RFlurbiprofen failed to demonstrate any clinical advantages [59, 60]. This failure is often ascribed to the drug’s inability to permeate the BBB and demonstrate significant pharmacodynamic characteristics [59, 60]. Since NSAIDs are likely the safest pharmacological choice for AD (in comparison to other agents), more resources should be employed to develop these drugs as novel therapy, and additional clinical trials are needed to investigate and validate their efficacy in AD patients.

6.3 Atypical Antipsychotics

Atypical antipsychotic medications have been widely utilized to treat behavioral and psychological dementia symptoms, such as hallucinations, aggression/agitation, delusions, oppositional conduct, and wandering that appear in AD patients [61, 62]. Due to the elevated risk of death in AD patients, these atypical antipsychotics are nevertheless accompanied by a “black box” warning from the FDA [61, 62]. In clinical trials, both olanzapine and risperidone at modest doses dramatically reduced psychological symptoms linked with AD [61, 62]. Antipsychotics, on the other hand, are associated with major adverse effects for instance somnolence, urinary tract infection, edema, and irregular gait [61, 62].

6.4

Recent studies indicate that antidepressants have a great deal of promise for treating AD patients who exhibit signs of depression. Antidepressant treatments are now being developed in several clinical studies to treat AD symptoms [62–64]. Citalopram, a selective serotonin reuptake inhibitor therapy that significantly reduced neuropsychiatric symptoms, especially agitation, in clinical investigations on AD patients, had the most notable outcomes [62–64]. The use of citalopram during AD has, however, been linked to several major side effects, such as a worsening of cognitive decline, a lengthened QTc interval, anorexia, weight loss, and sleeplessness [62–64]. However, limited study has been done utilizing these medications in AD; therefore further clinical studies are needed to develop antidepressants as a probable treatment for the disruptive and antagonistic behavior exhibited in AD [62–64].

Antidepressants

22

Vinay Kumar and Kunal Roy

6.5 HMG-CoA Reductase Inhibitors

A rate-regulating enzyme called 3-hydroxy-3-methylglutaryl-coenzyme A (HMG-CoA) reductase turns cholesterol-producing HMG-CoA into mevalonate [65, 66]. The progression of AD and high-plasma cholesterol is associated, according to numerous epidemiological research [65, 66]. Additionally, rabbit brains on a high-cholesterol diet showed pathology similar to AD. Cholesterol-induced AD-like pathology was evident in these rabbits and included several pathological markers for instance Aβ deposits, neurofibrillary tangles, apoptosis, microglia activation, and increased ventricular volume [65, 66]. In rabbits fed with a high-cholesterol diet, reducing circulating cholesterol was observed to reduce Aβ production and AD development [65, 66]. Additionally, statins and HMG-CoA reductase inhibitors reduced the generation of Aβ in ways that were dependent on lower cholesterol synthesis. Furthermore, studies show that statins improve endothelial function, increase cerebral blood flow, improve immunological modulation with an anti-inflammatory effect, decrease platelet aggregation, and have antioxidant activity [65, 66]. Simvastatin has also shown slight cognitive improvement in randomized, double-blind, placebo-controlled trials in AD patients, whereas atorvastatin has shown mild improvement in AD-associated cognitive and behavioral symptoms [65, 66]. For the development of a novel pharmacological therapy for the treatment of AD, additional clinical investigations utilizing new HMG-CoA inhibitors are required.

6.6 Hormone Replacement Therapy

Hormone replacement therapy is effective in the treatment of AD throughout the previous decade. Androgens influence many different brain functions, including visual-spatial memory and behavior [67]. Studies have shown that declining testosterone levels are associated with increased beta-amyloid levels in the brain, hyperphosphorylation of tau protein, and reduced neuronal survival [67]. The cognitive abilities of male AD patients (like spatial, verbal memory, and constructive ability) significantly improved after testosterone therapy [67]. Even though a small number of clinical research on hormone replacement treatment demonstrated positive outcomes, more mechanistic clinical investigations are needed to understand how androgens affect AD.

6.7 7-Methoxytacrine Derivative

It is well established that AD is a complex and multifactorial neurodegenerative disease with several pathogenic pathways that are interrelated [68]. To combat complex and multifactorial disorders like Alzheimer’s, numerous researchers are looking at the design of ligands that can act on numerous targets. Tacrine was the first AChEI to be approved by the Food and Drug Administration (FDA) in 1993 for the treatment of mild to moderate AD symptoms [68]. Despite having some promising pharmacokinetic characteristics, it was taken off the market because it had low selectivity

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

23

for AChEI and had negative side effects [68]. After that, efforts were made to develop derivatives of tacrine that wouldn’t have the first lethal side effects [68]. As a result, the tacrine analog 7-MEOTA was developed, having the same pharmacological effect as its parent molecule without the tacrine-related toxicity [68]. Korabecny et al. coupled 7-MEOTA with p-anisidine via an alkyl tether comprising thiourea or urea, in line with the MTD approach for AD [68, 69]. It was thought that by including thiourea or urea in the linker, AChEI’s inhibitory action might be increased [68, 69]. Tacrine-trolox and tacrine-scutellarin hybrids were developed as part of a group of pharmacological AD treatments based on the MTD strategy [68, 69]. These hybrid molecules were discovered to be particularly effective in suppressing AChE while also exhibiting low in vivo toxicity [68, 69].

7

Drugs in Clinical Trials for the Treatment of AD In ongoing clinical trials, scientists are developing and evaluating several potential treatments with different objectives, such as antiamyloid and anti-tau, neurotransmitter alteration, antineuroinflammation and neuroprotection, cognitive augmentation, and treatments to ease behavioral and psychiatric symptoms. In this review, we present the current state of clinical trials (Table 4) for AD as available from clinicaltrials.gov.

7.1 Reason for Failed Clinical Trials of Disease-Modifying Agents (DMA) for AD and Their Contribution to Current Research

For years, no medicine has been developed to prevent or treat memory-robbing conditions. The World Health Organization (WHO) reports that dementia is currently the seventh largest cause of death globally and one of the main causes of disability and reliance in older people [70]. According to data published in the Alzheimer’s Association journal in 2020, 121 unique drugs were being investigated in 136 studies to find a cure for AD (in the USA) [70]. This therapeutic development pipeline, gleaned from a review of the US FDA’s clinical trials registry, appears impressive until compared to other data [70]. Previously, researchers looked at 244 substances in 413 clinical trials between 2002 and 2012 and discovered a shocking 99.6% failure rate compared to 81% for cancer [70]. By the time a drug pipeline evaluation was published in 2020, no medicine had made it to the finish line. The USFDA, on the other hand, authorized Biogen’s beta-amyloid targeting medication Aduhelm in 2021, making it the first novel Alzheimer’s medicine approved in the USA in nearly 20 years [71]. The study data, however, sparked a heated dispute among scientists after the fast-track approval [71]. While that debate rages on, gantenerumab and crenezumab’s failure has called into question the methodology of depending too heavily on neutralizing beta-amyloid to combat Alzheimer’s disease, which is the strategy

Alzinova AB, CRST Oy VTBIO Co. LTD Virogenics, Inc., National Institute on Aging (NIA), Celerion Asceneuron Pty Ltd., Asceneuron S.A., Neuroscience Trials Australia Acumen Pharmaceuticals, National Institute on Aging (NIA) TrueBinding, Inc. NKGen Biotech, Inc. Mark Tuszynski, Case Western Reserve University, Ohio State University, University of California, San Diego Merck Sharp & Dohme LLC BeyondBio Inc. Merck Sharp & Dohme LLC IGC Pharma LLC, India Globalization Capital Inc Alnylam Pharmaceuticals APRINOIA Therapeutics, LLC, APRINOIA Therapeutics H. Lundbeck A/S

NCT05328115 ALZ-101

NCT05016427 VT301

NCT05318040 CMS121

NCT04759365 ASN51

NCT04931459 ACU193

NCT04920786 TB006

NCT04678453 SNK01

NCT05040217 AAV2-BDNF

NCT05227118 MK-8189 and MK-8189017

NCT04476303 BEY2153

NCT04730635 MK-0000-413

NCT04749563 IGC-AD1

NCT05231785 ALN-APP

NCT05344989 APNmAb005

NCT04149860 Lu AF87908

NCT04388982 Allogenic adipose MSC-Exos Ruijin Hospital, Cellular Biomedicine Group Ltd.

Johns Hopkins University

NCT04123314 Psilocybin

Sponsor/collaborators Mclean Hospital, Spier Family Foundation

Drug

NCT04075435 Cannabidiol solution

Clinical trial identifier

50

18–65

I/II

I

I

I

18 18–65

I

I

I

I

I

I

I

I

I

I

I

I

I

I

18–99

55–85

19–80

65–85

50–80

55–85

18–55

55–90

18–85

19–85

50–79

50–80

18–85

60–90

August 2022

November 22, 2022

January 2023

July 2025

July 2021

February 14, 2023

October 2021

November 14, 2023

March 1, 2025

December 2023

January 1, 2023

December 2022

December 31, 2022

December 17, 2022

April 1, 2022

July 31, 2023

December 30, 2023

January 11, 2023

Age (years) Phase Completion date

Table 4 Current status of anti-Alzheimer’s drugs in clinical trials (https://clinicaltrials.gov/, date of data collection: August 22, 2022)

24 Vinay Kumar and Kunal Roy

II II

50–89

55–80 45–85

Brigham and Women’s Hospital Stemedica Cell Technologies, Inc., Stemedica International SA Beth Israel Deaconess Medical Center Eli Lilly and Company University of Colorado, Denver, National Institute on Aging (NIA), Alzheimer’s Association, Partner Therapeutics, Inc. Capital Medical University, Xinjiang Uygur Pharmaceutical Co., Ltd.

Marc L Gordon, MD, Janssen Scientific Affairs, LLC, Northwell Health 55–85

50–85

Alzamend Neuro, Inc.

Janssen Research & Development, LLC Alector Inc., AbbVie IntelGenx Corp. reMYND Cognition Therapeutics Vivoryon Therapeutics N.V., Alzheimer’s Disease Cooperative Study (ADCS), National Institute on Aging (NIA) University Hospital, Toulouse, Foundation Plan Alzheimer The Israeli Medical Center for Alzheimer’s Janssen Research & Development, LLC Pharmazz, Inc.

NCT05040321 Sirtuin-NAD activator

NCT02833792 Allogeneic human mesenchymal stem cells

NCT03875638 Levetiracetam

NCT05063539 LY3372689

NCT04902703 GM-CSF/Sargramostim

NCT05269173 Flos Gossypii Flavonoids Tablet

NCT04070378 Daratumumab

NCT05307692 Seltorexant

NCT04592874 AL002

NCT03402503 Montelukast

NCT05478031 REM0046127

NCT04735536 CT1812

NCT03919162 PQ912

NCT03435861 Neflamapimod

NCT05239390 SCI-110

NCT04619420 JNJ-63733657

NCT04052737 PMZ-1620

60–85

18–90

50–85

50

50–85

55–85

50–85

60–80

60–85

50–90

55–80

55–85

50–80

II

II

II

II

II

II

II

II

II

II

II

II

II

II

I/II

I/II

I/II

NCT05363293 AL001

50–85

Hoffmann-La Roche

I/II

NCT04639050 RO7126209

55–85

Vaccinex Inc., Alzheimer’s Drug Discovery Foundation., Alzheimer’s Association

NCT04381468 SEMA4D blockade

(continued)

October 2022

November 5, 2025

June 29, 2023

June 30, 2021

May 31, 2023

December 31, 2022

June 7, 2023

December 2023

January 2024

April 19, 2023

June 1, 2023

December 30, 2023

July 2024

June 13, 2024

November 2023

June 2023

December 1, 2024

May 31, 2023

October 3, 2024

February 28, 2023

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 25

UCB Biopharma SRL, UCB Pharma Sunnybrook Health Sciences Centre, Alzheimer’s Drug Discovery Foundation, Weston Brain Institute Johns Hopkins University, Mclean Hospital, Miami Jewish Health AbbVie St. Joseph’s Hospital and Medical Center, Phoenix, National Institute on Aging (NIA), The Cleveland Clinic Vivoryon Therapeutics N.V., Nordic Bioscience A/S|Amsterdam UMC, location VUmc Longeveron Inc., bioRASI, LLC The University of Texas Health Science Center at San Antonio Johns Hopkins University CuraSen Therapeutics, Inc. Cognition Therapeutics Otsuka Pharmaceutical Co., Ltd. Anavex Life Sciences Corp., Anavex Australia Pty Ltd., Anavex Germany GmbH Otsuka Pharmaceutical Development & Commercialization, Inc. Suven Life Sciences Ltd. Shahid Sadoughi University of Medical Sciences and Health Services, National Institute for Medical Research Development (NIMAD), McMaster University

NCT04867616 Bepranemab

NCT02085265 Telmisartan vs. perindopril

NCT02792257 Dronabinol

NCT05291234 ABBV-916

NCT04032626 MCLENA-1

NCT04498650 PQ912

NCT05233774 Lomecel-B

NCT04629495 Rapamycin

NCT04601038 CORT108297

NCT05104463 CST-2032 and CST-107

NCT03507790 CT1812

NCT03620981 Brexpiprazole

NCT04314934 ANAVEX2-73-AD-004

NCT04464564 AVP-786

NCT05397639 Masupirdine

NCT04842552 Hydralazine

Sponsor/collaborators Dongzhimen Hospital, Beijing

Drug

NCT04780399 Yangxue Qingnao pills

Clinical trial identifier

Table 4 (continued)

49

50–90

50–90

55–85

55–90

50–85

50–80

55

55–89

60–85

50–80

50–89

50–90

60–95

50

50–80

65–85

October 24, 2023

December 2022

January 1, 2024

August 2024

September 29, 2023

July 2023

September 2024

December 30, 2024

May 2023

September 2023

July 2025

December 30, 2024

III

III

III

December 20, 2023

January 2025

December 31, 2024

II/III July 31, 2024

II/III March 2023

II

II

II

II

II

II

II

II

II

II

II

II

Age (years) Phase Completion date

26 Vinay Kumar and Kunal Roy

18

Hoffmann-La Roche Imperial College London Sunnybrook Health Sciences Centre, Alzheimer’s Drug Discovery Foundation, Weston Brain Institute Athira Pharma Hoffmann-La Roche Eli Lilly and Company JHSPH Center for Clinical Trials, National Institute on Aging (NIA) Novo Nordisk A/S Alzheon Inc., National Institute on Aging (NIA) Hyundai Pharmaceutical Co., LTD. University of California, San Diego Daewoong Bio Inc. Green Valley (Shanghai) Pharmaceuticals Co., Ltd. Green Valley (Shanghai) Pharmaceuticals Co., Ltd. Biogen

NCT03444870 Gantenerumab

NCT03116126 Noradrenergic with guanfacine

NCT04516057 Nabilone

NCT04488419 ATH-1017

NCT05256134 Gantenerumab

NCT05463731 LY3372993

NCT03108846 Escitalopram

NCT04777409 Semaglutide

NCT04770220 ALZ-801 in APOE4/4

NCT04229927 BPDO-1603

NCT03703856 Memantine

NCT05383183 Choline alfoscerate

NCT05181475 GV-971

NCT05058040 GV-971

NCT05310071 Aducanumaba

Approved by FDA (June 7, 2021) but still in controversy

a

50–85

Eli Lilly and Company

NCT05026866 Donanemab

III

III

III

III

III

III

III

III

III

III

III

60–85

IV

IV

IV

IV

50–85

October 31, 2026

June 2024

June 2025

December 9, 2025

June 30, 2024

IV 50–83

July 31, 2024

April 26, 2026

March 1, 2023

III

III

August 2022

November 15, 2025

October 13, 2028

October 2022

October 25, 2025

December 31, 2022

October 27, 2026

November 8, 2027

October 2026

October 2023

November 2024

January 2023

III

45

50–80

55–85

18–109 III

60–85

60–80

55–85

55

45

50–90

65–80

50–85

Green Valley (Shanghai) Pharmaceuticals Co., Ltd.

NCT04520412 GV-971

50–87

50

60–85

Cassava Sciences, Inc., Premier Research Group plc

˜ © gional de Recherche University Hospital, Lille, Groupement InterrA Clinique et d’Innovation, Laboratory of excellence DISTALZ| ˜ ©gion Nord-Pas de Calais, France, Meo coffee RA

BioVie Inc.

NCT04994483 Simufilam 100 mg

NCT04570085 CAFFEINE

NCT04669028 NE3107

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 27

28

Vinay Kumar and Kunal Roy

used in the bulk of Alzheimer’s research. We reviewed the underlying mechanisms of clinical trials, tried to understand the reason why prior clinical trials failed, and analyzed the future trend of AD clinical trials. Here we have mentioned some points to justify the reason behind the failure of previous clinical trials: 1. Insufficient knowledge of the complex pathophysiology of AD: inaccurate selection of leading treatment targets and inconsistency in the drug dosages 2. Insufficient knowledge of the complex pathophysiology of AD: late instigation of treatments during the progression of AD development 3. New biomarkers of Aβ metabolism and aggregation: CSF Aβ38 and plasma BACE1 4. Vascular system’s new biomarkers: heart-type fatty acidbinding protein (hFABP) 5. Newly identified biomarkers of inflammation and glial activation: CSF YKL-40 6. Newly identified biomarkers for synaptic dysfunction: CSF SNAP-25 and synaptotagmin 7. Newly identified biomarkers for α-synuclein pathology: CSF α-synuclein 8. Newly identified biomarkers for TDP-43 pathology: plasma TDP-43 9. Iron metabolism-related newly identified biomarkers: CSF ferritin 10. Oxidative stress biomarkers 11. Other neuronal proteins as newly identified biomarkers: CSF visinin-like protein 1 (VILIP-1), CSF, and plasma NF-L

8

Nanomaterials for the Treatment of AD The intricacy of the therapeutic approach to AD stems not only from its unknown origin and a lack of effective treatments but also from limited access to the affected organ. The BBB, blood-CSF barrier (BCSFB), and ependymal barrier are three biological barriers that regulate the entry of the majority of chemicals into the brain and hence considerably restrict drug availability [72–74]. The BBB, blood-CSF barrier (BCSFB), and ependymal barrier are three biological barriers that regulate the entry of the majority of chemicals into the brain and hence considerably restrict drug availability [72–74]. The gastrointestinal barrier and hepatic first-pass effect must be considered when administering medicines orally [72– 74]. Similarly, drug half-lives in the body are greatly shortened by

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

29

clearance processes and efflux pumps, which lowers their pharmacological potency [72–74]. Aside from approachability issues, drugs delivered to the CNS and other organs must possess certain biologic features that impart high bioavailability, which is not always present [72–74]. Even though their pharmacological activity may be significant, many medications suffer from physicochemical drawbacks, for instance, limited solubility, short stability, and a high molecular weight that greatly lower their bioavailability and, as a result, their ultimate therapeutic efficacy [72–74]. Other obstacles that CNS interventions confront include the presence of peripheral side effects and the difficulty in determining and maintaining the therapeutic threshold over time [72–74]. To address these challenges, the therapeutic potential of drug-loaded nanocarriers in several neurodegenerative disorders has been investigated. The high surface-to-volume ratio of nanocarriers and the ability for surface functionalization with selected ligands are the two most explored characteristics of these devices for drug delivery to the CNS [72–74]. Additionally, the primary goal of nanomedicine up till now has been Aβ targeting. Three nanotechnology approaches have been employed to specifically target and modify senile plaques: (i) genetic regulation and/or blockage of Aβ-peptide synthesis; (ii) inhibition and/or delay of the Aβ-nucleation-dependent process; and (iii) clearance of already formed Aβ plaques [72–74]. To transfer BACE1-siRNA into the brain and suppress BACE expression, Alvarez-Erviti et al. [75] used naturally innocuous exosome nanocarriers. This prevented the cleavage of APP and, consequently, the overproduction of Aβ peptide [76]. Recent outcomes concerning the most commonly used nanocarrier and selective nanoparticle-based delivery systems targeting AD experiments are given in Tables 5 and 6.

9

Natural Compounds a Promising Treatment Strategy Against AD Currently, herbs are among the most effective medications used to treat and slow the progression of a variety of illnesses, including diabetes, cancer, and neurological conditions like AD [77, 78]. Due to their effectiveness and lack of side effects, natural products have recently gained a lot of popularity as supplements or alternative therapies. More than 80% of medications were produced directly or indirectly from natural compounds before the emergence of post-genomic high-efficiency screening technologies [77, 78]. Herbal drugs are molecules produced from natural herbal sources that may be less toxic and have fewer adverse effects than synthetic medications, as well as being less expensive. According to studies, natural products have been used in over half of the treatments produced since 1994 [77, 78]. Some herbal items and phytochemicals contain anticholinesterase, anti-inflammatory,

Mesoporous silica NPs (MSN)

N-cetyltrimethyl ammonium bromide- Rivastigmine hydrogen functionalized with succinic tartrate anhydride and 3-aminopropyltriethoxysilane N-cetyltrimethylammonium bromide Metal chelator 5-chloro4-hydroxy-7and tetraethoxysilane functionalized iodoquinoline with gold nanoparticle

CLPFFD peptide

Gold nanoparticles of 5 nm with PEG; Anthocyanin

Gold (Au) NPs

Gold colloids-rods and spheres

Curcumin derivative Sphingomyelin and cholesterol functionalized with dimyristoylphosphatidic acid (PA) or bifunctionalized with PA and modified apolipoprotein E-derived peptide PLGA [poly (lactic-co-glycolic acid)]Peptide Aβ functionalized with anti-transferrin receptor monoclonal antibody (OX26) and anti-Aβ Cholesterol and soybean Curcumin and nerve phosphatidylcholine functionalized growth factor with surface wheat germ agglutinin (WGA) and cardiolipin (CL)

Ref.

[73]

[73]

Neuronal cell death/ cholinergic systems

BBB transcytosis, amyloid cascades PC12 rat adrenal medulla cells/ endothelial cell line

[73]

Simulated gastric and body fluids and neuroblastoma SH-SY5Y cell line viability

Mouse brain endothelial cells/Aβ1-42 Amyloid cascades and tau mouse model hyperphosphorylation Porcine brain capillary endothelial cells Amyloid cascades

[73]

[73]

Human neuroblastoma cell line/AD rat model

BBB transcytosis, amyloid cascades and tau hyperphosphorylation

[73]

Aβ, cholinergic dysfunction [73]

AD targeted

Porcine brain capillary endothelial cells BBB transcytosis, amyloid cascades

Blood plasma and cerebrospinal fluid

Potential drug candidate Investigation model

Liposomal

Nanoparticle form Carrier material

Table 5 Nanocarrier-based systems for targeting Aβ, Tau proteins, and mitochondrial therapy in AD

30 Vinay Kumar and Kunal Roy

Iminodiacetic acid

Epigallocatechin-3-gallate Amyloid fibrillation experiments(EGCG) in vitro and neuroblastoma SH-SY5Y cell line viability

Double transgenic AD mice/ neuroblastoma and brain capillary endothelial cell line Colchicine induced AD rats

Single-walled carbon nanotube

Hydrophobic pyridylphenylene dendrimers

Iron crystal structure; functionalized with PEG Protein-capped (PC)-Fe3O4 and PC-CdS nanoparticle

Pristine multi-walled (MW)CNTs; phospholipids and polysorbates

Dendrimers

Metallic NPs

Carbon nanotubes (CNTs)

Berberine

Amyloid cascades acetyl cholinesterase enzyme

(continued)

[74]

[73]

Tau cascades

Beta-amyloid-induced AD in Wistar rats BBB transcytosis, cholinergic systems

[73]

Amyloid cascades

[73]

[73]

Inhibit acetyl cholinesterase [73] enzyme

Amyloid cascades Inclusion bodies of ovine prion protein representing amyloid protein aggregates

Coarse-grained nanoparticle model and implicit/explicit lipid models

Amyloid fibrillation experimentsin vitro; No toxicity on neuroblastoma cells; PC metal nanoparticles; inhibition and disaggregation PCFe3O4 and PC-CdS of Tau nanoparticles

Iron oxide

o-Phenylene diamine

Tunable zero-dimension

[73]

Cholinergic and oxidative stress systems Amyloid cascades

[73]

[73]

[73]

[73]

Amyloid cascades and tau tangles

Zinc-mediated Aβ42 aggregation/ Amyloid cascades human neuroblastoma SH-SY5Y cell line viability; SH-SY5Y cell line; human glioblastoma Oxidative stress astrocytoma; immortalized mouse hippocampal cell line; 5XFAD transgenic AD mouse model

Clitoria ternatea-mediated synthesis of Clitoria ternatea The learning and memory capacity of graphene quantum dots graphene quantum dots ctGQDs in rats was improved in (ctGQDs) radial arm and water Morris maze assays

N-isopropyl acrylamide and Nt- butyl acrylamide-monomers; acrylic acid

1,2-distearoyl-sn-glycero-3Cerium(III) acetate phosphoethanolamine-N-[methoxy (polyethylene glycol)-2000]; TPP conjugated 1,2-distearoyl-snglycero-3-phosphoethanolamine-N[amino(polyethylene glycol)-2000] Dendrigraft poly-l-lysines and RVG29 peptide and polyethylene glycol (PEG) BACE1-AS shRNA gene Chitosan Piperine

Glycidyl methacrylate

Carbon dots

Graphene quantum dots (GQDs)

Polymeric NPs

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 31

Solid lipid NPs (SLNs)

RVG-9R/BACE1 siRNA

SH-SY5Y neuroblastoma cell line

Isoproterenol-induced cognitive deficits in rats Induced pluripotent mouse stem cells (iPSCs)

Human epithelial adenocarcinoma (Caco-2) cells Human brain-like endothelial cells

Potential drug candidate Investigation model

Cetyl palmitate and functionalized with Resveratrol/grape seed monoclonal antibody (OX26 mAb) extract Galantamine Glyceryl behenate lipids hydrobromide Nerve growth factor Heparin-conjugated stearic acid; (NGF) stearylamine-cationic lipid; esterquat 1 Cetyl palmitate Rapamycin (Rp)

Lipids and chitosan

Nanoparticle form Carrier material

Table 5 (continued)

[74]

[74]

Neuronal cell death

Mammalian target of rapamycin (mTOR) signaling pathway

[74]

[74]

[74]

Ref.

BBB transcytosis and Amyloid cascades Cholinergic systems

Amyloid cascades

AD targeted

32 Vinay Kumar and Kunal Roy

Phosphatidic acid/ cardiolipin

Retro-inverso peptide inhibitor (RIOR2, rGffvlkGr) + maleimidepoly (ethylene glycol)

1,2-Distearoyl-sn-glycero3-phosphoethanolamine (DSPE) conjugated polyethylene glycol with active succinimidyl ester (DSPE-PEGNHS) + Congo Red

Liposome (Shirasu porous glass + cholesterol)

Retro-inverso peptide inhibitor nanoparticles

Iron oxide

Cell

110 ± 6

131 ± 43

250–350

Inhibitors of aggregation of the Alzheimer’s Aβ peptide H2O2-responsive therapy

APPSWE/PS1dE9 transgenic Mice

APPSWE transgenic mice

APP/PS1 transgenic mice injection over 3 weeks

Model

Particle size (nm)

Modulate tau 102 ± 2 phosphorylation and glycogen synthase kinase 3 activities

Anti-transferrin monoclonal Curcumin antibody + poly(ethylene glycol

Nanoliposome (1,2-distearoyl-snglycero-3phosphocholine; cholesterol)

Bioactive molecule

Ligand-targeted

Nanoparticle material

Table 6 Selective nanoparticle-based delivery methods for AD treatment

Interfered with Aβ aggregation and neurotoxicity

[73]

[73]

[73]

[73]

Ref.

(continued)

Inhibited formation of Aβ oligomers and fibrils in vitro

Reduced Aβ clearance

Retardation of Aβ fibril formation

Pharmacological effect

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 33

Tanshinone IIA

Adenosine

Nerve growth factor

Negatively charged AuNPs ~30

Poly(ethylene glycol)

–

B6 peptide (transferrin substitute PEG)

Pokysorbate 80

–

Cationic bovine serum albumin

Lipidic (squalene)

Polylactic acid

Poly (butylcyanoacrylate)

Carboxyl-conjugated AuNPs (negative charged)

250 ± 30

Pharmacological effect

Ref.

–

[73]

[73]

[73]

Disrupted Aβ fibrillation and [73] fragmented the fibrils already formed

Rats with acute Reversed scopolaminescopolamine-induced induced amnesia and amnesia improve recognition and memory

Enhanced drug uptake in the brain ameliorated learning impairments, cholinergic disruption, and loss of hippocampal neurons

Mouse 2 h MCAO and Decreased infarct volume; increased neurological 22 h reperfusion 24 h deficit scores of permanent MCAO

Mouse 2 h MCAO and Decreased infarct volume, [73] 24 h reperfusion neurological deficit and caspase-3 activity Mouse 2 h MCAO and Decreased infarct volume; [73] 24 h reperfusion increased motor function deficit scores Rat 2 h MCAO and Decreased infarct volume, [73] 24 h reperfusion neurological deficit, neutrophil infiltration and neuronal apoptosis

Model

Nanoparticles angiopep-2- 118.3 ± 7.8 Mice injected with aggregated Aβ1–40 conjugated, 125 NAP (NAPVSIPQ)-loaded (NAP: neuroprotective peptide)

120

114 ± 14

Z-DEVD-FMK and bFGF 747 ± 42

Z-DEVD-FMK (caspase-3 650 ± 2 inhibitor)

Transferrin receptor antibody poly (ethylene glycol) Transferrin receptor antibody

Particle size (nm)

Chitosan

Bioactive molecule

Ligand-targeted

Nanoparticle material

Table 6 (continued)

34 Vinay Kumar and Kunal Roy

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

35

antioxidant, and neuroprotective effects that make them attractive for neurological diseases [77, 78]. The use of appropriate animal models in numerous scientific research has revealed the positive effects of natural products in the treatment of AD [77, 78]. Several natural compounds, including vitamins C and E, luteolin, melatonin, curcumin, quercetin, resveratrol, huperzine A, and rosmarinic acid, have been demonstrated to be beneficial in Alzheimer’s disease patients [77, 78]. Table 7 briefly shows the promising natural products that can prevent and treat neurodegeneration in AD.

10

Multitarget-Directed Ligand (MTDL): A Promising Strategy for AD Multitarget-directed ligands (MTDLs), a novel trend in drug design and discovery, have emerged since the year 2000 [79, 80]. Such a strategy looks to be especially useful in the treatment of complex disorders like AD. Caproctamine, one of the earliest examples of purposefully developed MTD, demonstrated synergistic cholinergic activity against AD by antagonizing presynaptic muscarinic acetylcholine M2 autoreceptors and inhibiting the AChE enzyme [79, 80]. This technique has been gradually used in AD drug discovery over the last two decades, and many papers have highlighted the advantages of the MTD approach over classical target-specific medicines (TSMs) and their combinations, while rarely emphasizing its limitations and shortcomings [79, 80]. But so far, this strategy has fallen short of expectations and has only produced two clinical candidates (ladostigil and NP-61), both of which were unsuccessful in clinical trials [79, 80]. Due to the consistently disappointing findings of clinical studies, novel candidate medications generally seem to be a long way from being approved (Table 8). The evaluation of clinical trials needs to be changed; thus there is a big demand for it. It is also recommended that continuous outcome measurements and discrete clinical state ratings be integrated to boost statistical power in clinical trials relative to any single endpoint, in addition to a more exact allocation of patients to trial groups. In Fig. 3, we have highlighted nine major targets linked with AD, which are AChE, β-amyloid aggregation, BACE-1, GSK-3β, MAOs, metal ions in the brain, NMDA receptor, 5-hydroxytryptamine (5-HT) receptors, the third subtype of histamine receptor (the H3 receptor), and phosphodiesterase (PDEs). Additionally, 11 multitarget design strategies have been categorized by the involvement of AChE (AChE and BACE-1, AChE and GSK-3β, AChE and MAOs, AChE and metal ions, AChE and NMDA receptor, AChE and 5-HT receptors, AChE and H3 receptor, AChE and PDEs) and without the involvement of AChE (BACE-1 and GSK-3β, MAO-B and metal ions, PDEs and metal ions), which were reported in recent years [81–110] for improvement of AD therapy.

Prevents tau phosphorylation, Reduces Aβ Reduces Aβ plaque accumulation, Improves mitochondrial function, Inhibits acetylcholinesterase activity Reduces the cholinergic effects, inhibits oxidative stress Improvement the symptoms by reducing the number of amyloid plaques Reduces Aβ plaque aggregation and activates microglia in the hippocampus and cortex

Reduces Aβ aggregation, hyperphosphorylation APPOSK mice (Aβ oligomer model), Tg2576 [77] of tau mice (AD model), and tau609 mice (tauopathy model)

Scutellaria genus

Huperzine A

Galanthus woronowii

Curcuma longa

Grapes and mulberries

–

Berberis, rhizoma coptidis or Reduces amyloid plaque accumulation and a reduction in AD phenotypic pathology Mahonia aquifolium

Glycine in vitro

Green tea

Tea

Baicalein

Alkaloids

Galantamine

Curcumin

Resveratrol

Rifampicin

Berberine

Betaine

EGCG

Catechin

Inhibits Aβ fibrillogenesis

Inhibits the oxidative stress in AD and declines the neurons’ apoptosis

Decrease homocysteine levels and Aβ toxicity

Improvement in memory and learning and reducing fibrillar amyloid accumulation

Tea and celery

In vitro

[77]

[77]

[77]

Caenorhabditis elegans AD model APP/PS1 transgenic mice

[77]

[77]

[77]

[77]

[77]

[77]

[77]

TgCRND8 mouse

AβPP/PS-1 transgenic mouse

APP/PS1 transgenic mice

Male C57BL/6Hsd mice, APPswe/PS1d, E9 mouse

Transgenic APPswe/PS1dE9 AD mice

APP/PS1mice, Tg2576 transgenic mice

APP/PS1 double transgenic AD mouse

[77]

Apigenin

APPsw/Tg 2576 mice

Reduces chronic oxidative stress, improves memory and anxiety-related behavioral disorders

Punica granatum Linn.

Ref.

Phenol

Animal model

Mechanism

Source

Compound

Table 7 List of selected promising natural compounds and crude plant extracts for the treatment of AD

36 Vinay Kumar and Kunal Roy

Prevents age-related cognitive impairment

–

DHA

[78]

The therapeutic potential of piperine ameliorates Male C57BL/6 mice oxidative-nitrosative stress, restores neurotransmission, and reduces neuroinflammation

Gardenia jasminoides Ellis

Piper longum

Strawberry, apple, Fisetin treatment regulates the neuronal survival Male wild-type C57BL/6N mice persimmon, grape, onion, PI3K/Akt/GSk-3β signaling and prevents and cucumber neuroinflammation and neurodegeneration CN-SLNs with their improved bioavailability are Adult male Sprague Dawley rats an effective armament against Aβ-induced oxidative stress

Ginger

Zingiber officinale

Genipin

Piperine

Fisetin

Passion flower

Induces neurite growth by activating neuronal PC12h cells, Neuro2a cells nitric oxide synthase (nNOS), protects hippocampal neurons from Aβ poisoning, and prevents Aβ peptide-induced neurotoxicity

Marine bryozoan

Bryostatin-1

Chrysin

[78]

Reduces AchE activity and Lipid peroxidation

Apples and onions Reduces amyloid plaques In vitro

APP/PS1 transgenic mouse model

Reduces oxidative stress and increases cognitive Zebrafish AD model, 3xTg-AD mice, and function, reduces extracellular Aβ non-transgenic (non-Tg) mice

[78]

[78]

[78]

[77]

[77]

[77]

Quercetin

AD flies

Reduces oxidative stress, AChE activity, and accumulation of Aβ42 peptides, inhibition of ROS

Broccoli

Luteolin

[77]

Non-transgenic AD rats Improvements in insulin signaling and ameliorating cognitive dysfunction, Increases p-GSK3β and inhibits AChE activity

[77]

[77]

Cinnamaldehyde Cinnamon

Female Wistar rats

Decreases tau phosphorylation, THY-Tau22 mice pro-inflammatory and oxidative stress markers

Coffee

Caffeine

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 37

38

Vinay Kumar and Kunal Roy

Table 8 Multitarget drugs for AD treatment in clinical trials (https://clinicaltrials.gov/, Date of data collection: October 6, 2022) Clinical trial identifier

Drug

Phase Mechanism of action on targets

NCT02913664 Losartan + amlodipine + atorvastatin III

Angiotensin II receptor blocker: losartan Calcium channel blocker: amlodipine Cholesterol agent: atorvastatin

NCT03533257 AMX0035 (sodium phenylbutyrate and tauroursodeoxycholic acid combination)

II

Chemical chaperone to inhibit endoplasmic reticulum stress responses. (Sodium phenylbutyrate) Naturally occurring bile acid to tackle mitochondrial dysfunction. (tauroursodeoxycholic acid)

NCT02547818 ALZT-OP1 (combination of NCT04570644 cromolyn and ibuprofen)

III II

Mast cell stabilizer (cromolyn) Anti-inflammation (ibuprofen)

NCT03790709 ANAVEX2-73 (blarcamesine) NCT04314934 NCT02756858

III III II

Sigma-1 receptor agonist M1 receptor agonist and M2 receptor antagonist GSK-3β inhibitor

NCT03393520 AVP786 (combination of NCT02446132 dextromethorphan and quinidine) NCT04464564 NCT04408755

III III III III

Sigma-1 receptor agonist (dextromethorphan) NMDA receptor antagonist (dextromethorphan)

NCT03620981 Brexpiprazole NCT03594123 NCT03548584 NCT03724942

III III III III

D2 receptor agonist 5-HT receptor agonist

NCT02008357 Gantenerumab and solanezumab

III

Monoclonal antibody directed at plaques and oligomers (gantenerumab) Monoclonal antibody directed at monomers (solanezumab)

NCT04063124 Dasatinib + quercetin (combination therapy)

II

Tyrosine kinase inhibitor (dasatinib) Flavonoid with antioxidant and anti-Aβ fibrilization properties (quercetin)

NCT02033941 Grapeseed extract

II

Polyphenolic compound with antioxidant property Anti-oligomerization

NCT03062449 L-serine

II

Synthesis of sphingolipids and phosphatidylserine The precursor of D-serine, a co-agonist of NMDARs

NCT03867253 ORY-2001 (vafidemstat)

II

LSD1 inhibitor MAO-B inhibitor (continued)

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

39

Table 8 (continued) Clinical trial identifier

Drug

Phase Mechanism of action on targets

NCT02085265 Telmisartan + perindopril

II

Angiotensin II receptor blocker (telmisartan) Angiotensin-converting enzyme inhibitor (perindopril)

NCT03748303 Allopregnanolone

I

Growth hormones to promote neurogenesis Positive allosteric GABAARs modulators

5-HT 5-hydroxytryptamine, GABAARs γ-aminobutyric acid type A receptors, GSK-3β glycogen synthase kinase 3, LSD1 lysine-specific histone demethylase 1A, MAO-B monoamine oxidase B, NMDA N-methyl-D-aspartate, NMDARs N-methyl-D-aspartate

Fig. 3 Brief connections between AD and 9 major targets and 11 multitarget design strategies based on the targets. (Note: red arrow denotes the multitarget strategies involving AChE and the black arrow denotes the multitarget strategies without AChE involvement)

11

Conclusions and Prospects One of the most difficult tasks in the area of CNS drug research is the exploration of therapeutic strategies for AD. Without a doubt, there is a definite need for improved pharmacological therapy to treat Alzheimer’s patients. It is extremely important in discovering

40

Vinay Kumar and Kunal Roy

new clinical therapies for AD patients since these symptoms are distressing for the patients and have a substantial influence on their lives. Even far, the causes of AD are not completely clear, even though the disease appears to begin in middle age and advances silently for many years, and clinical symptoms of dementia do not appear until the condition is advanced. The primary emphasis of drug discovery efforts has been to interfere with the amyloid pathway, which includes preventing Aβ peptide formation and aggregation or increasing Aβ peptide elimination. It is possible that because AD is so diverse, a multimodal strategy is required to halt disease progression. Recently, an anti-amyloid antibody drug named aducanumab (Aduhelm™) developed by Biogen and Eisai got approval (7 June 2021) as a treatment for AD from the US Food and Drug Administration (FDA). The study data, however, sparked a heated dispute among scientists after the fast-track approval of Aduhelm. It is anticipated that over 126 drugs, which are currently undergoing various stages of clinical trials, would soon be available on the market. Current nontherapeutic strategies in AD include Aβ targeting, metal ions binding, cholinesterase inhibition, neuroprotection, and estrogen therapy based on the many AD etiology hypotheses. The inclusion of active pharmacological compounds as well as targeted molecules can help to improve these nanomaterials. Since the AD is a multifactorial disease, the new medication design strategy should concentrate on multitarget directed ligands rather than single-target ligands because therapies based on single-target ligand target strategies failed to demonstrate pharmacodynamic effects during extensive clinical trials on AD patients. Recently, based on this multitarget-directed ligand concept, the 7-methoxytacrine (7-MEOTA) derivative was developed, which has given rise to new hope in the treatment of AD. However, a multitarget strategy is currently being implemented in the clinic and clinical trials. The fifth medication for moderate-to-severe Alzheimer’s patients to receive FDA approval combines memantine with one of the cholinesterase inhibitors (donepezil). Blarcamesine, a multifunctional medication that acts as a sigma-1 and muscarinic dual agonist as well as a GSK-3 inhibitor, is now in phase 3 clinical trials for the treatment of AD. Table 8 provides an overview of multitarget therapeutics, which are primarily the combination of numerous medicines with various anti-AD effects and multitarget drugs that are undergoing clinical trials for AD. There are now 13 multitargeting drugs being tested in 22 clinical trials for the treatment of AD, including 6 drugs in phase 3, 6 drugs in phase 2, and 1 drug in phase 1. ANAVEX2-73 is intended to influence synaptic dysfunction, cholinergic neurotransmission, and tauopathy via regulating the sigma-1 receptor, muscarinic receptors, and GSK-3β. Many therapeutic targets are known to play roles in numerous disease pathways. Therefore, therapeutic modulations of these targets may be advantageous in treating AD via a variety

Recent Progress in the Treatment Strategies for Alzheimer’s Disease

41

of modes of action. Given the complexity of AD pathophysiology, multifunctional medicines developed with many target possibilities could result in a milestone in AD therapeutic research. Future clinical research may benefit from a library of lead compounds that have undergone preclinical testing for various diseases and multitarget therapies. The development of multifunctional drugs may offer promise for the treatment of AD; however, there is no royal road to overcoming AD. Finally, the authors believe that this chapter will give readers a fundamental overview of AD pathogenesis, multiple targets of AD, disease-modifying drugs currently on the market, and the range of research being done on therapeutic approaches for AD.

Acknowledgments VK thanks the Indian Council of Medical Research (ICMR), New Delhi (File No: BMI/11(03)/2022, IRIS Cell No.: 2021-8243, dated: 13/05/2022) for financial support in the form of a Research Associateship (RA). References 1. Samanta S, Ramesh M, Govindaraju T (2022) Chapter 1: Alzheimer’s is a multifactorial disease. In: Alzheimer’s disease: recent findings in pathophysiology. Diagnostic and therapeutic modalities, pp 1–34. https://doi.org/ 10.1039/9781839162732-00001 2. Alzheimer’s Association (2021) Alzheimer’s disease facts and figures. Alzheimers Dement 17(3):327–406 ˇ arna´ M, Bennett 3. Onyang IG, Jauregui GV, C JP, Stokin GB (2021) Neuroinflammation in Alzheimer’s disease. Biomedicine 9(5):524. h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / biomedicines9050524 4. Tatulian SA (2022) Challenges and hopes for Alzheimer’s disease. Drug Discov Today. https://doi.org/10.1016/j.drudis.2022. 01.016 5. Gauthier S, Rosa-Neto P, Morais JA, Webster C (2021) World Alzheimer report 2021: journey through the diagnosis of dementia. Alzheimer’s Disease International, London 6. Srivastava S, Ahmad R, Khare SK (2021) Alzheimer’s disease and its treatment by different approaches: a review. Eur J Med Chem 216: 113320. https://doi.org/10.1016/j.ejmech. 2021.113320 7. Kumar V, Ojha PK, Saha A, Roy K (2020) Exploring 2D-QSAR for prediction of beta-

secretase 1 (BACE1) inhibitory activity against Alzheimer’s disease. SAR QSAR Environ Res 31(2):87–133. https://doi.org/10. 1080/1062936X.2019.1695226 8. Rajasekhar K, Govindaraju T (2018) Current progress, challenges and future prospects of diagnostic and therapeutic interventions in Alzheimer’s disease. RSC Adv 8(42): 23780–23804. https://doi.org/10.1039/ C8RA03620A 9. Lei G, Zhong MB, Larry Z, Bin Z, Dongming C (2022) Sex differences in Alzheimer’s disease: insights from the multiomics landscape. Biol Psychiatry 91(1):61–71. https://doi. org/10.1016/j.biopsych.2021.02.968 10. Konstantina GY, Papageorgiou SG (2020) Current and future treatments in Alzheimer disease: an update. J Cent Nerv Syst 12:1–12. h t t p s : // d o i . o r g / 1 0 . 1 1 7 7 / 1179573520907397 11. Kejing L, Ji N, Zhang X, Qiao W, Tang Z, Gou X (2019) Drug development for Alzheimer’s disease. J Drug Target 27(2):164–173. https://doi.org/10.1080/1061186X.2018. 1474361 12. Roy K (ed) (2018) Computational modeling of drugs against Alzheimer’s disease. Springer New York, Springer Science + Business Media, LLC, part of Springer Nature 2018.

42

Vinay Kumar and Kunal Roy

https://doi.org/10.1007/978-1-49397404-7 13. Zhang B, Zhao J, Wang Z, Guo P, Liu A, Du G (2021) Identification of multi-target antiAD chemical constituents from traditional Chinese medicine formulae by integrating virtual screening and in vitro validation. Front Pharmacol 12:709607. https://doi.org/10. 3389/fphar.2021.709607 14. Nadeem MS, Khan JA, Rashid U (2021) Fluoxetine and sertraline based multitarget inhibitors of cholinesterases and monoamine oxidase-A/B for the treatment of Alzheimer’s disease: synthesis, pharmacology and molecular modeling studies. Int J Biol Macromol 193:19–26. https://doi.org/10.1016/j. ijbiomac.2021.10.102 15. Brunetti L, Leuci R, Carrieri A, Catto M, Occhineri S, Vinci G, Piemontese L (2022) Structure-based design of novel donepezillike hybrids for a multi-target approach to the therapy of Alzheimer’s disease. Eur J Med Chem 237:114358. https://doi.org/ 10.1016/j.ejmech.2022.114358 16. Ajala A, Uzairu A, Shallangwa GA, Abechi SE (2022) 2D QSAR, design, docking study and ADMET of some N-aryl derivatives concerning inhibitory activity against Alzheimer disease. Future J Pharm Sci 8(1):1–14. https://doi.org/10.1186/s43094-02200420-w 17. Ambure P, Bhat J, Puzyn T, Roy K (2019) Identifying natural compounds as multitarget-directed ligands against Alzheimer’s disease: an in silico approach. J Biomol Struct Dyn 37(5):1282–1306. https://doi.org/10. 1080/07391102.2018.1456975 18. Kumar V, Saha A, Roy K (2020) In silico modeling for dual inhibition of acetylcholinesterase (AChE) and butyrylcholinesterase (BuChE) enzymes in Alzheimer’s disease. Comput Biol Chem 88:107355. https://doi. org/10.1016/j.compbiolchem.2020. 107355 19. Kumar A, Nisha CM, Silakari C, Sharma I, Anusha K, Gupta N, Kumar A (2016) Current and novel therapeutic molecules and targets in Alzheimer’s disease. J Formos Med Assoc 115(1):3–10. https://doi.org/10.1016/j. jfma.2015.04.001 20. Karran E, De Strooper B (2022) The amyloid hypothesis in Alzheimer disease: new insights from new therapeutics. Nat Rev Drug Discov 21(4):306–318 21. Ma C, Hong F, Yang S (2022) Amyloidosis in Alzheimer’s disease: pathogeny, etiology, and related therapeutic directions. Molecules

27(4):1210. https://doi.org/10.3390/ molecules27041210 22. Cho Y, Bae HG, Okun E, Arumugam TV, Jo DG (2022) Physiology and pharmacology of amyloid precursor protein. Pharmacol Ther:108122. https://doi.org/ 10.1016/j.pharmthera.2022.108122 23. Mashal Y, Abdelhady H, Iyer AK (2022) Comparison of tau and amyloid-β targeted immunotherapy nanoparticles for Alzheimer’s disease. Biomol Ther 12(7):1001. https:// doi.org/10.3390/biom12071001 24. Aillaud I, Funke SA (2023) Tau aggregation inhibiting peptides as potential therapeutics for Alzheimer disease. Cell Mol Neurobiol 43(3):951–961 25. Gonza´lez A, Singh SK, Churruca M, Maccioni RB (2022) Alzheimer’s disease and tau self-assembly: in the search of the missing link. Int J Mol Sci 23(8):4192. https://doi.org/ 10.3390/ijms23084192 26. Kumar K, Kumar A, Keegan RM, Deshmukh R (2018) Recent advances in the neurobiology and neuropharmacology of Alzheimer’s disease. Biomed Pharmacother 98:297–307. https://doi.org/10.1016/j.biopha.2017. 12.053 27. Bozzo F, Mirra A, Carrı` MT (2017) Oxidative stress and mitochondrial damage in the pathogenesis of ALS: new perspectives. Neurosci Lett 636:3–8. https://doi.org/10.1016/j. neulet.2016.04.065 28. Bai R, Guo J, Ye XY, Xie Y, Xie T (2022) Oxidative stress: the core pathogenesis and mechanism of Alzheimer’s disease. Ageing Res Rev:101619. https://doi.org/10.1016/ j.arr.2022.101619 29. de Oliveira J, Kucharska E, Garcez ML, Rodrigues MS, Quevedo J, MorenoGonzalez I, Budni J (2021) Inflammatory cascade in Alzheimer’s disease pathogenesis: a review of experimental findings. Cell 10(10):2581. https://doi.org/10.3390/ cells10102581 30. Whitson HE, Colton C, El Khoury J, Gate D, Goate A, Heneka MT, Terrando N (2022) Infection and inflammation: new perspectives on Alzheimer’s disease. Brain Behav Immun Health:100462. https://doi.org/10.1016/j. bbih.2022.100462 31. Zeng L, Wang M, Zhou J, Wang X, Zhang Y, Su P (2022) A hypothesis: retrotransposons as a relay of epigenetic marks in intergenerational epigenetic inheritance. Gene 817: 146229. https://doi.org/10.1016/j.gene. 2022.146229

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 32. Millan MJ (2022) The epigenetic dimension of Alzheimer’s disease: causal, consequence, or curiosity. Dialogues Clin Neurosci. https://doi.org/10.31887/DCNS.2014.16. 3/mmillan 33. Wang SC, Oelze B, Schumacher A (2008) Age-specific epigenetic drift in late-onset Alzheimer’s disease. PLoS One 3(7):e2698. https://doi.org/10.1371/journal.pone. 0002698 34. Bakulski KM, Dolinoy DC, Sartor MA, Paulson HL, Konen JR, Lieberman AP, Rozek LS (2012) Genome-wide DNA methylation differences between late-onset Alzheimer’s disease and cognitively normal controls in human frontal cortex. J Alzheimers Dis 29(3):571–588. https://doi.org/10.3233/ JAD-2012-111223 35. Malaguarnera M, Ferri R, Bella R, Alagona G, Carnemolla A, Pennisi G (2004) Homocysteine, vitamin B12 and folate in vascular dementia and in Alzheimer disease. Clin Chem Lab Med 42(9):1032–1035. https://doi.org/10. 1515/CCLM.2004.208 36. Ding H, Dolan PJ, Johnson GV (2008) Histone deacetylase 6 interacts with the microtubule-associated protein tau. J Neurochem 106(5):2119–2130. https://doi.org/ 10.1111/j.1471-4159.2008.05564.x 37. Tang YP, Gershon ES (2022) Genetic studies in Alzheimer’s disease. Dialogues Clin Neurosci. https://doi.org/10.31887/DCNS. 2003.5.1/yptang 38. Uddin MS, Hasana S, Hossain M, Islam M, Behl T, Perveen A, Ashraf GM (2021) Molecular genetics of early-and late-onset Alzheimer’s disease. Curr Gene Ther 21(1):43–52. h t t p s : // d o i . o r g / 1 0 . 2 1 7 4 / 1566523220666201123112822 39. Chen ZR, Huang JB, Yang SL, Hong FF (2022) Role of cholinergic signaling in Alzheimer’s disease. Molecules 27(6):1816. h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / molecules27061816 40. McNerney M, Heath A, Narayanan S, Yesavage J (2022) Repetitive transcranial magnetic stimulation improves brain-derived neurotrophic factor and cholinergic signaling in the 3xTgAD mouse model of Alzheimer’s disease. J Alzheimers Dis (Preprint):1–9. https://doi.org/10.3233/JAD-215361 41. Giacobini E, Cuello AC, Fisher A (2022) Reimagining cholinergic therapy for Alzheimer’s disease. Brain. https://doi.org/10. 1093/brain/awac096 42. Kumar V, De P, Ojha PK, Saha A, Roy K (2020) A multi-layered variable selection

43

strategy for QSAR modeling of butyrylcholinesterase inhibitors. Curr Top Med Chem 20(18):1601–1627. https://doi.org/10. 2174/1568026620666200616142753 43. Kumar V, Saha A (2020) Chemometric modeling of structurally diverse carbamates for the inhibition of acetylcholinesterase (AChE) enzyme in Alzheimer’s disease. Int J Quant Struct Prop Relatsh 5(3):6–60. https://doi. org/10.4018/IJQSPR.2020070102 44. Aaldijk E, Vermeiren Y (2022) The role of serotonin within the microbiota-gut-brain axis in the development of Alzheimer’s disease: a narrative review. Ageing Res Rev:101556. https://doi.org/10.1016/ j.arr.2021.101556 45. Nykamp MJ, Zorumski CF, Reiersen AM, Nicol GE, Cirrito J, Lenze EJ (2022) Opportunities for drug repurposing of serotonin reuptake inhibitors: potential uses in inflammation, infection, cancer, neuroprotection, and Alzheimer’s disease prevention. Pharmacopsychiatry 55(01):24–29. https://doi.org/ 10.1055/a-1686-9620 46. Abd-Elrahman KS, Ferguson SS (2022) Noncanonical metabotropic glutamate receptor 5 signaling in Alzheimer’s disease. Annu Rev Pharmacol Toxicol 62:235–254. https://doi. org/10.1146/annurev-pharmtox021821-091747 47. Sulkowski G, Wencel PL, Da˛browskaBouta B, Struzynska L, Strosznajder R (2022) Alterations in the transcriptional profile of genes related to glutamatergic signalling in animal models of Alzheimer’s disease. The effect of fingolimod. Folia Neuropathol 60(1):10–23. https://doi.org/10.5114/fn. 2022.114302 48. Pinky PD, Pfitzer JC, Senfeld J, Hong H, Bhattacharya S, Suppiramaniam V, Reed MN (2022) Recent insights on glutamatergic dysfunction in Alzheimer’s disease and therapeutic implications. Neuroscientist:10738584211069897. h t t p s : // d o i . o r g / 1 0 . 1 1 7 7 / 10738584211069897 49. Huang Q, Zhang C, Qu S, Dong S, Ma Q, Hao Y, Shi Y (2022) Chinese herbal extracts exert neuroprotective effect in Alzheimer’s disease mouse through the dopaminergic synapse/apoptosis signaling pathway. Front Pharmacol 13. https://doi.org/10.3389/ fphar.2022.817213 50. Gloria Y, Ceyze´riat K, Tsartsalis S, Millet P, Tournier BB (2021) Dopaminergic dysfunction in the 3xTg-AD mice model of Alzheimer’s disease. Sci Rep 11(1):1–11

44

Vinay Kumar and Kunal Roy

51. Kemppainen N, Laine M, Laakso MP, Kaasinen V, Na˚gren K, Vahlberg T, Rinne JO (2003) Hippocampal dopamine D2 receptors correlate with memory functions in Alzheimer’s disease. Eur J Neurosci 18(1): 149–154. https://doi.org/10.1046/j. 1460-9568.2003.02716.x 52. Yu ZY, Yi X, Wang YR, Zeng GH, Tan CR, Cheng Y, Liu YH (2022) Inhibiting α1adrenergic receptor signaling pathway ameliorates AD-type pathologies and behavioral deficits in APPswe/PS1 mouse model. J Neurochem 161(3):293–307. https://doi. org/10.1111/jnc.15603 53. Bekdash RA (2021) The cholinergic system, the adrenergic system and the neuropathology of Alzheimer’s disease. Int J Mol Sci 22(3):1273. https://doi.org/10.3390/ ijms22031273 54. Goodman AM, Langner BM, Jackson N, Alex C, McMahon LL (2021) Heightened hippocampal β-adrenergic receptor function drives synaptic potentiation and supports learning and memory in the TgF344-AD rat model during prodromal Alzheimer’s disease. J Neurosci 41(26):5747–5761. https://doi. org/10.1523/JNEUROSCI.0119-21.2021 55. Li X, Li J, Huang Y, Gong Q, Fu Y, Xu Y, Li J (2022) The novel therapeutic strategy of vilazodone-donepezil chimeras as potent triple-target ligands for the potential treatment of Alzheimer’s disease with comorbid depression. Eur J Med Chem 229:114045. https://doi.org/10.1016/j.ejmech.2021. 114045 56. Vaz M, Silva V, Monteiro C, Silvestre S (2022) Role of aducanumab in the treatment of Alzheimer’s disease: challenges and opportunities. Clin Interv Aging 17:797. https:// doi.org/10.2147/CIA.S325026 57. Pritam P, Deka R, Bhardwaj A, Srivastava R, Kumar D, Jha AK, Jha SK (2022) Antioxidants in Alzheimer’s disease: current therapeutic significance and future prospects. Biology 11(2):212. https://doi.org/10. 3390/biology11020212 58. Kabir MT, Rahman MH, Shah M, Jamiruddin MR, Basak D, Al-Harrasi A, Abdel-Daim MM (2022) Therapeutic promise of carotenoids as antioxidants and anti-inflammatory agents in neurodegenerative disorders. Biomed Pharmacother 146:112610. https://doi.org/10. 1016/j.biopha.2021.112610 59. Brod SA (2022) Anti-inflammatory agents: an approach to prevent cognitive decline in Alzheimer’s disease. J Alzheimers Dis 85(2): 457–472. https://doi.org/10.3233/ JAD-215125

60. Galimi R (2022) Interaction between antiAlzheimer’s disease drugs and antipsychotic agents in the treatment of behavioral and psychological symptoms: extrapyramidal side effects. Adv Neurol Neurosci 5(2):108–119 61. Phelps EB, Swantek S (2022) Effectiveness of atypical antipsychotic drugs in patients with Alzheimer’s disease. In: Essential reviews in geriatric psychiatry. Springer, Cham, pp 313–317 62. Mokhtari T (2022) Targeting autophagy and neuroinflammation pathways with plantderived natural compounds as potential antidepressant agents. Phytother Res. https:// doi.org/10.1002/ptr.7551 63. He Y, Li H, Huang J, Huang S, Bai Y, Li Y, Huang W (2021) Efficacy of antidepressant drugs in the treatment of depression in Alzheimer disease patients: a systematic review and network meta-analysis. J Psychopharmacol 35(8):901–909. https://doi.org/10. 1177/02698811211030181 64. Correia AS, Vale N (2021) Antidepressants in Alzheimer’s disease: a focus on the role of mirtazapine. Pharmaceuticals 14(9):930. https://doi.org/10.3390/ph14090930 65. Rosoff DB, Bell AS, Jung J, Wagner J, Mavromatis LA, Lohoff FW (2022) Mendelian randomization study of PCSK9 and HMG-CoA reductase inhibition and cognitive function. J Am Coll Cardiol 80(7): 653–662 66. Cheraghzadeh M, Nazeri Z, Mohammadi A, Azizidoost S, Aberomand M, Kheirollah A (2021) Amyloid Beta sharply increases HMG-CoA reductase protein levels in astrocytes isolated from C57BL/6 mice. Gene Rep 23:101070. https://doi.org/10.1016/j. genrep.2021.101070 67. Cardinali CA, Martins YA, Torra˜o AS (2021) Use of hormone therapy in postmenopausal women with Alzheimer’s disease: a systematic review. Drugs Aging 38(9):769–791 68. Mitra S, Muni M, Shawon NJ, Das R, Emran TB, Sharma R, Sweilam SH (2022) Tacrine derivatives in neurological disorders: focus on molecular mechanisms and neurotherapeutic potential. Oxid Med Cell Longev. https:// doi.org/10.1155/2022/7252882 69. Makhaeva GF, Kovaleva NV, Boltneva NP, Rudakova EV, Lushchekina SV, Astakhova TY, Richardson RJ (2022) Bis-amiridines as acetylcholinesterase and butyrylcholinesterase inhibitors: N-functionalization determines the multitarget anti-Alzheimer’s activity profile. Molecules 27(3):1060. https://doi.org/ 10.3390/molecules27031060

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 70. Yiannopoulou KG, Anastasiou AI, Zachariou V, Pelidou SH (2019) Reasons for failed trials of disease-modifying treatments for Alzheimer disease and their contribution in recent research. Biomedicine 7(4):97. h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / biomedicines7040097 71. Esang M, Gupta M (2021) Aducanumab as a novel treatment for Alzheimer’s disease: a decade of hope, controversies, and the future. Cureus 13(8). https://doi.org/10.7759/ cureus.17591 72. Pange SS, Patwekar M, Patwekar F, Alghamdi S, Babalghith AO, Abdulaziz O, Mallick J (2022) A potential notion on Alzheimer’s disease: nanotechnology as an alternative solution. J Nanomater. https://doi. org/10.1155/2022/6910811 73. Nguyen TT, Nguyen TTD, Nguyen TKO, Vo TK (2021) Advances in developing therapeutic strategies for Alzheimer’s disease. Biomed Pharmacother 139:111623. https://doi.org/ 10.1016/j.biopha.2021.111623 74. Zeng H, Qi Y, Zhang Z, Liu C, Peng W, Zhang Y (2021) Nanomaterials toward the treatment of Alzheimer’s disease: recent advances and future trends. Chin Chem Lett 32(6):1857–1868. https://doi.org/10. 1016/j.cclet.2021.01.014 75. Alvarez-Erviti L, Seow Y, Yin H, Betts C, Lakhal S, Wood MJ (2011) Delivery of siRNA to the mouse brain by systemic injection of targeted exosomes. Nat Biotechnol 29(4):341–345 76. Cano A, Turowski P, Ettcheto M, Duskey JT, Tosi G, Sa´nchez-Lo´pez E, Boada M (2021) Nanomedicine-based technologies and novel biomarkers for the diagnosis and treatment of Alzheimer’s disease: from current to future challenges. J Nanobiotechnology 19(1):1–30 77. Noori T, Dehpour AR, Sureda A, SobarzoSanchez E, Shirooie S (2021) Role of natural products for the treatment of Alzheimer’s disease. Eur J Pharmacol 898:173974. https:// doi.org/10.1016/j.ejphar.2021.173974 78. Mani R, Sha Sulthana A, Muthusamy G, Elangovan N (2022) Progress in the development of naturally derived active metabolites-based drugs: potential therapeutics for Alzheimer’s disease. Biotechnol Appl Biochem. https:// doi.org/10.1002/bab.2317 79. Zhang P, Xu S, Zhu Z, Xu J (2019) Multitarget design strategies for the improved treatment of Alzheimer’s disease. Eur J Med Chem 176:228–247. https://doi.org/10. 1016/j.ejmech.2019.05.020

45

80. Benek O, Korabecny J, Soukup O (2020) A perspective on multi-target drugs for Alzheimer’s disease. Trends Pharmacol Sci 41(7): 434–445. https://doi.org/10.1016/j.tips. 2020.04.008 81. Zhu Y, Xiao K, Ma L, Xiong B, Fu Y, Yu H, Shen J (2009) Design, synthesis and biological evaluation of novel dual inhibitors of acetylcholinesterase and β-secretase. Bioorg Med Chem 17(4):1600–1613. https://doi. org/10.1016/j.bmc.2008.12.067 82. Ferna´ndez-Bachiller MI, Pe´rez C, Monjas L, Rademann J, Rodrı´guez-Franco MI (2012) New Tacrine–4-Oxo-4 H-chromene hybrids as multifunctional agents for the treatment of Alzheimer’s disease, with cholinergic, antioxidant, and β-amyloid-reducing properties. J Med Chem 55(3):1303–1317. https://doi. org/10.1021/jm201460y 83. Mohamed T, Yeung JC, Vasefi MS, Beazely MA, Rao PP (2012) Development and evaluation of multifunctional agents for potential treatment of Alzheimer’s disease: application to a pyrimidine-2, 4-diamine template. Bioorg Med Chem Lett 22(14):4707–4712. https://doi.org/10.1016/j.bmcl.2012. 05.077 84. Viayna E, Sola I, Bartolini M, De Simone A, ˜ oz-Torrero Tapia-Rojas C, Serrano FG, Mun D (2014) Synthesis and multitarget biological profiling of a novel family of rhein derivatives as disease-modifying anti-Alzheimer agents. J Med Chem 57(6):2549–2567. https://doi. org/10.1021/jm401824w 85. Domı´nguez JL, Ferna´ndez-Nieto F, Castro M, Catto M, Paleo MR, Porto S, Sussman F (2015) Computer-aided structurebased design of multitarget leads for Alzheimer’s disease. J Chem Inf Model 55(1): 135–148. https://doi.org/10.1021/ ci500555g 86. Hui AL, Chen Y, Zhu SJ, Gan CS, Pan J, Zhou A (2014) Design and synthesis of tacrine-phenothiazine hybrids as multitarget drugs for Alzheimer’s disease. Med Chem Res 23(7):3546–3557 87. Jiang XY, Chen TK, Zhou JT, He SY, Yang HY, Chen Y, Sun HP (2018) Dual GSK-3β/ AChE inhibitors as a new strategy for multitargeting anti-Alzheimer’s disease drug discovery. ACS Med Chem Lett 9(3):171–176. https://doi.org/10.1021/acsmedchemlett. 7b00463 88. Bolea I, Juarez-Jimenez J, de los Rı´os C, Chioua M, Pouplana R, Luque FJ, Samadi A (2011) Synthesis, biological evaluation, and molecular modeling of donepezil and N-[(5-(benzyloxy)-1-methyl-1 H-indol-2-yl)

46

Vinay Kumar and Kunal Roy

methyl]-N-methylprop-2-yn-1-amine hybrids as new multipotent cholinesterase/monoamine oxidase inhibitors for the treatment of Alzheimer’s disease. J Med Chem 54(24): 8251–8270. https://doi.org/10.1021/ jm200853t 89. Wang L, Esteban G, Ojima M, BautistaAguilera OM, Inokuchi T, Moraleda I, Unzeta M et al (2014) Donepezil+ propargylamine+ 8-hydroxyquinoline hybrids as new multifunctional metal-chelators, ChE and MAO inhibitors for the potential treatment of Alzheimer’s disease. Eur J Med Chem 80: 543–561. https://doi.org/10.1016/j. ejmech.2014.04.078 90. Wu MY, Esteban G, Brogi S, Shionoya M, Wang L, Campiani G, Marco-Contelles J (2016) Donepezil-like multifunctional agents: design, synthesis, molecular modeling and biological evaluation. Eur J Med Chem 121:864–879. https://doi.org/10.1016/j. ejmech.2015.10.001 91. Sang Z, Pan W, Wang K, Ma Q, Yu L, Liu W (2017) Design, synthesis and biological evaluation of 3, 4-dihydro-2 (1H)-quinoline-Oalkylamine derivatives as new multipotent cholinesterase/monoamine oxidase inhibitors for the treatment of Alzheimer’s disease. Bioorg Med Chem 25(12):3006–3017. https://doi.org/10.1016/j.bmc.2017. 03.070 92. Lu C, Zhou Q, Yan J, Du Z, Huang L, Li X (2013) A novel series of tacrine–selegiline hybrids with cholinesterase and monoamine oxidase inhibition activities for the treatment of Alzheimer’s disease. Eur J Med Chem 62: 745–753. https://doi.org/10.1016/j. ejmech.2013.01.039 93. Xu Y, Zhang J, Wang H, Mao F, Bao K, Liu W, Li J (2018) Rational design of novel selective dual-target inhibitors of acetylcholinesterase and monoamine oxidase B as potential anti-Alzheimer’s disease agents. ACS Chem Neurosci 10(1):482–496. https://doi.org/10.1021/acschemneuro. 8b00357 94. Xu YX, Wang H, Li XK, Dong SN, Liu WW, Gong Q, Mao F (2018) Discovery of novel propargylamine-modified 4-aminoalkyl imidazole substituted pyrimidinylthiourea derivatives as multifunctional agents for the treatment of Alzheimer’s disease. Eur J Med Chem 143:33–47. https://doi.org/10. 1016/j.ejmech.2017.08.025 95. Ferna´ndez-Bachiller MI, Pe´rez C, Gonza´lezMunoz GC, Conde S, Lo´pez MG, Villarroya M, Rodrı´guez-Franco MI (2010) Novel tacrine- 8-hydroxyquinoline hybrids

as multifunctional agents for the treatment of Alzheimer’s disease, with neuroprotective, cholinergic, antioxidant, and coppercomplexing properties. J Med Chem 53(13): 4927–4937. https://doi.org/10.1021/ jm100329q 96. Li SY, Wang XB, Xie SS, Jiang N, Wang KD, Yao HQ, Kong LY (2013) Multifunctional tacrine–flavonoid hybrids with cholinergic, β-amyloid-reducing, and metal chelating properties for the treatment of Alzheimer’s disease. Eur J Med Chem 69:632–646. https://doi.org/10.1016/j.ejmech.2013. 09.024 97. Xie SS, Wang XB, Li JY, Yang L, Kong LY (2013) Design, synthesis and evaluation of novel tacrine–coumarin hybrids as multifunctional cholinesterase inhibitors against Alzheimer’s disease. Eur J Med Chem 64:540–553. https://doi.org/10.1016/j.ejmech.2013. 03.051 98. Liu Q, Qiang X, Li Y, Sang Z, Li Y, Tan Z, Deng Y (2015) Design, synthesis and evaluation of chromone-2-carboxamido-alkylbenzylamines as multifunctional agents for the treatment of Alzheimer’s disease. Bioorg Med Chem 23(5):911–923. https://doi. org/10.1016/j.bmc.2015.01.042 99. Yan J, Hu J, Liu A, He L, Li X, Wei H (2017) Design, synthesis, and evaluation of multitarget-directed ligands against Alzheimer’s disease based on the fusion of donepezil and curcumin. Bioorg Med Chem 25(12): 2946–2955. https://doi.org/10.1016/j. bmc.2017.02.048 100. Rosini M, Simoni E, Minarini A, Melchiorre C (2014) Multi-target design strategies in the context of Alzheimer’s disease: acetylcholinesterase inhibition and NMDA receptor antagonism as the driving forces. Neurochem Res 39(10):1914–1923 101. Makhaeva GF, Lushchekina SV, Boltneva NP, Sokolov VB, Grigoriev VV, Serebryakova OG, Bachurin SO (2015) Conjugates of γ-Carbolines and Phenothiazine as new selective inhibitors of butyrylcholinesterase and blockers of NMDA receptors for Alzheimer disease. Sci Rep 5(1):1–11 102. Rochais C, Lecoutey C, Gaven F, Giannoni P, Hamidouche K, Hedou D, Dallemagne P (2015) Novel multitarget-directed ligands (MTDLs) with acetylcholinesterase (AChE) inhibitory and serotonergic subtype 4 receptor (5-HT4R) agonist activities as potential agents against Alzheimer’s disease: the design of donecopride. J Med Chem 58(7): 3172–3187. https://doi.org/10.1021/acs. jmedchem.5b00115

Recent Progress in the Treatment Strategies for Alzheimer’s Disease 103. Liu W, Wang H, Li X, Xu Y, Zhang J, Wang W, Li J (2018) Design, synthesis and evaluation of vilazodone-tacrine hybrids as multitarget-directed ligands against depression with cognitive impairment. Bioorg Med Chem 26(12):3117–3125. https://doi.org/ 10.1016/j.bmc.2018.04.037 104. Huang W, Tang L, Shi Y, Huang S, Xu L, Sheng R, Hu Y (2011) Searching for the multi-target-directed ligands against Alzheimer’s disease: discovery of quinoxaline-based hybrid compounds with AChE, H3R and BACE 1 inhibitory activities. Bioorg Med Chem 19(23):7158–7167. https://doi.org/ 10.1016/j.bmc.2011.09.061 105. Mao F, Wang H, Ni W, Zheng X, Wang M, Bao K, Li J (2018) Design, synthesis, and biological evaluation of orally available firstgeneration dual-target selective inhibitors of acetylcholinesterase (AChE) and phosphodiesterase 5 (PDE5) for the treatment of Alzheimer’s disease. ACS Chem Neurosci 9(2): 328–345. https://doi.org/10.1021/ acschemneuro.7b00345 106. Prati F, De Simone A, Bisignano P, Armirotti A, Summa M, Pizzirani D, Cavalli A (2015) Multitarget drug discovery for Alzheimer’s disease: triazinones as BACE-1 and GSK-3β inhibitors. Angew Chem 127(5):

47

1598–1602. https://doi.org/10.1002/ ange.201410456 107. Di Martino RMC, De Simone A, Andrisano V, Bisignano P, Bisi A, Gobbi S, Belluti F (2016) Versatility of the curcumin scaffold: discovery of potent and balanced dual BACE-1 and GSK-3β inhibitors. J Med Chem 59(2):531–544. https://doi.org/10. 1021/acs.jmedchem.5b00894 108. Xie S, Chen J, Li X, Su T, Wang Y, Wang Z, Li X (2015) Synthesis and evaluation of selegiline derivatives as monoamine oxidase inhibitor, antioxidant and metal chelator against Alzheimer’s disease. Bioorg Med Chem 23(13):3722–3729. https://doi.org/10. 1016/j.bmc.2015.04.009 109. Su T, Zhang T, Xie S, Yan J, Wu Y, Li X, Luo HB (2016) Discovery of novel PDE9 inhibitors capable of inhibiting Aβ aggregation as potential candidates for the treatment of Alzheimer’s disease. Sci Rep 6(1):1–14 110. Wang Z, Wang Y, Wang B, Li W, Huang L, Li X (2015) Design, synthesis, and evaluation of orally available clioquinol-moracin M hybrids as multitarget-directed ligands for cognitive improvement in a rat model of neurodegeneration in Alzheimer’s disease. J Med Chem 58(21):8616–8637. https://doi.org/10. 1021/acs.jmedchem.5b01222

Part II Recent Advances in Computational Modeling of Anti-Alzheimer Drugs

Chapter 2 Understanding the Mechanisms of Amyloid Beta (Aβ) Aggregation by Computational Modeling Praveen P. N. Rao, Yusheng Zhao, and Rui Huang Abstract The misfolding, aggregation, and formation of complex self-assemblies of amyloid-beta (Aβ) are the key events in the pathophysiology of Alzheimer’s disease (AD). In the last two decades, research in this field has focused on discovering novel inhibitors of Aβ aggregation as potential therapies to treat Alzheimer’s disease. Computational modeling studies using the solved 3D structures of Aβ40 and Aβ42 peptides have led to a mechanistic understanding of the structural requirements to design novel inhibitors of Aβ aggregation. In contrast, molecules that can modulate and promote Aβ fibrillogenesis are far less explored. Such molecules are valuable chemical tools which can be used to study and understand the mechanisms of Aβ fibrillogenesis that can ultimately lead to the discovery of novel diagnostics and therapeutics. This chapter describes a systematic in silico study of Aβ42 self-assemblies and their interactions with the capped VQIVYK hexapeptide, to understand the mechanisms of Aβ42 fibrillogenesis. Computational techniques including molecular docking and nanoscale molecular dynamics (MD) simulations were carried out in explicit water. The interactions of the capped VQIVYK peptide in the Aβ42 oligomer model (peptide/Aβ molar ratio 1:1) suggest that it can prevent the conversion of Aβ42 oligomer into higher-order species by binding in a rigid, flat region at the N-terminus. The binding also leads to significant perturbations in the backbone structure of the oligomer assembly. In contrast, in the Aβ42 fibril model (peptide/Aβ molar ratio 2:1), two molecules of the capped VQIVYK peptide are able to bind at two symmetric and well-defined binding pockets. The binding of capped VQIVYK peptide blocks the solvent-exposed binding pockets, and it has a stabilizing effect on Aβ42 fibril assembly. The binding does not perturb the backbone structure of the fibril assembly based on the Cα RMSD calculations, which suggests that the flexible capped VQIVYK peptide can promote Aβ42 fibrillogenesis and exhibits better binding to higher-order structures such as Aβ42 fibrils. This study also demonstrates the significance of carrying out computational studies using appropriate molar ratios of Aβ42 binding ligands to understand the complex molecular mechanisms of Aβ42 fibrillogenesis, which serves as a novel computational strategy to study and design novel modulators of Aβ42 fibrillogenesis. Key words Amyloid-beta, Alzheimer’s disease, Capped VQIVYK peptide, Amyloid-beta-42 oligomer, Amyloid-beta-42 fibril, Molecular docking, Solvent-accessible surface area, Nanoscale molecular dynamics, Root mean square deviation, Root mean square fluctuations

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_2, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

51

52

1

Praveen P. N. Rao et al.

Introduction Alzheimer’s disease (AD) is a devastating neurodegenerative disorder, which is a major concern across the globe [1–4]. It is anticipated that the coming years will see a rapid rise in AD due to increasing life span and aging population worldwide [3, 5]. Despite the tremendous scientific advances in the field, the current treatment options for AD are very limited and do not offer permanent cure [6–9]. The recent controversial approval of the first antibody therapy aducanumab (Aduhelm®) by the US FDA has squarely brought the focus back on the amyloid cascade hypothesis of AD [10–12]. Aducanumab is a human IgG1 monoclonal antibody that targets the accumulated amyloid (Aβ) aggregates in the AD patient brain [10]. The amyloid cascade hypothesis of AD was proposed 30 years ago which posits that the misprocessing of amyloid precursor protein (APP) via secretases, results in the formation of Aβ peptides that undergo aggregation and deposit in the form of Aβ plaques, one of the characteristic hallmarks of AD brain [13– 17]. The APP gets cleaved by secretase enzymes to form two major types of Aβ peptides, known as Aβ40 and Aβ42. Among these two Aβ peptides, the longer peptide Aβ42 is known to be more toxic and tends to undergo rapid aggregation [14, 17– 24]. There are two broad approaches to design novel anti-AD therapies that target the amyloid cascade—(i) developing molecules as inhibitors of either β- or γ-secretase enzymes and (ii) developing molecules that can bind and prevent the misfolding and selfassembly of Aβ peptides into various soluble and insoluble toxic aggregates, including dimers, trimers, tetramers, higher-order oligomers, protofibrils, and fibrils [25–28]. The former approach was widely investigated to discover novel AD therapies. However, this field has fallen out of favor due to the failure of potential drug candidates during clinical trials [26]. The latter approach has led to the discovery of a wide variety of molecular libraries including small molecules and peptides that are able to inhibit the aggregation cascade of Aβ. These molecules have demonstrated impressive in vitro and in vivo activity profiles as potential anti-AD agents although none of them has entered the clinics [29–36]. Nevertheless, many of these Aβ aggregation inhibitors such as orange G are widely used as model compounds to study the mechanisms of Aβ misfolding and aggregation processes [30, 31]. In this regard, the misfolding and aggregation of Aβ into various types of selfassemblies is a complex process and the mechanisms involved are still not clearly understood. Toward this end, chemical tools/ molecular probes are being developed that are able to bind to various forms of Aβ aggregates, which can act as kinetic inhibitors, or stabilize the Aβ-assembly or promote Aβ disaggregation [37– 40]. These probes serve as valuable tools that help us understand

Understanding Aβ42 Oligomer and Fibril Binding

53

the complex molecular mechanism of Aβ misfolding, aggregation, and self-assembly and discover novel diagnostics and therapeutics for AD. 1.1 Tau-Derived Hexapeptide VQIVYK as a Novel Tool to Study Aβ Aggregation

2

The hexapeptide 306VQIVYK311 is part of the microtubule-binding region of the tau protein (441 amino acids) that is implicated in AD pathophysiology [41–43]. The VQIVYK peptide is known to undergo spontaneous self-assembly to form β-sheet-rich steric zippers also known as paired helical filaments (PHF6) [41, 44, 45]. The VQIVYK derived steric zippers are used as models to study aggregation mechanisms and to develop potential AD therapies [30, 44–48]. Interestingly, our previous research demonstrated that the capped VQIVYK peptide (Fig. 1) was capable of promoting the aggregation of both Aβ40 and Aβ42 peptides [49]. These activities were demonstrated by in vitro biophysical studies including thioflavin T-based fluorescence aggregation kinetics and transmission electron microscopy experiments [49]. This was a highly unusual result, as most of the work in this field tends to discover Aβ aggregation inhibitors as either novel agents or chemical tools to study the mechanisms of Aβ misfolding and aggregation pathways [29–38]. Our studies also demonstrated that the capped hexapeptide VQIVYK is able to rescue neuronal hippocampal cells from Aβ-induced toxicity despite promoting Aβ fibrillogenesis in vitro [49]. These observations suggest that VQIVYK binding to Aβ may alter the self-assembly pathways and the conformations of Aβ aggregates to reduce their toxicity to hippocampal neuronal cells. Furthermore, this experimental evidence also shows that the capped VQIVYK peptide can be used as a novel chemical tool to study the mechanisms of Aβ aggregation. Therefore, we investigated the interactions of the capped VQIVYK peptide with Aβ42 oligomer and fibril models (Fig. 1). We carried out computational modeling studies to determine the potential binding sites, binding interactions, and amino acids involved in binding and promoting Aβ42 aggregation. The 3D structure coordinates of the VQIVYK peptide and Aβ42 (Fig. 1) were used for molecular docking and molecular dynamics studies [30, 50]. These computational studies provide new insights on the ability of the VQIVYK peptide in promoting Aβ42 aggregation and its application in studying the molecular mechanisms of Aβ42 aggregation process and discovering novel diagnostics and therapeutics.

Materials The computational simulations were carried out using the Discovery Studio 2020 Client—Structure Based Design (SBD) software, version v20.1.0.19295 from Dassault Systemes Biovia Corp, France. The PC used to run the simulation was a Dell Optiplex

54

Praveen P. N. Rao et al.

Fig. 1 (a) Structure of the capped VQIVYK peptide. (b) Ribbon structures of the S-shaped Aβ42 oligomer and fibril models (pdb id: 5KK3). The N- and C-termini are color coded with blue and red, respectively

3040 PC with an Intel(R) Core (TM) i5-6500 CPU@ 3.2 GHz processor. The Discovery Studio (DS) SBD suite was used to prepare 3D structures of the capped VQIVYK, Aβ42 oligomer, and fibril models, to perform molecular docking and molecular dynamics (MD), and to generate topological surfaces. The molecular docking studies were carried out using either LibDock or CDOCKER algorithms. The LibDock algorithm samples the conformational space of the ligands based on shape matching to find polar and apolar hotspots or ligand binding sites on the protein [51]. A scoring function, LibDock score is used to rank the ligand binding poses, whereas the CDOCKER is a CHARMm-based docking algorithm which carries out flexible ligand-based molecular docking. It is a grid based computational algorithm that incorporates molecular dynamics (MD) for docking. The docked ligand poses obtained were ranked using the energy-based function CDOCKER Energy and CDOCKER Interaction Energy [52]. The solvent-accessible surface area (SASA) of the receptor surfaces of both Aβ42 oligomer and fibrils was calculated using the View Interactions command under the Receptor-Ligand

Understanding Aβ42 Oligomer and Fibril Binding

55

Interactions module in DS. The MD simulations of the capped VQIVYK docked to Aβ42 oligomer and fibril models were carried out, using the nanoscale molecular dynamics (NAMD) package in DS. The NAMD is a versatile algorithm that is used to carry out parallel MD simulations of small, medium, and large biomolecular systems [53–55].

3

Methods

3.1 Preparation of the Capped VQIVYK Peptide

The 3D structural coordinates of uncapped VQIVYK were extracted from the solved X-ray crystal structure (pdb id 3OVL) [30]. The C- and N-terminals of the VQIVYK peptide were capped with the amide and acetyl groups, respectively, using the Build and Edit Protein command under the Macromolecules module in DS. The capped VQIVYK peptide was prepared for molecular docking using the Prepare Protein command under the Macromolecules command in DS using the CHARMm force field at pH 7.4, a dielectric constant of 10, ionic strength of 0.145 M NaCl, and an energy cutoff of 0.9 kcal/mol. This 3D model was used to carry out molecular docking studies (Fig. 1a).

3.2 Preparation of Aβ42 Oligomer Model

The 3D structural coordinates of Aβ42 fibril (pdb id: 5KK3) were used to build the oligomer model of Aβ42 for molecular docking. This 3D structure was solved using magic angle spinning (MAS) nuclear magnetic resonance (NMR) spectroscopy [50]. The Aβ42 fibril conformer 5KK3_model_1, was used to build the oligomer model. The oligomer model was created by extracting the Aβ42 pentamer assembly, consisting of five β-strands that form the S-shaped β-sheet assembly (Fig. 2a). The pentamer assembly was used for docking as previous studies have shown that Aβ42 exists as either pentamers or hexamers in solution [56, 57]. The Aβ42 pentamer assembly was prepared for molecular docking study using the Prepare proteins command, under the Macromolecules module in DS that considers protonation sites and adds hydrogen atoms to the protein structure using the CHARMm force field at pH 7.4, a protein dielectric constant of 10, ionic strength of 0.145 M NaCl, and an energy cutoff of 0.9 kcal/mol. In the next step, a 20 Å radius binding site sphere was created centered around Leu17 from the central Aβ42 strand (strand 3) in the pentamer assembly (Fig. 2a) using the Define and Edit Binding Site command under the Receptor-Ligand Interactions module in DS. This assembly was used to carry out molecular docking studies.

56

Praveen P. N. Rao et al.

Fig. 2 (a) Binding site sphere in the Aβ42 oligomer (pentamer) model. (b) Docked complex of the capped VQIVYK peptide in the Aβ42 oligomer (pentamer) model. (c) 2D interaction map of the capped VQIVYK peptide docked in the Aβ42 oligomer (pentamer) model. The docked capped VQIVYK peptide is shown as stick cartoon 3.3 Molecular Docking of the Capped VQIVYK in the Aβ42 Oligomer Model

The capped VQIVYK peptide was docked on to the prepared Aβ42 pentamer assembly (peptide/Aβ molar ratio 1:1), possessing a 20 Å binding site sphere, using the LibDock algorithm under the Receptor-Ligand Interactions module in DS. The docking simulation parameters considered include 100 polar/apolar hotspots in the Aβ42 pentamer assembly, a docking tolerance of 0.25 Å, and a distance dependent dielectric constant. The CHARMm force field was used for the docking simulation. A maximum of 255 low-energy conformations of the ligand, capped VQIVYK, was investigated for docking (FAST conformation method). The binding poses obtained were further energy minimized using the Smart minimizer command (1000 steps, RMS gradient of 0.001 kcal/mol) available in the LibDock algorithm. The LibDock score was used to identify the top ranked binding pose of the capped VQIVYK in the Aβ42 pentamer assembly, and the binding interactions were evaluated by considering various polar and nonpolar contacts.

Understanding Aβ42 Oligomer and Fibril Binding

57

The top binding pose obtained (LibDock score = 157.18) shows that the capped VQIVYK was oriented perpendicular to the fiber axis on the surface of the S-shaped Aβ42 pentamer assembly in a region comprising of His13 to Leu17 at the N-terminus (Fig. 2b). The capped VQIVYK was oriented in an extended conformation and was in contact with the N-terminal solvent-exposed region of the Aβ42 pentamer assembly. These contact points were dominated by polar interactions. The capped amide of VQIVYKCONH2 undergoes hydrogen-bonding interactions with the amide side chain of Gln15 (strand 4, distance = 1.6 Å). The charged lysine amine side chain of the capped VQIVYK undergoes two hydrogenbonding interactions with the amide side chain of Gln15 (strand 2, distance ~ 1.6 Å). The aromatic ring of tyrosine of VQIVYK was in hydrophobic contact with Leu34 side chains (strands 4 and 5, distance < 5.0 Å) at the C-terminus, while the tyrosine hydroxyl group forms a hydrogen bond with the side chain amide of Gln15 (strand 4, distance = 2.1 Å). Interestingly the hydrophobic valine and isoleucine side chains of VQIVYK are not in contact with the Aβ42 surface and instead were exposed to the solvent (Fig. 2b, c). It should be noted that the docking studies were carried out in vacuum in the absence of solvent. Subheading 3.6 describes the MD simulation of capped VQIVYK–Aβ42 oligomer complex in explicit water. The glutamine and valine side chains of VQIVYK were in a polar region and form hydrogen bonds with Gln15 side chains (strands 2 and 3, distance < 2.1 Å). These studies show that the terminal regions of the capped VQIVYK peptide were in contact with the Aβ42 pentamer surface and that binding of the capped VQIVYK reduces the solvent exposure of the polar N-terminal region of the Aβ42 pentamer assembly (Fig. 2b). Furthermore, the interaction between the tyrosine in VQIVYK with Leu34 in Aβ42 also helps reduce the solvent exposure of the hydrophobic groove on the oligomer surface and has a stabilizing effect on the capped VQIVYK-Aβ42 oligomer complex. 3.4 Preparation of Aβ42 Fibril Model and Molecular Docking of the Capped VQIVYK Peptide in the Aβ42 Fibril Model

The cross-β-sheet Aβ42 fibril assembly model was built for molecular docking study by extracting individual Aβ42 strands from the solved structure (pdb id: 5KK3, conformer 5KK3_model_1) [50]. A cross-β-sheet fibril model consisting of a total of ten (decamer) Aβ42 strands was created with each oligomer unit consisting of Aβ42 pentamer assembly (Fig. 3a). The Aβ42 fibril assembly was prepared for molecular docking study using the Prepare proteins command, under the Macromolecules module in DS, that considers protonation sites and adds hydrogen atoms to the protein structure using the CHARMm force field at pH 7.4, a protein dielectric constant of 10, ionic strength of 0.145 M NaCl, and an energy cutoff of 0.9 kcal/mol. Interestingly, unlike the Aβ42 oligomer model, the Aβ42 fibril model reveals the presence of two symmetrical ligand binding pockets on the surface of individual oligomer

58

Praveen P. N. Rao et al.

Fig. 3 (a) Binding site sphere in the Aβ42 fibril model. (b) Two molecules of the capped VQIVYK peptide docked in the Aβ42 fibril model. (c) 2D interaction map of the capped VQIVYK peptide docked in the Aβ42 fibril model. The docked capped VQIVYK peptide is shown as stick cartoon

units, spanning the region between Glu11 and Glu22, closer to the N-terminus (Fig. 3a). This observation suggests the intriguing possibility that two ligand molecules interact with Aβ42 simultaneously at the two binding sites (peptide/Aβ molar ratio 2:1). Therefore, in order to understand the mechanism by which the capped VQIVYK peptide promotes Aβ42 aggregation, the molecular docking studies were carried out using two molecules of the capped VQIVYK peptide in the Aβ42 fibril model (peptide/Aβ molar ratio 2:1). Initially a 20 Å binding site sphere was created in the Aβ42 fibril model, by selecting the amino acids Glu11 and Glu22 from the central strand (strand 3) in one of the oligomer units (Fig. 3a) with one molecule of the capped VQIVYK peptide docked using the CDOCKER algorithm. The CDOCKER algorithm is available under the Receptor-Ligand Interactions module in

Understanding Aβ42 Oligomer and Fibril Binding

59

DS and is based on a simulated annealing protocol. The docking parameters include 2000 heating steps, 700 K target temperature, 5000 cooling steps with a target temperature of 300 K and CHARMm force field for docking. The top ten ligand poses obtained were ranked by evaluating the CDOCKER and CDOCKER interaction energies between the capped VQIVYK peptide and the Aβ42 fibril model. Furthermore, polar and nonpolar interactions between the peptide and the amino acids in Aβ42 fibril model were studied. In the next step, the top ranked binding pose of capped the VQIVYK peptide docked in the Aβ42 fibril model was used to create another 20 Å binding site sphere in the other oligomer unit, to dock a second molecule of the capped VQIVYK peptide. The amino acids Glu11 and Glu22 from the central strand (strand 3) in this oligomer unit were used to create the second ligand binding site in the Aβ42 fibril model. Docking of the second molecule of capped VQIVYK peptide was carried out using the CDOCKER algorithm as described before, and the top ranked docking poses were analyzed. Figure 3b shows two molecules of the capped VQIVYK peptide docked at the two symmetric binding pockets in the Aβ42 fibril model (peptide/Aβ molar ratio 2:1). The top two binding poses obtained were analyzed (Pose 1, CDOCKER Energy = -119.41 kcal/mol, CDOCKER Interaction Energy = -94.67 kcal/mol; Pose 2, CDOCKER Energy = 115.68 kcal/mol, CDOCKER Interaction Energy = 96.72 kcal/mol), which showed that the capped VQIVYK peptide is flexible and can exhibit two different binding modes in the binding pockets of Aβ42 fibril model. Strikingly, the binding pockets in the Aβ42 fibril model resembles the traditional convex active site pockets, as commonly seen in enzymes, whereas the binding site present in the Aβ42 oligomer model is rigid and flat and has a narrow groove. The contrast between the binding sites in Aβ42 fibril and oligomer models suggests that the flexible, capped VQIVYK peptide probably binds the Aβ42 fibril assemblies with higher affinity. The top docked pose of the capped VQIVYK underwent a number of interactions in the binding pocket lined by Glu11Glu22 of Aβ42 strands (Fig. 3b, c). The lysine side chain of the capped VQIVYK forms two salt bridges with Glu22 carboxylate (strand 3, distance = 1.8 and 2.4 Å, respectively). The carbonyl group of the tyrosine in VQIVYK forms a hydrogen bond with Lys16 of Aβ42 (strand 2, distance = 2.5 Å). The carbonyl group of valine in VQIVYK undergoes hydrogen-bonding interaction with the side chain of Lys16 of Aβ42 (strand 3, distance = 2.4 Å) and is also in contact with Val18 of Aβ42 (strand 3, distance < 5 Å) via hydrophobic interactions. The carbonyl group of the isoleucine in VQIVYK is in contact with the charged amine side chain of Lys16 of Aβ42 (strand 3, distance = 1.7 Å). The glutamine amide side chain of VQIVYK formed multiple hydrogen-bonding interactions with the side chains of Glu11 and Lys16 of Aβ42 (strands 4 and

60

Praveen P. N. Rao et al.

Table 1 Comparative analysis of the docked poses of the top ranked capped VQIVYK peptide bound to Aβ42 assemblies using scoring functions Scoring function

Peptide-Aβ42 oligomer complex

Peptide-Aβ42 fibril complex

LigScore1

6.87

6.41

LigScore2

6.89

6.21

–PLP1

104.99

77.74

–PLP2

100.97

73.26

–PMF

50.69

161.57

–PMF04

32.59

119.32

LigScore, PLP (Piecewise Linear Potential), PMF (Potential Mean Force) and PMF04 (Potential Mean Force 04) were calculated using the Receptor-Ligand Interactions module in Discovery Studio Client—Structure Based Design software (Dassault Systemes Biovia Corp, France)

5, distance = 2.2–2.5 Å). These investigations show that two molecules of the capped VQIVYK peptide are able to bind in the two symmetric binding pockets present in the Aβ42 fibril model. This study also suggests that the capped VQIVYK peptide exhibits better binding in the Aβ42 fibril model compared to that in the Aβ42 oligomer model. Furthermore, the docked poses of the capped VQIVYK peptide in the oligomer and fibril models were ranked together using various empirical and knowledge-based scoring functions and compared (Table 1). It should be noted that the capped VQIVYK peptide-Aβ42 fibril complex showed higher scores for PMF and PMF04 functions compared to the capped VQIVYK peptide-Aβ42 oligomer complex. These two scoring functions capture solvation and entropic effect of the ligandprotein complex [58], which again shows the favorable interaction of the capped VQIVYK peptide in the Aβ42 fibril binding pockets. In order to understand the differences in the capped VQIVYK binding to either the Aβ42 oligomer or Aβ42 fibril models, we calculated the SASA (Fig. 4a, b). The results show that the flat and rigid groove of Aβ42 oligomer is more exposed to solvent (surface area = 523.6 Å2) as compared to the binding pocket present in the Aβ42 fibril model (surface area = 519.8 Å2). It is noted that the differences in the SASA between the Aβ42 oligomer and fibril models are not significant to draw definite conclusions. In contrast, the binding mode of capped VQIVYK in the Aβ42 oligomer model is unable to reduce or prevent the solvent exposure of the N-terminal region of Aβ42 oligomer. This can have a destabilizing effect on the ligand-Aβ42 oligomer complex, whereas the binding of the capped VQIVYK in the Aβ42 fibril model essentially blocks the solvent accessibility of the N-terminal region which can stabilize the ligand-Aβ42 oligomer complex. These results further suggests that

Understanding Aβ42 Oligomer and Fibril Binding

61

Fig. 4 (a) The solvent-accessible surface area (SASA) in the Aβ42 oligomer model. (b) The solvent-accessible surface area (SASA) in the Aβ42 fibril model. Blue regions represent solvent-accessible surface, and the green regions represent Aβ42 regions not exposed to the solvent. The docked capped VQIVYK peptide is shown as stick cartoon

the capped VQIVYK exhibits better binding in the Aβ42 fibril model and can stabilize the fibril assembly indicating the ability of the capped VQIVYK peptide in promoting Aβ42 aggregation. 3.5 Preparation of the Docked Complex of Capped the VQIVYK Peptide and Aβ42 Oligomer Model for MD Simulation

The top pose of the capped VQIVYK peptide in complex with the Aβ42 oligomer model was soaked with TIP3P water containing 3574 water molecules under explicit boundary conditions (orthorhombic cell shape and minimum distance of 7 Å from the boundary) and 0.145 M NaCl salt concentration (14 sodium and 9 chloride atoms), using the Solvate command under the

62

Praveen P. N. Rao et al.

Fig. 5 (a) The solvated (TIP3P water model), capped VQIVYK peptide-Aβ42 oligomer docked complex used in MD simulation. (b) The solvated (TIP3P water model), capped VQIVYK peptide-Aβ42 fibril docked complex used in MD simulation. The capped VQIVYK peptides are shown as stick cartoon

Simulation module in DS. The solvated complex was further energy minimized using the Smart minimizer algorithm for 5000 steps (0.1 kcal/mol RMS gradient) using CHARMm force field (Fig. 5a). A similar protocol was also used to prepare the unliganded Aβ42 oligomer model for MD simulation. 3.6 MD Simulation of the Docked Complex of the Capped VQIVYK Peptide and Aβ42 Oligomer Model

The energy minimized, solvated, capped VQIVYK peptide-Aβ42 oligomer complex was subjected to 30 ps of heating/cooling cycle using a time step of 2 fs; target temperature of 300 K; a lower and higher nonbond cutoff distance of 10 Å and 12 Å, respectively; and spherical cutoff for electrostatistics, under SHAKE constraints and CHARMm force field using the Simulation module in DS. The .rst file obtained after the heating/cooling was used to carry out a 300 ps equilibration using a time step of 2 fs; target temperature of 300 K; a lower and higher nonbond cutoff distance of 10 Å and 12 Å, respectively; spherical cutoff for electrostatistics; and Leapfrog Verlet dynamics integrator, under SHAKE constraints and CHARMm force field. Then, a 5 ns production run was carried out using the NAMD program, available under the Simulation module of DS using CHARMm 36 force field. The .rst file obtained

Understanding Aβ42 Oligomer and Fibril Binding

63

Fig. 6 (a) The Cα RMSD fluctuations during the 1–5 ns MD simulation of the unliganded and liganded (docked capped VQIVYK peptide) Aβ42 oligomer model. (b) The RMSF fluctuations of key regions in the unliganded and liganded (docked capped VQIVYK peptide) Aβ42 oligomer complex during the 5 ns MD simulation. (c) The total energy of the unliganded and liganded Aβ42 oligomer model during the 5 ns MD simulation. (d) The electrostatic energy of the unliganded and liganded Aβ42 oligomer model during the 5 ns MD simulation

from the previous equilibration run was used to carry out the production run. A time step of 2 fs, the target temperature of 300 K, a lower and higher nonbond cutoff distance of 10 Å and 12 Å, particle mesh Ewald (PME) function for electrostatistics, and the constant-temperature and constant-pressure (NPT) ensemble parameters were used for the production run. The trajectories obtained after the 5 ns MD simulation for the unliganded and liganded Aβ42 oligomer were analyzed. The Cα root mean square deviation (RMSD) analysis of the unliganded and liganded Aβ42 oligomer shows that the flexibility of Aβ42 pentamer was reduced after binding to the capped VQIVYK peptide (Fig. 6a). The average RMSD difference between the unliganded and liganded Aβ42 pentamer was ~1.3 Å throughout the 5 ns MD simulation. This shows that upon binding, the capped VQIVYK peptide was able to stabilize the Aβ42 pentamer assembly. Furthermore, as observed during the docking study (Fig. 2b, c), the valine and isoleucine side chains of the capped VQIVYK peptide were exposed to solvent and did not undergo any contact with the Aβ42 pentamer assembly during the 5 ns MD simulation. Both the unliganded and liganded Aβ42 pentamer Cα RMSD were stable

64

Praveen P. N. Rao et al.

during the 1–5 ns MD simulation. In the next analysis, we calculated the root mean square fluctuations (RMSF) to determine the flexibility of specific regions in the Aβ42 pentamer assembly before and after binding of the capped VQIVYK peptide during the 5 ns MD simulation (Fig. 6b). This analysis clearly shows that the ligand binding regions at the N- and C-termini become less flexible in the capped VQIVYK peptide bound Aβ42 pentamer complex. For example, the N-terminal region spanning the amino acids His13, His14, and Gln15 (from strands 1–5) become less flexible in the ligand bound Aβ42 pentamer, with a RMSF that is ~0.83 Å lower compared to that in the unliganded Aβ42 pentamer. It was also interesting to note that the nonpolar amino acids at the C-terminus, both Leu34 and Met35 (from strands 2–5), exhibited less flexibility after ligand binding, RMSF ~0.81 Å lower than that in the unliganded Aβ42 pentamer. This is expected since the capped VQIVYK peptide also undergoes hydrophobic contact with the Leu34 in the Aβ42 pentamer model. The other regions of the Aβ42 pentamer containing polar amino acids Glu11, Glu22, and Asp23 were also flexible (Fig. 6b). This is not surprising as these amino acid chains were exposed to the solvent. The comparison of the total energy of the unliganded and ligand bound Aβ42 pentamer during the 5 ns MD simulation also confirms that the binding of the capped VQIVYK peptide at the N- and C-terminal interface of the Aβ42 oligomer model has a stabilizing effect on the complex and the total energy of the system was lowered (Fig. 6c). Furthermore, the MD simulation shows that in the unliganded and liganded Aβ42 oligomer model, electrostatic interactions represent major type of forces involved (Fig. 6d). In summary, these computational studies suggest that the capped VQIVYK peptide can bind at the solvent-exposed N-terminal region, which can have a stabilizing effect on the oligomer complex and the potential to prevent and inhibit further oligomer elongation to form fibrils. One should note that the capped VQIVYK peptide is flexible and can exhibit multiple modes of binding. Yet, due to the S shape of the Aβ42 oligomer, it can bind only at one binding site lined by the N(His13, His14, and Gln15) and C-terminal (Leu34, Met35) amino acids. MD studies also show that the binding of the capped VQIVYK peptide can lead to perturbations in the protein backbone in the Aβ42 oligomer model (Fig. 6a), which suggests that ligand binding would change the global conformation of the pentamer assembly and reduce or prevent further oligomer elongation. However, it should be noted that the binding of the capped VQIVYK peptide is reversible and may not be strong in the Aβ42 oligomer assembly, as the binding site surface is flat and rigid (Fig. 2b).

Understanding Aβ42 Oligomer and Fibril Binding

65

3.7 Preparation of the Docked Complex of the Capped VQIVYK Peptides and Aβ42 Fibril Model for MD Simulation

The top poses of two capped VQIVYK peptides in complex with the Aβ42 fibril model were soaked with TIP3P water containing 5283 water molecules under explicit boundary conditions (orthorhombic cell shape and minimum distance of 7 Å from the boundary) and 0.145 M NaCl salt concentration (24 sodium and 14 chloride atoms), using the Solvate command under the Simulation module in DS. The solvated complex was further energy minimized using the Smart minimizer algorithm for 5000 steps (0.1 kcal/mol RMS gradient) and the CHARMm force field (Fig. 5b). A similar protocol was also used to prepare the unliganded Aβ42 fibril model for MD simulation.

3.8 MD Simulation of the Docked Complex of the Capped VQIVYK Peptides and Aβ42 Fibril Model

The energy minimized, solvated, capped VQIVYK peptide-Aβ42 fibril complex (containing two molecules of the capped VQIVYK peptide) was subjected to 30 ps of heating/cooling cycle, using a time step of 2 fs; target temperature of 300 K; a lower and higher nonbond cutoff distance of 10 Å and 12 Å, respectively; and spherical cutoff for electrostatistics, under SHAKE constraints and CHARMm force field. These steps were carried out using the Simulation module in DS. The .rst file obtained after the heating/cooling was used to carry out a 300 ps equilibration using a time step of 2 fs; target temperature of 300 K; a lower and higher nonbond cutoff distance of 10 Å and 12 Å, respectively; spherical cutoff for electrostatistics; and Leapfrog Verlet dynamics integrator, under SHAKE constraints and CHARMm force field. The 5 ns production run was carried out using the NAMD program, available under the Simulation module of DS using CHARMm 36 force field. The .rst file obtained, from the previous equilibration run, was used to carry out the production run. A time step of 2 fs, target temperature of 300 K, a lower and higher nonbond cutoff distance of 10 Å and 12 Å, particle mesh Ewald (PME) function for electrostatistics, and the constant-temperature, constant-pressure (NPT) ensemble parameters were used for the production run. The MD simulation trajectories of the unliganded and liganded Aβ42 fibril were analyzed by comparing the Cα RMSD and RMSF (Fig. 7a, b). Interestingly, the backbones of both the unliganded and the liganded Aβ42 fibril did not exhibit any significant difference in their RMSD values during the 5 ns simulation (Fig. 7a). This shows the rigid and stable nature of the tightly formed Aβ42 fibril assembly, which is not perturbed after the binding of two molecules of the capped VQIVYK peptide. This observation also indicates that upon ligand binding, there is no significant change in the global conformation of the fibril assembly. In contrast, calculating the RMSF values between the liganded and unliganded Aβ42 fibril does reveal perturbations in some regions of the fibril assembly (Fig. 7b). For example, the N-terminal amino acids Glu11, Val12, and His13 of strands 1, 8, and 10, the turn region consisting

66

Praveen P. N. Rao et al.

Fig. 7 (a) The Cα RMSD fluctuations during the 1–5 ns MD simulation of the unliganded and liganded (docked capped VQIVYK peptide) Aβ42 fibril model. (b) The RMSF fluctuations of key regions in the unliganded and liganded (two molecules of the docked capped VQIVYK peptide) Aβ42 fibril complex during the 5 ns MD simulation. (c) The total energy of the unliganded and liganded Aβ42 fibril model during the 5 ns MD simulation. (d) The electrostatic energy of the unliganded and liganded Aβ42 fibril model during the 5 ns MD simulation

of Glu22, Asp23, Val24, Gly25, Ser26, Asn27, Lys28, and Gly29 of strands 1–10 and the C-terminal region consisting of Ala30, Val36, Gly37, Gly38, Val39, Val40, and Ala42 of strands 1–6, exhibited flexibility (range 1.3–3.0 Å), upon ligand binding at the symmetric binding pockets. It was worth noting that ligand binding leads to increased fluctuations in all the amino acids in strand 1 (~5.0 Å). In summary, MD simulation shows that the capped VQIVYK peptide primarily interacts with the amino acid side chains of the binding pockets and doesn’t cause changes in the protein backbone conformation, as the Cα RMSD was conserved in the liganded and unliganded Aβ42 fibril assembly (Fig. 7a). The total energy of the liganded and unliganded fibril complex (Fig. 7c) shows that binding of the capped VQIVYK peptide leads to a further reduction in the total energy of the complex compared to the unliganded Aβ42 fibril assembly, suggesting that ligand binding has a stabilizing effect on the Aβ42 fibril assembly. Furthermore, MD simulation shows that the electrostatic energy was the major contributor in the tightly packed unliganded Aβ42 fibril assembly and that upon ligand binding, there was an increased contribution to the total energy of the complex via electrostatic

Understanding Aβ42 Oligomer and Fibril Binding

67

interactions (Fig. 7d). This is expected as the capped VQIVYK peptide interacts with charged, polar amino acids Glu11, Lys16, and Glu22 (strands 2, 3, 4, and 5) in the fibril binding pockets (Fig. 3b, c). These MD studies of the Aβ42 fibril assembly demonstrate that, as such, the Aβ42 fibril is very stable and ligand binding doesn’t cause significant perturbations in the tightly packed β-sheet assembly. The MD trajectories also show that with bound ligand, the solvent-exposed binding sites of Aβ42 fibril are shielded from bulk water which is the driving force for ligand binding. Furthermore, the MD trajectories shows that local interactions of the capped VQIVYK peptide with the amino acid side chains, lining the binding pockets (Fig. 7b), play a major role in the peptide-Aβ42 binding interaction, while the tightly packed protein backbone is not perturbed (Fig. 7a). These MD investigations show that capped VQIVYK peptide binding stabilizes the Aβ42 fibril assembly via local interactions at the binding pockets and promotes Aβ42 fibrillogenesis as previously observed [49].

4

Conclusions This chapter reports computational methods to investigate the aggregation mechanisms of Aβ42 by carrying out systematic investigations in the oligomer and fibril models. A known promoter of Aβ42 fibrillogenesis, the capped VQIVYK hexapeptide, was used to study its interactions with the Aβ42 pentamer as a representative model of Aβ42 oligomer and a decamer, as a representative model of Aβ42 fibril. Molecular docking study shows that the Aβ42 oligomer exhibits one binding site on its surface and that this site was the solvent-exposed N-terminal region, whereas the Aβ42 fibril model presents two symmetric binding sites, where two molecules of the capped VQIVYK can bind simultaneously and stabilize the fibril assembly. Moreover, the binding of the capped VQIVYK peptide blocks the solvent exposure of this binding site in the fibril model which is not the case in the oligomer model. This computational observation supports the experimental evidence that demonstrates the ability of the capped VQIVYK peptide in promoting Aβ42 fibrillogenesis [49]. Molecular dynamics studies demonstrate that binding of the capped VQIVYK peptide in the Aβ42 oligomer model was stable and that ligand binding leads to significant perturbation in the backbone assembly. These observations indicate that the conformational change in the oligomer assembly is the key factor in preventing further elongation of the fibril and inhibition of Aβ42 fibrillogenesis. In contrast, binding of the capped VQIVYK peptides in the Aβ42 fibril assembly did not perturb the protein backbone conformation significantly, which in turn did not change the global conformation of the Aβ42 fibril assembly. This suggests that ligand binding doesn’t affect the β-sheet assembly in the Aβ42

68

Praveen P. N. Rao et al.

fibril which might contribute to Aβ42 fibrillogenesis, instead of fibril inhibition or disaggregation [49]. These studies suggest that (i) appropriate models of Aβ42 should be evaluated by computational techniques to understand the mechanisms of Aβ42 aggregation and (ii) novel chemical tools can be designed as promoters of Aβ42 aggregation. The in silico experiments described here will assist in the discovery of novel diagnostics, disease mechanisms, and therapies for AD.

5

Notes 1. The 3D structure of Aβ42 (PDB ID 5KK3) was used to carry out molecular docking and molecular dynamics studies. This 3D structure was solved using the solid-state magic angle spinning (MAS) NMR method [50]. 2. In the Aβ42 file (pdb id: 5KK3) used for molecular docking and molecular dynamics, each strand consists of amino acids 11–42. The first ten amino acids (residue 1–10) at the N-terminus are missing and were not resolved due to their disordered nature. 3. While carrying out the molecular docking studies of the capped VQIVYK peptide in the Aβ42 oligomer model, initial studies using the CDOCKER algorithm did not yield any poses. Therefore, LibDock algorithm was used. Similarly, while docking the capped VQIVYK peptide in the Aβ42 fibril model using the LibDock algorithm, no poses were obtained. Therefore, the CDOCKER algorithm was used. This demonstrates the differences in the shape, size, and surfaces of binding site between the Aβ42 oligomer and fibril models and the limitations of either docking algorithms. 4. We used five strands (pentamer) of Aβ42 as the oligomer model and ten strands (decamer) of Aβ42 as the fibril model to conduct molecular docking and molecular dynamics experiments. The number of strands used in the simulation experiments can be increased depending on the available computational power, research questions, hypothesis considered, and the type of computational software used.

Acknowledgments The PPNR would like to thank the School of Pharmacy and the Office of Research (NSERC-RIF and CIHR-RIF), University of Waterloo, for the financial support. RH was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) grant (RGPIN-2020-05066).

Understanding Aβ42 Oligomer and Fibril Binding

69

References 1. Masters CL, Bateman R, Blennow K, Rowe CC, Sperling RA, Cummings JL (2015) Alzheimer’s disease. Nat Rev Dis Primers 1:15056 2. Knopman DS, Amieva H, Petersen RC, Chetelat G, Holtzman DM, Hyman BT, Nixon RA, Jones DT (2021) Alzheimer’s disease. Nat Rev Dis Primers 7:33 3. Nichols E, Vos T (2021) Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the global burden of disease study 2019. Alzheimers Dement 17:e051496 4. (2022) 2022 Alzheimer’s disease facts and figures. Alzheimers Dement 18:700–789 5. Monfared AAT, Byrnes MJ, White LA, Zhang Q (2022) Alzheimer’disease: epidemiology and clinical progression. Neurol Ther 11:553– 569 6. Cummings J, Lee G, Zhong K, Fonesca J, Taghva K (2021) Alzheimer’s disease drug development pipeline: 2021. Alzheimers Dement (N Y) 7:e12179 7. Bokhoven PV, Wilde AD, Vermunt L, Leferink PS, Heetveld S, Cummings J, Scheltens P, Vijverberg EGB (2021) The Alzheimer’s disease drug development landscape. Alzheimers Res Ther 13:186 8. Yiannopoulou KG, Papageorgiou SG (2020) Current and future treatments in Alzheimer’s disease: an update. J Cent Nerv Syst Dis 12:1– 12 9. Frozza RL, Lourenco MV, Felice FGD (2018) Challenges for Alzheimer’s disease therapy: insights from novel mechanisms beyond memory defects. Front Neurosci 12:37 10. Sevigny J, Chiao P, Bussiere T, Weinreb PH, Williams L, Maier M, Dunstan R, Salloway S, Chen T, Ling Y, O’Gorman J, Qian F, Arastu M, Li M, Chollate S, Brennan MS, Quintero-Monzon O, Scannevin RH, Arnold HM, Engber T, Rhodes K, Ferrero J, Hang Y, Mikulskis A, Grimm J, Hock C, Nitsch RM, Sandrock A (2016) The antibody aducanumab reduces Aβ plaques in Alzheimer’s disease. Nature 537:50–56 11. Cummings J, Aisen P, Apostolova LG, Atri A, Salloway S, Weiner M (2021) Aducanumab: appropriate use and recommendation. J Prev Alzheimers Dis 8:398–410 12. Fillit H, Green A (2021) Aducanumab and the FDA – where are we now? Nat Rev Neurol 17: 129–130 13. Hardy JA, Higgins GA (1992) Alzheimer’s disease: the amyloid cascade hypothesis. Science 256:184–185

14. Selkoe DJ, Hardy J (2016) The amyloid hypothesis of Alzheimer’s disease at 25 years. EMBO Mol Med 8:595–608 15. Haas C, Selkoe D (2022) If amyloid drives Alzheimer disease, why have anti-amyloid therapies not yet slowed cognitive decline? PLoS Biol 20:e3001694 16. Karran E, Stropper BD (2022) The amyloid hypothesis in Alzheimer disease: new insights from new therapeutics. Nat Rev Drug Discov 21:306–318 17. O’Brien RJ, Wong PC (2011) Amyloid precursor protein processing and Alzheimer’s disease. Annu Rev Neurosci 34:185–204 18. Bitan G, Kirkitadze MD, Lomakin A, Teplow DB (2003) Amyloid β-protein (Aβ) assembly: Aβ40 and Aβ42 oligomerize through distinct pathways. Proc Natl Acad Sci U S A 100:330– 335 19. Yan Y, Wang C (2006) Abeta42 is more rigid than abeta40 at the C terminus: implications for a beta aggregation and toxicity. J Mol Biol 364:853–862 20. Finder VH, Vodopivec I, Nitsch RM, Glockhuber R (2010) The recombinant amyloid-beta peptide Aβ1-42 aggregates faster and is more neurotoxic than synthetic Aβ1-42. J Mol Biol 396:9–18 21. Nirmalraj PN, List J, Battacharya S, Howe G, Xu L, Thompson D, Mayer M (2020) Complete aggregation pathway of amyloid β (1-40) and (1-42) resolved on atomically clean interface. Sci Adv 6:eaa6014 22. Linse S (2019) Mechanism of amyloid protein aggregation and the role of inhibitors. Pure Appl Chem 91:211–229 23. Willbold D, Strodel B, Schroder GF, Hoyer W, Heise H (2021) Amyloid-type protein aggregation and prion-like properties of amyloids. Chem Rev 121:8285–8307 24. Almeida ZL, Brito RMM (2020) Structure and aggregation mechanisms in amyloids. Molecules 25:1195 25. Ghosh AK, Osswald HL (2014) BACE1 (β-secretase) inhibitors for the treatment of Alzheimer’s disease. Chem Soc Rev 43:6765– 6813 26. Satir TM, Agholme L, Karlsson A, Karlsson M, Karila P, Illes S, Bergstrom P, Zetterberg H (2020) Partial reduction of amyloid β production by β-secretase inhibitors does not decrease synaptic transmission. Alzheimers Res Ther 12: 63

70

Praveen P. N. Rao et al.

27. Zhao J, Liu X, Xia W, Zhang Y, Wang C (2020) Targeting amyloidogenic processing of APP in Alzheimer’s disease. Front Mol Neurosci 4:137 28. Gu K, Li Q, Lin H, Zhu J, Mo J, He S, Jiang X, Sun H (2017) Gamma secretase inhibitors: a patent review (2013 – 2015). Expert Opin Ther Pat 27:851–866 29. Belluti F, Rampa A, Gobbi S, Bisi A (2013) Small-molecule inhibitors/modulators of amyloid-β peptide aggregation and toxicity for the treatment of Alzheimer’s disease: a patent review (2010–2012). Expert Opin Ther Pat 23:581–596 30. Landau M, Sawaya MR, Faull KF, Laganowsky A, Jiang L, Sievers SA, Liu J, Barrio JR, Eisenberg D (2011) Towards a pharmacophore for amyloid. PLoS Biol 9:e1001080 31. Jiang L, Liu C, Leibly D, Landau M, Zhao M, Hughes MP, Eisenberg DS (2013) Structurebased discovery of fiber-binding compounds that reduce the cytotoxicity of amyloid beta. Elife 2:e00857 32. Mohamed T, Shakeri A, Rao PPN (2016) Amyloid cascade in Alzheimer’s disease: recent advances in medicinal chemistry. Eur J Med Chem 113:258–272 33. Bu XL, Rao PPN, Wang YJ (2016) Antiamyloid aggregation activity of natural compounds: implications for Alzheimer’s drug discovery. Mol Neurobiol 53:3565–3575 34. Habchi J, Chai S, Limbocker L, Mannini B, Ahn M, Perni M, Hansson O, Arosio P, Kumita JR, Challa PK, Cohen SIA, Linse S, Dobson CM, Knowles TPJ, Vendruscolo M (2017) Systematic development of small molecules to inhibit specific microscopic steps of Aβ42 aggregation in Alzheimer’s disease. Proc Natl Acad Sci U S A 114:E200–E208 35. Jokar S, Khazaei S, Behnammanesh H, Shamloo A, Erfani M, Beiki D, Bavi O (2019) Recent advances in the design and applications of amyloid-β peptide aggregation inhibitors for Alzheimer’s disease therapy. Biophys Rev 11: 901–925 36. Henning-Knechtel A, Kumar S, Wallin C, Krol K, Warmlander SKTS, Jarvet J, Esposito G, Kirmiziatin S, Graslund A, Hamilton AD, Magzoub M (2020) Designed cellpenetrating peptide inhibitors of amyloid-beta aggregation and cytotoxicity. Cell Rep Phys Sci 1:100014 37. Young LM, Ashcroft AE, Radford SE (2017) Small molecule probes of protein aggregation. Curr Opin Chem Biol 39:90–99 38. Aliyan A, Cook NP, Marti AA (2019) Interrogating amyloid aggregates using fluorescent probes. Chem Rev 119:11819–11856

39. Landrieu I, Dupre E, Sinnaeve D, Hajjar LE, Smet-Nocca C (2022) Deciphering the structure and formation of amyloids in neurodegenerative diseases with chemical biology tools. Front Chem 10:886382 40. Sarkany Z, Rocha F, Damas AM, MacedoRibeiro S, Martins PM (2019) Chemical kinetic strategies for high-throughput screening of protein aggregation modulators. Chem Asian J 14:500–508 41. Von Bergen M, Friedoff P, Biernat J, Heberle J, Mandelkow EM, Mandelkow E (2000) Assembly of tau protein into Alzheimer paired helical filaments depends on a local sequence motif (306VQIVYK311) forming beta structure. Proc Natl Acad Sci U S A 97:5129–5134 42. Perez M, Santa-Maria I, Tortosa E, Cuadros R, Valle MD, Hernandez F, Moreno FJ, Avila J (2007) The role of the VQIVYK peptide in tau protein phosphorylation. J Neurochem 103: 1447–1460 43. Ganguly P, Do TD, Larini L, LaPointe NE, Sercel AJ, Shade MF, Feinstein SC, Bowers MT, Shea JM (2015) Tau assembly: the dominant role of PHF6 (VQIVYK) in microtubule binding region repeat R3. J Phys Chem B 119: 4582–4593 44. Sawaya MR, Sambashivan S, Nelson R, Ivanova MI, Sievers SA, Apostol MI, Thompson MJ, Balbirnie M, Wiltzius JJW, McFarlane HT, Madesen A, Riekel C, Eisenberg D (2007) Atomic structures of amyloid cross-spines reveal varied steric zippers. Nature 447:453– 457 45. Zheng J, Liu C, Sawaya MR, Vadla B, Khan S, Woods RJ, Eisenberg D, Goux WJ, Nowick JS (2011) Macrocyclic β-sheet peptides that inhibit the aggregation of a tau-proteinderived hexapeptide. J Am Chem Soc 133: 3144–3157 46. Zheng J, Baghkhanian AM, Nowick JS (2013) A hydrophobic surface is essential to inhibit the aggregation of a tau-protein-derived hexapeptide. J Am Chem Soc 135:6846–6852 47. Seidler P, Boyer DR, Rodriguez JA, Sawaya MR, Cascio D, Murray K, Gonen T, Eisenberg DS (2017) Structure-based inhibitors of tau aggregation. Nat Chem 10:170–176 48. Mohamed T (2013) Tau-derived-hexapeptide 306VQIVYK311 aggregation inhibitors: nitrocatechol moiety as a pharmacophore in drug design. ACS Chem Neurosci 18:1559–1570 49. Mohamed T, Gujral SS, Rao PPN (2018) Tau derived hexapeptide AcPHF6 promotes betaamyloid (A) fibrillogenesis. ACS Chem Neurosci 9:773–782

Understanding Aβ42 Oligomer and Fibril Binding 50. Colvin MT, Silvers R, Ni QZ, Sergeyev I, Rosay M, Donovan KJ, Michael B, Wall J, Linse S, Griffin RG (2016) Atomic resolution structure of monomorphic 42 amyloid fibrils. J Am Chem Soc 138:9663–9674 51. Diller DJ, Merz KM (2001) High throughput docking for library design and library prioritization. Proteins 43:113–124 52. Wu G, Roberson DH, Brooks CL III, Vieth M (2003) Detailed analysis of grid-based molecular docking: a case study of CDOCKER- a CHARMm-based MD docking algorithm. J Comput Chem 24:1549–1562 53. Philips JC, Braun R, Wang W, Gumbart J, Tajkhorshid E, Villa E, Chipot C, Skeel RD, Kale L, Schulten K (2005) Scalable molecular dynamics with NAMD. J Comput Chem 26: 1781–1802 54. Philips JC, Hardy DJ, Maia JD, Stone JE, Ribeiro JV, Bernardi RC, Buch R, Fiorin G, Henin J, Jiang W, McGreevy R, Melo MCR, Radak BK, Skeel RD, Singharoy A, Wang Y, Roux B, Aksimentiev A, Luthey-Schulten Z, Kale LV, Schulten K, Chipot C, Tajkhorshid E (2020) Scalable molecular dynamics on CPU

71

and GPU architectures with NAMD. J Chem Phys 153:044130 55. NAMD Scalable Molecular Dynamics, Theoretical and Computational Biophysics Group in the Beckman Institute for Advanced Science and Technology at the University of Illinois at Urbana-Champaign. http://www.ks.uiuc. edu/Research/namd/ 56. Bitan G, Kirkitadze MD, Lomakin A, Vollers SS, Benedek GB, Teplow DB (2003) Models of amyloid seeding in Alzheimer’s disease and scrapie: mechanistic truths and physiological consequences of the time-dependent solubility of amyloid proteins. Proc Acad Natl Sci U S A 100:330–335 57. Wolff M, Haagen BZ, Decker C, Barz B, Schneider M, Biehl R, Radulescu A, Strodel B, Willbold D, Nagel-Steger L (2017) Aβ42 pentamers/hexamers are the smallest detectable oligomers in solution. Sci Rep 7: 2493 58. Xue M, Zheng M, Xiong B, Li Y, Jiang H, Shen J (2010) Knowledge-based scoring functions in drug design. 1. Developing a target-specific method for kinase-ligand interactions. J Chem Inf Model 50:1378–1386

Chapter 3 Recent Advances in Computational Modeling of BACE1 Inhibitors as Anti-Alzheimer Agents Konstantinos D. Papavasileiou, Francesco Dondero, Georgia Melagraki, and Antreas Afantitis Abstract A growing number of people worldwide are being affected by aging-associated neurodegenerative illnesses, the most prevalent of which, Alzheimer’s, is defined by progressive neuronal death and synaptic loss in the human brain and can be brought on by both genetic and environmental risk factors. The beta-site amyloid precursor protein (APP) cleaving enzyme 1 (BACE1) is the major beta secretase for the generation of amyloid-β peptides in the neurons, which – according to the amyloid hypothesis – results in the formation of amyloid plaques. Therefore, in order to avert the accumulation of beta-amyloid and (per the amyloid hypothesis) delay or prevent the progression of Alzheimer’s disease, the creation of BACE1 small-molecule inhibitors consists of one of the principal pharmaceutical routes. Using computer-aided drug design, inhibitors for the BACE1 biomolecular target connected to Alzheimer’s disease have been effectively created. In this chapter, the recent developments in the computational modeling search of novel BACE1 inhibitors are discussed. Key words Alzheimer’s disease, Computer-aided drug design, β-secretase, Computational structurebased design, Molecular docking, Molecular dynamics, Computational ligand-based design, QSAR, Cheminformatics, Machine learning

1

Introduction One of the main causes of dementia in the elderly is Alzheimer’s disease (AD), an irreversible, progressive neurodegenerative brain ailment for which there is presently no therapy [1, 2]. Millions of individuals throughout the world are suffering from AD, which has severe effects on both the patients and their relatives. According to estimations from the Alzheimer’s association, approximately 6.5 million Americans at the age of 65 and above are living with AD today [3], and recent global estimates including persons with AD dementia, prodromal AD, and preclinical AD reported a prevalence of 22% of all people above 50 years [4]. Families, healthcare systems, and society as a whole are severely financially burdened by the

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_3, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

73

74

Konstantinos D. Papavasileiou et al.

care and support of AD patients. Therefore, many research efforts have been made in recent decades to understand the origins of AD so that safe and effective pharmaceutical agents can be developed [1]. Despite the extensive funding and research invested into AD or other forms of dementia, the underlying causes are still unclear, but the evidences point to a combination of risk factors including the person’s lifestyle, genetics, and environmental factors such as exposure to toxins and pollution [5, 6]. Moreover, the lack of effective medical treatments and inadequate diagnostic technologies is expected to further amplify AD’s negative societal impact. Thus, understanding the molecular pathogenesis of AD is essential for the creation of better diagnostic and treatment methods [5]. AD is a neurodegenerative disorder of the central nervous system (CNS) [7, 8], associated with abnormal amyloid-β (Aβ) metabolism [9], hyperphosphorylation of tubulin-associated unit (Tau) [10, 11], oxidative stress [12, 13], reactive glial [14], microglial changes [15], and other pathological abnormalities [7, 8]. Even though substantial improvements in research have led to new understandings of AD’s pathogenesis, the disease’s molecular pathways are complicated and poorly understood [16]. The focus of research for the creation of AD treatments is on the Aβ and Tau pathways that result in amyloid plaques and neurofibrillary tangles (NFTs), respectively [17]. The transmembrane enzyme β-secretase or β-site APP cleaving enzyme I (BACE1) is involved in the amyloidogenic pathway [18], and its inhibition is being regarded as a possible treatment method for the development of AD drugs [19]. BACE1 was first identified in 1999 [20]. It is a membrane-anchored aspartic protease ubiquitously expressed in the neurons, with a higher activity in the Golgi apparatus, trans-Golgi network (TGN), secretory vesicles and endosomes [20–22], and optimal enzymatic activity in an acidic pH [23]. Chromosomal localization studies facilitated the identification of BACE1’s homologue, BACE2 [24]. BACE1 is involved in the rate-limiting step of the cleavage process of the amyloid precursor protein (APP), which results in the generation of the neurotoxic amyloid (Aβ) protein after BACE1 completes its function. Insoluble Aβ aggregates that are formed cause plaque buildup and neurodegeneration (Fig. 1) [25]. Therefore, inhibition of APP proteolysis by BACE1 to decrease the concentration of neurotoxic Aβ peptides is considered one of the most important therapeutic approaches for AD, making BACE1 an alluring target [2, 19, 20, 26]. Effective clinical aspartic protease inhibitors were initially discovered for other therapeutic targets, like the treatment of hypertension and the human immunodeficiency virus (HIV) [27]. This development, coupled with the elucidation of BACE1 first crystal structure in 2000 [28], supplied vital data for using structure-based

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

75

Fig. 1 Representation BACE1 function in the amyloidogenic pathway. BACE1 cleaves APP to form Aβ peptides. (Figure reproduced with permission from Ref. [25])

(SB) drug design and at first appeared to simplify the overall procedure for creation of BACE1 inhibitors. Unfortunately, this expectation was quickly unrealized for several reasons, mainly due to the reduced binding efficacy of the candidate compounds in BACE1’s catalytic pocket, their selectivity toward other aspartic proteases and not meeting the strict requirements for CNS penetration [26]. Despite these obstacles, significant progress has been achieved recently in the development of BACE1 small-molecule inhibitors that exhibit improved pharmacokinetic (PK) and bloodbrain barrier (BBB) permeation profiles. Milestones in the development of small-molecule BACE1 inhibitors [2, 19, 26] many of which reached late stages of clinical trials include the identification of acyl guanidine-based compounds [29], followed by the discovery of aminothiazine- and aminooxazoline-based compounds [30] and the fluorinated iminothiadiazinane dioxide-based compounds best represented by Merck’s verubecestat (MK-8931) [31]. The latter was the most advanced BACE1 inhibitor for a long time before it was discontinued due to an unsatisfactory risk/benefit ratio after two significant phase III trials, a fate followed by other BACE1 inhibitors that showed great potential in early development stages [32]. Because of the high failure rate of drug candidates in clinical trials [33], more extensive efforts in the creation of novel small-molecule inhibitors are necessary [34]. The intricacy of the molecular pathways involved in the course of AD renders the discovery of small-molecule BACE1 inhibitors a difficult and time-consuming process [35]. This is an issue hampering all aspects of AD treatment. For example, aducanumab (branded as Aduhelm)

76

Konstantinos D. Papavasileiou et al.

[36] is the first new drug approved by the US Food and Drug Administration for Alzheimer’s disease in 18 years [37]. In this regard, computer-aided drug design (CADD) techniques are particularly effective tools for the rational selection of hit compounds and hit-to-lead optimization. CADD includes ligandbased design methods like quantitative structure-activity relationships (QSAR), descriptor-based QSAR, and pharmacophore mapping as well as computational structure-based design methods like molecular docking and dynamics [38]. In order to fully comprehend and make predictions on the binding mechanisms and energetics of small-molecule BACE1 inhibitors, CADD harnesses the power of computers, mathematics, and statistical mechanics [39]. Several examples of successful CADD application with respect to AD can be found in literature [40, 41]. In this chapter, the structure and function of BACE1 as a druggable target involved in AD is discussed, along with several recent CADD application examples for the development of BACE1 inhibitors and modulators as anti-Alzheimer agents.

2

BACE1: Structure and Active Site Characteristics More than 300 BACE1 crystal structures with and without an inhibitor in the active site have been resolved [42]. BACE1 comprises 501 amino acids [43] having a large two-lobe catalytic domain, the crystal structure of which (385 residues) was first determined in 2000 (PDB code: 1FKN) [28] and has the typical conserved folding of aspartic proteases [44] (Fig. 2). The catalytic domain is characterized by a substrate binding site, namely, the “cleft,” located centrally between the N-terminal lobe (residues 1–150) and C-terminal lobes (residues 151–385), which features two catalytic aspartate amino acids, namely, Asp32 and Asp228 [45]. A combination of molecular docking and molecular dynamics (MD) simulations was integral in the elucidation of the catalytic dyad protonation state [46, 47]. Like other aspartic proteases, a water molecule bridges the catalytic residues in the BACE1 apo structure [48]. The BACE1 active site can be further subdivided into a total of 11 subsites that offer an increased number of available interactions, mainly with peptidic substrates, as 3 of these subsites are rarely occupied by small molecular inhibitors [42]. The substrate binding site is sensitive to pH changes, which are correlated with conformational changes that affect its activity [49]. The substrate binding “cleft” is shielded by a long, flexible β-hairpin loop (residues 67–77) that is often referred to as the “flap,” which is positioned at the N-terminal lobe and controls substrate access through a conformational change (Fig. 3) [45, 49]. A rich network of hydrogen bonds in the BACE1’s flap region stabilizes a flap-open apo structure conformation [50], and

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

77

Fig. 2 The Homo sapiens three-dimensional BACE1 structure [28]. The N- (cyan) and C- (lime) terminal lobes, the “flap” (blue), the bound ligand (brown), and the catalytic dyad of Asp32/Asp228 (red) are illustrated. (Figure was prepared by means of VMD [142])

Fig. 3 Superimposed BACE1 crystal structures with (green, PDB ID: 3TPP) and without (orange, PDB ID: 3TPJ) a bound inhibitor (grey) reveal the “flap” conformational change. The catalytic aspartate residues are also shown (red)

when a substrate is bound, it assumes a flap-closed or a flap-open conformation, depending on the substrate characteristics [29]. Other important residues involved in substrate binding include Leu30, Tyr71, and Phe108 located near the flap region [42]. The remarkable flexibility exhibited by BACE1 has also been supported by MD simulation studies [51].

3 CADD of Anti-Alzheimer BACE1 Inhibitors: Techniques and Recent Developments This section will summarize recent CADD developments in search for new BACE1 inhibitors. Historically, the breakthrough in BACE1 inhibitor development is considered the discovery of OM99-2 and OM00-3 substrate-based inhibitors (Fig. 4) [52, 53]; the elucidation of their binding mode with the active

78

Konstantinos D. Papavasileiou et al.

Fig. 4 Chemical structures of the OM99-2 and OM00-3 inhibitors

site of the enzyme was a turning point in future inhibitor design [54]. Since then, CADD techniques have been extensively employed and for a comprehensive review of past efforts (the interested reader is referred to previous works [19, 39]). In the following sections, recent studies on CADD design of novel BACE1 inhibitors are presented, roughly organized according to the main computational methods used. 3.1 Virtual Screening-Based Approaches 3.1.1 Molecular DockingMolecular Simulations

Virtual screening (VS), molecular docking and MD simulations combined with Molecular Mechanics Generalized Born Surface Area (MM-GBSA) [55, 56] calculations are important tools in CADD studies for the research of BACE1 inhibitors. Briefly, VS is a computational technique complementary to high-throughput screening (HTS) whose primary objective is to facilitate the quick and affordable evaluation of large databases to screen lead compounds for the discovery of new drugs [40, 57]. Molecular docking calculations are used for the prediction of the most favorable interaction of small-molecules (ligands) within a protein (receptor) binding site by scanning the conformational space using a scoring function that uses a semiempirical free energy force field to rank and discriminate the bound conformations [58]. The force field is a mathematical expression describing the potential energy of a system of atoms or molecules as the sum of individual classical potential energy terms accounting for the bonded (bonds, bond angles, and dihedral angles) and nonbonded interactions (van der Waals and electrostatic), designed to reproduce the behavior of the system [59]. Classical force fields are commonly utilized in MD

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

79

simulations that are applied in large atomic and molecular systems to calculate and predict various macroscopic properties using statistical mechanics. Configurations of a molecular system are generated sequentially by integrating Newton’s laws of motion, and as a result, trajectories are obtained in the phase space, which determine the change in the positions and velocities of the atoms/molecules of the system with time. Statistical analysis of the resulting trajectories allows the calculation of a series of characteristic structural and thermodynamic properties of a system [59]. MD trajectory analysis with methods like the Molecular Mechanics PoissonBoltzmann Surface Area (MM-PBSA) and MM-GBSA allows for the estimation of relative protein-ligand binding affinities by decomposing the binding free energy into individual energy terms accounting for the gas phase molecular mechanics energy, the polar and nonpolar contributions to the solvation free energy and the conformational entropy [55, 56]. As an application example of the aforementioned methods, notopterol (a furan coumarin from Notopterygium incisum, Fig. 5) was identified from VS to possess simultaneous inhibitory activity toward both BACE1 (IC50: 26.01 μM) and glycogen synthase kinase-3 beta (GSK3β) (IC50: 1 μM) [60]. Molecular docking with biological evaluations of cardamonin, pinocembrin, and pinostrobin (Fig. 5) – i.e., Boesenbergia rotunda flavonoids – showed cardamonin as the strongest BACE1 inhibitor (cardamonin, pinocembrin, and pinostrobin IC50 values of 4.35 ± 0.38, 27.01 ± 2.12, and 28.44 ± 1.96 μM, respectively), while none of the other tested compounds exhibited binding in BACE1’s active site, illustrating noncompetitive inhibitory activity for all three compounds [61]. Another molecular docking and experimental investigation identified three synthesized flavone derivatives exhibiting significant biological effects on both acetylcholinesterase (AChE) and BACE1 [62]. It was also observed that the three compounds with the lowest half maximal inhibitory concentration (IC50) (viz., B3, 3.98; D5, 1.66; and D6, 1.58 μM) on BACE1 are in correlation with the calculated docking scores [62]. Molecular docking calculations along with kinetic studies were employed to investigate sulforaphane (an isothiocyanate found in cruciferous vegetables, Fig. 5) against BACE1 [63]. Results revealed that sulforaphane activity not only was sixfold in potency (IC50 value of 2.80 ± 0.19 μM) compared to well-known positive controls resveratrol and quercetin (IC50 18.10 ± 0.03 and 15.04 ± 0.87 μM, respectively) but also displayed selective and noncompetitive BACE1 inhibitory activity by developing van der Waals interactions with other BACE1 binding sites [63]. In a molecular docking, MD simulation, liquid chromatography, and high-resolution electrospray mass spectrometry study, Morus macroura was investigated as a food supplement for AD management [64]. It was found that among 29 phytochemicals, resveratrol

Fig. 5 Selected structures of potential small-molecule BACE1 inhibitors discussed in this chapter

80 Konstantinos D. Papavasileiou et al.

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

81

together with chrysin exhibited comparable inhibitory activities toward BACE1 (IC50 16.78 ± 0.9 and 21.34 ± 1.1 μM, respectively) [64]. A molecular docking-based VS with structure-based (SB) and ligand-based (LB) pharmacophore model protocol employed to design and screen several drug-like compound databases identified 13 novel hit BACE1 inhibitors (peptidomimetic, indole, coumarin, piperazine, piperidine, and 1,3,5 triazine-derivatives, Fig. 5), further subjected to molecular docking, in silico filtering and biochemical experiments to predict their blood-brain barrier (BBB) crossing potential [65]. The most promising BACE1 hit detected (viz., compound 11) showed an IC50 value of 15 μM [65]. In another molecular docking and in vivo study, it was revealed that melatonin and its derivatives displayed potent inhibitory activity for BACE1 [66]. Recently, molecular docking, MD simulations and post-MD analyses showed one of these melatonin derivatives to exhibit strong allosteric BACE1 binding and considerable stability at eight different subsites [67]. More advanced MD methods such as multiple replica (MR) accelerated molecular dynamics (aMD) [68, 69] simulations combined with principal component analysis (PCA) [70] were used to investigate the effect of BACE1 disulfide bonds (SSBs) on the binding of three inhibitors (viz., 3KO, 3KT, and 779), showing that SSB breaking impacts their binding modes as well as the structural flexibility and dynamics of inhibitor-BACE1 complexes [71]. A combination of MR Gaussian aMD (MR-GaMD) [72, 73] simulations and the MM-GBSA calculations showed a strong pH-dependent protonation effect on the structural flexibility, correlated motions, dynamic behavior, and binding energetics of CS9 [74], C6U [75], and 6WE [76] inhibitors on BACE1, following the experimental IC50 value trends [77]. BACE1 and BACE2 are structurally very similar [78]; hence designing BACE1 inhibitors with high selectivity is a particularly demanding and laborious task. Recently [79], a possible BACE1 inhibitor, namely, C28 [80], was reported to exhibit greater selectivity toward BACE1 than BACE2 compared to inhibitors Lanabecestat (AZD3293) [81] and AZD3839 [82]. To this end, classical (cMD), aMD simulations, and free energy calculations revealed similar binding affinities of AZD3293 to BACE1 and BACE2 and higher binding affinity of AZD3839 and C28 to BACE1 than BACE2. The underlying mechanisms of these findings are attributed to differences in electrostatic interactions patterns associated with different energy barriers, hence aiding the rational design of more potent BACE1 selective inhibitors [79]. Another molecular docking and MD simulation study of several potential BACE1 and BACE2 ligands revealed that 8 (55E, 6Z2, 6Z5, BRW, F1B, GVP, IQ6, and X37) showed favorable binding toward both, with 6Z5 having the best binding potential [83]. Molecular docking and MD

82

Konstantinos D. Papavasileiou et al.

simulation studies combined with clustering analysis and phylogenetic studies identified 11-oxotigogenin (Fig. 5) as the most promising inhibitor against BACE1 [84]. Compounds from Cajanus cajan and Citrus reticulata plants were investigated for their BACE1 and AChE inhibitory properties using in silico methods to identify genistin, naphthalen-2-yl-acetic acid, 6-hydroxy-6-methyl-cyclodecyl ester, and vitexin (Fig. 5) as potent binders that also passed the oral druggability test, while naphthalen-2-yl-acetic acid and 6-hydroxy-6-methyl-cyclodecyl ester exhibited BBB permeation [85]. Molecular docking, MD simulations, and free energy calculations on the daidzin, genistin, mangiferin, puerarin, and tuberosin phytochemicals (Fig. 5) from Pueraria tuberosa showed that they are all potent binders within the BACE1 active site [86]. Molecular docking calculations predicted the binding modes of b-sitosterol and stigmasterol in the BACE1 binding site, with b-sitosterol being more favorable compared to stigmasterol. Othman et al. [87] reported docking, absolute binding free energy calculations, and MD simulations on seven amide alkaloids, namely, N-trans-feruloyl-3-methoxytyramine (1), N-trans-feruloyltyramine (2), S-(-)N-trans-feruloyl normetanephrine (3), S-(-)-N-trans-feruloyloctopamine (4), R-(+)-N-trans-feruloyloctopamine (5), N-trans-caffeoyltyramine (6), and S-(-)-3-(4-hydroxy-3-methoxyphenyl)-N[2-(4-hydroxyphenyl)-methoxyethyl]acrylamide (7), from halophytic plants Bassia indica and Agathophora alopecuroides, with compounds 1, 2, and 7 displaying strong BACE1 inhibition (IC50 < 6 mg mL-1) [87]. The inhibitory activities of 83 endophyte-derived compounds and standard BACE1 inhibitors were assessed by means of molecular docking, MD simulations, druglikeness, absorption-distribution-metabolism-excretion-toxicity (ADMET) and BBB properties, showing the remarkable inhibitory activity of asperflavin, ascomfurans C, camptothecin, and corynesidone A against BACE1, with corynesidone A being safe and able to transverse the BBB [88]. In order to prioritize candidates for BACE1 inhibitors, do Bomfim et al. [89] performed a hierarchical VS by pharmacophore model and molecular docking against 216,833 molecules contained in several databases. Four molecules were finally selected and evaluated for mutagenic potential, tested against the descriptors on Lipinski’s Ro5 [90] and Veber rules [91], and were subjected to MD simulations to finally identify ZINC01589617 as a potential candidate for biological tests [89]. A docking study of more than 4000 naturally occurring compounds in the Vietnamese plants (VIETHERB) database combined with steered MD (SMD) simulations to show that myricetin, quercetin, and hydroxysafflor are remarkably better binders than the 23I BACE1 inhibitor [92]. SB docking screening approaches, Lipinski Ro5, and ADMET predictions along with MD simulations

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

83

and MM-GBSA calculations were employed for a pharmacological activity analysis of 876 bioactive Medhya Rasayana plant compounds, to finally identify convolidine and N-(4-hydroxybutyl) phthalimide (Fig. 5) as potential BACE1 inhibitors [93]. A series of newly synthesized 6-chloro-N′-(substituted benzylidene)nicotinohydrazide were investigated experimentally and in silico to identify a single compound (viz., P5) exhibiting high inhibition rates with IC50 values of 0.205 ± 0.008 μM and 0.027 ± 0.001 μM against BACE1 and Aβ AChE, respectively, also exhibiting high BBB permeation [94]. Molecular docking studies identified curcumin (Fig. 5), a natural flavonoid with potent antioxidant and antiaging properties as a candidate BACE1 inhibitor [95]. Finally, Gupta et al. [96] performed VS, molecular docking, MD simulations, and binding free energy calculations to find five potential BACE1 inhibitors (B1, C21H23N5O2; B2, C21H23N5O; B3, C22H30FN3O3; B4, C19H21FN4O2; and B5, C22H26N4O) from the Asinex chemical library database [96]. 3.1.2 Ligand- and Structure-Based Design: Cheminformatics

Another important CADD technique for the production of smallmolecule inhibitors particularly when a receptor is not accessible is ligand-based (LB) design, as accurately described in previously published review articles [39, 97]. Because of the plethora of BACE1 crystal structures, hybrid SB virtual screening techniques including both SB and LB design were developed for discovering possible BACE1 inhibitors [98]. The application of QSAR approaches has previously resulted in the successful development of structure-activity relationship models with potential uses in binding affinity predictions of prospective BACE1 inhibitors [99]. Recently, QSAR models combined with molecular docking, MD simulations, MM-GBSA calculations, virtual screening, and pharmacophore modeling led to the discovery of natural compounds as BACE1 inhibitors that were screened for anti-amyloidogenic activity [100]. Iwaloye et al. [101] performed VS of ~33,000 natural compounds from the NPASS database [102] based on molecular docking, ADME/TOX, and QSAR analysis to identify four natural compounds (NPC469686, NPC262328, NPC29763, and NPC86744) as novel potential BACE1 inhibitors [101]. A five descriptor QSAR model was developed to finally screen five flavonols (isorhamnetin, syringetin, galangin, tamarixetin, rhamnetin) and two flavanonols (dihydromyricetin, taxifolin) (Fig. 5) natural compounds as potent BACE1 inhibitors [103]. Another QSARbased VS study on a 26,467 food compounds database coupled with MD simulations and MM-GBSA calculations led to the identification of 4-(3,4-dihydroxyphenyl)-2-hydroxy-1H-phenalen-1one as a hit BACE1 inhibitor [104]. 3D-pharmacophore, 2D-QSAR, and molecular docking in silico models were employed for the VS of a library containing more than three million curcumin and flavonoid derivatives, to finally identify 47 substances

84

Konstantinos D. Papavasileiou et al.

(2 curcumins and 45 flavonoids) with remarkable predicted pIC50 values against AChE and BACE1 (from 4.24 to 5.11 and from 4.52 to 10.27, respectively), as confirmed by in vitro assays [105]. A LB pharmacophore design produced a 64 molecule ensemble with BACE1 inhibitory activity, to finally select 3 molecules (viz., F3S4-m, F2S4-m, and F2S4-p) after molecular docking calculations, cheminformatics analyses, and in silico predicted toxicity screening for synthesis and evaluation, with the former 2 displaying BACE1 inhibition (IC50 15.97 and 8.38 μM, respectively) [106]. In another study, a toxicity evaluation with the admetSAR online database [107, 108] along with molecular docking calculations and ADME prediction indicated quercetin as a possible BACE1 inhibitor [109]. A new robust ligand-based predictive model for BACE1 inhibitory data was developed, validated according to Organization for Economic Co-operation and Development (OECD) principles as part of the Enalos Chemoinformatics Cloud Platform [110]. The online interface of the service has been deliberately developed for simplicity and user-friendliness, allowing individuals with no informatics background to readily utilize the BACE LB models and profit from the provided forecasts and outcomes, thus allowing the interested user to submit and virtually screen one or several compounds (Fig. 6) [110, 111]. Three different options are available for submitting a structure that include the following: (i) drawing a structure with the available sketcher – compounds can be easily generated and modified to create a set of structures that can be first visualized and then submitted; (ii) submitting the SMILES notation for one or many compounds at the form available; and (iii) submitting an .sdf file including a batch of compounds (Fig. 6). After importing the structures with one of the options described, the BACE workflow will run after the submit button is pressed. When structures are submitted, the results page will appear, which includes a class prediction for each of the structures submitted, and an indication of whether this prediction can be considered as reliable or not, based on the domain of applicability. This web service dedicated to the proposed model can easily facilitate the virtual screening of new structures that fall within its domain of applicability. In general, the Enalos Cloud Platform [112] currently hosts 23 predictive models released as web services for a broad spectrum of material design and development, drug discovery, virtual screening of chemical substances, nanosafety, and safe-by-design (nano)materials applications (Fig. 7) and is being utilized by several EU research projects. As it has already been pointed out, the development of a predictive model might turn out unusable unless it is delivered as a user-friendly tool to ensure sustainability. Based on this, the platform is an easy-to-use portal where web services are arranged by categories of use

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

85

Fig. 6 Screenshot of the Enalos Chemoinformatics Cloud Platform input page for the BACE ligand-based predictive model [111]

(chemoinformatics, nanoinformatics, image analysis, exposure, and biokinetics models) and projects so that users can quickly explore and choose the tool of their preference, easily use the models, and benefit from the produced predictions and results (Fig. 7). 3.1.3

Machine Learning

Modern machine learning (ML), the foundation of artificial intelligence (AI), has had a significant influence on all branches of research, including chemistry [113, 114]. Compared to classic computational approaches, new ML methods based on deep neural networks and representation learning tend to deliver superior prediction quality [113]. In this context, Singh et al. [115] performed a classification analysis on 3,536 different BACE1 inhibitors taken from the binding DB database by extracting two types of descriptors, namely, molecular property (Mordred) and fingerprints (PubChem, MACCS, and KRFP), on the basis of which ML algorithms like Naı¨ve Bayesian (NB), nearest known neighbors (kNN), support vector machine (SVM), random forest (RF), and gradientboosted algorithms (XGB) were applied to develop classification models. The BACE1 inhibitors were divided into 11 subgroups, and their structural properties were studied, identifying important fragments shared between active and inactive BACE1 drugs, thus developing a model for building and virtual screening [115]. In a recent example of combining ML and deep learning (DL) approaches with VS for the identification of potential BACE1 small-molecule inhibitors, a dataset containing 57 AChE and 53 BACE1 N-benzyl piperidine derivatives was used for multitarget directed ligand-based 2D-QSAR model development using

86

Konstantinos D. Papavasileiou et al.

Fig. 7 The integrated in silico Enalos Cloud Platform toolset of innovative chem-, bio-, and nano- informatics models and services [112]

five different classes of molecular descriptors, namely, structural (Cl element counts, number of rotatable bonds), electrotopological (number of oxygen atoms connected with one double bond), electronic (induced dipole moment along the z axis), and spatial (extent of molecule shadows) for BACE1 [116]. Linear, genetic function approximation (GFA), nonlinear, SVM and artificial neural network (ANN) ML methods were used to show that these molecular descriptors should be further utilized in the rational design of multi-targeted anti-lead Alzheimer’s compounds [116], highlighting the strength and great promise that this approach holds in the future of BACE1 inhibitor discovery. ML methods are commonly utilized to analyze chemical compound ligand activity toward putative target proteins like BACE1. The exploration of highly selective ligands is critical for the creation of novel pharmaceuticals with improved safety. The lack of data on true negative compound-protein interactions (i.e., molecules that do not bind to relevant proteins) is key to building such predictive models for inhibitor efficacy [117]. To address this challenge, a graph convolution neural network (GCNN) approach was suggested by Miyazaki et al. [117], aiming to thoroughly investigate natural compounds targeting BACE1 with reduced off-target effect toward cathepsin D. Results unveiled significantly different BACE1 and cathepsin D ligand distributions on the density map, which is likely to hasten the search for new candidates for highly selective AD therapeutics [117]. Lastly, based on multiple-property optimization via gradient descent in the latent space, a generative network

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

87

complex (GNC) was developed to automatically generate many novel BACE1 inhibitors, as well as thousands alternatives to commercially available pharmaceuticals, including ceritinib, ribociclib, acalabrutinib, idelalisib, dabrafenib, macimorelin, enzalutamide, and panobinostat [118]. 3.2 Other TechniqueBased Methods 3.2.1 Quantum Mechanical Approaches

Quantum mechanical (QM) approaches comprise ab initio, semiempirical, empirical, and density functional theory (DFT) methods [59, 119–122] and have been extensively employed in the elucidation of small peptide BACE1 modulators as well as in binding energy predictions of potential inhibitors [123]. The estimation of binding affinities with high accuracy is very important; hence QM approaches are growing in popularity in computational drug design and development. Briefly, QM methods aim to approximate the wave function and determine the electronic structure of a molecule or polyatomic system by solving the electronic Hamiltonian of the Schro¨dinger equation. DFT methods use the electron density instead of the wave function that describes the system under study to calculate its electronic energy [121]. QM methods allow the accurate calculation of many molecular properties such as equilibrium structures, vibrational frequencies, dipole moments, binding free energies, reaction paths, etc. but are computationally very expensive and time-consuming and are therefore limited to molecular systems comprising a small number of atoms. Recently, DFT coupled with molecular docking calculations were used with in vitro and in vivo experiments to investigate BACE1 inhibition by 3,150 phytochemicals from almost 25 different plants [124]. These calculations together with ADMET studies revealed seven phytochemicals (shinflavanone, glabrolide, glabrol, and prenyllicoflavone A from Glycyrrhiza glabra, macleanine from Huperzia serrata, 3a-dihydro-cadambine from Uncaria rhynchophylla, and volvalerelactone B from Valeriana officinalis) having high BACE1 inhibitory activity [124]. In another study, two potent BACE1 inhibitors, namely, AM-6494 [125], a newly reported potent BACE1 inhibitor picked for preclinical considerations, and Umibecestat (CNP-520) [126, 127], recently discontinued at human trials, were investigated using DFT and our own N-layered integrated molecular orbital and molecular mechanics (ONIOM) [128] calculations to illustrate AM-6494 as more favorable toward BACE1 inhibition (ΔGbind = -62.849 kcal mol-1) than CNP-520 (ΔGbind = -33.463 kcal mol-1), with the calculated binding free energy reproducing the in vivo inhibition trends (IC50 values of 0.4 and 11 nM, respectively) [129]. Gnanaraj et al. [130] investigated karanjin – a furanoflavonoid isolated primarily from Pongamia pinnata – with extensive docking, MD simulations, frontier molecular orbitals (FMOs), and DFT calculations, using the Lipinski’s rule of five (Ro5) [90] and ADMET to show that it could be considered as a suitable therapeutic lead [130]. Lastly, a

88

Konstantinos D. Papavasileiou et al.

nonlinear optical (NLO) responses computational investigation of the active phytochemicals of the Clitoria ternatea at the B3LYP/6311G++(d, p) level of theory coupled with docking and MD simulations identified the kaempferol glycoside Clitorin as the most active and inhibiting compound toward BACE1 [131].

4

Future Outlook Alzheimer’s disease has unmet medical requirements, particularly in terms of disease-modifying medicines. For many years, the role of Aβ amyloid in AD pathogenesis has been central to drug discovery efforts, leading to a substantial number of drugs targeting BACE1. Unfortunately, the majority of these drugs have failed in clinical trials [43] for reasons that are being actively investigated. It has long been hypothesized – and is now being thoroughly explored – that multitarget medication therapy is probably preferable to single-target therapy for treating complicated neurological illnesses like AD. It is inevitable to adopt new strategies for the CADD of new drugs against AD considering the associated factor range responsible for the initiation, progression, and severity of the disease. However, this increases the complexity of their pathophysiology and is linked to the ineffectiveness of the available therapeutic tools [39]. In a recent example, molecular docking, MD simulations, artificial neural networks, and multilinear regression models were employed for the virtual screening of 20,397 small compounds (MW < 600) extracted from the ZINC database [132]. Three potential multifunctional drug candidates acting simultaneously toward AChE, SERT, BACE1, and GSK3β protein targets were proposed [132]. Another promising aspect involves CADD design for drug delivery systems (DDS) in order to ensure improved effectiveness and minimal adverse effects of novel BACE1 pharmaceuticals. In this regard, highly porous inorganic materials such as metal-organic frameworks (MOFs) are currently being investigated as DDS candidates for the administration of various pharmaceuticals, owing to their simple preparation, regulated release, and organ-targeting advantages [133]. For example, recent studies showed that magnolol – a phenolic natural product – exhibits neuroprotective properties through degradation of Aβ amyloid plaques in PC-12 cells [134] and prevention of behavioral impairments and neuropathological findings in transgenic mice models (TgCRDN8) [135]. The inhibitory activity, bioavailability, and BBB properties of magnolol against BACE1 and AlCl3 were significantly enhanced by the use of UiO-66(Zr) MOF as determined by in silico molecular docking, MD simulation studies, and extensive in vitro evaluations, to confer that MOFs are promising DDS platforms for poorly bioavailable drugs [136].

Recent Advances in Computational Modeling of BACE1 Inhibitors. . .

89

It has already been discussed in this chapter’s introduction that environmental neurotoxicants have been linked to neurodegenerative diseases including Parkinson’s and Alzheimer’s and is currently another intriguing field of scientific research. BACE1 comprises a significant pathogenic target of environmentally induced neurotoxicity, with neurotoxicants such as metals, pesticides, herbicides, fungicides, polyfluoroalkyl compounds, heterocyclic aromatic amines, advanced glycation end products, and acrolein being prime examples that can modulate BACE1 [137]. In this category, per- and poly-fluoroalkyl substances (PFASs) are present in a wide variety of industrial and consumer applications, the majority of which having unknown hazardous potential in terms of bioactivity, bioaccumulation, and toxicity. Several studies reported on neurotoxicity of PFAS and particularly on their effects on neurotransmission [138]. In this respect, it was recently attempted to develop a powerful ML-based QSAR model for the prediction of 4,730 PFASs bioactivity from the OECD report [139], where a total of 6 different data sets including BACE1 [140] were utilized for the construction of the PFAS-specific data [141]. Overall, future CADD developments in the field of BACE1 inhibitors are extremely promising, paving the way in providing innovative means and improved understanding for the generation of a diversified and expanded arsenal of pharmaceuticals against Alzheimer’s disease.

5

Conclusions Alzheimer’s disease is a neurodegenerative illness causing neuronal death and synaptic degeneration in the human brain. The Aβ pathway, which leads to amyloid plaques, is considered one of the primary biological pathways for the development of Alzheimer’s therapeutics. The BACE1 protein is a key center target at the heart of the amyloid pathway. The principal BACE1 structural features and functions that render it an attractive target for computer-aided drug design methodologies aimed at the production of novel small-molecule inhibitors were presented. Recent examples of state-of-the-art CADD techniques and approaches employed in the development of BACE1 inhibitors were also summarized.

Acknowledgments K. D. Papavasileiou, F. Dondero, and A. Afantitis acknowledge financial support by the EU H2020 project Scenarios (ID: 101037509). K. D. Papavasileiou and A. Afantitis also acknowledge the EU H2020 project EthnoHERBS (ID: 823973) as well as the ENTERPRISES/0618/0122 project, which was co-funded by

90

Konstantinos D. Papavasileiou et al.

the European Regional Development Fund and the Republic of Cyprus through the Research and Innovation Foundation. References 1. Francis PT, Palmer AM, Snape M, Wilcock GK (1999) The cholinergic hypothesis of Alzheimer’s disease: a review of progress. J Neurol Neurosurg Psychiatry 66(2):137. https://doi.org/10.1136/jnnp.66.2.137 2. Moussa-Pacha NM, Abdin SM, Omar HA et al (2020) BACE1 inhibitors: current status and future directions in treating Alzheimer’s disease. Med Res Rev 40(1):339–384. https://doi.org/10.1002/med.21622 3. Alzheimer’s Association (2019) 2019 Alzheimer’s disease facts and figures. Alzheimers Dement 15(3):321–387. https://doi.org/ 10.1016/j.jalz.2019.01.010 4. Gustavsson A, Norton N, Fast T et al (2022) Global estimates on the number of persons across the Alzheimer’s disease continuum. Alzheimer’s Dement:1–13. https://doi.org/ 10.1002/alz.12694 5. Nichols E, Szoeke CEI, Vollset SE et al (2019) Global, regional, and national burden of Alzheimer’s disease and other dementias, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol 18(1):88–106. https://doi.org/10. 1016/S1474-4422(18)30403-4 6. Collaborators GBDN (2019) Global, regional, and national burden of neurological disorders, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol 18(5):459–480. https://doi.org/10.1016/S1474-4422(18) 30499-X 7. Wang J, Gu BJ, Masters CL, Wang YJ (2017) A systemic view of Alzheimer disease – insights from amyloid-beta metabolism beyond the brain. Nat Rev Neurol 13(11): 703. https://doi.org/10.1038/nrneurol. 2017.147 8. Wang J, Gu BJ, Masters CL, Wang YJ (2017) A systemic view of Alzheimer disease – insights from amyloid-beta metabolism beyond the brain. Nat Rev Neurol 13(10): 612–623. https://doi.org/10.1038/ nrneurol.2017.111 9. Chen L, Xu S, Wu T et al (2019) Abnormal platelet amyloid-β precursor protein metabolism in SAMP8 mice: evidence for peripheral marker in Alzheimer’s disease. J Cell Physiol 234(12):23528–23536. https://doi.org/10. 1002/jcp.28921

10. Sˇimic´ G, Babic´ Leko M, Wray S et al (2016) Tau protein hyperphosphorylation and aggregation in Alzheimer’s disease and other tauopathies, and possible neuroprotective strategies. Biomol Ther 6(1):6. https://doi. org/10.3390/biom6010006 11. Alonso AD, Cohen LS, Corbo C et al (2018) Hyperphosphorylation of tau associates with changes in its function beyond microtubule stability. Front Cell Neurosci 12:338. https://doi.org/10.3389/fncel.2018.00338 12. Huang WJ, Zhang X, Chen WW (2016) Role of oxidative stress in Alzheimer’s disease. Biomed Rep 4(5):519–522. https://doi. org/10.3892/br.2016.630 13. Uttara B, Singh AV, Zamboni P, Mahajan RT (2009) Oxidative stress and neurodegenerative diseases: a review of upstream and downstream antioxidant therapeutic options. Curr Neuropharmacol 7(1):65–74. https://doi. org/10.2174/157015909787602823 14. Bouvier DS, Jones EV, Quesseveur G et al (2016) High resolution dissection of reactive glial nets in Alzheimer’s disease. Sci Rep 6(1): 2 4 5 4 4 . h t t p s : // d o i . o r g / 1 0 . 1 0 3 8 / srep24544 15. Tejera D, Heneka TM (2016) Microglia in Alzheimer’s disease: the good, the bad and the ugly. Curr Alzheimer Res 13(4): 370–380. https://doi.org/10.2174/ 1567205013666151116125012 16. Solas M, Puerta E, Ramirez MJ (2015) Treatment options in Alzheimer s disease: the GABA story. Curr Pharm Des 21(34): 4960–4971. https://doi.org/10.2174/ 1381612821666150914121149 17. Kametani F, Hasegawa M (2018) Reconsideration of amyloid hypothesis and tau hypothesis in Alzheimer’s disease. Front Neurosci 12: 25. https://doi.org/10.3389/fnins.2018. 00025 18. Cole SL, Vassar R (2007) The Alzheimer’s disease β-secretase enzyme, BACE1. Mol Neurodegener 2(1):22. https://doi.org/10. 1186/1750-1326-2-22 19. Cruz-Vicente P, Passarinha LA, Silvestre S, Gallardo E (2021) Recent developments in new therapeutic agents against Alzheimer and Parkinson diseases: in-silico approaches. Molecules 26(8):2193. https://doi.org/10. 3390/molecules26082193

Recent Advances in Computational Modeling of BACE1 Inhibitors. . . 20. Vassar R, Bennett BD, Babu-Khan S et al (1999) β-secretase cleavage of Alzheimer’s amyloid precursor protein by the transmembrane aspartic protease BACE. Science 286(5440):735–741. https://doi.org/10. 1126/science.286.5440.735 21. Hussain I, Powell D, Howlett DR et al (1999) Identification of a novel aspartic protease (Asp 2) as β-secretase. Mol Cell Neurosci 14(6): 419–427. https://doi.org/10.1006/mcne. 1999.0811 22. Sinha S, Anderson JP, Barbour R et al (1999) Purification and cloning of amyloid precursor protein beta-secretase from human brain. Nature 402(6761):537–540. https://doi. org/10.1038/990114 23. Dislich B, Lichtenthaler SF (2012) The membrane-bound aspartyl protease BACE1: molecular and functional properties in Alzheimer’s disease and beyond. Front Physiol 3:8. https://doi.org/10.3389/fphys.2012. 00008 24. Saunders AJ, Kim T-W, Tanzi RE (1999) BACE maps to chromosome 11 and a BACE homolog, BACE2, reside in the obligate Down syndrome region of chromosome 21. Science 286:1255–1255. https://doi.org/ 10.1126/science.286.5443.1255a 25. Zolezzi JM, Bastı´as-Candia S, Santos MJ, Inestrosa NC (2014) Alzheimer’s disease: relevant molecular and physiopathological events affecting amyloid-β brain balance and the putative role of PPARs. Front Aging Neurosci 6:176. https://doi.org/10.3389/fnagi. 2014.00176 26. Maia MA, Sousa E (2019) BACE-1 and γ-secretase as therapeutic targets for Alzheimer’s disease. Pharmaceuticals 12(1):41. https://doi.org/10.3390/ph12010041 27. Hamada Y, Kiso Y (2016) New directions for protease inhibitors directed drug discovery. Biopolymers 106(4):563–579. https://doi. org/10.1002/bip.22780 28. Hong L, Koelsch G, Lin X et al (2000) Structure of the protease domain of memapsin 2 (beta-secretase) complexed with inhibitor. Science 290(5489):150–153. https://doi. org/10.1126/science.290.5489.150 29. Cole DC, Manas ES, Stock JR et al (2006) Acylguanidines as small-molecule β-secretase inhibitors. J Med Chem 49(21):6158–6161. https://doi.org/10.1021/jm0607451 30. Malamas MS, Barnes K, Hui Y et al (2010) Novel pyrrolyl 2-aminopyridines as potent and selective human β-secretase (BACE1) inhibitors. Bioorg Med Chem Lett 20(7):

91

2068–2073. https://doi.org/10.1016/j. bmcl.2010.02.075 31. Scott JD, Li SW, Brunskill APJ et al (2016) Discovery of the 3-imino-1,2,4-thiadiazinane 1,1-dioxide derivative verubecestat (MK-8931)–a β-site amyloid precursor protein cleaving enzyme 1 inhibitor for the treatment of Alzheimer’s disease. J Med Chem 59(23):10435–10450. https://doi.org/10. 1021/acs.jmedchem.6b00307 32. Dabur M, Loureiro JA, Pereira MC (2022) The current state of amyloidosis therapeutics and the potential role of fluorine in their treatment. Biochimie 202:123. https://doi.org/ 10.1016/j.biochi.2022.08.003 33. McDade E, Voytyuk I, Aisen P et al (2021) The case for low-level BACE1 inhibition for the prevention of Alzheimer disease. Nat Rev Neurol 17(11):703–714. https://doi.org/ 10.1038/s41582-021-00545-1 34. Graham WV, Bonito-Oliva A, Sakmar TP (2017) Update on Alzheimer’s disease therapy and prevention strategies. Annu Rev Med 68(1):413–430. https://doi.org/10.1146/ annurev-med-042915-103753 35. Cummings J, Lee G, Nahed P et al (2022) Alzheimer’s disease drug development pipeline: 2022. Alzheimers Dement 8(1):e12295. https://doi.org/10.1002/trc2.12295 36. Sevigny J, Chiao P, Bussie`re T et al (2016) The antibody aducanumab reduces Aβ plaques in Alzheimer’s disease. Nature 537(7618):50–56. https://doi.org/10. 1038/nature19323 37. Mullard A (2021) Landmark Alzheimer’s drug approval confounds research community. Nature 594(7863):309–310. https:// doi.org/10.1038/d41586-021-01546-2 38. Prieto-Martı´nez FD, Lo´pez-Lo´pez E, Eurı´dice Jua´rez-Mercado K, Medina-Franco JL (2019) Computational drug design methods – current and future perspectives. In: Roy K (ed) In silico drug design. Academic Press, pp 19–44. https://doi.org/10.1016/B9780-12-816125-8.00002-X 39. Mouchlis VD, Melagraki G, Zacharia LC, Afantitis A (2020) Computer-aided drug design of beta-secretase, gamma-secretase and anti-tau inhibitors for the discovery of novel Alzheimer’s therapeutics. Int J Mol Sci 21(3):703. https://doi.org/10.3390/ ijms21030703 40. Baig MH, Ahmad K, Rabbani G et al (2018) Computer aided drug design and its application to the development of potential drugs for neurodegenerative disorders. Curr

92

Konstantinos D. Papavasileiou et al.

Neuropharmacol 16(6):740–748. https:// d o i . o r g / 1 0 . 2 1 7 4 / 1570159X15666171016163510 41. Salman MM, Al-Obaidi Z, Kitchen P et al (2021) Advances in applying computer-aided drug design for neurodegenerative diseases. Int J Mol Sci 22(9). https://doi.org/10. 3390/ijms22094688 42. Hu H, Chen Z, Xu X, Xu Y (2019) Structurebased survey of the binding modes of BACE1 inhibitors. ACS Chem Neurosci 10(2): 880–889. https://doi.org/10.1021/ acschemneuro.8b00420 43. Calcoen D, Elias L, Yu X (2015) What does it take to produce a breakthrough drug? Nat Rev Drug Discov 14(3):161–162. https:// doi.org/10.1038/nrd4570 44. Dunn BM (2002) Structure and mechanism of the pepsin-like family of aspartic peptidases. Chem Rev 102(12):4431–4458. https://doi. org/10.1021/cr010167q 45. Gorfe AA, Caflisch A (2005) Functional plasticity in the substrate binding site of β-secretase. Structure 13(10):1487–1498. https://doi.org/10.1016/j.str.2005.06.015 46. Park H, Lee S (2003) Determination of the active site protonation state of beta-secretase from molecular dynamics simulation and docking experiment: implications for structure-based inhibitor design. J Am Chem Soc 125(52):16416–16422. https://doi. org/10.1021/ja0304493 47. Herna´ndez-Rodrı´guez M, Correa-Basurto J, Gutie´rrez A et al (2016) Asp32 and Asp228 determine the selective inhibition of BACE1 as shown by docking and molecular dynamics simulations. Eur J Med Chem 124:1142– 1154. https://doi.org/10.1016/j.ejmech. 2016.08.028 48. Patel S, Vuillard L, Cleasby A et al (2004) Apo and inhibitor complex structures of BACE (beta-secretase). J Mol Biol 343(2): 407–416. https://doi.org/10.1016/j.jmb. 2004.08.018 49. Shimizu H, Tosaki A, Kaneko K et al (2008) Crystal structure of an active form of BACE1, an enzyme responsible for amyloid β protein production. Mol Cell Biol 28(11): 3663–3671. https://doi.org/10.1128/ MCB.02185-07 50. Hong L, Tang J (2004) Flap position of free memapsin 2 (beta-secretase), a model for flap opening in aspartic protease catalysis. Biochemistry 43(16):4689–4695. https://doi. org/10.1021/bi0498252 51. Xu Y, Li M-J, Greenblatt H et al (2012) Flexibility of the flap in the active site of BACE1 as

revealed by crystal structures and molecular dynamics simulations. Acta Cryst D 68(1): 1 3 – 2 5 . h t t p s : // d o i . o r g / 1 0 . 1 1 0 7 / S0907444911047251 52. Ghosh AK, Shin D, Downs D et al (2000) Design of potent inhibitors for human brain memapsin 2 (beta-secretase). J Am Chem Soc 122(14):3522–3523. https://doi.org/10. 1021/ja000300g 53. Turner RT 3rd, Koelsch G, Hong L et al (2001) Subsite specificity of memapsin 2 (beta-secretase): implications for inhibitor design. Biochemistry 40(34):10001–10006. https://doi.org/10.1021/bi015546s 54. Ghosh AK, Osswald HL (2014) BACE1 (β-secretase) inhibitors for the treatment of Alzheimer’s disease. Chem Soc Rev 43(19): 6765–6813. https://doi.org/10.1039/ C3CS60460H 55. Wang W, Kollman PA (2001) Computational study of protein specificity: the molecular basis of HIV-1 protease drug resistance. Proc Natl Acad Sci U S A 98(26):14937–14942. https://doi.org/10.1073/pnas.251265598 56. Wang W, Kollman PA (2000) Free energy calculations on dimer stability of the HIV protease using molecular dynamics and a continuum solvent model. J Mol Biol 303(4): 567–582. https://doi.org/10.1006/jmbi. 2000.4057 57. Kim KH, Kim ND, Seong BL (2010) Pharmacophore-based virtual screening: a review of recent applications. Expert Opin Drug Discov 5(3):205–222. https://doi. org/10.1517/17460441003592072 58. Halperin I, Ma B, Wolfson H, Nussinov R (2002) Principles of docking: an overview of search algorithms and a guide to scoring functions. Proteins 47(4):409–443. https://doi. org/10.1002/prot.10115 59. Leach AR (2001) Molecular modelling: principles and applications. Prentice Hall, New York 60. Jiang X, Lu H, Li J et al (2020) A natural BACE1 and GSK3β dual inhibitor Notopterol effectively ameliorates the cognitive deficits in APP/PS1 Alzheimer’s mice by attenuating amyloid-β and tau pathology. Clin Transl Med 10(3):e50. https://doi.org/10.1002/ ctm2.50 61. Youn K, Jun M (2019) Biological evaluation and docking analysis of potent BACE1 inhibitors from Boesenbergia rotunda. Nutrients 11(3):662. https://doi.org/10.3390/ nu11030662 62. Tran T-S, Tran T-D, Tran T-H et al (2020) Synthesis, in silico and in vitro evaluation of

Recent Advances in Computational Modeling of BACE1 Inhibitors. . . some flavone derivatives for acetylcholinesterase and BACE-1 inhibitory activity. Molecules 25(18):4064. https://doi.org/10.3390/ molecules25184064 63. Youn K, Yoon J-H, Lee N et al (2020) Discovery of sulforaphane as a potent BACE1 inhibitor based on kinetics and computational studies. Nutrients 12(10):3026. https://doi. org/10.3390/nu12103026 64. El-Hawary SS, Sayed AM, Issa MY et al (2021) Anti-Alzheimer chemical constituents of Morus macroura Miq.: chemical profiling, in silico and in vitro investigations. Food Funct 12(17):8078–8089. https://doi.org/ 10.1039/D1FO01177D 65. Coimbra JRM, Baptista SJ, Dinis TCP et al (2020) Combining virtual screening protocol and in vitro evaluation towards the discovery of BACE1 inhibitors. Biomol Ther 10(4): 5 3 5 . h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / biom10040535 66. Panyatip P, Tadtong S, Sousa E, Puthongking P (2020) BACE1 inhibitor, neuroprotective, and neuritogenic activities of melatonin derivatives. Sci Pharm 88(4):58. https://doi.org/ 10.3390/scipharm88040058 67. Ugbaja SC, Lawal IA, Abubakar BH et al (2022) Allostery inhibition of BACE1 by psychotic and meroterpenoid drugs in Alzheimer’s disease therapy. Molecules 27(14): 4 3 7 2 . h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / molecules27144372 68. Markwick PR, Pierce LC, Goodin DB, McCammon JA (2011) Adaptive accelerated molecular dynamics (Ad-AMD) revealing the molecular plasticity of P450cam. J Phys Chem Lett 2(3):158–164. https://doi.org/10. 1021/jz101462n 69. Miao Y, Nichols SE, Gasper PM et al (2013) Activation and dynamic network of the M2 muscarinic receptor. Proc Natl Acad Sci U S A 110(27):10982–10987. https://doi.org/ 10.1073/pnas.1309755110 70. Hayward S, Kitao A, Go ¯ N (1995) Harmonicity and anharmonicity in protein dynamics: a normal mode analysis and principal component analysis. Proteins Struct Funct Genet 23(2):177–186. https://doi.org/10.1002/ prot.340230207 71. Chen J, Yin B, Wang W, Sun H (2020) Effects of disulfide bonds on binding of inhibitors to β-amyloid cleaving enzyme 1 decoded by multiple replica accelerated molecular dynamics simulations. ACS Chem Neurosci 11(12): 1811–1826. https://doi.org/10.1021/ acschemneuro.0c00234

93

72. Miao Y, Feher VA, McCammon JA (2015) Gaussian accelerated molecular dynamics: unconstrained enhanced sampling and free energy calculation. J Chem Theory Comput 11(8):3584–3595. https://doi.org/10. 1021/acs.jctc.5b00436 73. Miao Y, McCammon JA (2016) Graded activation and free energy landscapes of a muscarinic G-protein-coupled receptor. Proc Natl Acad Sci U S A 113(43):12162–12167. h t t p s : // d o i . o r g / 1 0 . 1 0 7 3 / p n a s . 1614538113 74. Iserloh U, Wu Y, Cumming JN et al (2008) Potent pyrrolidine- and piperidine-based BACE-1 inhibitors. Bioorg Med Chem Lett 18(1):414–417. https://doi.org/10.1016/j. bmcl.2007.10.116 75. Fujimoto K, Matsuoka E, Asada N et al (2019) Structure-based design of selective β-site amyloid precursor protein cleaving enzyme 1 (BACE1) inhibitors: targeting the flap to gain selectivity over BACE2. J Med Chem 62(10):5080–5095. https://doi.org/ 10.1021/acs.jmedchem.9b00309 76. Wu Y-J, Guernon J, Shi J et al (2016) Discovery of S3-truncated, C-6 heteroaryl substituted aminothiazine β-site APP cleaving enzyme-1 (BACE1) inhibitors. J Med Chem 59(18):8593–8600. https://doi.org/10. 1021/acs.jmedchem.6b01012 77. Chen J, Zhang S, Wang W et al (2021) Binding of inhibitors to BACE1 affected by pH-dependent protonation: an exploration from multiple replica Gaussian accelerated molecular dynamics and MM-GBSA calculations. ACS Chem Neurosci 12(14): 2591–2607. https://doi.org/10.1021/ acschemneuro.0c00813 78. Sun X, Wang Y, Qing H et al (2005) Distinct transcriptional regulation and function of the human BACE2 and BACE1 genes. FASEB J 19(7):739–749. https://doi.org/10.1096/ fj.04-3426com 79. Li S, Zhao H, Li J et al (2021) A series of molecular modeling techniques to reveal selective mechanisms of inhibitors to β-site amyloid precursor protein cleaving enzyme 1 (BACE1) and β-site amyloid precursor protein cleaving enzyme 2 (BACE2). J Biomol Struct Dyn 39(8):2824–2837. https://doi. org/10.1080/07391102.2020.1754917 80. Johansson P, Kaspersson K, Gurrell IK et al (2018) Toward β-secretase-1 inhibitors with improved isoform selectivity. J Med Chem 61(8):3491–3502. https://doi.org/10. 1021/acs.jmedchem.7b01716 81. Eketjall S, Janson J, Kaspersson K et al (2016) AZD3293: a novel, orally active BACE1

94

Konstantinos D. Papavasileiou et al.

inhibitor with high potency and permeability and markedly slow off-rate kinetics. J Alzheimers Dis 50(4):1109–1123. https://doi.org/ 10.3233/JAD-150834 82. Jeppsson F, Eketjall S, Janson J et al (2012) Discovery of AZD3839, a potent and selective BACE1 inhibitor clinical candidate for the treatment of Alzheimer disease. J Biol Chem 287(49):41245–41257. https://doi. org/10.1074/jbc.M112.409110 83. Jabir NR, Rehman MT, Alsolami K et al (2021) Concatenation of molecular docking and molecular simulation of BACE-1, γ-secretase targeted ligands: in pursuit of Alzheimer’s treatment. Ann Med 53(1): 2332–2344. https://doi.org/10.1080/ 07853890.2021.2009124 84. Kushwaha P, Singh V, Somvanshi P et al (2021) Identification of new BACE1 inhibitors for treating Alzheimer’s disease. J Mol Model 27(2):58. https://doi.org/10.1007/ s00894-021-04679-3 85. Adewole KE, Ishola AA (2021) BACE1 and cholinesterase inhibitory activities of compounds from Cajanus cajan and Citrus reticulata: an in silico study. In Silico Pharmacol 9(1):14. https://doi.org/10.1007/s40203020-00067-6 86. Ahuja A, Tyagi PK, Tyagi S et al (2021) Potential of Pueraria tuberosa (Willd.) DC. to rescue cognitive decline associated with BACE1 protein of Alzheimer’s disease on Drosophila model: an integrated molecular modeling and in vivo approach. Int J Biol Macromol 179:586–600. https://doi.org/ 10.1016/j.ijbiomac.2021.03.032 87. Othman A, Sayed AM, Amen Y, Shimizu K (2022) Possible neuroprotective effects of amide alkaloids from Bassia indica and Agathophora alopecuroides: in vitro and in silico investigations. RSC Adv 12(29): 18746–18758. https://doi.org/10.1039/ D2RA02275C 88. Falade AO, Adewole KE, Ishola AA et al (2022) Computational studies on the cholinesterase, beta-secretase 1 (BACE1) and monoamine oxidase (MAO) inhibitory activities of endophytes-derived compounds: towards discovery of novel neurotherapeutics. J Biomol Struct Dyn:1–15. https://doi.org/ 10.1080/07391102.2022.2035255 89. do Bomfim MR, Barbosa DB, de Carvalho PB et al (2022) Identification of potential human beta-secretase 1 inhibitors by hierarchical virtual screening and molecular dynamics. J Biomol Struct Dyn:1–15. https://doi.org/10. 1080/07391102.2022.2069155

90. Lipinski CA (2004) Lead- and drug-like compounds: the rule-of-five revolution. Drug Discov Today Technol 1(4):337–341. https://doi.org/10.1016/j.ddtec.2004. 11.007 91. Veber DF, Johnson SR, Cheng HY et al (2002) Molecular properties that influence the oral bioavailability of drug candidates. J Med Chem 45(12):2615–2623. https://doi. org/10.1021/jm020017n 92. Nhung NT, Duong N, Phung HTT et al (2022) In silico screening of potential β-secretase (BACE1) inhibitors from VIETHERB database. J Mol Model 28(3): 60. https://doi.org/10.1007/s00894-02205051-9 93. Nadh AG, Revikumar A, Sudhakaran PR, Nair AS (2022) Identification of potential lead compounds against BACE1 through in-silico screening of phytochemicals of Medhya rasayana plants for Alzheimer’s disease management. Comput Biol Med 145:105422. https://doi.org/10.1016/j.compbiomed. 2022.105422 ¨ zkay Y et al (2022) 94. Tok F, Sag˘lık BN, O Design, synthesis, biological activity evaluation and in silico studies of new nicotinohydrazide derivatives as multi-targeted inhibitors for Alzheimer’s disease. J Mol Struct 1265:133441. https://doi.org/10. 1016/j.molstruc.2022.133441 95. Rahman MA, Shuvo AA, Bepari AK et al (2022) Curcumin improves D-galactose and normal-aging associated memory impairment in mice: in vivo and in silico-based studies. PLoS One 17(6):e0270123. https://doi. org/10.1371/journal.pone.0270123 96. Gupta S, Parihar D, Shah M et al (2020) Computational screening of promising betasecretase 1 inhibitors through multi-step molecular docking and molecular dynamics simulations - Pharmacoinformatics approach. J Mol Struct 1205:127660. https://doi.org/ 10.1016/j.molstruc.2019.127660 97. Nino H, Garcia-Pintos I, Rodriguez-Borges JE et al (2011) Review of synthesis, biological assay and QSAR studies of beta-secretase inhibitors. Curr Comput Aided Drug Des 7(4): 263–275. https://doi.org/10.2174/ 157340911798260322 98. Vijayan RS, Prabu M, Mascarenhas NM, Ghoshal N (2009) Hybrid structure-based virtual screening protocol for the identification of novel BACE1 inhibitors. J Chem Inf Model 49(3):647–657. https://doi.org/10. 1021/ci800386v 99. Nastase AF, Boyd DB (2012) Simple structure-based approach for predicting the

Recent Advances in Computational Modeling of BACE1 Inhibitors. . . activity of inhibitors of beta-secretase (BACE1) associated with Alzheimer’s disease. J Chem Inf Model 52(12):3302–3307. https://doi.org/10.1021/ci300331d 100. Das S, Chakraborty S, Basu S (2019) Hybrid approach to sieve out natural compounds against dual targets in Alzheimer’s disease. Sci Rep 9(1):3714. https://doi.org/10. 1038/s41598-019-40271-9 101. Iwaloye O, Elekofehinti OO, Momoh AI et al (2020) In silico molecular studies of natural compounds as possible anti-Alzheimer’s agents: ligand-based design. Netw Model Anal Health Inform Bioinform 9(1):54. https://doi.org/10.1007/s13721-02000262-7 102. Zeng X, Zhang P, He W et al (2018) NPASS: natural product activity and species source database for natural product research, discovery and tool development. Nucleic Acids Res 46(D1):D1217–D1222. https://doi.org/ 10.1093/nar/gkx1026 103. Das S, Majumder T, Sarkar A et al (2020) Flavonoids as BACE1 inhibitors: QSAR modelling, screening and in vitro evaluation. Int J Biol Macromol 165:1323–1330. https://doi. org/10.1016/j.ijbiomac.2020.09.232 104. Mukerjee N, Das A, Jawarkar RD et al (2022) Repurposing food molecules as a potential BACE1 inhibitor for Alzheimer’s disease. Front Aging Neurosci 14. https://doi.org/ 10.3389/fnagi.2022.878276 105. Tran T-S, Le M-T, Tran T-D et al (2020) Design of curcumin and flavonoid derivatives with acetylcholinesterase and beta-secretase inhibitory activities using in silico approaches. Molecules 25(16):3644. https://doi.org/10. 3390/molecules25163644 106. Garcı´a Marı´n ID, Camarillo Lo´pez RH, Martı´nez OA et al (2022) New compounds from heterocyclic amines scaffold with multitarget inhibitory activity on Aβ aggregation, AChE, and BACE1 in the Alzheimer disease. PLoS One 17(6):e0269129. https://doi.org/10. 1371/journal.pone.0269129 107. Cheng F, Li W, Zhou Y et al (2012) admetSAR: a comprehensive source and free tool for assessment of chemical ADMET properties. J Chem Inf Model 52(11):3099–3105. https://doi.org/10.1021/ci300367a 108. Cheng F, Li W, Zhou Y et al (2019) Correction to “admetSAR: a comprehensive source and free tool for assessment of chemical ADMET properties”. J Chem Inf Model 59(11):4959. https://doi.org/10.1021/acs. jcim.9b00969

95

109. Hasan MM, Khan Z, Chowdhury MS et al (2022) In silico molecular docking and ADME/T analysis of Quercetin compound with its evaluation of broad-spectrum therapeutic potential against particular diseases. Inform Med Unlocked 29:100894. https:// doi.org/10.1016/j.imu.2022.100894 110. Varsou D-D, Tsoumanis A, Afantitis A, Melagraki G (2020) Enalos cloud platform: nanoinformatics and cheminformatics tools. In: Roy K (ed) Ecotoxicological QSARs. Springer US, New York, pp 789–800. https://doi.org/10.1007/978-10716-0150-1_31 111. Enalos Chemoinformatics Cloud Platform: BACE ligand-based predictive model. NovaMechanics Ltd. http://www.enaloscloud. novamechanics.com/EnalosWebApps/ BACE/ 112. Enalos Cloud Platform. NovaMechanics Ltd. https://novamechanics.com/services-tools/ enalos-cloud-platform/ 113. Tetko IV, Klambauer G, Clevert D-A et al (2022) Artificial intelligence meets toxicology. Chem Res Toxicol 35(8):1289–1290. https://doi.org/10.1021/acs.chemrestox. 2c00196 114. Chen H, Engkvist O, Wang Y et al (2018) The rise of deep learning in drug discovery. Drug Discov 23(6):1241–1250. https://doi. org/10.1016/j.drudis.2018.01.039 115. Singh R, Ganeshpurkar A, Ghosh P et al (2021) Classification of beta-site amyloid precursor protein cleaving enzyme 1 inhibitors by using machine learning methods. Chem Biol Drug Des 98(6):1079–1097. https:// doi.org/10.1111/cbdd.13965 116. Dhamodharan G, Mohan CG (2022) Machine learning models for predicting the activity of AChE and BACE1 dual inhibitors for the treatment of Alzheimer’s disease. Mol Divers 26(3):1501–1517. https://doi.org/ 10.1007/s11030-021-10282-8 117. Miyazaki Y, Ono N, Huang M et al (2020) Comprehensive exploration of target-specific ligands using a graph convolution neural network. Mol Inf 39(1–2):1900095. https:// doi.org/10.1002/minf.201900095 118. Gao K, Nguyen DD, Tu M, Wei GW (2020) Generative network complex for the automated generation of drug-like molecules. J Chem Inf Model 60(12):5682–5698. https://doi.org/10.1021/acs.jcim.0c00599 119. Szabo A, Ostlund NS (2012) Modern quantum chemistry: introduction to advanced electronic structure theory. Dover Publications, New York

96

Konstantinos D. Papavasileiou et al.

120. Tsuneda T (2014) Density functional theory in quantum chemistry. Springer, Tokyo 121. Koch W, Holthausen MC (2015) A CHEMIST’S guide to density functional theory. Wiley, New York 122. Jensen F (2016) Introduction to computational chemistry, 3rd edn. Wiley, New Jersey 123. Frush EH, Sekharan S, Keinan S (2017) In silico prediction of ligand binding energies in multiple therapeutic targets and diverse ligand sets—a case study on BACE1, TYK2, HSP90, and PERK proteins. J Phys Chem B 121(34):8142–8148. https://doi.org/10. 1021/acs.jpcb.7b07224 124. Arif N, Subhani A, Hussain W, Rasool N (2020) In silico inhibition of BACE-1 by selective phytochemicals as novel potential inhibitors: molecular docking and DFT studies. Curr Drug Discov Technol 17(3): 397–411. https://doi.org/10.2174/ 1570163816666190214161825 125. Pettus LH, Bourbeau MP, Bradley J et al (2020) Discovery of AM-6494: a potent and orally efficacious β-site amyloid precursor protein cleaving enzyme 1 (BACE1) inhibitor with in vivo selectivity over BACE2. J Med Chem 63(5):2263–2281. https://doi.org/ 10.1021/acs.jmedchem.9b01034 126. Neumann U, Ufer M, Jacobson LH et al (2018) The BACE-1 inhibitor CNP520 for prevention trials in Alzheimer’s disease. EMBO Mol Med 10(11):e9316. https:// doi.org/10.15252/emmm.201809316 127. Machauer R, Lueoend R, Hurth K et al (2021) Discovery of umibecestat (CNP520): a potent, selective, and efficacious β-secretase (BACE1) inhibitor for the prevention of Alzheimer’s disease. J Med Chem 64(20): 15262–15279. https://doi.org/10.1021/ acs.jmedchem.1c01300 128. Dapprich S, Koma´romi I, Byun KS et al (1999) A new ONIOM implementation in Gaussian98. Part I. The calculation of energies, gradients, vibrational frequencies and electric field derivatives. J Mol Struct THEOCHEM 461-462:1–21. https://doi.org/10. 1016/S0166-1280(98)00475-8 129. Ugbaja SC, Sanusi ZK, Appiah-Kubi P et al (2021) Computational modelling of potent β-secretase (BACE1) inhibitors towards Alzheimer’s disease treatment. Biophys Chem 270:106536. https://doi.org/10.1016/j. bpc.2020.106536 130. Gnanaraj C, Sekar M, Fuloria S et al (2022) In silico molecular docking analysis of Karanjin against Alzheimer’s and Parkinson’s diseases as a potential natural lead molecule for new

drug design. Dev Ther Mol 27(9):2834. h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / molecules27092834 131. Lakhera S, Devlal K, Rana M, Celik I (2022) Study of nonlinear optical responses of phytochemicals of Clitoria ternatea by quantum mechanical approach and investigation of their anti-Alzheimer activity with in silico approach. Struct Chem 34:1–16. https:// doi.org/10.1007/s11224-022-01981-5 132. Ivanova L, Karelson M, Dobchev DA (2020) Multitarget approach to drug candidates against Alzheimer’s disease related to AChE, SERT, BACE1 and GSK3β protein targets. Molecules 25(8):1846. https://doi.org/10. 3390/molecules25081846 133. Sayed E, Haj-Ahmad R, Ruparelia K et al (2017) Porous inorganic drug delivery systems—a review. AAPS PharmSciTech 18(5):1507–1525. https://doi.org/10. 1208/s12249-017-0740-2 134. Xie Z, Zhao J, Wang H et al (2020) Magnolol alleviates Alzheimer’s disease-like pathology in transgenic C. elegans by promoting microglia phagocytosis and the degradation of betaamyloid through activation of PPAR-γ. Biomed Pharmacother 124:109886. https:// doi.org/10.1016/j.biopha.2020.109886 135. Xian Y-F, Qu C, Liu Y et al (2020) Magnolol ameliorates behavioral impairments and neuropathology in a transgenic mouse model of Alzheimer’s disease. Oxidative Med Cell Longev 2020:5920476. https://doi.org/10. 1155/2020/5920476 136. Santos J, Quimque MT, Liman RA et al (2021) Computational and experimental assessments of magnolol as a neuroprotective agent and utilization of UiO-66(Zr) as its drug delivery system. ACS Omega 6(38): 24382–24396. https://doi.org/10.1021/ acsomega.1c02555 137. Syeda T, Cannon JR (2021) Environmental exposures and the etiopathogenesis of Alzheimer’s disease: the potential role of BACE1 as a critical neurotoxic target. J Biochem Mol Toxicol 35(4):e22694. https://doi.org/10. 1002/jbt.22694 138. Brown-Leung JM, Cannon JR (2022) Neurotransmission targets of per- and polyfluoroalkyl substance neurotoxicity: mechanisms and potential implications for adverse neurological outcomes. Chem Res Toxicol 35(8): 1312–1333. https://doi.org/10.1021/acs. chemrestox.2c00072 139. OECD (2018) Toward a new comprehensive global database of per- and polyfluoroalkyl substances (PFASs): summary report on updating the OECD 2007 list of per- and

Recent Advances in Computational Modeling of BACE1 Inhibitors. . . polyfluoroalkyl substances (PFASs). Series on Risk Management, no. 39. Paris, France 140. Subramanian G, Ramsundar B, Pande V, Denny RA (2016) Computational modeling of β-secretase 1 (BACE-1) inhibitors using ligand based approaches. J Chem Inf Model 56(10):1936–1949. https://doi.org/10. 1021/acs.jcim.6b00290 141. Cheng W, Ng CA (2019) Using machine learning to classify bioactivity for 3486 per-

97

and polyfluoroalkyl substances (PFASs) from the OECD list. Environ Sci Technol 53(23): 13970–13980. https://doi.org/10.1021/ acs.est.9b04833 142. Humphrey W, Dalke A, Schulten K (1996) VMD: visual molecular dynamics. J Mol Graph Model 14(1):33–38. https:// doi.org/10.1016/0263-7855(96)00018-5

Chapter 4 Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents Thamires Quadros Froes, Deyse Brito Barbosa, Mayra Ramos do Bomfim, Franco Henrique Andrade Leite, and Marcelo Santos Castilho Abstract Since BACE-1’s role in Alzheimer’s disease (AD) etiology was elucidated, inhibitors of this enzyme have been pursued as primary targets for the treatment of this ailment. Despite several hurdles have been overcome and several BACE-1 inhibitors progressed to clinical trials, none has reached the market until now. The synergy of ligand- and structure-based molecular modeling tools proved essential to gain a deeper understanding of BACE-1 conformational flexibility, its allosteric regulation, as well as identify novel molecular scaffolds that explore cryptic and allosteric sites in BACE-1 or have improved BACE-1/ BACE-2 selectivity profile. This chapter showcases recent contributions to the above achievements and highlights the challenges ahead. Key words BACE-1, Alzheimer, Modeling, CADD, Virtual screening

1

Introduction Alzheimer’s disease (AD) is responsible for 70% of dementia cases worldwide, affecting approximately 40 million people [1]. In 2019 it costed a trillion dollars to the US healthcare system, and the amount required to treat AD patients is expected to double by 2030 [2]. Despite AD’s economic and social impacts, currently available drugs only deal with AD symptoms, such as behavioral disturbances and difficulties in performing daily activities [3], as they focus on inhibiting cholinesterase enzymes and N-methyl-Daspartate receptor [4, 5]. Although such an approach positively impacts the patients’ quality of life, they do not cure nor prevent the disease progression. Thus, there is a clear need to develop disease-modifying therapies for the underlying disease pathogenesis. According to the amyloid hypothesis [6], the pathological hallmarks of AD are due to the presence of amyloid plaques, which are produced by the proteolytic cleavage of amyloid precursor protein

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_4, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

99

100

Thamires Quadros Froes et al.

(APP) into β-amyloid peptide (Aβ) [7, 8]. Since BACE-1’s role in APP processing was elucidated, inhibitors of this enzyme have been pursued as primary targets for the treatment of AD. Initial efforts relied on peptidomimetics [9], such as OM99-2, but the low bioavailability and blood-brain barrier permeability issues this class of compounds faced hindered their use in vivo. Computational tools were instrumental in overcoming such limitations and guiding scaffold hopping to nonpeptide-like BACE-1 inhibitors that progressed to the clinical trial (Table 1) [10]. Although many of those drug candidates were considered safe in phase 1 clinical trials, their effectiveness in phase 3 was not satisfactory [11]. Moreover, some BACE-1 inhibitors negatively impacted the patients’ cognition [5]. Aiming to overcome the poor efficacy/toxicity profile of previous BACE-1 inhibitors, it is essential to improve compounds’ BACE-1/BACE-2 selectivity profile and gain a deeper understanding of BACE-1 allostery, as this information can guide the development of novel drug candidates. Another approach is to develop dual-acting drugs (multitarget drug design for Alzheimer’s disease). Either way, computational tools will be instrumental in surpassing the limitations found in BACE-1 inhibitors developed so far. For that reason, following a brief description of computational tools that have been exploited in this endeavor (Subheading 2), this chapter will showcase significant contributions to BACE-1 inhibitors’ design (Subheading 3) since 2016.

2

Methods

2.1 Structure- and Ligand-Based Virtual Screening Tools

Virtual screening has become a promising tool for discovering lead compounds since it reduces drug development time and costs [12] and allows large libraries of compounds to be screened, and the most promising compounds to be prioritized for biological evaluation [13]. Virtual screening approaches can be guided either by ligand-based (e.g., pharmacophore models) or structure-based (e.g., molecular docking) information.

2.1.1 Pharmacophore Models

As long as a few bioactive compounds with a common mechanism of action (and similar binding profile) are known, pin-pointing equivalent stereo-electronic requirements shared by all of them leads to the identification of pharmacophore features. Those features can be exploited to build pharmacophore models, which will make it possible to select compounds with dissimilar scaffolds that still satisfy the pharmacophore restraints and, thus, are expected to bind the same target [14]. Compared to molecular docking (next subheading 2.1.2), the main advantage of pharmacophore models is avoiding the entropic and enthalpic terms required to calculate the binding energy. A simple superposition of putative hits to pharmacophore features, often depicted as spheres, is used instead.

MW (g/mol) 437.4

513.8

Chemical structure

(Elenbecestat)

CNP-520 (Umibecestat)

2.9

1.4

2

2

12

9

3

4

103

128

LogP HBD HBA RB PSA

Physical-chemical properties

Table 1 BACE-1 inhibitors that entered clinical trials between 2015 and 2022

Phase 2/3

Phase 3

Maximum phase

Lack of efficacy

Lack of efficacy

Failure reason

(continued)

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents 101

0.6

2

409.4

367.4

MK-8931 (Verubecestat)

JNJ-54861911 (Atabecestat)

3

2

2

1

6

7

4

3

3

3

130

126

72.9

LogP HBD HBA RB PSA

412.5

MW (g/mol)

Physical-chemical properties

AZD-3293 (Lanabecestat)

Chemical structure

Table 1 (continued)

Phase 2/3

Phase 3

Phase 3

Maximum phase

Liver toxicity

Lack of efficacy

Lack of efficacy

Failure reason

102 Thamires Quadros Froes et al.

–

– 390.4

320.4

LY-3323795a

LY-2886721

LY-2811376

2.1

1.5

1.3

498.5

LY-3202626

1

2

–

2

6

7

–

11

2

3

–

5

89.5

115

–

57

Phase 1

Phase 2

Phase 1

Phase 2

Retinal toxicity in rats

Liver toxicity

(continued)

Negative effects on cognition

Negative effects on cognition

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents 103

–

–

Adapted from Rombouts et al. [10] a Structure not available

PF-05297909

–

–

BI-1181181/VTP-37948a

a

3.2

431.4

AZD-3839

–

–

1

–

–

7

–

–

4

–

–

77

LogP HBD HBA RB PSA

MW (g/mol)

Physical-chemical properties

Chemical structure

Table 1 (continued)

Phase 1

Phase 1

Phase 1

Maximum phase

Lack of central programmed death effects

Skin toxicity

QTc prolongation

Failure reason

104 Thamires Quadros Froes et al.

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents

105

However, it is already established that compounds fulfilling the pharmacophore requirements might not be active due to steric clashes related to auxophore moieties. Another limitation of ligand-based pharmacophore models is that bioactive conformations often are unknown [15]. To overcome such weaknesses, either structure-based pharmacophore models can be built, or docking tools must be employed to filter out compounds bumping to residues within the active site [16–19]. When structure-based pharmacophore models are built from the intermolecular interactions found in one or more crystallographic protein-ligand complex(es), the positions of interacting residues can be employed as restraints (“excluded volumes”) in the pharmacophore model. 2.1.2

Molecular Docking

2.2 Structure- and Ligand-Based Ligand Optimization Tools

Molecular docking is a computational strategy used to predict the binding mode of one ligand to its macromolecular target [20]. When this analysis is carried out for millions of ligands, we call it a virtual screening. In both cases, molecular docking can be split into conformational sampling and binding energy evaluation steps. In the first step, the macromolecule is held rigid or with flexibility restricted to a few selected residues. In contrast, the ligand conformational search is carried out by random (e.g., genetic and simulated annealing algorithms) or systematic (e.g., rotation of single bonds by a fixed angle, incremental search) methods [21, 22]. The lack of proper protein flexibility is considered a major limitation in predicting biologically significant binding modes [23]. The second step relies on force field-based, empirical, or knowledge-based equations to predict the binding energy of the ligand-macromolecule complex [24]. Yet, most of these scoring functions have little, if any, correlation to the biological property [25–28]. Improved correlations are achieved when continuous solvation models such as MM-PBSA [29–31], or inhomogeneous solvent models [32], and free-energy perturbation [33–35] are taken into account. However, the computational cost (time) required for such approaches reduces their use in most virtual screening campaigns [31]. Another major limitation of implicit solvation models is that they do not appropriately describe water molecules’ participation in complex stabilization [36]. Drug discovery and development is a very expensive and timeconsuming process [37, 38]. For that reason, computer-aided drug design (CADD) approaches have been widely employed to reduce the impact of traditional drug discovery experiments. Among the tools available to rationalize the molecular modifications required to improve the lead compound potency and selectivity profile quantitative structure-activity relationship (QSAR)

106

Thamires Quadros Froes et al.

models and molecular dynamics (MD) simulations have a significant impact on the development of BACE-1 inhibitors. On the other hand, the analysis of BACE-1 hot spots has not been fully exploited. 2.2.1

QSAR

Quantitative structure-activity relationship (QSAR) models strive to depict the chemical features that contribute to potency or selectivity through a mathematical equation, using different types of descriptors (1D, 2D, 3D, 4D), according to which QSAR models are classified. Hence, 2D QSAR methods rely on topological descriptors that are not influenced by the conformation of the ligands [39]. In fact, from a historical point of view, QSAR modeling is a ligand-based approach, as 3D structural information was very scarce in the 1960s when this approach was invented. On the other hand, molecular interaction field descriptors (3D descriptors) provide QSAR models (e.g., CoMFA, comparative molecular field analysis) that are deeply influenced by the molecular alignment of the ligands as well as their spatial arrangement [40]. Ideally, the molecular alignment would have all ligands in their bioactive conformation. However, low energy, stable conformations, or docked poses are also reasonable surrogates for the bioactive conformation [40].

2.2.2 Molecular Dynamics

To circumvent the limitations described in Subheading 2.1.2, molecular dynamics (MD) have been applied not only to account for water molecules within the binding site explicitly but also to explore how fast internal motions’ (e.g., side-chain rotamer shift), or even domain movements’, impact on ligand binding [41, 42]. Using Newton’s laws of motion [43], MD describes the conformational flexibility of the whole system (protein-protein or ligand-protein complexes) for a certain amount of time [44].

2.2.3 Hot Spot and Druggability Analysis

Hot spots are regions that can contribute significant amounts of binding energy [45] and can be identified experimentally through alanine scanning mutagenesis and crystallography techniques [46]. Another approach is the computational method of FTMap, in which small organic solvent probes are positioned on a dense grid around the surface of the protein, and favorably interacting probes are identified, energy-minimized, and clustered. Each cluster, ranked according to its average energy, contains different partially overlapping probes, thus forming a consensus site (CS) [47]. CS ≥13 probes are generally considered hot spots, and those with 7.0 kcal/mol cutoff led to the selection of 893 fragments, which were subjected to a second virtual screening with the extra precision (XP) score function. Among the bestranked compounds, those hydrogen-bonding to R307 (82 fragments) and capable of performing additional interactions at the secondary site of BACE-1 (6 fragments) were selected to be linked to huprine Y to develop dual-acting inhibitors that do not target the BACE-1 active site (Fig. 3c). 3.4 Hot Spot-Guided Optimization of BACE1 Inhibitors

Both virtual and high-throughput screening campaigns afford many hits against BACE-1, which require further potency or selectivity optimization to be considered as promising lead compounds. Even though docking provides several hints on how to increase steric and electronic complementarity, some compounds adopt different binding profiles after their groups are modified. It has been proposed that fragments overlapping strong hot spots are prone to keep their binding profile even in larger ligands [74]. On the other hand, Rathi and coworkers [75] proposed that whenever the same moiety, from different ligands occupy the same binding site at multiple structures, it lies in a hot or warm spot, depending on the fraction of ligands’ occupation at each position within the binding site; high-occupancy regions are

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents

115

considered hot spots, whereas low-occupancy regions are deemed “warm-spots.” Within the database of fragments that are grown into larger ligands, assembled by Wakefield and coworkers [76], there are two examples of fragments (2AQ and 8AP, A and B in Fig. 4) that were developed into lead-like compounds (A5 and B3, respectively), with conservation of the fragment binding profile within each series. Despite the apparent chemical similarity of the 2-amine naphthalene ring (A) and the 2-amine phenyl moiety of B, these rings are flipped within the BACE-1 active site. As a result, these two series have unique structure-activity relationships that render highpotency (i.e., A5 – pIC50 = 9.15) or low-potency (i.e., B3 pIC50 = 5.38) inhibitors. Although the information provided by hot spots analysis was not employed during the fragment-to-lead optimization steps, a retrospective analysis sheds light on the reasons for these results. It can provide novel insights on how to develop BACE-1 inhibitors with improved ligand efficiency. For instance, 2AQ and 8AP are located inside a druggable hot spot (A6 and B4, Fig. 4). Still, when these fragments are grown to larger compounds, they explore the hot spot differently: B series mainly explored the “south part” of the hot spot, and in doing so, there is a 100-fold improvement in the biological property. However, the analysis of the ligands regarding the hot spot occupation reveals that a fraction of the ligand lies outside the hot spot (fractional occupation – FO) (B5 and B6 in Fig. 4); when the moiety attached to 2AQ explores the south part of the hot spot (A2 and A7), a similar trend is observed. When the “north part” of the hot spot is also explored (A8 and A9), additional gain in potency is observed, despite a small fraction of the ligands remaining outside the hot spots. This sort of analysis cannot be employed to pinpoint which interactions are responsible for the potency gain. Still, it is quite useful to guide the fragment-to-lead optimization strategy. To fully exploit this approach, one must explain why the “south” is less promising than the “north” portion of the hot spot. One reasonable explanation was given by Hall and coworkers [50], according to which occupation of dense areas within the hot spots (density-correlation efficiency – DCE), calculated with FTMap,1 improves the compound’s ligand efficiency. In order to avoid compounds’ size-bias, the authors expressed the correlation they found as the ratio between the fragment-DCE (fragment-LE) and each analog-DCE (analog-LE). Using this strategy (Fig. 4C1), it can be seen that all ligands, but A2, follow a straight-line (r2 = 0.97) correlation. As there is some criticism regarding LE, Froes and coworkers [49] extended this analysis to the binding

1 Hot spot density is calculated as the number of molecular probes (organic solvents) within a defined radius from the ligand’s atoms.

116

Thamires Quadros Froes et al.

Fig. 4 Optimization BACE-1 inhibitors (A1–A9, B1–B6) and correlation between the ligand efficiency of 2AQ AND 8AP and its analogs versus occupancy of BACE-1 hot spots (C). (C) The plots show the correlation between both the ligand efficiency (LE) and the binding efficiency index (BEI) versus fractional overlap (FO)

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents

117

efficiency index (BEI), when they analyzed human DHODH inhibitors. In case this alternative efficiency metric is employed, a high correlation (r2 = 0.88) is still observed for the BACE-1 inhibitors reported in Fig. 4. Moreover, those authors suggested that FO would be a good surrogate for DCE. This appears to be the case here too (Fig. 4C2), as the correlation between Fragment-Fo and Analog-Fo ratio is high whether LE (r2 = 0.94) or BEI (r2 = 0.84) ratios are employed. 3.5 QSAR-Guided Optimization of BACE1 Inhibitors

According to the Organization for Economic Co-operation and Development [77] guidelines, QSAR models, whenever possible, must provide a mechanistic interpretation of the results. Hence, descriptor selection can be considered a vital step as long as it helps identify those variables most important for the response prediction [78]. The definition of suitable molecular descriptors to explain the biological activity, among the many thousands available [79], is considered by many researchers a key step in QSAR model development [80]. However, descriptor selection constitutes a nondeterministic polynomial time (NP-hard) problem. Consequently, any algorithm that solves this problem affords solutions that explore only a small fraction of the huge molecular descriptors combinatorial space available. To overcome this limitation, Ponzoni and coworkers [81] resorted to hybridized QSAR models, whose descriptor redundancy was eliminated by a backward elimination strategy. The best classificatory QSAR model shows some, but not all, molecular descriptors seen in the QSAR model published by Gupta and coworkers (H1e, H6m, and GGI7), along with molecular descriptors selected with DELPHOS (Mor31p, nCrs, and N-069) (Table 2). One drawback of this model is that 3D molecular descriptors were calculated for energy-minimized conformations. On the other hand, Palakurti and Vadrevu [82] relied on conformations that fit a five-feature pharmacophore model (acceptor, hydrophobic, hydrophobic, positive, aromatic ring – AHHPR) to develop a QSAR model for a small series (training set = 40 compounds) of structurally diverse BACE-1 inhibitors. The QSAR model displays reasonable statistical parameters (r2 = 0.90, r2 test set 0.77) and allows an easy interpretation of which fields have a positive (blue) or negative (red) contribution to potency (Fig. 5). Another study that exploits 2D-QSAR along with pharmacophores models to shed light on BACE-1 inhibition requirements was carried out by Kumar and coworkers [80] using a chemically diverse dataset containing 98 BACE-1 inhibitors, available from BindingDB database [84]. After the removal of salts and mixtures and chemical structure normalization, carried out with RDKit and biological data curation carried out with a KNIME workflow, K-Medoids clustering was employed to split the whole dataset into training (n = 76) and test (n = 22) sets. As expected, the

118

Thamires Quadros Froes et al.

Table 2 QSAR models for BACE-1 inhibitorsa

Model type

a

Dataset

Independent variables

Best model statistics parameters

External validation Y-random % CC 53%

Classificatory 126 High activity (IC50 < 1000 nM) 89 low activity (IC50 > 1000 nM)

Mor31p nCrs N-069 H1e H6m GGI7

CC 85% AUC 0.88

Classificatory 126 High activity (IC50 < 1000 nM) 89 low activity (IC50 > 1000 nM

H1e H6m GGI7 RDF080m

CC 79% AUC 0.82

Classificatory 653 active (IC50 < 100 nM) 825 inactive (IC50 > 100 nM)

E-state fingerprint

CC 75% AUC 0.81

Classificatory 653 active (IC50 < 100 nM) 825 inactive (IC50 > 100 nM)

Distance matrices from E-state fingerprint

CC 100% AUC 1.00

Regression

1478 compounds E-state fingerprint (pIC50 2.54–9.49)

r2 0.34

Regression

1478 compounds Extended fingerprint (pIC50 2.54–9.49)

r2 0.70

Regression

1478 compounds Distance matrices from extended fingerprint (pIC50 2.54–9.49)

r2 0.97 (SVM) r2 0.97 (complextree) r2 0.99 (GPR)

SE 0.96–1.00 SP 1.00

r2 0.97–1.00 r2 o > 0.97 q2 < 0.1

CC correct classification, Y-random %CC correct classification following random Y values permutation, AUC area under ROC curve, SE sensitivity, SP specificity, r2 o correlation coefficient when forcing an intercept value equal to zero, q2 correlation coefficient for models following random Y values permutation. Brief description of independent variables meaning: Mor31p-3D-MoRSe (Molecule Representation of Structures Based on Electron diffraction) of signal 31, weighted by atomic polarizabilities; nCrs atom-count descriptor calculated as the number of ring secondary C (sp3); and N-069 atom-centered fragment calculated by the ratio Ar-NH2/X-NH2, which represents the number of substructures in which a sp3 nitrogen atom is connected by a simple bond to an electronegative atom or an aromatic substituent; H1e and H6m are GETAWAY (Geometric Topology and Atom Weights Assembly) descriptors. The first represents the autocorrelation index of lag 1 weighted by Sanderson electronegativity, whereas the second represents the autocorrelation index of lag 6 weighted by atomic mass; GGI7 – 2D autocorrelation descriptor of order 7, weighted by the topological charge; RDF080m – radial distribution function – 080/weighted by atomic mass and E-state fingerprint, molecular fingerprint calculated with electrotopological-state indexes for the atoms in a molecule [82]. E-states values are influenced by the steric and electronic effects of the surrounding atoms

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents

119

Fig. 5 Schematic 2D representation of molecular interaction field with a positive (blue) or negative (red) contribution (a) Acceptor group. (b) Hydrophobic. (c) Positively charged. (d) Aromatic ring) to the potency of a diverse set of nanomolar BACE-1 inhibitors (n = 58, pIC50 7.45–8.72) [82], overlaid on the pharmacophore features (green sphere, hydrophobic center; red sphere, acceptor group; red ring, aromatic ring; blue sphere positive center). The molecular interaction fields show a good superposition to the pharmacophore features

training set was employed to build 2D-QSAR models (3PCs, r2 = 0.83 and r2 Pred = 0.85), which highlight the importance of (A) hydrogen bond donor atoms (e.g., OH, –NH2, –SH, etc.), descriptor “ETA_dEpsilon_D”; (B) the presence/absence of N-S fragment at the topological distance 3, descriptor “B03[N-S]”; (C) lone electron pairs forming resonance interactions with an aromatic system, descriptor “ETA_Beta_ns_d”; and (D) two atom-centered fragment descriptors: H-047 (number of H atoms attached to C1(sp3)/C°(sp2), where the superscript represents the formal oxidation number of a carbon atom, calculated as the sum of the conventional bond orders with electronegative atoms) and C-033 (number of the R–CH – X fragments in a molecule). Interestingly, the authors employed the test set to build a pharmacophore model (AHHR), which corroborates the importance of ETA_Beta_ns_d and C-033 descriptors to BACE-1 inhibitors, and has reasonable accuracy (73.43%), precision (65.38%), sensitivity (68%), and specificity (76.92%) to differentiate active (IC50 < 1000 nM) from less active (IC50 > 1000 nM) compounds.

120

Thamires Quadros Froes et al.

Zhang and coworkers [85] resorted to molecular docking to generate putative bioactive conformations that were employed “as it is” (alignment 1) or superposed on their maximum common substructure (alignment 2) to build CoMFA and CoMSIA models for 55 cyclic sulfone hydroxyethyl amine derivatives (training set 46, Test set 9). Although the best 3D QSAR models (alignment 2) display excellent statistical parameters for the training set (CoMFA r2 = 0.91 SEE = 0.21, CoMSIA r2 = 0.97, SEE 0.20), the performance of the CoMFA model for the test set is rather disappointing (r2 = 0.58). On the other hand, CoMSIA shows good predictive power (r2 = 0.78). These authors also compared their 3D QSAR models to hologram QSAR, a 2D QSAR approach. Strangely, the statistical parameters for the training set with alignment 1 (r2 = 0.98, SEE 0.17) are different from those obtained with alignment 2 (r2 = 0.94, SEE 0.24), despite the same parameters being employed for model generation (fragment distinction A/B/Ch/DA, fragment sizes 4–7, hologram length 257). Ruiz and Nieto [86] have shown that while classificatory/ qualitative QSAR models for BACE-1 inhibitors are highly dependent on the descriptor employed (support vector machine classification models with sensibility ranging from 0.74 (E-state fingerprint) to 0.84 (PubChem fingerprint) and specificity ranging from 0.75 (E-state fingerprint) to 0.87 (standard fingerprint), the use of relative distances, calculated from those same descriptors, affords models with improved statistics parameters (sensibility and specificity equal to 1.0 for any of the descriptors employed. However, this approach is not amenable to external validation, as the relative distance matrices for the training (TR) and test (TE) set have different dimensions (TR × TR and TE × TR, respectively) and, consequently, the information about the variables representing the relationship between the molecules belonging the test set is lost. Nevertheless, bootstrapping validation hints that using relative distance matrices affords models with sensibility and specificity higher than 0.95. Similar results are achieved with continuous QSAR models: fingerprint-based descriptors afford models with r2 ranging from 0.34 (E-state fingerprint) to 0.70 (standard fingerprint), whereas relative distance matrices afford models with r2 above 0.95, wherever descriptor is exploited to calculate the relative distance matrix. Y-scrambling (25 random permutations of the activity values) supports the correlation between the activity value of the molecules and the relative distances.

4

Conclusion and Perspectives The great number of papers dedicated to BACE-1 inhibitors development underscores the relevance of this target for AD. Besides exploiting virtual screening strategies to uncover inhibitors with novel molecular scaffolds, a great deal of effort has been dedicated

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents

121

to understanding how BACE-1/BACE-2 dissimilar flexibility can lead to drug candidates with improved toxicity profile. The identification of allosteric and cryptic pockets within BACE-1 provides new opportunities to modulate BACE-1 catalytic activity, whereas pharmacophore and QSAR models continuously shed light on steric and electronic features required for each class of BACE-1 inhibitor. Bringing together all this knowledge is crucial, but probably not enough, to develop BACE-1 inhibitors with a significant impact on AD patients. On the other hand, it should be sufficient to guide the development of dual-acting compounds with improved efficacy and suitable safety to be useful as disease-modifying agents. References 1. World Health Organization (2021) Dementia 2. Lynch C (2020) World Alzheimer Report 2019: attitudes to dementia, a global survey. Alzheimers Dement 16:38255. https://doi. org/10.1002/alz.038255 3. Vaz M, Silvestre S (2020) Alzheimer’s disease: recent treatment strategies. Eur J Pharmacol 887:173554. https://doi.org/10.1016/j. ejphar.2020.173554 4. Najafi Z, Mahdavi M, Saeedi M, KarimpourRazkenari E, Asatouri R, Vafadarnejad F, Moghadam FH, Khanavi M, Sharifzadeh M, Akbarzadeh T (2017) Novel tacrine-1,2,3-triazole hybrids: in vitro, in vivo biological evaluation and docking study of cholinesterase inhibitors. Eur J Med Chem 125:1200–1212. https://doi.org/10.1016/j.ejmech.2016. 11.008 5. Huang LK, Chao SP, Hu CJ (2020) Clinical trials of new drugs for Alzheimer disease. J Biomed Sci 27:1–13. https://doi.org/10. 1186/s12929-019-0609-7 6. Mohamed T, Shakeri A, Rao PPN (2016) Amyloid cascade in Alzheimer’s disease: recent advances in medicinal chemistry. Eur J Med Chem 113:258–272. https://doi.org/10. 1016/j.ejmech.2016.02.049 7. Mamelak M (2017) Energy and the Alzheimer brain. Neurosci Biobehav Rev 75:297–313. https://doi.org/10.1016/j.neubiorev.2017. 02.001 8. Youn K, Jun M (2019) Biological evaluation and docking analysis of potent BACE1 inhibitors from boesenbergia rotunda. Nutrients 11: 1 – 1 3 . h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / nu11030662 9. Atta A, Darwish K, Elgawish M, Moustafa S, Helal M (2021) Recent insight into BACE1 as a potential target for treatment of Alzheimer’s disease. Rec Pharm Biomed Sci 5:100–111.

https://doi.org/10.21608/rpbs.2021. 55758.1087 10. Rombouts F, Kusakabe K, Hsiao C-C, Gijsen HJM (2021) Small-molecule BACE1 inhibitors: a patent literature review (2011 to 2020). Expert Opin Ther Pat 31:25–52. https://doi.org/10.1080/13543776.2021. 1832463 11. Ugbaja SC, Lawal IA, Abubakar BH, Mushebenge AG, Lawal MM, Kumalo HM (2022) Allostery inhibition of BACE1 by psychotic and meroterpenoid drugs in Alzheimer’s disease therapy. Molecules 27. https://doi.org/ 10.3390/molecules27144372 12. Rizzuti B, Grande F (2020) Virtual screening in drug discovery: a precious tool for a stilldemanding challenge. Elsevier Inc. 13. Carpenter KA, Huang X (2018) Machine learning-based virtual screening and its applications to Alzheimer’s drug discovery: a review. Curr Pharm Des 24:3347–3358. https://doi. o r g / 1 0 . 2 1 7 4 / 1381612824666180607124038 14. Mascarenhas AMS, de Almeida RBM, de Araujo Neto MF, Mendes GO, da Cruz JN, dos Santos CBR, Botura MB, Leite FHA (2020) Pharmacophore-based virtual screening and molecular docking to identify promising dual inhibitors of human acetylcholinesterase and butyrylcholinesterase. J Biomol Struct Dyn:1–10. https://doi.org/10.1080/ 07391102.2020.1796791 15. Kutlushina A, Khakimova A, Madzhidov T, Polishchuk P (2018) Ligand-based pharmacophore modeling using novel 3D pharmacophore signatures. Molecules 23:1–14. https:// doi.org/10.3390/molecules23123094 16. Razzaghi-Asl N, Sepehri S, Ebadi A, Miri R, Shahabipour S (2015) Molecular docking and quantum mechanical studies on biflavonoid

122

Thamires Quadros Froes et al.

structures as BACE-1 inhibitors. Struct Chem 26:607–621. https://doi.org/10.1007/ s11224-014-0523-2 17. Liu S, Fu R, Zhou LH, Chen SP (2012) Application of consensus scoring and principal component analysis for virtual screening against β-secretase (BACE-1). PLoS One 7. https:// doi.org/10.1371/journal.pone.0038086 18. Manoharan P, Chennoju K, Ghoshal N (2015) Target specific proteochemometric model development for BACE1 – protein flexibility and structural water are critical in virtual screening. Mol BioSyst 11:1955–1972. https://doi.org/10.1039/C5MB00088B 19. Kumar A, Roy S, Tripathi S, Sharma A (2016) Molecular docking based virtual screening of natural compounds as potential BACE1 inhibitors: 3D QSAR pharmacophore mapping and molecular dynamics analysis. J Biomol Struct Dyn 34:239–249. https://doi.org/10.1080/ 07391102.2015.1022603 20. McNutt AT, Francoeur P, Aggarwal R, Masuda T, Meli R, Ragoza M, Sunseri J, Koes DR (2021) GNINA 1.0: molecular docking with deep learning. J Cheminform 13:1–20. https://doi.org/10.1186/s13321-02100522-2 21. Guedes IA, de Magalha˜es CS, Dardenne LE (2014) Receptor-ligand molecular docking. Biophys Rev 6:75–87. https://doi.org/10. 1007/s12551-013-0130-2 22. Fan J, Fu A, Zhang L (2019) Progress in molecular docking. Quant Biol 7:83–89. https://doi.org/10.1007/s40484-0190172-y 23. Crampon K, Giorkallos A, Deldossi M, Baud S, Steffenel LA (2022) Machine-learning methods for ligand–protein molecular docking. Drug Discov Today 27:151–164. https://doi. org/10.1016/j.drudis.2021.09.007 24. Pagadala NS, Syed K, Tuszynski J (2017) Software for molecular docking: a review. Biophys Rev 9:91–102. https://doi.org/10.1007/ s12551-016-0247-1 25. Wang B, Ng HL (2020) Deep neural network affinity model for BACE inhibitors in D3R Grand Challenge 4. J Comput Aided Mol Des 34:201–217. https://doi.org/10.1007/ s10822-019-00275-z 26. Kumar S, Kim MH (2021) SMPLIP-Score: predicting ligand binding affinity from simple and interpretable on-the-fly interaction fingerprint pattern descriptors. J Cheminform 13:1– 17. https://doi.org/10.1186/s13321-02100507-1 27. El Khoury L, Santos-Martins D, Sasmal S, Eberhardt J, Bianco G, Ambrosio FA, Solis-

Vasquez L, Koch A, Forli S, Mobley DL (2019) Comparison of affinity ranking using AutoDock-GPU and MM- GBSA scores for BACE-1 inhibitors in the D3R Grand Challenge 4. J Comput Aided Mol Des 33:1011– 1020. https://doi.org/10.1007/s10822019-00240-w.Comparison 28. Wang Z, Sun H, Yao X, Li D, Xu L, Li Y, Tian S, Hou T (2016) Comprehensive evaluation of ten docking programs on a diverse set of protein–ligand complexes: the prediction accuracy of sampling power and scoring power. Phys Chem Chem Phys 18:12964–12975. https://doi.org/10.1039/C6CP01555G 29. Thompson DC, Humblet C, JosephMcCarthy D (2008) Investigation of MM-PBSA rescoring of docking poses. J Chem Inf Model 48:1081–1091. https://doi. org/10.1021/ci700470c 30. Rastelli G, Rio AD, Degliesposti G, Sgobba M (2010) Fast and accurate predictions of binding free energies using MM-PBSA and MM-GBSA. J Comput Chem 31:797–810. https://doi.org/10.1002/jcc 31. Poli G, Granchi C, Rizzolio F, Tuccinardi T (2020) Application of MM-PBSA methods in virtual screening. Molecules 25:1–19. https:// doi.org/10.4018/978-1-60960-860-6.ch011 32. Balius TE, Fischer M, Stein RM, Adler TB, Nguyen CN, Cruz A, Gilson MK, Kurtzman T, Shoichet BK (2017) Testing inhomogeneous solvation theory in structurebased ligand discovery. Proc Natl Acad Sci U S A 114:E6839–E6846. https://doi.org/10. 1073/pnas.1703287114 33. Shivakumar D, Deng Y, Roux B (2009) Computations of absolute solvation free energies of small molecules using explicit and implicit solvent model. J Chem Theory Comput 5:919– 930. https://doi.org/10.1021/ct800445x 34. Michel J, Foloppe N, Essex JW (2010) Rigorous free energy calculations in structure-based drug design. Mol Inform 29:570–578. https://doi.org/10.1002/minf.201000051 35. Procacci P (2021) Methodological uncertainties in drug-receptor binding free energy predictions based on classical molecular dynamics. Curr Opin Struct Biol 67:127–134. https:// doi.org/10.1016/j.sbi.2020.08.001 36. Lohning AE, Levonis SM, Williams-Noonan B, Schweiker SS (2017) A practical guide to molecular docking and homology modelling for medicinal chemists. Curr Top Med Chem 1 7 : 2 0 2 3 . h t t p s : // d o i . o r g / 1 0 . 2 1 7 4 / 1568026617666170130110827 37. Marino M, Jamal Z, Siccardi MA (2022) Pharmaceutics, Treasure Island

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents 38. Cummings JL, Goldman DP, Simmons-Stern NR, Ponton E (2022) The costs of developing treatments for Alzheimer’s disease: a retrospective exploration. Alzheimers Dement 18:469– 477. https://doi.org/10.1002/alz.12450 39. Cherkasov A, Muratov EN, Fourches D, Varnek A, Baskin II, Cronin M, Dearden J, Gramatica P, Martin YC, Todeschini R, Consonni V, Kuz’Min VE, Cramer R, Benigni R, Yang C, Rathman J, Terfloth L, Gasteiger J, Richard A, Tropsha A (2014) QSAR modeling: where have you been? Where are you going to? J Med Chem 57: 4977–5010. https://doi.org/10.1021/ jm4004285 40. Kim KH, Greco G, Novellino E (1998) A critical review of recent CoMFA applications. Perspect Drug Discovery Des 12(14):257–315 41. Choi SB, Yap BK, Choong YS, Wahab H (2018) Molecular dynamics simulations in drug discovery. Encycl Bioinforma Comput Biol ABC Bioinforma 1–3:652–665. https:// doi.org/10.1016/B978-0-12-809633-8. 20154-4 42. Mortier J, Rakers C, Bermudez M, Murgueitio MS, Riniker S, Wolber G (2015) The impact of molecular dynamics on drug design: applications for the characterization of ligandmacromolecule complexes. Drug Discov Today 20:686–702. https://doi.org/10. 1016/j.drudis.2015.01.003 43. Namba AM, Da Silva VB, Da Silva CHTP ˜ es (2008) Dinaˆmica molecular: Teoria e aplicac¸o em planejamento de fa´rmacos. Eclet Quim 33: 1 3 – 2 4 . h t t p s : // d o i . o r g / 1 0 . 1 5 9 0 / S0100-46702008000400002 44. Maginn EJ, Elliott JR (2010) Historical perspective and current outlook for molecular dynamics as a chemical engineering tool. Ind Eng Chem Res 49:3059–3078. https://doi. org/10.1021/ie901898k 45. Vajda S, Whitty A, Kozakov D (2015) Fragments and hot spots in drug discovery. Oncotarget 6:18740–18741. https://doi.org/10. 18632/oncotarget.4968 46. Brenke R, Kozakov D, Chuang GY, Beglov D, Hall D, Landon MR, Mattos C, Vajda S (2009) Fragment-based identification of druggable “hot spots” of proteins using Fourier domain correlation techniques. Bioinformatics 25: 621–627. https://doi.org/10.1093/bioinfor matics/btp036 47. Kozakov D, Grove LE, Hall DR, Bohnuud T, Mottarella S, Luo L, Xia B, Beglov D, Vajda S (2015) The FTMap family of web servers for determining and characterizing ligand binding hot spots of proteins. Nat Protoc 10:733–755.

123

https://doi.org/10.1038/nprot.2015. 043.The 48. Teixeira O, Lacerda P, Froes TQ, Nonato MC, Castilho MS (2021) Druggable hot spots in trypanothione reductase: novel insights and opportunities for drug discovery revealed by DRUGpy. J Comput Aided Mol Des 35:871– 882. https://doi.org/10.1007/s10822-02100403-8 49. Froes TQ, Zapata LCC, Akamine JS, Castilho MS, Nonato MC (2021) DHODH hot spots: an underexplored source to guide drug development efforts. Curr Top Med Chem 21: 2134–2154. https://doi.org/10.2174/ 1568026621666210804122320 50. Hall DR, Ngan CH, Zerbe BS, Kozakov D, Vajda S (2012) Hot spot analysis for driving the development of hits into leads in fragmentbased drug discovery. J Chem Inf Model 52: 1 9 9 – 2 0 9 . h t t p s : // d o i . o r g / 1 0 . 1 0 2 1 / ci200468p 51. Sarker SD, Nahar L, Miron A, Guo M (2020) Anticancer natural products, 1st edn. Elsevier Inc. 52. Wang Z, Sun H, Shen C, Hu X, Gao J, Li D, Cao D, Hou T (2020) Combined strategies in structure-based virtual screening. Phys Chem Chem Phys 22:3149–3159. https://doi.org/ 10.1039/C9CP06303J 53. Llanos MA, Gantner ME, Rodriguez S, Alberca LN, Bellera CL, Talevi A, Gavernet L (2021) Strengths and weaknesses of docking simulations in the SARS-CoV-2 era: the main protease (Mpro) case study. J Chem Inf Model 61:3758–3770. https://doi.org/10.1021/ acs.jcim.1c00404 54. Chen J, Wang J, Yin B, Pang L, Wang W, Zhu W (2019) Molecular mechanism of binding selectivity of inhibitors toward BACE1 and BACE2 revealed by multiple short molecular dynamics simulations and free-energy predictions. ACS Chem Neurosci 10:4303–4318. https://doi.org/10.1021/acschemneuro. 9b00348 55. Sabbah DA, Zhong HA (2016) Modeling the protonation states of β-secretase binding pocket by molecular dynamics simulations and docking studies. J Mol Graph Model 68:206– 215. https://doi.org/10.1016/j.jmgm.2016. 07.005 56. Razzaghi-Asl N, Karimi A, Ebadi A (2018) The potential of natural product vs neurodegenerative disorders: in silico study of artoflavanocoumarin as BACE-1 inhibitor. Comput Biol Chem 77:307–317. https://doi.org/10. 1016/j.compbiolchem.2018.10.015

124

Thamires Quadros Froes et al.

57. do Bomfim MR, Barbosa DB, de Carvalho PB, da Silva AM, de Oliveira TA, Taranto AG, Leite FHA (2022) Identification of potential human beta-secretase 1 inhibitors by hierarchical virtual screening and molecular dynamics. J Biomol Struct Dyn:1–15. https://doi.org/10. 1080/07391102.2022.2069155 58. Islam MA, Pillay TS (2018) B-secretase inhibitors for Alzheimer’s disease: identification using pharmacoinformatics. J Biomol Struct Dyn 37: 1–17. https://doi.org/10.1080/ 07391102.2018.1430619 59. Coimbra JRM, Baptista SJ, Dinis TCP, Silva MMC, Moreira PI, Santos AE, Salvador JAR (2020) Combining virtual screening protocol and in vitro evaluation towards the discovery of BACE1 inhibitors. Biomol Ther 10:1–22 60. Hilpert H, Guba W, Woltering TJ, Wostl W, Pinard E, Mauser H, Mayweg AV, RogersEvans M, Humm R, Krummenacher D, Muser T, Schnider C, Jacobsen H, Ozmen L, Bergadano A, Banner DW, Hochstrasser R, Kuglstatter A, David-Pierson P, Fischer H, Polara A, Narquizian R (2013) β-secretase (BACE1) inhibitors with high in vivo efficacy suitable for clinical evaluation in Alzheimer’s disease. J Med Chem 56:3980–3995. https:// doi.org/10.1021/jm400225m 61. Banner DW, Gsell B, Benz J, Bertschinger J, Burger D, Brack S, Cuppuleri S, Debulpaep M, Gast A, Grabulovski D, Hennig M, Hilpert H, Huber W, Kuglstatter A, Kusznir E, Laeremans T, Matile H, Miscenic C, Rufer AC, Schlatter D, Steyaert J, Stihle M, Thoma R, Weber M, Ruf A (2013) Mapping the conformational space accessible to BACE2 using surface mutants and cocrystals with Fab fragments, Fynomers and Xaperones. Acta Crystallogr Sect D Biol Crystallogr 69:1124– 1 1 3 7 . h t t p s : // d o i . o r g / 1 0 . 1 1 0 7 / S0907444913006574 62. Ostermann N, Eder J, Eidhoff U, Zink F, Hassiepen U, Worpenberg S, Maibaum J, Simic O, Hommel U, Gerhartz B (2006) Crystal structure of human BACE2 in complex with a hydroxyethylamine transition-state inhibitor. J Mol Biol 355:249–261. https://doi.org/10. 1016/j.jmb.2005.10.027 63. Tan JZA, Gleeson PA (2019) The role of membrane trafficking in the processing of amyloid precursor protein and production of amyloid peptides in Alzheimer’s disease. Biochim Biophys Acta Biomembr 1861:697–712. https:// doi.org/10.1016/j.bbamem.2018.11.013 64. Grant BJ, Skjærven L, Yao XQ (2021) The Bio3D packages for structural bioinformatics. Protein Sci 30:20–30. https://doi.org/10. 1002/pro.3923

65. Fujimoto K, Yoshida S, Tadano G, Asada N, Fuchino K, Suzuki S, Matsuoka E, Yamamoto T, Yamamoto S, Ando S, Kanegawa N, Tonomura Y, Ito H, Moechars D, Rombouts FJR, Gijsen HJM, Kusakabe KI (2021) Structure-based approaches to improving selectivity through utilizing explicit water molecules: discovery of selective β-secretase (BACE1) inhibitors over BACE2. J Med Chem 64:3075–3085. https://doi.org/10.1021/acs.jmedchem. 0c01858 66. Johansson P, Kaspersson K, Gurrell IK, B€ack E, Eketj€all S, Scott CW, Cebers G, Thorne P, McKenzie MJ, Beaton H, Davey P, Kolmodin K, Holenz J, Duggan ME, Budd Haeberlein S, Bu¨rli RW (2018) Toward β-secretase-1 inhibitors with improved isoform selectivity. J Med Chem 61:3491–3502. https://doi.org/10.1021/acs.jmedchem. 7b01716 67. Kornacker MG, Lai Z, Witmer M, Ma J, Hendrick J, Lee VG, Riexinger DJ, Mapelli C, Metzler W, Copeland RA (2005) An inhibitor binding pocket distinct from the catalytic active site on human β-APP cleaving enzyme. Biochemistry 44:11567–11573. https://doi.org/ 10.1021/bi050932l 68. Gutie´rrez M, Vallejos GA, Corte´s MP, Bustos C (2019) Bennett acceptance ratio method to calculate the binding free energy of BACE1 inhibitors: theoretical model and design of new ligands of the enzyme. Chem Biol Drug Des 93:1117–1128. https://doi.org/10. 1111/cbdd.13456 69. Egbert M, Jones G, Collins MR, Kozakov D, Vajda S (2022) FTMove: a web server for detection and analysis of cryptic and allosteric binding sites by mapping multiple protein structures. J Mol Biol 434:167587. https:// doi.org/10.1016/j.jmb.2022.167587 70. Wang W, Liu Y, Lazarus RA (2013) Allosteric inhibition of BACE1 by an exosite-binding antibody. Curr Opin Struct Biol 23:797–805. https://doi.org/10.1016/j.sbi.2013.08.001 71. Gutie´rrez LJ, Andujar SA, Enriz RD, Baldoni HA (2014) Structural and functional insights into the anti-BACE1 Fab fragment that recognizes the BACE1 exosite. J Biomol Struct Dyn 32:1421–1433. https://doi.org/10.1080/ 07391102.2013.821024 ˜ oz72. Di Pietro O, Juarez-Jimenez J, Mun Torrero D, Laughton CA, Javier Luque F (2017) Unveiling a novel transient druggable pocket in BACE-1 through molecular simulations: conformational analysis and binding mode of multisite inhibitors. PLoS One 12:1–

Modeling of BACE-1 Inhibitors as Anti-Alzheimer’s Agents 22. https://doi.org/10.1371/journal.pone. 0177683 73. Le Guilloux V, Schmidtke P, Tuffery P (2009) Fpocket: an open source platform for ligand pocket detection. BMC Bioinf 10:1–11. https://doi.org/10.1186/1471-210510-168 74. Kozakova D, Hall DR, Jehle S, Luo L, Ochiana SO, Jones EV, Pollastri M, Allen KN, Whitty A, Vajda S (2015) Ligand deconstruction: why some fragment binding positions are conserved and others are not. Proc Natl Acad Sci U S A 112:E2585–E2594. https://doi.org/10. 1073/pnas.1501567112 75. Rathi PC, Ludlow RF, Hall RJ, Murray CW, Mortenson PN, Verdonk ML (2017) Predicting “hot” and “warm” spots for fragment binding. J Med Chem 60:4036–4046. https://doi.org/10.1021/acs.jmedchem. 7b00366 76. Wakefield AE, Yueh C, Beglov D, Castilho MS, Kozakov D, Keseru¨ GM, Whitty A, Vajda S (2020) Benchmark sets for binding hot spot identification in fragment-based ligand discovery. J Chem Inf Model 60:6612–6623. https://doi.org/10.1021/acs.jcim.0c00877 77. Organisation for Economic Co-operation and Development (OECD) (2004). (Q)SARs on the Principles for the Validation of (Q)SARs (OECD Series on Testing and Assessment No. 49). OECD Publishing. Available in: https://one.oecd.org/document/env/jm/ mono(2004)24/en/pdf 78. De P, Kar S, Ambure P, Roy K (2022) Prediction reliability of QSAR models: an overview of various validation tools. Arch Toxicol 96: 1279–1295. https://doi.org/10.1007/ s00204-022-03252-y 79. Danishuddin, Khan AU (2016) Descriptors and their selection methods in QSAR analysis: paradigm for drug design. Drug Discov Today 21:1291–1302. https://doi.org/10.1016/j. drudis.2016.06.013 80. Kumar V, Ojha PK, Saha A, Roy K (2020) Exploring 2D-QSAR for prediction of beta-

125

secretase 1 (BACE1) inhibitory activity against Alzheimer’s disease. SAR QSAR Environ Res 31:87–133. https://doi.org/10.1080/ 1062936X.2019.1695226 81. Ponzoni I, Sebastia´n-Pe´rez V, Martı´nez MJ, Roca C, De la Cruz Pe´rez C, Cravero F, Vazquez GE, Pa´ez JA, Dı´az MF, Campillo NE (2019) QSAR classification models for predicting the activity of inhibitors of beta-secretase (BACE1) associated with Alzheimer’s disease. Sci Rep 9:1–13. https://doi.org/10.1038/ s41598-019-45522-3 82. Palakurti R, Vadrevu R (2017) Pharmacophore based 3D-QSAR modeling, virtual screening and docking for identification of potential inhibitors of β-secretase. Comput Biol Chem 68: 107–117. https://doi.org/10.1016/j. compbiolchem.2017.03.001 83. Hall LH, Mohney B, Kier LB (1991) The electrotopological state: structure information at the atomic level for molecular graphs. J Chem Inf Comput Sci 31:76–82. https://doi.org/ 10.1021/ci00001a012 84. Gilson MK, Liu T, Baitaluk M, Nicola G, Hwang L, Chong J (2016) BindingDB in 2015: a public database for medicinal chemistry, computational chemistry and systems pharmacology. Nucleic Acids Res 44:D1045– D1053. https://doi.org/10.1093/nar/ gkv1072 85. Zhang S, Lin Z, Pu Y, Zhang Y, Zhang L, Zuo Z (2017) Comparative QSAR studies using HQSAR, CoMFA, and CoMSIA methods on cyclic sulfone hydroxyethylamines as BACE1 inhibitors. Comput Biol Chem 67:38–47. https://doi.org/10.1016/j.compbiolchem. 2016.12.008 ´ (2018) 86. Luque Ruiz I, Go´mez-Nieto MA QSAR classification and regression models for β-secretase inhibitors using relative distance matrices. SAR QSAR Environ Res 29:355– 383. https://doi.org/10.1080/1062936X. 2018.1442879

Chapter 5 Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents Priyanka De and Kunal Roy Abstract Alzheimer’s disease (AD) is a progressive neurodegenerative disease distinguished by memory loss, cognitive dysfunction, impaired functional abilities, and behavioral changes. Being the most common form of senile dementia, AD can be characterized by the presence of two types of neuropathological hallmarks: neurofibrillary tangles (NFTs) and senile plaques (SP). The phosphorylation of tau is controlled and regulated by a group of kinase and phosphatase enzymes, making their systemic balance to be an important issue. Disruption of this equilibrium leads to tau hyperphosphorylation followed by tau aggregation. Inhibition of specific tau kinases, therefore, is a potential strategy to reverse tau pathology. However, new drug discovery comes with its own challenges involving high cost of experimentation, resources, and manpower. Thus, to combat these drawbacks, computational methods like pharmacophore modeling, molecular docking, molecular dynamic (MD) simulation, binding energy analysis, and QSAR are used for screening and prediction of new targets. Besides these, such techniques allow the application of available structural information for generating novel molecules contributing to the rational design of inhibitors. In the present book chapter, we have extensively reviewed different tau kinases, their systemic roles, and mechanism of tau phosphorylation relevant to cause AD. Also, the chapter encompasses different computational studies carried out in the last 4 years on various protein kinases in search of potential antiAlzheimer’s agents. Key words Alzheimer’s disease (AD), Tau protein, Tau phosphorylation, Protein kinases inhibitors, In silico modeling, Molecular docking, Molecular docking simulation

1

Introduction Alzheimer’s disease (AD) is the most common form of senile dementia characterized by two neuropathological hallmarks: extracellular deposition of neuritic plaques and intracellular neurofibrillary tangles (Fig. 1) [1]. AD is distinguished by progressive loss of cognitive functions, loss of memory, behavioral changes, and loss of functional abilities. According to the World Health Organization (WHO), currently, more than 55 million people live with dementia

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_5, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

127

128

Priyanka De and Kunal Roy

Fig. 1 The physiological structure of the brain and neurons in healthy brain and Alzheimer’s disease (AD) brain (with Aβ plaques and NFTs)

worldwide, and there are nearly ten million new cases occurring every year [2]. This number is expected to reach over 150 million by 2050 [3, 4]. AD can broadly be classified based on the onset age of the disease or occurring out of any hereditary changes or genetic mutation as (i) early-onset AD and (ii) late-onset AD. Familial AD (FAD) or early-onset AD is a result of hereditary mutations and represents almost 2% of reported cases, which sometimes occur as early as 40 years of age. Sporadic AD is more prevalent which is further subdivided into early-onset and late-onset types [1, 5, 6]. In FAD, the mutation of three genes, namely, amyloid precursor protein (APP, chromosome 21), presenilin 1 (PS1, chromosome 14), and presenilin 2 (PS2, chromosome 1), triggers amyloid beta generation and neurofibrillary tangle (NFT) formation. In sporadic or late-onset AD, a relevant number of patients are found to be a carrier of e4 allele of a lipid transport protein apolipoprotein E gene (ApoE gene, chromosome 19) [7]. The understanding of AD etiology is an essential criterion to effectively diagnose and treat the disease. Although the exact cause of AD is still unknown, a number of hypotheses have been reviewed. These include the amyloid beta (Aβ) hypothesis, tau hypothesis, cholinergic hypothesis, mitochondrial cascade hypothesis, dendritic hypothesis, metabolic hypothesis, and another that implicate oxidative stress and neuroinflammation [8]. The amyloid hypothesis suggests the generation, aggregation, and deposition of Aβ peptides (Aβ 1–42) triggering the primary event in AD neurodegeneration and neurotoxicity [6, 9]. Tau hypothesis is characterized by hyperphosphorylation of tau protein leading to the formation of insoluble paired helical

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

129

filaments (PHF) which form neurofibrillary tangles (NFTs). Thus, tau-centered treatments are promising targets to inhibit the phosphorylation and/or aggregation of the tau protein [10]. Currently, only five treatment options are approved to treat the cognitive symptoms of AD. Four of these five standard-of-care treatments providing symptomatic relief include three cholinesterase inhibitors (donepezil, galantamine, and rivastigmine) and one N-methyl-Daspartate receptor antagonist (memantine) [11–15]. Thus, it has been a priority to develop drugs that can slow down the progression of the disease. The conventional treatment of AD includes six categories of drugs acetylcholinesterase inhibitors (AchEI), N-methyl-D-aspartate (NMDA) receptor antagonists, monoamine oxidase inhibitors (MAOI), antioxidants, metal chelators, and anti-inflammatory agent [16]. However, scientists and researchers are focusing on new lead compounds targeting different regulatory enzymes, proteins, and pathways. In this regard, protein kinases are a growing drug target class for disorders in peripheral tissues, but the development of kinase-targeted therapies for central nervous system (CNS) diseases remains a challenge, largely owing to issues associated specifically with CNS drug discovery. Kinases belong to the enzyme group “transferases,” as they are responsible for transferring the phosphate group from high-energy donor molecules (e.g., adenosine triphosphate (ATP) or guanosine triphosphate (GTP)) to a definite substrate [17]. The role of different protein kinases in the AD treatment is exhaustively reported in various literature [18, 19]. Tau protein kinases described here are grouped into two main classes: proline-directed protein kinases (PDPK), and nonproline-directed protein kinases (non-PDPK). PDPK includes glycogen synthase kinase (GSK) and cyclin-dependent protein kinase (CDK), p38α mitogen-activated protein kinase (MAPK), and c-Jun N-terminal kinases (JNK). On the other hand, non-PDPKs include dual-specificity tyrosine phosphorylation-regulated kinase (DYRK), casein kinase, calcium calmodulin-activated protein kinase II (CaM kinase II), and protein kinase A (PKA). 1.1 Tau Protein and Microtubule Destabilization

Tau protein belongs to a crucial group of proteins referred to as microtubule-associated proteins (MAPs), which play a major role in stabilizing the microtubules in the axon. The microtubules play an indispensable function as the structural and functional backbone of the neurons [20]. The prime role of tau proteins is to provide microtubule (MT) stabilization in the distal regions of the neurons primarily at the axon and dendrites [21]. The tau protein has additional functions to perform, viz., engaging with enzymes and structures like RNA and presenilin 1 (PS1) [22, 23]. Abnormal aggregation and accumulation of these proteins in the distal regions are the most common pathologies allied with various neurodegenerative diseases mainly AD. Microtubules play a pivotal role in various cellular processes cell division, cell morphogenesis, and

130

Priyanka De and Kunal Roy

Fig. 2 Sequential process of formation of neurofibrillary tangles (NFTs) through the generation of dimer, trimer, oligomer, and paired helical filaments

other intracellular transporting and trafficking [24]. It also functions as a mediator for organelle transport mainly at the axonal junction assisted by motor proteins like kinesin (plus-end-directed) and dynein (minus-end-directed) accelerating the signal transduction at the neuronal ends. The stability of MTs is ensured by the tau protein which is composed of a carboxy-terminal (C-terminal) on one end, and the other end of the protein is composed of two zones, i.e., a positively charged proline residue and acidic amino acid terminal (N-terminal). The tau protein attaches itself through its tubulin binding residue with the MT to provide its structural configuration [25, 26]. Tau phosphorylation at a normal level controls its action to bind to the MTs and encourage their assembly. However, in case of increased modification of tau, i.e., in a hyperphosphorylated state (as in the case of AD), the biological activity of tau is lost [27]. Hyperphosphorylation modulates the charge and conformation of the tau protein, exposing its MT binding domain ultimately leading to accumulation, oligomerization, and selfaggregation [28, 29]. In terms of tau protein phosphorylation potential, it has been reported that the longest variant of tau (440 amino acids) holds about 80 serine or threonine phosphorylation sites [30]. The abnormal hyperphosphorylation of tau is caused as a result of the upregulation of tau kinases or downregulation of tau phosphatases [31]. Hyperphosphorylation leads to tau aggregation with the formation of insoluble paired helical filaments (PHFs) and subsequently formation of neurofibrillary tangles (NFTs), hampering the binding of tau to MTs and neuronal death (Fig. 2). Protein kinases which are responsible for

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

131

hyperphosphorylation can be categorized into three groups: (a) proline-directed protein kinase (PDPK), (b) non-prolinedirected protein kinase (Non-PDPK), and (c) tyrosine protein kinases (TPK). 1.2 Proline-Directed Protein Kinases (PDPK) 1.2.1 Glycogen synthase kinase-3 (GSK-3)

1.2.2 Cyclin-Dependent Protein Kinases (CDK)

GSK-3 is a widely expressed, highly conserved serine/threonine protein kinase that regulates various physiological and pathological processes. It is considered as the key kinase responsible for the phosphorylation of glycogen synthase during glycogen accumulation thereby controlling glycogen metabolism, cell cycle regulation, and cell proliferation [32]. In mammalian cells, GSK-3 is encoded by two genes, namely, GSK-3α and GSK-3β, and among these two GSK-3β plays a vital role in AD pathogenesis. GSK-3β plays a crucial role in the pathogenesis of AD implying the generation of AD-associated pathological lesions including amyloid beta aggregation, tau fibril formation, and neuronal loss [33, 34]. In AD pathophysiology, tau hyperphosphorylation commences due to kinase-phosphatase imbalance in the brain ascertaining GSK-3β to be one of the key kinases involved [35, 36]. Inhibition of GSK-3 is controlled by phosphorylation at a serine residue near the N-terminal (Ser9 of GSK-3β and Ser21 of GSK-3α) [33]. On the activation of GSK3, the tyrosine site (Tyr216 of GSK-3β and Tyr279 of GSK-3α) will be autophosphorylated [37]. Overexpression of GSK-3β potentiates tau phosphorylation and disassembles from microtubules, leading to neuronal transport alterations and hippocampal neurodegeneration [38, 39]. Cyclin-dependent kinases (CDKs), a family of proline-directed serine-threonine kinases, have a central role in controlling various cell cycle events and also serve to integrate diverse growth regulatory signals. These are activated upon the association of a regulatory partner, mainly cyclins; however, non-cyclins proteins are also able to trigger its activity [40]. CDKs control different stages of cell cycle and transcription process. Among almost ten different types of CDKs present in the mammalian body, CDK5 is responsible for the phosphorylation of tau and its aggregation, leading to AD [41]. CDK5 possesses almost 60% homology to human CDKs like CDK 1, 2, 4, and 6 and cell division cycle kinase 2 (CDC2). In general, CDK5 is associated with the development of CNS and neuronal motility such as neuron migration and differentiation and memory consolidation. Under any pathological stimulus, CDK5 hyperactivation occurs leading to hyperphosphorylation of various substrates like tau and neurofilament, amyloid precursor protein (APP) [41]. Activation of CDK5 is controlled by two related protein cofactors, p35 and p39, which are particularly localized in the CNS. These two proteins are short-lived, have a short-life, and are susceptible to ubiquitin-mediated proteasome degradation, implicating CDK5 activity is firmly controlled over a specific time period. Among these two cofactors, p35 is more good as an

132

Priyanka De and Kunal Roy

Fig. 3 Deregulation mechanism of CDK5 enzyme through cleavage of p35

Fig. 4 Examples of CDK5 inhibitors

activator protein which is composed of 307 amino acids with 35 kDa mass. Cleavage of p35 into the truncated form p25 by calpain, a calcium dependent protease, delocalizes and deregulates CDK5 [42] (Fig. 3). This abnormality of CDK5 caused by p25 causes NFTs [43, 44]. CDK5 inhibitors like roscovitine, indirubin3′-oxime, and aloisine have found potential used in the treatment of AD (Fig. 4). 1.2.3 p38α MitogenActivated Protein Kinase (p38α MAPK)

The critical role of mitogen-activated protein kinases (MAPKs) in controlling cellular mechanisms and in the diseases pathogenesis of AD is well recognized. MAPKs are a class of serine/threonine kinase which are activated by multiple protein kinases in response to extracellular stimuli by dual phosphorylation [45]. MAPKs are a group of three distinct family of proteins, the extracellular signal-

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

133

Fig. 5 p38α MAPK inhibitors

regulated protein kinases (ERKs), the c-Jun N-terminal kinases (JNKs), and the p38 MAPKs (p38α, p38β, p38δ, p38γ), and two upstream components, MAPK kinase and MAPK kinase kinase (MKKK). Activated MKKK functions in the phosphorylation of the next kinase, i.e., an MAPK kinase (MKK); the latter in turn phosphorylates a Thr-X-Tyr motif in the activation loop of MAPK. This explains the dual phosphorylation mechanism of MKK. MAPKs are the final kinases in the three-kinase module which phosphorylate substrates on serine and threonine residues [46]. MAPKs are known to be involved in controlling oxidative stress signaling and cell cycle, particularly in neuronal survival thereby contributing to AD pathogenesis. It is found that active ERKs are highly present in intracellular NFTs marking early tau deposition. p38α isoform of MAPKs have emerged as one of the most important target in CNS disease treatment. Increased neuronal p38α MAPK activation potentiates proinflammatory cytokine hyperproduction and acts as a possible driving force for the AD progression. p38α MAPK in the glial cells and synaptic junction contributes to the phosphorylation and abnormal functioning of tau [47]. p38α MAPK inhibitors can be classified in five main classes according to their location of binding. Some representative classes are shown in Fig. 5.

134

Priyanka De and Kunal Roy

Fig. 6 Common small molecule inhibitors of c-jun N-terminal kinase enzyme 1.2.4 c-Jun N-Terminal Kinases (JNK)

c-Jun N-terminal kinase (JNK) is a group of MAP kinase that was first characterized as stress-activated protein kinase. It has become a focus inhibitor in screening approaches owing to its important role in the regulation of inflammatory responses, cell proliferation, and apoptosis [48]. JNKs are encoded by three dissimilar genes, among which jnk1 and jnk2 are expressed ubiquitously, while jnk3 is selectively expressed in the brain [49]. JNK readily phosphorylates Thr205 and Ser422, which are more highly phosphorylated in Alzheimer’s disease compared with GSK-3 [50]. c-Jun acts as an inducible component of the activator protein-1 (AP-1) transcription factor, the main nuclear substrate of JNK, mediating its bipartite-fashioned response leading to neurodegeneration. Axonal injury leads to overexpression of c-Jun thus leading to hyperphosphorylation of tau caused by JNK [51]. Some common small molecule inhibitors of c-Jun N-terminal kinase are shown in Fig. 6.

1.3 Non-prolineDirected Kinases (NPDKs)

Non-proline-directed kinases include dual-specificity tyrosine phosphorylation and regulated kinase-1A/2 (DYRK1A/2), casein kinase 1α/1δ/1ε/2 (CK1α/1δ/1ε/2), Ca2+/calmodulin-dependent protein kinase II (CaMKII), and protein kinase A (PKA).

1.3.1 Dual-Specificity Tyrosine PhosphorylationRegulated Kinases (DYRKs)

“Dual-specificity Yak-related kinases” (DYRKs) are serine/threonine kinases, also known as dual-specificity tyrosine phosphorylation-regulated kinases, are related to the yak protein obtained from Saccharomyces cerevisiae. These kinases belong to the CMGC family of the eukaryotic kinome which includes cyclindependent kinases (CDKs), mitogen-activated protein kinases (MAP kinases), glycogen synthase kinases (GSKs), and Ccd2-like kinases (CLKs). DYRKs are further subclassified into two classes: (a) human class I, DYRK1A and DYRK1B, and (b) human class II,

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

135

DYRK2, DYRK3, and DYRK4 according to sequences outside the kinase domain [52–54]. The two classes share a common central domain and an adjacent N-terminal DYRK homology box but have different extended N- and C- terminal regions. Class I DYRKs are primarily located in the nucleus, whereas class II kinases are predominantly found in the cytoplasm. DYRK have the tendency to autophosphorylate an intramolecular tyrosine in their activation loop [53–55], and after activation they further function by phosphorylating the substrate on the serine-threonine amino acid residues [56]. Researchers identified that the hyperactivity of DYRK1A leads to neurodegeneration, cognitive dysfunction, abnormal brain maturation, and early-onset Alzheimer’s disease. DYRK1A hyperphosphorylates tau protein at Tyr 212 residues in the neuroblastoma cells which leads to the generation of neurofibrillary tangles. The “dual-specificity” nature of this kinase is due to phosphorylation of the tau protein at various serine or threonine residues and autophosphorylation at tyrosine 321 [57]. 1.3.2 (CKs)

Casein Kinases

1.3.3 Ca2+/CalmodulinDependent Protein Kinase II (CaMKII)

Casein kinases (CKs) are monomeric serine/threonine kinases, which are associated with diverse cellular mechanisms involving DNA repair, cell division, cellular signaling, vesicular trafficking, and membrane transport [58]. Broadly classified into types, casein kinase 1 (CK1) and casein kinase 2 (CK2) regulate the microtubule dynamics through tau phosphorylation [59]. Among the seven isoforms of casein kinase (CK), only α, δ, and ε isoforms could be involved in the pathology of AD [60]. In case of CK2, three isoforms (α, α’, and β) were identified which requires ATP or GTP as energy or as phosphate group donors [61]. CK1 has been suggested to have a greater role in tau pathology in AD brain. CK1δ expression is markedly upregulated in AD. As observed, the protein levels are elevated by 2.4-fold, 33-fold and ninefold in case of CK1α, CK1δ, and CK1ε, respectively [62]. CK1δ overphosphorylates tau at multiple sites disrupting microtubule binding. An overexpression of CK1ε contributes to tau pathology by dysregulation of phosphorylation and exon 10 splicing of tau [63]. These tau pathology mechanisms contributed by CKs make them a promising target for therapeutic intervention. Ca2+/calmodulin-dependent protein kinase II (CaMKII) is a broadly distributed metazoan Ser/Thr protein kinase that is important in neuronal signaling. Disruption of calcium homeostasis and downregulation of CaMKII coincide with pathological phosphorylation of tau in AD brains [64]. These dysfunctions elicit tau hyperphosphorylation and amyloid-β accumulation. These phenomena are further confirmed by a Drosophila model where CaMKII promoted neurodegeneration caused by tau phosphorylation at the Ser262/356 AD sites [65]. CaMKII also binds to N-methyl-Daspartic acid (NMDA) receptor when phosphorylated at the

136

Priyanka De and Kunal Roy

threonine residue [66]. CaMKII potentiates the induction of longterm potentiation of synaptic transmission (LTP), which regulates memory formation [67]. Autophosphorylated CaMKII can work in a Ca2+- independent manner, regulated by a negative feedback mechanism [68, 69]. Thus, hyperphosphorylation of intracellular tau and disruption of Ca2+ homeostasis may induce a selfperpetuating harmful loop to promote neurodegeneration being a major contributor of AD pathology. 1.3.4 Protein Kinase A (PKA)

Also known as cyclic AMP (cAMP)-dependent protein kinase, protein kinase A (PKA) is a tetrameric holoenzyme composed of two regulatory subunits and two catalytic subunits (R1α, R2 α, R1β, and R2β) [70]. Upon activation by cAMP, PKA holoenzyme is dissociated which triggers the translocation of fraction of the C subunit to the nucleus [71]. Two molecules of cAMP are required for the activation of PKA, and this cAMP-PKA complex binds to several serine and threonine residues of tau inhibiting its tubulin binding affinity [72]. It is found that 17 out 25 tau phosphorylation sites are targeted by PKA in AD brains [30, 73]. Computational approaches or in silico methods have shown great potential toward the design of new hits/leads for complex diseases by extensively facilitating the drug discovery process against individual targets [74]. The extensive use of these techniques has helped in the prediction of target chemicals by simulating their mechanism of action. The search and design of new compounds include computational approaches like homology modeling, molecular docking [75], molecular dynamics simulation [76], pharmacophore mapping [77], structure-activity relationship (SAR), and quantitative structure-activity relationship (QSAR) [78], machine learning [79], and a combination of these methods. All these in silico methods are elaborately explained in Chapter 7 of this book. In the current chapter, an exhaustive study of various in silico approaches relating to AD drug discovery has been addressed. The following sections contain various computational studies reported worldwide in the last few years (from 2019 onward) for the development of novel small molecules as kinase inhibitors for AD.

2 Computational Studies on Glycogen Synthase Kinase (GSK) and Its Inhibitors as Potential Anti-Alzheimer’s Agents Iwaloye et al. [80] intended to discover novel inhibitors targeted against GSK-3β from natural phytoconstituents of Melissa officinalis with the assistance of computational techniques. They have performed molecular docking analysis, induced fit docking, binding energy calculation studies using MM-GBSA method, and estimation of Lipinski’s rule of five (Fig. 7). Further, the

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

137

Fig. 7 Computational techniques used by Iwaloye et al. [80] to discover novel inhibitors targeted against GSK-3β from natural phytoconstituents of Melissa officinalis

phytochemicals’ pIC50 was predicted using machine learning algorithm using AutoQSAR. The crystal structure of GSK-3β with PDB ID: 1UV5, was procured from protein databank and preprocessed for molecular docking analysis. Prediction of binding affinity of the novel inhibitors was performed by using the induced fit docking (IFD) method employing Glide and Refinement module in Prime. It was documented that the docking score of hit compounds ranged from -7.289 to 17.284 kcal/mol. Melissa officinalis-derived phytochemical luteoin 3′-O-β-d-glucoronopyranoside(VI) (-17.284 kcal/mol) was found to attain highest bindingenergy with luteoin 7-O-β-d-glucoronopyranoside (V) (-17.199 kcal/mol), chlorogenic acid (-15.650 kcal/mol), and salvianolic acid F (-14.285 kcal/mol) following it. On the other hand, the co-crystallized structure of the protein attained the least binding energy, suggesting the hits derived from Melissa officinalis are potential targets against GSK-3β as GSK-3β inhibitors. Induced fit docking further allowed for more accurate prediction of binding affinity by protein rotation on binding to a ligand. Luteoin 3′-O-b-D-glucoronopyranoside(VI) continued to show most favorable binding with IFD binding score of -716.889 kcal/mol. In IFD analysis, naringenin (-716.819 kcal/mol) and apigenin (715.304 kcal/mol) showed more favorable interaction than other compounds. The most highly active compound luteoin 3′-O-β-dglucoronopyranoside(VI) was found to interact with Val135,

138

Priyanka De and Kunal Roy

Asn64, Lys183, and Gly68 of GSK-3β protein with hydrogen bond interaction. Other important interactions are listed in the source article. Validation of the docking score using MMGBSA analysis was affirmed by analyzing the lowest-energy poses predicted by scoring function. Surprisingly, the compounds with good docking score possessed favorable binding energy also in terms of different binding free energies like van der Waals (ΔGvdw), Coulomb interaction (ΔGColulomb), Hydrogen bond (ΔGHbond), and lipophilic energy (ΔGsolLipo). Furthermore, a rigorous screening process continued using machine learning-based prediction of pIC50 using AutoQSAR module of Schrodinger. This generated a promising model with goodness of fit R2 = 0.8467 and predictivity Q2 = 0.8069. Several compounds under this study showed good interaction and binding free energy; however, only three compounds including salvianolic acid C, ellagic acid, and naringenin surpassed satisfactory pIC50 values. Jiang et al. [81] designed and synthesized a series of novel pyridine thiazole-containing benzylpiperidine hybrids as dual inhibitors of GSK-3β/AChE. To better acknowledge the ligandprotein interactions of the designed compounds, they have performed molecular docking analysis using Discovery Studio software. One representative compound GD29, upon docking with GSK-3β, sits on the ATP catalytic site of GSK-3β, where it interacts with Lys85, Tyr134, Val135, and Cys199 amino acids of the enzyme. Further, intermolecular interaction forming hydrogen bonds between pyridine with Val135, and carboxamide with Lys85, Met101, and Phe201 gave crucial information for improving binding affinity. Moreover, MD simulation studies unveil that Lys85, Tyr134, Val135, and Cys199 amino acids play significant roles in the protein-ligand interactions. Amino acids Lys85, Lyr134, Val135, and Cys199 contributed significantly to the binding affinity with binding energies of -1.9, -2.2, -1.9, and -2.3 Kcal/mol, respectively. The most optimum candidate GD29 was further identified by in vitro pharmacokinetic assays. The inhibition of GSK-3β preventing tau hyperphosphorylation and NFT formation was studied by Eskandarzadeh et al. [82]. The research encompasses iridoid glycosides having a marked effect on improving memory and cognitive impairment thereby controlling Alzheimer’s disease. These phytochemicals were extracted from the common snowberry Symphoricarpos albus that consists of loganin, secologanin, and loganetin. In this work, AutoDock Vina 4.2 software was chosen for accurate prediction of binding sites of GSK-3β protein. The selected protein with PDB ID: 1UV5 contains a 6-bromine dirubio-3-oxime inhibitor (BRW1383) that inhibits the GSK-3β protein at ATP-binding site competitively. The most crucial amino acids identified that control

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

139

Fig. 8 Computational techniques used by Elangovan and her coworkers [83] to investigate the ability of small molecules to inhibit GSK-3β

the kinase behavior are ASP133, ASP200, and VAL135. The binding energies in the molecular docking of loganin, loganetin, and secologanin with GSK3-β protein were -7.15, -5.43, and 4.98 kcal/mol, respectively. It was found that loganin binds more toughly at the active site of the protein compared to other ligands as evidenced from the intermolecular interactions (hydrogen bonds) and the bond lengths. Loganin forms hydrogen bond interactions which were formed with Asn64, Asp200, and Val135, and hydrophobic (HYP) interactions were formed with Cys199, Asp133, Ala83, Val110, Leu18, Leu132, Val70, Lys85, Ile62, and Gly63 amino acid residues. MD simulation studies including RMSD, RMSF, Rg, SASA, hydrogen bond analyses, and MMPBSA suggested that loganin-GSK-3β complex is more stable than the other two complexes. Though loganin and loganetin possessed similar binding energy, the latter showed better drug likeliness. Elangovan and her coworkers [83] investigated the ability of small molecules to inhibit GSK-3β through virtual screening, ADME studies, induced fit docking (IFD), molecular dynamics simulation, and binding free energy calculation. The authors have screened three large databases: (a) ZINC database, 384,000 compounds; (b) Enamine database, 83,000; and (c) Ottava database: 150000. A meticulous study of molecular interactions between database compounds and GSK-3β protein was enacted with a rigorous screening protocol as explained in Fig. 8. The molecular docking results showed that two compounds 6966

140

Priyanka De and Kunal Roy

(N-[(2-fluorophenyl)methyl]-3-(1H-indol-3-yl) propanamide) and 6961 (N-benzyl-3-(1H-indol-3-yl) propanamide) bind at the ATP-binding domain of the receptor forming strong hydrogen bond interaction with Val135 and Asp133 amino acid residues. These compounds also exhibit a better docking score of 9.05 kcal/mol and -8.11 kcal/mol, respectively, than the GSK-3β II inhibitor (-6.73 kcal/mol). Root mean square deviation (RMSD) analysis from the MD simulation showed the docked complexes reached a stable state and have a good interaction with the protein residues. Further, RMSD and RMSF (root mean square fluctuation) were comparatively less throughout the simulation period as compared with the co-crystal. The main results point to the compounds (6966 and 6961) as potential ATP-competitive inhibitors for GSK-3β that could be developed to lessen the effects of AD by inhibiting GSK-3β at the active site. The selectivity mechanism of GSK3β inhibition at the molecular level was explained by Zhu et al. [84] by applying a systematic computational method, combining 3D-QSAR, molecular docking, molecular dynamic simulations, and free energy calculations. A dataset of 31 compounds was chosen for the development of the CoMFA model based on the GSK3β inhibitory activity. This model gave reliable predictive ability with a high linear correlation R2pred . The three-dimensional contour maps give potential ideas about the structural features essential for receptor binding. For understanding the binding mode, two types of molecular docking protocols were applied, i.e., the CDOCKER (CD) module and flexible docking (FD). These docking methods were further compared, and the FD protocol was found to be more accurate for further MD simulation studies. The dynamic binding process of studied GSK-3β inhibitors was studied using the RMSD variation of the complex structure concerning the initial structure skeleton atoms during the simulation process. The system was found to reach an equilibrium state within 10 ns having an average of 1.5 Å RMSD fluctuation. Some important features observed during the simulation period include (a) significantly strong non-polarity, (b) favorable van der Waals force and electrostatic terms for inhibitor binding to GSK3β, and (c) unfavorable polar solvation term for binding in all four complexes. Important inhibitors showing the best binding were found to have a similar pattern of interactions with the active site residues like Val70, Asp133, Tyr134, Val135, and Arg141 of GSK3β. Further, to comprehend the selectivity mechanism between GSK3β and cyclin-dependent kinase (CDK), the inhibitor specificity was determined using two complexes of the same inhibitor (Cpd23).

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

141

3 Computational Studies on Cyclin-Dependent Kinase-5 (CDK-5) Inhibitors as Potential Anti-Alzheimer’s Agents The hyperphosphorylation of tau proteins and the formation of plaques and tangles, which are presumed to be one of the primary causes of Alzheimer’s disease, are caused by the cyclin-dependent kinase (CDK5) and its activator p25 forming a stable complex. Hence, the pathological CDK5-p25 complex is an inspiring therapeutic target for AD. Tamareddy et al. [85] computationally studied the use of small peptides in inhibiting the pathological CDK5p25 complex effectively while crossing the blood-brain barrier. In this study, a 24-residue peptide, p5, has shown selective inhibition toward the pathological complex in vivo. The authors have studied the mechanism of inhibition of the studied peptide p5 using four binding modes (c1, c2, c3, c4) along with previously developed binding modes outside the CDK5-p25 binding pocket. The binding modes were almost stable throughout the 500 ns simulation except for c3 and c2_r2, which were removed from further analysis. Another important observation identified was the binding shift of c1 and c4 binding modes which might be due to sampling inaccuracies accompanying the binding prediction method. Conformational changes in the secondary structure of peptide were recognized. The c2 mode structure of peptide formed a coil shape in place of the helical structure without notably changing its binding with CDK5. Besides this, in modes c1 and c4, a reduction in the percentage of the peptide helical structure was identified. The authors have utilized three computational tools to understand the binding modes, viz., dynamic cross-correlation (DCC) maps, principal component analysis (PCA), and average root mean square fluctuation (RMSF). The DCC maps indicate the presence of allosteric effects in all binding modes except for c4_r1_a. PCA analysis was used to study collective motions. Along with DCC, the PCA analysis identified significant p5-induced dynamic motions in most of the binding modes, which were compatible with allosteric effects. The RMSF characterizing the fluctuations of all the residues about an average position identified allosteric effects in c1_r3, c2_r1, and c2_r3. From the stable binding modes identified, c2 mode is the only one without helical secondary structure for p5 allowing increased flexibility and more p5 residues to be available for binding with CDK5. Zeb et al. [86] have applied computational approaches and identified pyrrolidine-2,3-dione derivatives as novel inhibitors of Cdk5/p25. They have employed a pharmacophore model generated from previously known inhibitors of Cdk5 and ultimately used those as 3D queries in the virtual screening of drug-like databases to recognize competent drug-like hit candidates against the Cdk5/ p25. Furthermore, molecular docking, molecular dynamics

142

Priyanka De and Kunal Roy

simulation, and binding free energy calculations identified four pyrrolidine-2,3-dione derivatives as novel inhibitors of Cdk5/ p25. Pharmacophore modeling was performed showing five main features, viz., hydrogen bond acceptor (HBA), hydrogen bond donor (HBD), hydrophobic (HYP), hydrophobic aromatic (HYA), and ring aromatic (RAR), and the ten hypotheses generated had ranking scores of range from 55.82 to 75.17 kcal/mol. First four hypotheses exhibited a combination of two hydrogen bond acceptor (HBA) features, one hydrogen bond donor (HBD) feature, and one hydrophobic (HYP) feature. These combinations give a picture that the chosen candidates can make good interactions with CDK5. Molecular docking was performed using Cdk5/ p25 in complex with roscovitine (PDB ID: 1UNL) in the GOLD package. The inhibitors were found to form hydrogen bond interaction with either of Ile10, Phe80, Glu81, Phe82, Cys83, Gln85, Asp86, and Asn144 amino acid residues of CDK5. In total, a group of 12, 38, and 41 compounds were screened from the NCI, Asinex, and Specs databases, respectively. Successfully docked inhibitors were further used for MD simulation for 30 ns time scale. From MD simulation and binding free energy calculation, four best candidates were screened as good inhibitors of CDK5/p25 complex. An RMSD evaluation of Cα atoms and backbone atoms of Cdk5/p25 revealed that, as a complex, the candidate molecules were stable and exhibited the lowest RMSD. Further, the hit candidates (Hit compound 1–4) showed stable consistency of the H-bond formation with the Cys83 residue of CDK5 at the ATP-binding domain. It was found that Hit 1, Hit 2, and Hit 4 also interact with Asn144 residue making H-bond. Hydrophobic bonds are also formed mostly with Ile10, Gly11, Glu12, Gly13, Val18, Ala31, Lys33, Phe80, Phe82, Asp84, Gln85, Asp86, Lys89, and Leu133 amino acid residues. The authors have also performed binding free energy analysis using MM/PBSA approach which proposed that each candidate inhibitor had the lowest binding free energy with the CDK5/p25 complex supporting their strong binding. Thus, these newly identified four hit pyrrolidine-2,3dione derivatives (Fig. 9) may alleviate the tau-associated Alzheimer’s pathology through CDK5/p25 inhibition. Advani and Kumar [87] performed computational analysis of anticancer natural compounds against cyclin-dependent kinase-5 (CDK5) enzyme. The work involves the use of molecular docking studies, MD simulation, and ADMET profiling to determine the nature of binding to CDK5. The enzyme structure of CDK5 with PDB ID: 1UNL, was retrieved from the protein data bank, and it contained roscovitine as a bounded ligand. Molecular docking showed that this bound ligand interacts with CDK5 with a docking score -8.7 Kcal/mol and adjusts itself in a hydrophobic docking site of CDK5 containing Ile10, Gly11, Glu12, Gly13, Val18, Ala31, Lys33, Val64, Phe80, Phe82, Asp84, Gln85, Asp86,

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

143

Fig. 9 Identification of pyrrolidine-2,3-dione derivatives as CDK5/p25 inhibitors using computational approaches as proposed by Zeb et al. [86]. (Adapted from Zeb et al. [86])

Lys89, Gln130, Leu133, and Ala143 residues. Three natural compounds, apigenin (-8.5 kcal/mole), salvianolic acid (-9.3 kcal/ mole), and silymarin (-8.7 kcal/mole) showed good docking scores. Other two compounds, cryptotanshinone and tanshinone, obtained a good docking score but had no noteworthy interactions. These were then removed from further evaluation. MD simulation was performed for a span of 50 ns, and among all the 27 compounds used for MD simulation, three compounds apigenin, salvianolic acid, and silymarin were found to be the best. RMSD evaluation of three compounds showed steady systems with slight deviations not more than 0.3 nm. RMSF analysis showed varying fluctuations for the compounds with apigenin having the lowest. ADMET profiling of the three best compounds (i.e., apigenin, salvianolic acid, and silymarin) showed that physicochemical property values were slightly higher in the case of salvianolic acid and silymarin, but for apigenin, an acceptable score was obtained within the threshold range. Computational analysis suggested apigenin and salvianolic acid as potential CDK5 inhibitors for the management of AD pathophysiology. Garkani-Nejad and Ghanbari [88] applied molecular modeling techniques including 3D-QSAR and molecular docking to study triazolyl thiophenes as CDK5/P25 inhibitors. A dataset of 112 methyl-linked cyclohexyl thiophenes with triazole inhibitors was used for 3D-QSAR model generation with CoMFA and CoMSIA analyses. The models generated provide a high degree of

144

Priyanka De and Kunal Roy

confidence and strong prediction capability. For CoMFA, r 2 = 0:980, q 2 = 0:539, r 2pred = 0:968, and for CoMSIA, r 2 = 0:967, q 2 = 0:558, r 2pred = 0:945. The CoMSIA model generated was a combination of five descriptors including hydrophobic (H), hydrogen bond donor (D) and acceptor (A), steric (S), and electrostatic (E) fields. Thus, based on the CoMFA and CoMSIA model contour maps, particular regions for steric, electrostatic, hydrophobic, and H-bond interactions were recognized to augment the bioactivity. Molecular docking investigated the binding interactions of the highest active compound (N-(3-{[2-(2-chloroacetamido)-4,5,6,7-tetrahydro-1-benzothiophen-3-yl]methyl}-5[({N′-[(1Z)-phenylmethylidene]hydrazinecarbonyl}methyl)disulfanyl]-4H-1,2,4-triazol-4-yl)benzamide) to find important residues with two amino acids SER229 and LYS177 and four hydrogen bond interaction formation. Further, indole ring formed two interactions with SER229 and R1 position and two interactions with LYS177. These observations can help in the design of new potent CDK5/p25 inhibitors.

4 Computational Studies on p38α Mitogen-Activated Protein Kinase (MAPK) Inhibitors as Potential Anti-Alzheimer’s Agents El Aissouq et al. [89] applied computational approaches like 2D QSAR and molecular docking techniques to a novel series of 37 pyridinyl imidazole derivatives as p38α MAPK inhibitors. The prime aim was to find a correlation between the chemical structures and inhibitory activity of the p38α MAPK. Herein, a ligand-based drug-designing approach was employed by developing a 2D-QSAR model on the IC50 values of the pyridinyl imidazole series of compounds. The QSAR modeling was performed by using descriptors calculated from the Molecular Operating Environment (MOE) software. Internal and external validations were performed by cross-validation method and taking six compounds in the validation set. The developed QSAR model could explain 85% of the total variance in the training set and internal cross-validation yielded 77% of the determination coefficient (r 2cv ). External validation with six compounds in the test set resulted in 0.92 value for test set determination coefficient (r 2test ). The binding mode of these pyridinyl imidazole derivatives and p38MAPK receptor was understood by molecular docking. The most active compound (N-{4-[2-(benzylsulfanyl)-5-(4-fluorophenyl)-1H-imidazol-4-yl]pyridin-2-yl}-2(4-fluorophenyl)acetamide) was found to interact with a low binding energy (-10.6 kcal/mol) to form conventional hydrogen bond interactions with Lys 53, Gly 170, Asp 168, and His 148 amino acid residues. From the different features explained in the QSAR and docking methods, the authors have proposed few novel

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

145

pyridinyl imidazole derivatives keeping in mind the descriptors PEOE_RPC+ (the relative positive partial charge) and SMR_VSA2 (the sum of van der Waals surface area such that Ri (molar refractivity for atom “i” is in the range 0.26–0.35)). In addition to this, molecular docking emphasized the relevance of conventional hydrogen bond interactions to improve the inhibitory activity of the p38α MAPK. From these findings, the authors designed three new compounds whose predicted pIC50 was higher than the most active compound in the dataset. Moreover, the low binding energy in molecular docking reflect the stability of these compounds in the binding pocket of the p38α MAPK receptor. Khan et al. [90] performed computational analysis using a series of dibenzepinones, dibenzoxepines, and benzosuberones against p38α MAP kinase. They have utilized pharmacophore modeling, 3D-QSAR, and molecular docking in silico methods to understand the chemical features responsible for inhibitory action. A dataset of 67 compounds with IC50 values ranging from 0.003 to 6.80 μM (pIC50: 8.523 to 5.167) was used for the in silico studies. In pharmacophore analysis, the best fitted model (DDHRR.8) with R2 = 0.97687, Q2 = 0.9458 and F = 330 presented two aromatic rings, hydrogen bond donors, and one hydrophobic center. The interstitial site distances was a major attribute established in the alignment of pharmacophore hypothesis DDHRR.8 as evidenced from the highly active (pIC50 > 7.99) and inactive (pIC50 < 5.40) compounds. DDHRR.8 model was characterized by 6 PLS factor QSAR model whose validation criteria was further analyzed using important prediction and validation metrics. The authors have also checked the domain of applicability and performed randomization test. From contour map analysis, important features of the training compounds were identified including the hydrogen bond donor nature, hydrophobic character, and electron-withdrawing group. This work further included molecular docking and binding free energy analysis study by using MAP kinase enzyme with PDB ID: 3UVP. The docking score (Glide score) of most of the compounds was in agreement with the pIC50 values of the dataset compounds. The docking studies confirmed the molecular interactions were majorly contributed by hydrophobic and aromatic groups due to the presence of a phenyl ring and a thiophene moiety. The interacting residues at the active site of docking were Lys 53, Glu 71, Phe 169, and Asp 168. Hydrophobic interactions were found occurring at Val 105, Ala 51, Val 52, Leu 104, Leu 86, Ile 146, Leu 74, Leu 75, Met 78, Phe 169, Ile 84, Leu 167, Leu 108, Val 30, and Val 38 residues. It was observed in the compound shown in Fig. 10 that the phenyl and thiophene moiety was embedded in the hydrophobic pocket (activity-increasing feature), whereas the amide, ketone, and halogen groups were present in the hydrophilic pocket. Sulfur in the thiophene moiety is an important attribute that plays a major in ligand stabilization the receptor binding zone

146

Priyanka De and Kunal Roy

Fig. 10 Features responsible for p38α MAP kinase as identified by Khan et al. [90] Table 1 Important docking interactions observed by Khan et al. [90] for the most potent molecule (Fig. 9) Type of interactions

Ligand group

Interacting amino acid residue

Hydrophobic interactions

Phenyl and thiophene moieties

Val 105, Ala 51, Val 52, Leu 104, Leu 86, Ile 146, Leu 74, Leu 75, Met 78, Phe 169, Ile 84, Leu 167, Leu 108, Val 30, and Val 38

Polar interactions

Suberone ring

Ser 32, Ser 37, His 107, Thr 106, Asn 155, and Ser 154

Hydrogen bond CONH bridge interaction

Backbone of amino acid Asp 168 and the side chain of Glu 71

π-π interaction

Thiophene ring

Phe 169

Salt bridge

Phenyl ring

Lys 53

(Fig. 10). Suberone ring (cycloheptanone) interacts with Ser 32, Ser 37, His 107, Thr 106, Asn 155, and Ser 154 forming polar interactions due to the effect of solvent exposure. Some other interactions are listed in Table 1 below. Further, the authors have analyzed the molecular deviation over time through RMSD, RMSF, and other simulation parameters over a course of 10 ns trajectories. RMSD became stable at around 1.5 Å after 3 ns of simulation. Amino acid residues Glu 71, Met 109, Gly 110, and Asp 168 play important roles in p38 MAPK regulation with interaction fractions 0.6, 0.9, and 1.0, respectively. These findings will be immensely beneficial in AD drug development for dibenzepinone-, dibenzoxepine-, and benzosuberone-based p38α MAP kinase inhibitors. ˇ ivadinovic´ and coworkers [91] have applied several in silico Z techniques on pyridinylimidazole derivatives acting as dual inhibitors of p38α MAPK and C-Jun N-terminal kinase 3 (JNK3). They have collected a data of 60 pyridinylimidazoles derivatives with a

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

147

recognized inhibition effect on both JNK3 and p38 α MAPK. The Monte Carlo optimization method was used for calculating conformation independent models using optimal descriptors, derived both from a local graph and the SMILES notation invariants. They have combined genetic algorithm along with multiple linear regressions (MLRs) to obtain QSAR models based on different 0D, 1D, and 2D descriptors. A rigorous analysis of a variety of statistical parameters was performed to check the QSAR models’ robustness and predictive potential. From the QSAR models generated, important structural features improving the pIC50 were documented. The presence of carbon atom or a methyl group (“C............”) and the presence two connected carbon atoms or ethyl group (“C... C.......”) increased the pIC50 values. Other important features regulating (increasing) the pIC50 values were (a) methyl group addition to benzene leading to branching (“c...1...(...” or “c...C.......”), (b) oxygen atom (“O...........”), (c) methoxy group (“O...C.......”), and (d) hydroxyl/methoxy group bounded to benzene (“c...O... C...”). Other important SMILES descriptors signifying the presence of bromine and amine groups include: “Br..........”, “HALO00100000”, “++++ Br–N === ”, “++++ Br–O === ”, N...........”, “N...(.......”, “N...(...C...”,“c...N.......”, and “c...N... (....” Molecular docking of the highly active compounds showed good binding forming hydrogen bonds with ARG67, GLU71, and PHE169 amino acids at the active site. The molecular docking score ranged from -164.889 to -146.103. Further, these compounds were checked for their drug likeliness using SwissADME web service to predict ADME parameters, pharmacokinetic properties, the drug-like nature, and medicinal chemistry friendliness of the designed molecules in order to support drug discovery. The methodologies presented in this research can be put to use to develop new drugs for neurodegenerative diseases like AD. Shen et al. [92] performed a machine learning-based virtual screening to identify p38α inhibitors from a natural products library in search of novel drug lead scaffolds. They have collected a dataset of 4509 compounds against p38α from the ChEMBL, DUDE databases, and an in-house NP 42 library. The training dataset was passed through a similarity check to fit the chemical space of the NP library. Chemical space analysis was performed using the principle component analysis (PCA) method including features like molecular weight (MW), Log (ALogP), the topological polar surface area (TPSA), the number of hydrogen bond donors (HBDs), the number of hydrogen bond acceptors (HBA), and the number of rotatable bonds (RotB). Similarity screening was accomplished by the Tanimoto index (TI) based on Morgan fingerprints using RDKit. Machine learning approaches including k-nearest neighbors (kNN), random forest (RF), and support vector machine (SVM) with RBF kernel were employed for model generation. Six classifiers (kNN_2D, kNN_ECFP, RF_2D, RF_ECFP,

148

Priyanka De and Kunal Roy

21 SVM_2D, and SVM_ECFP) combining two sets of descriptors (2D molecular descriptors and Morgan extended-connectivity fingerprints (ECFP6)) were built. In the kNN method, the number of neighbors affected the performance of kNN_2D more than that of kNN_ECFP. The optimized number of neighbors (k) was set to k = 8 for modeling. In case of the RF models, the maximum depth of the tree (max_depth) and the number of trees (n_estimators) affected the performance of the models (RF_2D and RF_ECFP). The optimized maximum depth was kept 25 and 30 for RF_2D and RF_ECFP, respectively. In the SVM analysis, C (the penalty for incorrect classification of training samples) and gamma (influence of a single training point) are controlling parameters. For SVM_2D, C = 5 and gamma = 10 were kept as the optimized setting, whereas for SVM_ECFP model, C and gamma were set to 1 and 0.5, respectively. The fivefold cross-validation showed that kNN_ECFP (AccuracyTrain = 0.946, AccuracyTest = 0.932, AUC = 0.966, MCC = 0.855), RF_ECFP (AccuracyTrain = 0.999, AccuracyTest = 0.933, AUC = 0.969, MCC = 0.857), and SVM_2D (AccuracyTrain = 0.979, AccuracyTest = 0.932, AUC = 0.964, MCC = 0.854) were the best classifier. The three best classifiers were used for the virtual screening of natural products of 3111 compounds as p38α inhibitors. Among these, 15 compounds were identified for inhibitory activity against p38α. Of these, one compound, picrasidine S (a dimeric alkaloid) revealed its IC50 of 34.14 μM which was extracted from Picrasma quassioides. This compound was further investigated by molecular docking with p38α with PDB ID: 2QD9. Docking analysis revealed hydrophobic interactions with Leu167, 25 Leu108, Gly110, Ala157, and Ala111 and one hydrogen bonding with the main chain of Met109. The binding however doesn’t give an idea of whether it is a type I or type II inhibitor. Picrasidine S is a nonATP-competitive or allosteric inhibitor that can be a good scaffold against p38α in place of common ATP-binding candidates present.

5 Computational Studies on c-Jun N-Terminal Kinase 3 (JNK3) Inhibitors as Potential Anti-Alzheimer’s Agents The quantitative structure-activity relationship of JNK3 inhibitors using 3D-CoMFA and CoMSIA models was studied by Liu et al. [93]. A dataset of JNK3 inhibitors having a common structure of N-(6-methylpyridn-3-yl)-3-(4-(3-phenylureido)-1H-pyrazol-1-yl) benzamide was used in the study for CoMFA and CoMSIA analysis. The statistical validation metrics for both CoMFA and CoMSIA models gave promising results. From a contour map analysis, it was found that bulky, electron-withdrawing, and steric groups at R1 substitution are beneficial for inhibitory activity. The benzene ring should contain a minor hydrophobic and donating accepter

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

149

Fig. 11 Features essential for good JNK3 inhibition identified by Liu et al. [93]. (a) Common core structure of N-(6-methylpyridn-3-yl)-3-(4-(3-phenylureido)-1H-pyrazol-1-yl) benzamide with R1 and R2 groups, (b) Structure of the highest active compound, (c) Designed JNK3 inhibitor

substituent for better activity. At the R2 position, bulky and hydrophobic groups are favorable to increase the potency (Fig. 11a). Based on QSAR analysis, the lead compound was chosen, and 11 new compounds were designed. These compounds were predicted using the 3D-QSAR model and were superimposed to the alignment considered as test set. Molecular docking analysis was performed taking the highest active compound (Fig. 11b) to find the important interactions necessary. Hydroxyl bonds formed hydrogen bond interactions with Gln75 and Arg227-Arg230. 3-NH of the side chain R1 of the lead compound formed H-bond with the carbonyl group of Arg227-Arg230, and core 4 carbonyl of the lead (Fig. 11b) formed H-bond with the -NH group of Gln75. Other designed compounds were also docked with JNK, and it was found designed compound (Fig. 11c) formed similar but firmer nature of hydrogen bond interactions as that of the lead compound. Thus, the newly developed compounds are promising c-JNK inhibitors suggesting that the 3D-QSAR models can be useful for the development novel inhibitors of JNK3. Jun et al. [94] recently synthesized 51 2-aryl-1-pyrimydyl-1, 4, 5, 6-tetrahydrocyclopenta[d]imidazole-5-carboxamide derivatives designed as novel JNK3 selective inhibitors. The inhibitory activities of these synthesized compounds were good with a few compounds having potent inhibition activities with IC50 of 0.716, 0.564, 0.397, and 0.779 nM. Further these, compounds prevented

150

Priyanka De and Kunal Roy

amyloid-induced neuronal death and inhibited c-Jun phosphorylation in primary rat neurons. Computational analysis involving molecular docking gave knowledge of the important molecular interactions. In general, four important observations were noted, (a) amino pyrimidine as a hinge binder with the Met149, (b) aryl groups in the hydrophobic pocket, (c) hydrogen bonding of carboxamide with Asn152, and (d) a water bridge hydrogen bond between the nitrogen atom at position 3 of imidazole and Lys93. Further, ADME prediction recommended the safety profile of the designed compounds. Thus, tetrahydrocyclopenta[d]imidazole scaffold can serve as a promising JNK3 inhibitor.

6 Computational Studies on Dual-Specificity Tyrosine Phosphorylation-Regulated Kinase-1A (DYRK1A) as Potential Anti-Alzheimer’s Agents Bhardwaj et al. [95] reported a series of pyrrolone-fused benzosuberene (PBS) compounds as prominent inhibitors of the DYRK1A receptor. They have reported the molecular interactions through molecular docking analysis by utilizing in-house synthesized PBS analogs using DS Visualizer v17.2. The important molecular interactions occurring between standard and highly active ligands (Ligand X1, Ligand X2, and Ligand X3) (Table 2) with the DYRK1A receptor are listed in Table 2. The binding conformations of the PBS ligand were perceived by comparing the docking interactions with the standard molecule. Further, they have performed MD simulations of the best-docked complexes to understand their stability and to determine the binding energy of complexes using the GROMACS v5.0.7. The RMSD plots confirmed that the ligand-receptor complex stabilized after 50 ns fluctuations with maintained RMSD between 0.2 nm to 0.5 nm for all the complexes. The binding free energies and interaction analysis were estimated through MM-PBSA including SASA, van der Waals, polar solvation, and electrostatic energy calculations for DYRK1A complexes. The chosen ligands and the standard compound showed high binding free energy and good binding efficiency at the ATP-binding site which makes them lead compounds to develop therapeutics against DYRK1A. Shahroz et al. [96] have successfully applied a molecular docking-based virtual screening computational technique on the molMall database compounds against the DYRK1A protein to find out potential inhibitors. The molMall database comprises 15,381 chemically diverse and unique classes of compounds which were screened against five sets of DYRK1A protein (PDB ID: 3ANQ, 4AZE, 2WO6, 4YLK, and 5A3X). From this screening stage, 54 small molecules were selected from each protein structure. A total of 40 common structures were procured for further binding pattern analysis. DYRK1A possesses a conserved ATP-binding zone

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

151

Table 2 Molecular docking results showing type of interactions and interacting residues involved as discussed by Bhardwaj et al. [95]

Compound Standard compound

Structure

Type of interactions

Amino acid residue

HB HB (with ester carbonyl oxygen) HB (with pyridine moiety) HB (with terminal amine)

GLY171 LYS188

Hinge residue LEU241 LEU241 The side chain of ASP307 and ASN292 GLU239

Halogen bond (with chlorine on phenyl group) SER242, Pi-Pi, PHE238, Pi-sigma, VAL173, Alky, ILE165, Pi-alkyl, VAL222, Vdw ALA186, and VAL306 Ligand X1

Hydrophobic LYS188 and gate residue PHE238 Pi-alkyl VAL22, ALA186, VAL306 Pi-sigma and Val173, Vdw LEU294

Ligand X2

HB

LEU241 (hinge residue) Alkyl MET240 interaction (hinge residue) VAL222, ALA186, VAL306, ILE165, VAL173 (continued)

152

Priyanka De and Kunal Roy

Table 2 (continued)

Compound

Structure

Type of interactions

Amino acid residue

Vdw

SER242, TYR243 (hinge residues) LYS188, ASP307 (conserved residues) PHE238 (conserved residue) VAL173

Pi-sigma

Ligand X3

HB

Vdw

Carbon HB

Pi-sigma

Pi-alkyl

LEU24 (hinge residues) MET240 and TYR243 (hinge residues) LYS188 and ASP307 (conserved residues) SER242 (hinge residues) PHE238 (conserved residues) VAL173 ILE165, ALA186, VAL306

for inhibitor interaction comprises amino acids residues: Phe238 (gatekeeper; gk), Glu239 (gk + 1), Leu241 (gk + 3), Lys188 (catalytic), Phe170, Ser242, Asn292, and Asp307. Further, through binding free energy calculation using MM/GBSA, they have selected one compound out of 32 compounds with ΔGbind less than -40 kcal/mol and used it for further analysis. The main objective of this work was to find out selective DYRK1A inhibitors which was obtained by docking-based virtual screening of these

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

153

32 compounds against 12 related protein kinases [PDB IDs: 4D2S (DYRK1B), 3KVW (DYRK2), 3O0G (CDK5/p25), 5W4W (CK1), 1Z57 (CLK1), 3RAW(CLK3), 5OY4 (GSK3β), 2PZY (MAPK2), 3RTP (MAPK10), 5VUA (PIM1), 5LCP (PKA), and 1DSY (PKC)]. The ligands with less than -5 Glide score and binding energy below -40 kcal/mol were deselected for further analysis. Among these 32 compounds, quercetin (molMall ID: 22067) was found to bind with 10 out of 12 proteins with a high Glide score and binding free energy. The authors have also calculated various physicochemical and pharmacokinetic properties of selected 15 hits, including the toxicity profile of 6 top compounds (molMall IDs: 9539, 11352, 15938, 19037, 21830, and 21878) in ProTox-II webserver. The selected six docked ligand-protein complexes were used for MD simulation study for 100 ns which ascertained their mechanism of interactions and stability in the ATP pocket of DYRK1A kinase. Out of these hits, three compounds (molmall IDs: 15938, 11352 and 21878) had affinity toward DYRK2, GSK3, CLK1, CLK3, and MAPK10 protein kinases also. In conclusion, the synthesis of the top six hit compounds along with their kinase profiling would validate their DYRK1A inhibition. Recently, Abduljelil et al. [97] have applied in silico techniques to investigate the inhibitory activities of 34 natural inhibitors of DYRK1A against AD. Initially, they have developed MLR-based QSAR model using MLRPlusValidation version 1.3 tool available from dtclab.webs.com/software-tools. The predictivity, strength, and reliability of the established model successfully passed the internal and external the unique criteria (R2 = 0:837, Q 2cv = 0:788 ). The selected descriptors include ALogP (Ghose-Crippen LogKow), apol (sum of the atomic polarizabilities), MATS6v (Moran autocorrelation of lag 6 weighted by van der Waals volume), GATS2m (Geary autocorrelation of lag 2 weighted by mass), and GATS2e (Geary autocorrelation of lag 2 weighted by Sanderson electronegativity). The study also includes the analysis of descriptor correlation, model randomization, and applicability domain (using the leverage approach) assessment. They have further analyzed the molecular interactions using molecular docking in AutoDock 4.2. Virtual screening of potential natural compounds against the crystal structure of human DYRK1A (PDB ID: 6A1G) gave an idea that compound Y (Fig. 12) binds best with the enzyme with three hydrogen bond interactions and six nonbonding interactions and binding energy of -12.8 kcal/mol. Comparatively, harmine (the standard compound) formed only two nonbonded interactions. It was observed that compound Y makes hydrogen bond interactions with Arg325 and Asn365 at the catalytic cleft and pi-cation interaction with Arg328 proving its stability at the receptor active site. The docking complex stability and structural flexibility were evaluated using MD simulation parameters (RMSD, RMSF, Rg, SASA, and

154

Priyanka De and Kunal Roy

Fig. 12 Interaction of compound Y and Harmine with human DYRK1A as observed by Abduljelil et al. [97]

ligand hydrogen bonds (H-bond number)) performed over 100 ns trajectory. The RMSD analysis showed the stability of the complex’s conformation, and it reached equilibrium after 20 ns. The binding energies calculated using MM-PBSA methods showed that harmine complex possesses weaker interaction than the compound Y complex. The drug likeliness was also predicted using Lipinski’s parameters including, log P, nHBA, nHBD, nRB, and MW which affirmed that compound with favorable interaction also showed good pharmacokinetic properties.

7 Computational Modeling Studies on Miscellaneous Kinases and Their Inhibitors as Anti-Alzheimer’s Agents 7.1 Casein Kinase Inhibitors

In an study, by Cescon et al. [98] have highlighted the vital importance of casein kinase 1δ (CK1δ) in the development of several neurodegenerative diseases. This study reported new potential CK inhibitors which can bind to the enzyme by a dual approach in silico/in vitro. The virtual screening has been done using an in-house chemical library previously designed for different targets. A consensus strategy was followed to evaluate the results of GOLD (Goldscore), PLANTS (chemplp), and Glide (standard precision) during molecular docking. Consensus docking improves the reliability of docking by following different strategies in a parallel manner. The minimum RMSD and average RMSD were considered for analysis. Additionally, interaction energy fingerprints (IEF) were applied to 19 ligands for a qualitative analysis of the molecular

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

155

Fig. 13 Workflow for hit compound identification as proposed by Cescon et al. [98]

interaction. Coupling geometry and IEF gave knowledge about the interactions at the hinge region of the kinase showing amino acid Leu 85 making two hydrogen bonds with many of the ligands. Further, the presence of aromatic moieties in most of the compounds as revealed by compound superimposition suggested the occurrence of hydrophobic interactions confirmed by the fingerprints. These pharmacophoric features were applied to each docking protocol to obtain three independent lists of screened compounds (Fig. 13). Finally, they have selected two hit compounds (IC50 = 15.22 ± 2.71 μM and IC50 = 12.95 ± 3.21 μM) containing pyrrolo[3,2-f]quinolinone moiety as key-scaffold (Fig. 13). These compounds were initially designed for a completely different target; however, they have shown good inhibitory activity against CK1δ proving that drug repurposing here is successful. However, these compounds need further investigation for their pharmacokinetic behavior. The application of fragment-based drug discovery (FBDD) in the identification of CS-1δ inhibitors was reported by Bolcato et al. [99]. They have screened a large fragment library of 272,000

156

Priyanka De and Kunal Roy

Fig. 14 IC50 values of seven hit compounds

compounds using different in silico approaches including molecular docking, molecular dynamics (MD), and pharmacophore filter. Like the previous study, this work also applies docking-based virtual screening using PLANTS-ChemPLP, GOLD-ChemScore, and Glide-SP which generated about 13.6 million poses. This large information was further filtered through a pharmacophore model generating three vital features: (a) two features, one hydrogen bond donor, and another hydrogen bond acceptor making interaction with Leu 85 amino acid of CS-1δ enzyme and (b) the presence of an aromatic group close to the hinge region. These pharmacophore filters were applied to all three docking methods and henceforth used for a consensus docking to filter out 840 docking poses only. This was followed by a post-docking refinement using MD simulation to filter out the fragments with unstable binding. Finally, RMSF data screened out 66 fragments for further enzyme analysis. The enzyme analysis calculated the IC50 values for compounds with a residual kinase activity of less than 40%. Seven hit compounds (compounds Z1 to Z7) were selected with half-maximal inhibitory concentrations ranging from 3.31 μM to 24.86 μM (Fig. 14). The recognition mechanism of the most active inhibitor (compound Z1; IC50 = 3.31 μM) was learned by supervised molecular dynamics simulations (SuMD). Here, it was found that Asp149 amino acid acts as an “electrostatic recruiter” for the amino-thiophene moiety of the ligand. On the other hand, vicinal Lys38 restricts ligand entrance into the binding core, caused due to electrostatic repulsion between the charged amino group of the amino acid side chain

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

157

and the uncharged amino group of the ligand. The binding pathway, as depicted by two conformations (S1 and S2), gave an important information. S1 showed that the ligand interacts with the backbone Lys85 with its amino-thiophene moiety and exposing its morpholine moiety to the solvent. S2 shows a bivalent hydrogen bond interaction with Lys 85 and the morpholine embedded in the hydrophobic pocket of Met80, Met82, Ile23, and the alkyl portion of the Lys38 side chain. It was also found the S2 pharmacophoric state was more stable than the S1 state. The selected seven scaffolds were tested using an enzymatic assay to determine the residual CS-1δ enzymatic activity resulting in IC50 values in low micromolar range (Fig. 14). Out of these seven compounds, five of them showed novel scaffolds for CK1δ, authenticating the proposed pathway to be of extreme benefit in CK1δ inhibitor discovery. 7.2 Ca2+/ CalmodulinDependent Protein Kinase II (CaMKII) Inhibitors

Eduful et al. [100] reported a hinge-binding scaffold hopping strategy to design new calcium/calmodulin-dependent protein kinase kinase 2 (CAMKK2) inhibitors. The work includes the synthesis of a series of 32 inhibitors containing 5,6-bicyclic, 6,6-bicyclic, and single-ring hinge-binding. The aim of these compounds’ synthesis was to retain CAMKK2 inhibition and improve azaindole (7-azaindole GSK650393) activity by modulating strong H-bond interactions occurring with kinase hinge-binding residues. To understand the binding modes, computational molecular docking analysis was performed between CAMKK2 (PDB ID: 6BKU) and important compounds. Molecular docking showed favorable and improved binding of many of the synthesized compounds. In addition, the physicochemical properties, drug likeliness, and selectivity of the compounds were also evaluated. Thus, this research involves the identification of potential CAMKK2 inhibitors scaffolds possessing alternate hinge-binding moieties.

7.3 Multitarget Kinase Inhibitors

Owing to the possible downregulation of a variety of kinase enzymes, AD can be controlled by using multitarget drugs. Simultaneous modulation of multiple protein kinases (PKs) is involved in abnormal tau phosphorylation, paving a way for multitarget kinase inhibitors to achieve a superior therapeutic effect. Demuro et al. [101] presented a report on the identification and characterization of ARN25068, a triple GSK-3β/FYN/ DYRK1A inhibitor to control tau-generated neuropathies. A computational approach involving docking simulation was used to study ARN25068, a 2,4-disubstituted pyrimidine thiophene core as a promising multikinase inhibitor. Firstly, the bound conformations were analyzed by docking at both GSK-3b and FYN ATP-binding pockets resulting in desirable docking scores (-39.3 and -39.5, respectively). On binding with GSK-3β, ARN25068 formed three hydrogen bonds between its amino pyrazole with the hinge residues V135 and D133 and hydrophobic interactions between pyrimidine core and I62, T138, R141, E137, and P136 residues (Fig. 15). Some other

158

Priyanka De and Kunal Roy

Fig. 15 Features responsible for multikinase inhibitory activity as identified by Demuro et al. [101]

hydrophobic interactions also occurred between the amino pyrazole ring and Y134, L188, and A83 residues, between the cyclopropyl group and residues L132 and C199, and between the benzylamine moiety and residues V70, D200, N186, N64, and G63. While interacting with FYN kinase, two hydrogen bonds are formed, and other hydrophobic interactions are formed with amino pyrazole, cyclopropyl, benzylamine, and pyrimidine groups. Crystallographic structure confirmed the computational poses and recognized essential structural features. Further, NanoBRET binding assays along with computational and X-ray crystallography analysis suggested proved the inhibitory activity of ARN25068 against DYRK1A in a single-digit micromolar range.

8 Conclusions and Future Prospects Current drug therapies of AD do not inhibit the unremitting neuropathology of the disease. Thus, new generation approaches strive to target the underlying disease mechanisms. One of such AD pathology includes tau deposition which is largely affected by a group of protein kinases (PKs). PKs produce aberrant tau phosphorylation, and thus its inhibition is one of the key strategies to elaborate therapies against AD. The current book chapter gives a detailed information about the computational processes involved in the discovery of new leads as inhibitors of the different protein kinases discussed. The short time requirement of in silico techniques are conducive for high-throughput screening of available drugs to identify them as potential protein kinase inhibitors acting against AD. Various computational methods involving molecular docking, molecular dynamics simulation, pharmacophore

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents

159

modeling, and QSAR have been studied to identify novel molecules as AD protein kinase inhibitors. With the advancement of these methods, kinase selectivity can be analyzed to offer better kinase activity against AD.

Acknowledgment PD thanks Indian Council of Medical Research for Research Associateship (File No: BMI/11(35)/2022). References 1. Chiang K, Koo EH (2014) Emerging therapeutics for Alzheimer’s disease. Annu Rev Pharmacol Toxicol 54:381–405 2. Dementia F sheet on (2003) Face sheet on Dementia. https://www.who.int/newsroom/fact-sheets/detail/dementia. Accessed on 08.01.2023 3. Lynch C (2020) World Alzheimer report 2019: attitudes to dementia, a global survey. Alzheimers Dement 16:e038255 4. Zeisel J, Bennett K, Fleming R (2020) World Alzheimer Report 2020: design, dignity, dementia: dementia-related design and the built environment. Alzheimer’s Dis Int 2 5. Huang Y, Mucke L (2012) Alzheimer mechanisms and therapeutic strategies. Cell 148:1204–1222 6. Hardy J, Selkoe DJ (2002) The amyloid hypothesis of Alzheimer’s disease: Progress and problems on the road to therapeutics. Science (80- ) 297:353–356 7. Bellenguez C, Grenier-Boley B, Lambert JC (2020) Genetics of Alzheimer’s disease: where we are, and where we are going. Curr Opin Neurobiol 61:40–48 8. Folch J, Petrov D, Ettcheto M et al (2016) Current research therapeutic strategies for Alzheimer’s disease treatment. Neural Plast 2016:8501693 9. Haass C, Kaether C, Thinakaran G, Sisodia S (2012) Trafficking and proteolytic processing of APP. Cold Spring Harb Perspect Med 2: a006270 10. West S, Bhugra P (2015) Emerging drug targets for Aβ and tau in Alzheimer’s disease: a systematic review. Br J Clin Pharmacol 80: 221–234 11. Aricept (2015) (donepezil hydrochloride). Full prescribing information. Eisai Inc., Woodcliff Lake 12. Exelon (2015) (rivastigmine tartrate). Full Prescribing Information. Novartis Pharmaceuticals Corporation, East Hanover

13. Exelon Patch (2016) (rivastigmine transdermal system). Full prescribing information. Novartis Pharmaceuticals Corporation, East Hanover 14. Razadyne (2016) (galantamine hydrobromide). Full prescribing information. Janssen Pharmaceuticals Inc., Titusville 15. Namenda XR (2014) (memantine hydrochloride). Full prescribing information. Forest Pharmaceuticals Inc., St. Louis 16. Mitra A, Dey B (2013) Therapeutic interventions in Alzheimer’s disease. In: Uday K (ed) Neurodegenerative diseases. Rijeka, Croatia, pp 291–317 17. Martin L, Latypova X, Wilson CM et al (2013) Tau protein kinases: Involvement in Alzheimer’s disease. Ageing Res Rev 12: 289–309 18. Alam J, Sharma L (2018) Potential enzymatic targets in Alzheimer’s: a comprehensive review. Curr Drug Targets 20:316–339 19. Turab Naqvi AA, Hasan GM, Hassan MI (2020) Targeting tau hyperphosphorylation via kinase inhibition: strategy to address Alzheimer’s disease. Curr Top Med Chem 20: 1059–1073 20. Cleveland DW, Hwo SY, Kirschner MW (1977) Physical and chemical properties of purified tau factor and the role of tau in microtubule assembly. J Mol Biol 116:227– 247 21. Das BC, Sribidya P, Devi PO et al (2018) The role of tau protein in diseases. Ann Adv Chem 2:001–016 22. Kampers T, Pangalos M, Geerts H et al (1999) Assembly of paired helical filaments from mouse tau: implications for the neurofibrillary pathology in transgenic mouse models for Alzheimer’s disease. FEBS Lett 451:39– 44 23. Chen Q, Yoshida H, Schubert D et al (2001) Presenilin binding protein is associated with neurofibrillary alterations in Alzheimer’s

160

Priyanka De and Kunal Roy

disease and stimulates tau phosphorylation. Am J Pathol 159:1597–1602 24. Dustin P (2012) Microtubules. Springer Science & Business Media 25. Ballatore C, Lee VMY, Trojanowski JQ (2007) Tau-mediated neurodegeneration in Alzheimer’s disease and related disorders. Nat Rev Neurosci 8:663–672 26. Binder LI, Frankfurter A, Rebhun LI (1985) The distribution of tau in the mammalian central nervous system. J Cell Biol 101: 1371–1378 27. Iqbal K, Liu F, Gong C (2017) Tau and neurodegenerative disease: the story so far. Nat Rev Neurol 12:15–27 28. Iqbal K, Gong C-X, Liu F, Novak M (2013) Hyperphosphorylation-induced tau oligomers. Front Neurol 4:112 29. Mcgregor G, Harvey J, Mainardi M et al (2018) Regulation of hippocampal synaptic function by the metabolic hormone, leptin: implications for health and neurodegenerative disease. Front Cell Neurosci 12:340 30. Sergeant N, Delacourte A, Bue´e L (2005) Tau protein as a differential biomarker of tauopathies. Biochim Biophys Acta Mol basis Dis 1739:179–197 31. Chun W, Johnson GV (2007) The role of tau phosphorylation and cleavage in neuronal cell death. Front Biosci 12:733–756 32. Cai Z, Zhao Y, Zhao B (2012) Roles of glycogen synthase kinase 3 in Alzheimer’s disease. Curr Alzheimer Res 9:864–879 33. Doble BW, Woodgett JR (2003) GSK-3: tricks of the trade for a multi-tasking kinase. J Cell Sci 116:1175–1186 34. Doble BW, Patel S, Wood GA et al (2007) Functional redundancy of GSK-3α and GSK-3β in Wnt/β-catenin signaling shown by using an allelic series of embryonic stem cell lines. Dev Cell 12:957–971 35. Ly PTT, Wu Y, Zou H et al (2013) Inhibition of GSK3β-mediated BACE1 expression reduces Alzheimer-associated phenotypes. J Clin Invest 123:224–235 36. Phiel CJ, Wilson CA, Lee VMY, Klein PS (2003) GSK-3α regulates production of Alzheimer’s disease amyloid-β peptides. Nature 423:435–439 37. Cole A, Frame S, Cohen P (2004) Further evidence that the tyrosine phosphorylation of glycogen synthase kinase-3 (GSK3) in mammalian cells is an autophosphorylation event. Biochem J 377:249–255

38. Avila J, Santa-Marı´a I, Pe´rez M et al (2006) Tau phosphorylation, aggregation, and cell toxicity. J Biomed Biotechnol 2006 39. Muyllaert D, Kremer A, Jaworski T et al (2008) Glycogen synthase kinase-3β, or a link between amyloid and tau pathology? Genes Brain Behav 7:57–66 40. Lim S, Kaldis P (2013) Cdks, cyclins and CKIs: roles beyond cell cycle regulation. Development 140:3079–3093 41. Liu S-L, Wang C, Jiang T et al (2016) The role of Cdk5 in Alzheimer’s disease. Mol Neurobiol 53:4328–4342 42. Asada A, Saito T, Hisanaga S-I (2012) Phosphorylation of p35 and p39 by Cdk5 determines the subcellular location of the holokinase in a phosphorylation-site-specific manner. J Cell Sci 125:3421–3429 43. Cruz JC, Tseng HC, Goldman JA et al (2003) Aberrant Cdk5 activation by p25 triggers pathological events leading to neurodegeneration and neurofibrillary tangles. Neuron 40:471–483 44. Noble W, Olm V, Takata K et al (2003) Cdk5 is a key factor in tau aggregation and tangle formation in vivo. Neuron 38:555–565 45. Zhu X, Lee HG, Raina AK et al (2002) The role of mitogen-activated protein kinase pathways in Alzheimer’s disease. Neurosignals 11: 270–281 46. Guo Y, Pan W, Liu S et al (2020) ERK/MAPK signalling pathway and tumorigenesis (review). Exp Ther Med 19:1997–2007 47. Dalrymple SA (2002) p38 mitogen activated protein kinase as a therapeutic target for Alzheimer’s disease. J Mol Neurosci 19:295–299 48. Chen YR, Tan TH (2000) The c-Jun N-terminal kinase pathway and apoptotic signaling (review). Int J Oncol 16:651–662 49. Mielke K, Herdegen T (2000) JNK and p38 stresskinases — degenerative effectors of signal-transduction-cascades in the nervous system. Prog Neurobiol 61:45–60 50. Reynolds CH, Utton MA, Gibb GM et al (1997) Stress-activated protein kinase/c-Jun N-terminal kinase phosphorylates τ protein. J Neurochem 68:1736–1744 51. Thakur A, Wang X, Siedlak SL et al (2007) c-Jun phosphorylation in Alzheimer disease. J Neurosci Res 85:1668–1673 52. Varjosalo M, Bjo¨rklund M, Cheng F et al (2008) Application of active and kinasedeficient kinome collection for identification of kinases regulating hedgehog signaling. Cell 133:537–548

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents 53. Becker W, Weber Y, Wetzel K et al (1998) Sequence characteristics, subcellular localization, and substrate specificity of DYRKrelated kinases, a novel family of dual specificity protein kinases*. J Biol Chem 273:25893– 25902 54. Becker W, Joost HG (1998) Structural and functional characteristics of Dyrk, a novel subfamily of protein kinases with dual specificity. Prog Nucleic Acid Res Mol Biol 62:1–17 55. Kentrup H, Becker W, Heukelbach J et al (1996) Dyrk, a dual specificity protein kinase with unique structural features whose activity is dependent on tyrosine residues between subdomains VII and VIII. J Biol Chem 271: 3488–3495 56. Kumar K, Man-Un Ung P, Wang P et al (2018) Novel selective thiadiazine DYRK1A inhibitor lead scaffold with human pancreatic b-cell proliferation activity. Eur J Med Chem 157:1005–1016 57. Himpel S, Panzer P, Eirmbter K et al (2001) Identification of the autophosphorylation sites and characterization of their effects in the protein kinase DYRK1A. Biochem J 359: 497–505 58. Gross SD, Anderson RA (1998) Casein kinase I: spatial organization and positioning of a multifunctional protein kinase family. Cell Signal 10:699–711 59. Li G, Yin H, Kuret J (2004) Casein kinase 1 delta phosphorylates tau and disrupts its binding to microtubules. J Biol Chem 279: 15938–15945 60. Knippschild U, Gocht A, Wolff S et al (2005) The casein kinase 1 family: participation in multiple cellular processes in eukaryotes. Cell Signal 17:675–689 61. Ahmad KA, Wang G, Unger G et al (2008) Protein kinase CK2 – a key suppressor of apoptosis. Adv Enzym Regul 48:179–187 62. Ghoshal N, Smiley JF, DeMaggio AJ et al (1999) A new molecular link between the fibrillar and granulovacuolar lesions of Alzheimer’s disease. Am J Pathol 155:1163–1172 63. Chen C, Gu J, Basurto-Islas G et al (2017) Up-regulation of casein kinase 1ε is involved in tau pathogenesis in Alzheimer’s disease. Sci Rep 71(7):1–15 64. Pierrot N, Ferrao Santos S, Feyt C et al (2006) Calcium-mediated transient phosphorylation of tau and amyloid precursor protein followed by intraneuronal amyloid-β accumulation*. J Biol Chem 281:39907– 39914 65. Oka M, Fujisaki N, Maruko-Otake A et al (2017) Ca2+/calmodulin-dependent protein

161

kinase II promotes neurodegeneration caused by tau phosphorylated at Ser262/356 in a transgenic Drosophila model of tauopathy. J Biochem 162:335–342 66. Griffith LC (2004) Regulation of calcium/ calmodulin-dependent protein kinase II activation by intramolecular and intermolecular interactions. J Neurosci 24:8394–8398 67. Wang JH, Kelly PT (1995) Postsynaptic injection of Ca2+/CaM induces synaptic potentiation requiring CaMKII and PKC activity. Neuron 15:443–452 68. Terry RD, Masliah E, Salmon DP et al (1991) Physical basis of cognitive alterations in alzheimer’s disease: synapse loss is the major correlate of cognitive impairment. Ann Neurol 30:572–580 69. Lucchesi W, Mizuno K, Giese KP (2011) Novel insights into CaMKII function and regulation during memory formation. Brain Res Bull 85:2–8 70. Liang Z, Liu F, Grundke-Iqbal I et al (2007) Down-regulation of cAMP-dependent protein kinase by over-activated calpain in Alzheimer disease brain. J Neurochem 103:2462– 2470 71. Shi J, Qian W, Yin X et al (2011) Cyclic AMP-dependent protein kinase regulates the alternative splicing of tau exon 10: a mechanism involved in tau pathology of Alzheimer disease. J Biol Chem 286:14639–14648 72. Amini E, Nassireslami E, Payandemehr B et al (2015) Paradoxical role of PKA inhibitor on amyloidβ-induced memory deficit. Physiol Behav 149:76–85 73. Hanger DP, Anderton BH, Noble W (2009) Tau phosphorylation: the therapeutic challenge for neurodegenerative disease. Trends Mol Med 15:112–119 74. Macalino SJY, Gosu V, Hong S, Choi S (2015) Role of computer-aided drug design in modern drug discovery. Arch Pharmacal Res 389(38):1686–1701 75. Morris GM, Lim-Wilby M (2008) Molecular docking. In: Methods in molecular biology. Humana Press, pp 365–382 76. Hollingsworth SA, Dror RO (2018) Molecular dynamics simulation for all. Neuron 99: 1129–1143 77. Yang SY (2010) Pharmacophore modeling and applications in drug discovery: challenges and recent advances. Drug Discov Today 15: 444–450 78. Roy K (2017) Advances in QSAR modeling. In: Applications in pharmaceutical, chemical, food, agricultural and environmental sciences. Springer, Cham

162

Priyanka De and Kunal Roy

79. Vamathevan J, Clark D, Czodrowski P et al (2019) Applications of machine learning in drug discovery and development. Nat Rev Drug Discov 18:463–477 80. Iwaloye O, Elekofehinti OO, Oluwarotimi EA et al (2020) Insight into glycogen synthase kinase-3β inhibitory activity of phyto-constituents from Melissa officinalis: in silico studies. Silico Pharmacol 8:1–13 81. Jiang X, Wang Y, Liu C et al (2021) Discovery of potent glycogen synthase kinase 3/cholinesterase inhibitors with neuroprotection as potential therapeutic agent for Alzheimer’s disease. Bioorg Med Chem 30:115940 82. Eskandarzadeh M, Kordestani-Moghadam P, Pourmand S et al (2021) Inhibition of GSK_3β by iridoid glycosides of snowberry (Symphoricarpos albus) effective in the treatment of Alzheimer’s disease using computational drug design methods. Front Chem 9: 709932 83. Elangovan ND, Dhanabalan AK, Gunasekaran K et al (2021) Screening of potential drug for Alzheimer’s disease: a computational study with GSK-3 β inhibition through virtual screening, docking, and molecular dynamics simulation. J Biomol Struct Dyn 39:7065– 7079 84. Zhu J, Wu Y, Xu L, Jin J (2019) Theoretical studies on the selectivity mechanisms of glycogen synthase kinase 3β (GSK3β) with pyrazine ATP-competitive inhibitors by 3DQSAR, molecular docking, molecular dynamics simulation and free energy calculations. Curr Comput Aided Drug Des 16:17– 30 85. Tammareddy T, Keyrouz W, Sriram RD et al (2022) Computational study of the allosteric effects of p5 on CDK5–p25 hyperactivity as alternative inhibitory mechanisms in neurodegeneration. J Phys Chem B 126:5033–5044 86. Zeb A, Kim D, Alam SI et al (2019) Computational simulations identify pyrrolidine-2,3dione derivatives as novel inhibitors of cdk5/ p25 complex to attenuate alzheimer’s pathology. J Clin Med 8:746 87. Advani D, Kumar P (2022) Computational analysis of natural compounds as cyclindependent kinase-5 inhibitors for Alzheimer’s and Parkinson’s disease. In: IEEE global conference on computing, power and communication technologies (GlobConPT), pp 1–6 88. Garkani Nejad Z, Ghanbari A (2021) Molecular modeling studies of TRIAZOLYL thiophenes as CDK5/P25 inhHibitors using 3D-QSAR and molecular docking. Iran J Anal Chem 8:29–38

89. El Aissouq A, Lachhab A, El Rhabori S et al (2022) Computer-aided drug design applied to a series of pyridinyl imidazole derivatives targeting p38α MAP kinase: 2D-QSAR, docking, MD simulation, and ADMET investigations. New J Chem 46:20786–20800 90. Khan MF, Verma G, Alam P et al (2019) Dibenzepinones, dibenzoxepines and benzosuberones based p38α MAP kinase inhibitors: their pharmacophore modelling, 3D-QSAR and docking studies. Comput Biol Med 110: 175–185 ˇ ivadinovic´ B, Stamenovic´ J, Z ˇ ivadinovic´ J 91. Z et al (2022) QSAR modelling, molecular docking studies and ADMET predictions of polysubstituted pyridinylimidazoles as dual inhibitors of JNK3 and p38α MAPK. J Mol Struct 1265:133504 92. Shen T, Tao Y, Liu B et al (2022) Machine learning assisted discovery of novel p38α inhibitors from natural products. Diabetes 5:21 93. Liu Y, Xie Y, Liu Y et al (2019) Insights into the c-Jun N-terminal kinase 3 (JNK3) inhibitors: CoMFA, CoMSIA analyses and molecular docking studies. Med Chem Res 28:1796– 1805 94. Jun J, Baek J, Kang D et al (2023) Novel C-Jun N-terminal kinase 3 inhibitors 1, 4, 5, 6-tetrahydrocyclopenta[D]imidazole-5carboxamide: design, synthesis, molecular docking, and biological evaluation as potential therapeutics for neurodegenerative disease. Synth Mol Docking Biol Eval as Potential Ther Neurodegener Diseases. Eur J Med Chem 245: 114917 95. Bhardwaj VK, Singh R, Sharma J et al (2020) Structural based study to identify new potential inhibitors for dual specificity tyrosinephosphorylation- regulated kinase. Comput Methods Prog Biomed 194:105494 96. Shahroz MM, Sharma HK, Altamimi ASA et al (2022) Novel and potential small molecule scaffolds as DYRK1A inhibitors by integrated molecular docking-based virtual screening and dynamics simulation study. Molecules 27:1159 97. Abduljelil A, Uzairu A, Shallangwa GA et al (2023) Natural inhibitors of DYRK1A as drug candidates against Alzheimer Disease: QSAR, molecular docking, molecular dynamics simulation and drug evaluation assessment. h t t p s : // d o i . o r g / 1 0 . 2 1 2 0 3 / r s . 3 . r s 2443598/v1 98. Cescon E, Cescon E, Bolcato G et al (2020) scaffold repurposing of in-house chemical library toward the identification of new casein kinase 1 δinhibitors. ACS Med Chem Lett 11: 1168–1174

Computational Modeling of Kinase Inhibitors as Anti-Alzheimer Agents 99. Bolcato G, Cescon E, Pavan M et al (2021) A computational workflow for the identification of novel fragments acting as inhibitors of the activity of protein kinase ck1δ. Int J Mol Sci 22:9741 100. Eduful BJ, O’Byrne SN, Temme L et al (2021) Hinge binder scaffold hopping identifies potent calcium/calmodulin-dependent

163

protein kinase kinase 2 (CAMKK2) inhibitor chemotypes. J Med Chem 64:10849–10877 101. Demuro S, Sauvey C, Tripathi SK et al (2022) ARN25068, a versatile starting point towards triple GSK-3β/FYN/DYRK1A inhibitors to tackle tau-related neurological disorders. Eur J Med Chem 229:114054

Chapter 6 Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors and Its Implications in Alzheimer’s Disease Arvind Kumar Jain, C. Karthikeyan, Piyush Trivedi, and Anita Dutt Konar Abstract Kinases/phosphatases are a novel class of enzymes that maintains the equilibrium of the amino acids’ optimum phosphorylation/dephosphorylation strategies in proteins, responsible for regulating normal cellular metabolism. However, a slight imbalance in the kinase signaling pathways leads to hyperphosphorylation of the residues, resulting in pathological disorders eventually leading to neurodegenerative diseases. Therefore, the design of scaffolds commonly coined as inhibitors, which could resist the pathway of hyperphosphorylation, invites immense interest from diverse segments of bioorganic and medicinal chemistry. Although in recent years, significant contributions have been made in this avenue, we feel that there is a lot more room for development, as science never fails to create surprises. In this book chapter, our effort lies to modulate two different aspects: Firstly, the underlying concepts of Alzheimer’s disease are discussed, along with the general structural aspect of kinases. Secondly, the discussion moves toward an overview of other groups’ reported inhibitors of the CMGC family of kinases, namely, DYRK1A, CLK1, GSK3β, and CDK5. Finally, our effort lied to modulate how we have utilized the existing knowledge from the literature, with the assistance of computer-aided drug design approaches and fished out potent and selective inhibitors against these targets. Overall, we anticipate that this chapter would provide a conceptual demonstration of the different protocols adopted toward the construction and implementation of these potential therapeutics against Alzheimer’s disease. Key words Hyperphosphorylation, Neurofibrillary tangles, Kinase inhibitors, Molecular modelling studies

1

Introduction

1.1 Outline About Alzheimer’s Disease

Alzheimer’s disease (AD), the most prevalent neurodegenerative disease, commonly observed in elderly patients is characterized by two main criteria, firstly as an extracellular accumulation of amyloid-beta (Aβ) peptides as amyloid plaques and secondly aggregation of hyperphosphorylated Tau proteins as neurofibrillary tangles (NFTs) [1–10]. Focusing on the latter aspect, Tau is a microtubuleassociated protein (MAP) highly expressed in the central nervous

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_6, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

165

166

Arvind Kumar Jain et al.

Fig. 1 The phosphorylation–dephosphorylation phenomena in normal cycle and diseased case

system (CNS) that possesses the ability to stabilize the neuronal cytoskeleton crucial for normal cell metabolism. Indeed kinases/ phosphatases are the novel class of enzymes that maintains the equilibrium of optimum phosphorylation/dephosphorylation phenomena in the consensus residues in proteins. However, slight disparity in the activity of the enzymes leads to hyperphosphorylation of the amino acids, resulting in the instability of MTs that gives rise to microtubule disassembly. Subsequently, the hyperphosphorylated Tau gets accumulated in the form of paired helical filaments (PHFs) which leads to the prohibition of intracellular transport. Eventually, neuronal degradation occurs leading to pathological disorders and finally to neurodegenerative diseases (Fig. 1) [1–10]. Figure 2 demonstrates a representation of a healthy and a diseased neuron. Henceforth, targeting Tau becomes a promising approach for the development of therapeutics against Alzheimer’s disease. But before stepping toward it, understanding the structural aspect of Tau seemed important. So we delved deeper in order to gain an insight into the general structural aspect of the protein kinases. 1.2 Structural Aspect of Tau

The X-ray crystallographic analysis revealed that the catalytic forum of kinases has been broadly categorized into two major subdomains, namely, the N-terminus “lobe” consisting of five β-strands and the larger C-terminal lobe with an α-helical conformation. These two lobes are tethered by a peptidyl fragment at the hinge region, forming a cleft (Fig. 3) [11, 12]. Interestingly the latter (the cleft) comprises two different pockets: (a) The front pocket contained residues, directly involved in the catalysis or the ATP binding. (b) On the contrary, the pocket commonly known as the hydrophobic zone, slightly behind the cleft, is responsible for supporting regulatory functions. The functions played by these pockets are as described: (a) In the front pocket, the adenine of ATP forms an H-bond with the amide backbone of the hinge region, initially (Fig. 3, shown in green) [11, 12]. (b) Secondly, a flexible

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

167

Fig. 2 The tentative representation of a healthy and diseased neuron

N-terminal loop, commonly known as the G-loop/P-loop, binds to the nontransferable α- and β-phosphate groups of ATP through ionic interactions. The activity of the dynamic G-loop depends on the state of the catalytic domain and the presence of ligands. Enzymatic activation is mainly governed by the orientation of the C-helix, which interacts with the C-terminal activation segment to alter the alignment of catalytic residues. The activation segment has multiple elements with numerous functions. Firstly, the conserved DFG sequence is

168

Arvind Kumar Jain et al.

Fig. 3 Structures of the catalytic domain of kinases responsible for introducing selectivity and potency in the inhibitors

crucial for the catalysis orients itself such that the aspartate residue binds directly to the magnesium ion cofactor orienting the phosphate of ATP for transfer. Under such conditions, the activation loop typically containing consensus residues phosphorylates the sites, critical to the regulation of catalytic activity by creating a network of strong hydrogen bonds with the catalytic active site residues and the phosphorylation acceptor binding site (Fig. 3). Studies revealed that the interaction at the hinge region is responsible for the potency whereas that at the activation loop is known for playing a key role in selectivity [11, 12]. In an effort to devise potential therapeutics for Alzheimer’s disease, in this book chapter, our effort modulates to analyze few members of the CMGC family, namely, DYRK1A, CLK1, GSK3β, and CDK5, which are mainly known to be responsible for the hyperphosphorylation of Tau [1–4]. In this perspective our approach lies to describe an overview of the inhibitors reported to date, the prime interactions responsible for the activity, and finally our journey to build the inhibitors for the particular receptors.

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

169

Table 1 Some of the residues at the active site of CLK1 and DYRK1A Receptors

167

168

172

175

189

191

206

225

CLK1

L

G

F

V

A

K

E

V

DYRK1A

I

G

F

V

A

K

E

V

Receptors

241

244

292

293

295

324

325

CLK1

F

L

E

N

L

V

D

DYRK1A

F

L

E

N

L

V

D

2

Design of Inhibitors for the Receptor DYRK1A/CLK1 It has been noted that at the active site, the receptors DYRK1A/ CLK1 differ only at 167th residue, which is leucine for CLK1 and isoleucine for DYRK1A, whereas the other residues are identical (Table 1). Inspired by the close homology, nearly 70.4% in these targets, we performed a thorough investigation of the potent inhibitors reported to date, with a common nucleus, which showed activity in both the kinases [13–15]. The inhibitors were designed with the idea that they would enter into intermolecular H-bonding with different residues at the active site of the receptor that has the probability of getting hyperphosphorylated. Thus reduction in the case of aberrant hyperphosphorylation would occur, leading to the development of therapeutics for Alzheimer’s disease. Scheme 1 shows the structures of some of the selected inhibitors of the target DYRK1A/CLK1 with their IC50 values. From the literature we found Lu’s molecule to be one of the most potent derivatives among others and selective for CLK1 (Scheme 1, XV). Therefore, we searched for the interactions present therein and found that in the receptor, Leu244, Phe 241, Phe 172, and Lys 191 were the main residues known for their correspondences with different parts of the inhibitors reported, which were responsible for conferring potency and selectivity in the molecules (Scheme 1) [16–27].

3

Our Design Approach and Contribution to CLK1/DYRK1A Inspired by the report from Prof. Lu’s group, we chose the heterocycle 1,2,4 triazole for our design keeping in mind the better biological relevance of the nucleus in comparison to the isomer reported (Fig. 4) [28]. Next, the hinge region in the receptors was characterized to be narrow. So we intended to select a moiety capable enough to pass on this narrow loop and maximize

170

Arvind Kumar Jain et al.

Scheme 1 Structures of some of the inhibitors of the target DYRK1A/CLK1 with their IC50 values

interactions at the active site. Therefore, we introduced flexibility in the form of a linker (hydrazine/amines) on either side of the core (Fig. 4) and started to play with various substituents which could function as H-bond donors/acceptors. As a control, the unsubstituted group (template) was taken to compare the effect of different substitutions on the target inhibition elicited by these analogs. Indeed, the substituents were extracted from two databases, namely, Specs collection database and may bridge screening collection database with approximately 375,816 and 10,676 substituents on similar nuclei. The objective for the selection of these substituents was to formulate a structure–activity relationship and also to evaluate the ability of these functions to harness additional interactions at the active site, especially with amino acid side chains flanking between the two phenyl rings. Studies reveal that triazole heterocycles that incorporate sulfur in the form of mercapto- and thione-substitution show more potency compared to their parent unsaturated derivatives [29]. Henceforth, we decided to include this functionality in our design as presented in Fig. 5. Thus, after the design, compounds I–XVI were synthesized and evaluated for their biological activity in both the receptors, namely, CLK1 and DYRK1A, at a final concentration of 10 mM. As described [28], indeed our results indicated that for the target CLK1, compound V was found to be the most potent with

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . . O

O S O O

N

O

N

O H N

Cl

S

NH O

171 N

S

N N

NH

O

N

N

O

N

N

O

O

S

N

Cl

O

(II): Ref.-17 IC50 DYRK1a-12nM CLK1- 19nM

(I): Ref.-16 IC50 DYRK1a-7000nM CLK1- 40nM

H 2N

CN

NH2

CO2C2H5

N

N

N

Cl

S

(VIII): Ref.-10 IC50 DYRK1A-260nM CLK1- 51nM

S N

CONH2

N

N

OH

(VII): Ref.-20 IC50 DYRK1A-55.2nM CLK1- 19.7nM

(VI): Ref.-19 IC50 DYRK1A-0.028nM CLK1- 0.048PM

H N

N

S

H3C

(V): Ref.-18 IC50 DYRK1a-62nM CLK1- 39nM

Br

NH

CH3

Cl

H 2N

N

(IX): Ref.-21 IC50 DYRK1A-2.6PM CLK1- 0.7PM

(X): Ref.-22 IC50 DYRK1a-1.8PM CLK1- 3.3PM

N

HOOC I N

N

(IV): Ref.-18 IC50 DYRK1a-14nM CLK1- 20nM

O

S N

NH2

N

(III): Ref.-17 IC50 DYRK1a-282nM CLK1- 68nM

Cl

NH

N

NO2

N O

NH2

N

NH2 N

(XII): Ref.-24 IC50 DYRK1a-0.20PM CLK1- 0.049PM

N

(XIII): Ref.-25 IC50 DYRK1a-0.62PM CLK1- 0.069PM

(XIV): Ref.-26 IC50 DYRK1a-80nM CLK1- 72nM

Fig. 4 Design strategy for the inhibitors for the receptors CLK1 and DYRK1A

Fig. 5 The structures of our designed inhibitors

N N H

Cl

(XI): Ref.-23 IC50 DYRK1a-6nM CLK1- 500nM

F

O

N N

N N

(XV): Ref.-27 IC50 DYRK1a-138 nM CLK1- 2 nM

172

Arvind Kumar Jain et al.

an IC50 of 0.97 μM in comparison to compound III (IC50 of 1.38 μM) and compound IV (IC50 of 2.58 μM) [28]. To validate our biological results, we took the help of molecular modelling where the compounds were docked into the X-ray crystal structures of CLK1 and DYRK1A, retrieved from the Protein Data Bank (PDB ID: 1Z57 and 3ANR, respectively) as reported [28, 30, 31]. The structure of the kinase inhibitor hymenialdisine (HMD) co-crystallized with CLK1 was used as the reference (Fig. 6a, b) [28, 30]. Our computational analysis for the receptors CLK1 has been presented in Fig. 6c–h. From the studies, it has been evident that in CLK1, when we observed from the direction of the heterocycle of compound V, we noticed that the thiol-functionalized triazole suitably adjusted at the entrance of the ATP binding site to form a H-bond with Glu169 [28]. Next, the two linkers assisted the heterocycle to be tethered, such that the molecule could approach closer to the hinge region establishing not only a H-bond with Asn 293 but also strong π-π interaction with Phe 241, the gatekeeper residue as described [28]. We feel that these few interactions have collectively attributed to the cause of such high potency and selectivity in the compounds. Also, we speculated that these correspondences would have never been observed without the intervention of flexible linkers [28]. However, in compounds III and IV, the interactions with all the main residues as listed above were absent which could be the reason for slightly lower potency in the compounds [28]. Next, the molecular modelling studies of the receptor DYRK1A, with the co-crystallized ligand harmine, revealed interactions with the residues Asp307, Glu203, Phe170, and Leu241 (Fig. 7) [31]. So, when we applied the same protocol as that for CLK1, in DYRK1A, and found that the 2,4-dihydroxy derivative (compound III) formed H-bonding interactions extensively over other non-covalent forces. As a result, the linkers twisted in such a way that the triazole nucleus was compelled to adopt a planar conformation such that favorable H-bonds with the Asn 244 and others could take place, accounting for its inhibitory activity in the micromolar range (IC50: 2.24 μM) [28]. These stabilizing interactions were absent in compound IV (positional isomer of compound III), the vital cause for a higher IC50 value than that of the former (IC50: 2.87 μM). Therefore, from the same series, we found compound V to be selective for CLK1 but compounds III and IV to be active against both the targets. Thus, through this work, we were able to focus not only on how the computer-aided drug design approach allowed us to develop a potential therapeutic against the receptors CLK1/ DYRK1A by mere changes of functionality but also validating the biological activity results in vitro.

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

173

Fig. 6 3D and 2D computation images of compounds III, IV, and V docked into the active site of CLK1 along with hymenaldisine (HMD) (a–h). (Represented with permission from ref. 28)

174

Arvind Kumar Jain et al.

Fig. 7 3D and 2D computation images of compounds III–IV docked into the active site DYRK1A with the reference ligand Harmine (a–f). (Represented with permission from Ref. [28])

4

Design of Inhibitors for the Receptors GSK3β and CDK5 Literature documentation revealed that CDK5 and GSK3β receptors are interrelated kinases that share a strong structural resemblance to each other [32–36]. Schemes 2 and 3 show the structures of some of the selected inhibitors of the target GSK3β/CDK5 with their IC50 values. The potential inhibitors reported to date (Schemes 2 and 3) indicated that for the receptors GSK3β and CDK5, Val 135 and Cys 83, respectively, were the main residues known to confer potency in the molecules [33–37]. However, the other important residues known for assisting in activity were Arg141/Thr 138 for GSK3β and Glu81/Phe80 for CDK5, respectively [4, 38–42].

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

175

Scheme 2 Structures of some of the selected inhibitors of the target GSk3β with their IC50 values

But in spite of such importance, most of the reported molecules failed to attain clinical trials. In an attempt to address the challenge, we primarily analyzed the residues involved at the hinge region and found that for GSK3β, the core was more hydrophobic in nature (Val 135, log P: -2.26), compared to CDK5 (Cys 83, log P: -2.49). So for better potency, a more hydrophobic scaffold should better interact with a more hydrophobic core and vice versa (less hydrophobic scaffold with the less hydrophobic core using the principle of “like dissolves like”) [44, 45]. Inspired by the preliminary result on quinoxalines from Professor Trivedi (Scheme 2; XX), we decided to select the same as our primary core and optimize the series [38]. As the first step, we attempted to introduce electron-donating and electronwithdrawing groups in the 6,7 positions of the nucleus, keeping other parameters of the reported molecule unchanged (Table 2). Next, we performed some DFT calculations where we calculated the zero-point vibrational energy and HOMO/LUMO difference of both the nucleus (Table 2). Interestingly we found that methyl groups at 6 and 7 positions were found to possess the least zero-point vibrational energy and maximum HOMO/LUMO difference. Our main thrust was to develop a kinase inhibitor suitable

176

Arvind Kumar Jain et al.

Scheme 3 Structures of some of the selected inhibitors of the target CDK5 with their IC50 values

for brain targeting. So, we calculated another parameter, the log P value, that gave an idea about the extent of hydrophobicity in a molecule. We learned that the higher the value, the more potential the inhibitor would have in its action. Now from the log P value, we confirmed that nucleus 1 with electron-donating methyl groups at the 6,7 positions possessed the maximum value (Table 2). So, with this idea, we decided to consider the series and attempted substituent orchestration in the other aromatic nucleus tethered to the linker by systematic enhancement/depletion of electron density (Fig. 8) [43–45]. We mainly tried to incorporate methoxy/ethoxy and halide as substituents, keeping in mind that the reported GSK3β/CDK5 was known to show promising effects with these substituents. The compounds were then synthesized and subjected to biological evaluation. Indeed our results revealed that in the

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

177

Table 2 DFT calculations of reported molecule, nucleus 1, and nucleus 2

Molecules

H

N

H N

H

N H

O

N

Zero-point (-) 178.14 (Kcal/mol) vibrational energy

Nucleus 2

Nucleus 1

Reported molecule

OCH3 H3C

N

H N

OH

N H

O

H3C

N

OCH3 O2N

N

H N

O2N

N H

O

OH

N

(-) 192.56 (Kcal/mol)

-180.664 (Kcal/mol)

HOMO/ LUMO

-0.13931 (eV)

-0.14206 (eV)

-0.11324 (eV)

Log P

2.07

3.17

1.6

OCH3 OH

receptor CDK5, compound III was found to be the most potent with an IC50 of 1.24 μM, unlike compounds VI (IC50 of 1.88 μM) and II (IC50 of 3.43 μM) (Fig. 8) [44, 45]. However, compounds IV and V, which are halide derivatives, were screened to be the most active GSK3β inhibitors with IC50 of 0.27 μM and 0.39 μM, respectively (Fig. 8) [44, 45]. To validate the inhibitory activities, we performed molecular modelling studies, where we took the help of indirubin-3′-oxime as the reference ligand and 1UNH as the grid for the receptor CDK5 [44]. Next we redocked compounds II, III, and VI using the same constraints as that of the reference ligand. However, for compound III (methoxy), with the highest activity (IC50 of 1.24 μM), we noted that in addition to other interactions, the H-bonding correspondences with Cys 83 carbonyl were observed (Fig. 9) [43– 45]. To our surprise, when the steric bulk of the methoxy group was increased to ethoxy (compound VI), the interaction with Cys 83 was missing, the probable reason for reduced activity (IC50 of 1.88 μM) (Fig. 9) [44, 45]. Next, we chose the grid 4AFJ and the co-crystallized ligand 5-aryl-4-carboxamide-1,3-oxazole, for the receptor GSK3β for our computational investigation. Thereafter, we redocked all the compounds using the same constraints as that of the reference ligand. Our studies revealed that for the bromo substituent, compound IV, and chloro substituent, compound V, interactions with the residues of the hinge region Val 135 and its neighbors Thr 138/Arg 141 were obtained. These two interactions were known to confer selectivity and potency in the molecules (Fig. 10) [44]. To check whether the halogens and hydroxyl were ample enough to bring about interesting interactions, compounds VII and VIII were designed. However, we failed to observe interactions in the hinge region and its surroundings.

178

Arvind Kumar Jain et al.

Fig. 8 Structures of the designed compounds I–IX of the quionoxaline series. (Represented with permission from Ref. [43])

Thus from the same series, we found that the halide derivatives were selectively GSK3β inhibitors (compounds IV and V), while alkoxy derivatives were CDK5 inhibitors (compounds III and VI) (Fig. 8). Thus, the importance of non-covalent interactions in generating a potential therapeutic against a particular receptor has been explicitly emphasized.

5

General Docking Protocol To investigate the possible binding mode in the ATP binding site of the target proteins, a molecular docking study was accomplished for the designed compounds. To achieve this, these molecules were docked in the X-ray crystal structure of CLK1, DYRK1A, CDK5, and GSK3β obtained from the Protein Data Bank at the Research Collaboratory for Structural Bioinformatics (RCSB) (http://www. rcsb.org) (PDB ID: 1Z57, 3ANR, 1UNH, and 4AFJ, respectively). The structures of the kinase inhibitor co-crystallized with CLK1, DYRK1A, CDK5, and GSK3β kinase were used as reference ligands

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

179

Fig. 9 3D and 2D computation images of compounds II, III, and VI docked into the active site CDK5 with the reference ligand indirubin-3′-oxime (a, c, e, g) 3D images (b, d, f, h) 2D images. (Represented with permission from Ref. [44])

180

Arvind Kumar Jain et al.

Fig. 10 3D and 2D computation images of compounds IV–V docked into the active site GSK3β with the reference ligand 5-aryl-4-carboxamide-1,3-oxazole (a, c, e) 3D images (b, d, f) 2D images. (Represented with permission from Ref. [45])

to define the binding site cavity [28, 43–45]. The structures of the designed compounds were constructed using the builder module of Maestro (v 9.3.5) along with the LigPrep module of the Schro¨dinger suite. LigPrep modifies bond orders according to their data and generates different conformers for the input structures. Each generated structure was subjected to energy minimization with the OPLS (optimized potential for liquid simulation) force field to eliminate the bond length and bond angles biased from the crystal

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

181

structure. Finally, the LigPrep produced structures with various ionization states, tautomers, stereochemistry, and ring conformations. The ligand structure obtained from LigPrep v 2.5 runs was given as the input for docking simulation at the ATP binding site of CLK1, DYRK1A, 1UNH, and 4AFJ [28, 43–45]. The structure of the target proteins was imported to the Maestro module (v9.3) available in the Schro¨dinger package (v2012), and the protein was optimized by using the Protein Preparation Wizard. This optimization includes adding hydrogens, assigning correct bond orders, and building disulfide bonds. The protonation state of all the ionizable residues was predicted by PROPKA provided in the Protein Preparation Wizard. Furthermore, the optimization automatically optimizes hydroxyl, Asn, Gln, and His states using ProtAssign Protein Preparation Wizard. The restrained minimization job was performed using the OPLS-2005 force field with converging heavy atoms restrained or frozen to RMSD 0.30 Å and hydrogens unrestrained [43–49]. The ATP binding site present in a refined protein model of CLK1, DYRK1A, CDK5, and GSK3β was used to generate the grid for docking the ligands. The prepared ligands were docked at the generated grid using the Extra Precision mode of the Glide program v 5.8 (Schro¨dinger, Inc., New York, 2012) with the default functions. The best docked conformations of designed compounds were chosen for further analysis. All computations were carried out on a Dell Inspiron Core i3 processor workstation with a Windows 7 operating system [43–49]. Notes The selection of pharmacophore was done based on the Craig plot. The ring scaffolds were selected based on a thorough study of a database, the literature present in the existing field, and applying knowledge from the preliminary work done. The PDB file was selected based on key interactions at the active site of the receptor. Before designing the ligands, an understanding of the active site is necessary which would allow designing the inhibitors with high potency and selectivity. Along this line, the dimensions of the designed inhibitors would be preferable if it is the same/comparable to that of the co-crystallized ligand. In this process, care should be taken such that Lipinski’s rule for calculation of H-bond donors/acceptors has been followed and hydrophobicity is maintained. Finally, the interaction of the designed ligand/cocrystallized ligand should be compared [43–49].

6

Conclusion In summary, the deadly challenge of Alzheimer’s disease (AD), mainly caused by hyperphosphorylation of Tau, has been widely observed among elderly patients in developing countries. Kinases

182

Arvind Kumar Jain et al.

are the set of enzymes that are responsible for the optimum phosphorylation strategies of the amino acids in proteins. However, slight imbalance in the kinase signaling pathways leads to hyperphosphorylation of the residues and eventually to neurodegenerative disorders. Although in recent years, significant contributions have been made in this avenue, we feel that there is a lot more room for development, as science never fails to create surprises. Therefore, through this chapter firstly we wanted to convey the underlying concepts of Alzheimer’s disease, along with the general structural aspect of kinases. Next, we wanted to communicate three messages that should be kept in mind while designing ATP-competitive inhibitors: 1. The mimetic should be designed in such a way that when it approaches the close proximity of the active site, it might enter into specific interactions with some precise residues of the active site of the ATP binding pocket, thereby inducing the inhibitory activities in the nanomolar range. So the introduction of an appropriate number of H-bond donors/acceptors would be advisable. 2. Next, the key interactions of the reference ligand at the active site should be identified. Then ligands with diversified pharmacophore might be chosen that possess the eligibility to induce comparable interactions as that of the reference ligand. In this way, selectivity might be obtained. 3. Then by fine-tuning the electron density around the identified pharmacophore in appropriate locations, potency might be improved. Using the above strategy, we were successful in devising a set of heterocyclic derivatives comprising of 6,7 dimethyl Quinoxalines that displayed excellent selectivity on the targets GSK3β and CDK5, depending on the substituents tethered to the aromatic core connected to the linker. Detailed experiments demonstrated that the mild hydrophobic derivatives, namely, compounds III and VI, were the most potent inhibitors in the CDK5 receptor (micromolar) (Fig. 9) while the more hydrophobic compounds IV and V were active against GSK3β (Fig. 10) over other kinases, namely, DYRK1A and CLK1, which might be due to favorable non-covalent interactions with the other aromatic nucleus of the compounds and Arg 141/Thr 138 residues of the receptor. Such interdependency of pharmacological parameters (inhibitory activities) as well as target selectivity as a function of substituent orchestration in the aromatic core, based on the extent of hydrophobicity, remains an important and a rare finding, in the light of inhibitor design, backed by DFT calculations and molecular modelling studies.

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . .

183

In a nutshell, the assistance of the molecular modelling approach would lead to further optimization of intelligent nucleus, thereby not only leading to the discovery of novel pharmacophores but also marked selectivity against a pool of kinases, indeed implementing a distinct road map toward the design of potential therapeutics against Alzheimer’s disease.

Acknowledgments AKJ wishes to thank UGC (MANF-2015-2017-MAD-50686), New Delhi, for financial support. ADK is grateful to UGC (F.4(55)/2014(BSR)/FRP), New Delhi, for financial support. All the authors are extremely thankful to Mr. Rishabh Ahuja and Mr. Vaibhav Shivhare, for extending extensive help in preparing the book chapter, without which it would have been impossible for the authors to submit the present version. References 1. Prokop A (2013) The intricate relationship between microtubules and their associated motor proteins during axon growth and maintenance. Neural Dev 8:17 2. Johnson GVW, Stoothoff WH (2004) Tau phosphorylation in neuronal cell function and dysfunction. J Cell Sci 117:5721–5729 3. Verwilst P, Kim HS, Kim S, Kang C, Kim JS (2018) Shedding light on tau protein aggregation: the progress in developing highly selective fluorophores. Chem Soc Rev 47:2249–2265 4. Martin L, Latypova X, Wilson CM, Magnaudeix A, Perrin M-L, Yardin C, Terro F (2013) Tau protein kinases: involvement in Alzheimer’s disease. Ageing Res Rev 12:289– 309 5. Gao Y-L, Wang N, Sun F-R, Cao X-P, Zhang W, Yu J-T (2018) Tau in neurodegenerative disease. Ann Transl Med 6:175–175 6. Pir GJ, Choudhary B, Mandelkow E (2017) Caenorhabditis elegans models of tauopathy. FASEB J 31:5137–5148 7. Ni R, Ji B, Ono M, Sahara N, Zhang M-R, Aoki I, Nordberg A, Suhara T, Higuchi M (2018) Comparative in vitro and in vivo quantifications of pathologic tau deposits and their association with neurodegeneration in tauopathy mouse models. J Nucl Med 59:960–966 8. Venkatramani A, Panda D (2019) Regulation of neuronal microtubule dynamics by tau: implications for tauopathies. Int J Biol Macromol 133:473–483

9. Zhang F, Zhong R, Cheng C, Li S, Le W (2021) New therapeutics beyond amyloid-β and tau for the treatment of Alzheimer’s disease. Acta Pharmacol Sin 42:1382–1389 10. Karthikeyan C, Jharia P, Waiker DK, Nusbaum AC, Amawi H, Kirwen EM, Christman R, Arudra SKC, Meijer L, Tiwari AK, Trivedi P (2017) N-(1H-Pyrazol-3-yl)quinazolin-4amines as a novel class of casein kinase 1δ/ε inhibitors: synthesis, biological evaluation and molecular modeling studies. Bioorg Med Chem Lett 27:2663–2667 11. Schwartz PA, Murray BW (2011) Protein kinase biochemistry and drug discovery. Bioorg Chem 39:192–210 12. Johnson LN, Lowe ED, Noble ME, Owen DJ (1998) The structural basis for substrate recognition and control by protein kinases 1. FEBS Lett 430:1–11 13. Walter A, Chaikuad A, Helmer R, Loae¨c N, Preu L, Ott I, Knapp S, Meijer L, Kunick C (2018) Molecular structures of cdc2-like kinases in complex with a new inhibitor chemotype. PLoS One 13:e0196761 14. Iqbal K, Grundke-Iqbal I (1995) Alzheimer abnormally phosphorylated tau is more hyperphosphorylated than the fetal tau and causes the disruption of microtubules. Neurobiol Aging 16:375–379 15. Colwill K, Feng LL, Yeakley JM, Gish GD, Ca´ceres JF, Pawson T, Fu X-D (1996) SRPK1 and Clk/Sty protein kinases show distinct substrate specificities for serine/arginine-rich

184

Arvind Kumar Jain et al.

splicing factors. J Biol Chem 271:24569– 24575 16. Mura´r M, Dobiasˇ J, Sˇramel P, Addova´ G, Hanquet G, Boha´cˇ A (2017) Novel CLK1 inhibitors based on N-aryloxazol-2-amine skeleton – a possible way to dual VEGFR2 TK/CLK ligands. Eur J Med Chem 126:754– 761 17. Coombs TC, Tanega C, Shen M, Wang JL, Auld DS, Gerritz SW, Schoenen FJ, Thomas CJ, Aube´ J (2013) Small-molecule pyrimidine inhibitors of the cdc2-like (Clk) and dual specificity tyrosine phosphorylation-regulated (Dyrk) kinases: development of chemical probe ML315. Bioorg Med Chem Lett 23: 3654–3661 18. Rosenthal AS, Tanega C, Shen M, Mott BT, Bougie JM, Nguyen D-T, Misteli T, Auld DS, Maloney DJ, Thomas CJ (2011) Potent and selective small molecule inhibitors of specific isoforms of Cdc2-like kinases (Clk) and dual specificity tyrosine-phosphorylation-regulated kinases (Dyrk). Bioorg Med Chem Lett 21: 3152–3158 19. Esvan YJ, Zeinyeh W, Boibessot T, Nauton L, The´ry V, Knapp S, Chaikuad A, Loae¨c N, Meijer L, Anizon F, Giraud F, Moreau P (2016) Discovery of pyrido[3,4-g]quinazoline derivatives as CMGC family protein kinase inhibitors: design, synthesis, inhibitory potency and X-ray co–crystal structure. Eur J Med Chem 118:170–177 20. Fedorov O, Huber K, Eisenreich A, Filippakopoulos P, King O, Bullock AN, Szklarczyk D, Jensen LJ, Fabbro D, Trappe J, Rauch U, Bracher F, Knapp S (2011) Specific CLK inhibitors from a novel chemotype for regulation of alternative splicing. Chem Biol 18:67–76 21. Lawson M, Rodrigo J, Baratte B, Robert T, Delehouze´ C, Lozach O, Ruchaud S, Bach S, Brion J-D, Alami M, Hamze A (2016) Synthesis, biological evaluation and molecular modeling studies of imidazo[1,2-a]pyridines derivatives as protein kinase inhibitors. Eur J Med Chem 123:105–114 22. Walter A, Chaikuad A, Loae¨c N, Preu L, Knapp S, Meijer L, Kunick C, Koch O (2017) Identification of CLK1 inhibitors by a fragment-linking based virtual screening. Mol Inform 36:1600123 23. Falke H, Chaikuad A, Becker A, Loae¨c N, Lozach O, Abu Jhaisha S, Becker W, Jones PG, Preu L, Baumann K, Knapp S, Meijer L, Kunick C (2015) 10-Iodo-11 H -indolo[3,2c]quinoline-6-carboxylic acids are selective inhibitors of DYRK1A. J Med Chem 58: 3131–3143

24. Deau E, Loidreau Y, Marchand P, Nourrisson M-R, Loae¨c N, Meijer L, Levacher V, Besson T (2013) Synthesis of novel 7-substituted pyrido [2′,3′:4,5]furo[3,2-d]pyrimidin-4-amines and their N-aryl analogues and evaluation of their inhibitory activity against Ser/Thr kinases. Bioorg Med Chem Lett 23:6784–6788 25. Zeinyeh W, Esvan YJ, Josselin B, Baratte B, Bach S, Nauton L, The´ry V, Ruchaud S, Anizon F, Giraud F, Moreau P (2019) Kinase inhibitions in pyrido[4,3-h] and [3,4-g] quinazolines: synthesis, SAR and molecular modeling studies. Bioorg Med Chem 27: 2083–2089 26. Nguyen TL, Fruit C, He´rault Y, Meijer L, Besson T (2017) Dual-specificity tyrosine phosphorylation-regulated kinase 1A (DYRK1A) inhibitors: a survey of recent patent literature. Expert Opin Ther Pat 27:1183– 1199 27. Sun Q-Z, Lin G-F, Li L-L, Jin X-T, Huang L-Y, Zhang G, Yang W, Chen K, Xiang R, Chen C, Wei Y-Q, Lu G-W, Yang S-Y (2017) Discovery of potent and selective inhibitors of Cdc2-like kinase 1 (CLK1) as a new class of autophagy inducers. J Med Chem 60:6337– 6352 28. Jain AK, Karthikeyan C, McIntosh KD, Tiwari AK, Trivedi P, DuttKonar A (2019) Unravelling the potency of 4,5-diamino-4 H -1,2,4 triazole-3-thiol derivatives for kinase inhibition using a rational approach. New J Chem 43: 1202–1215 29. Sˇermuksˇnyte A, Kantminiene K, Jonusˇkiene I, Tumosiene I, Petrikaite V (2022) The effect of 1,2,4-triazole-3-thiol derivatives bearing hydrazone moiety on cancer cell migration and growth of melanoma, breast, and pancreatic cancer spheroids. Pharmaceuticals 15: 1 0 2 6 . h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / ph15081026 30. Fant X, Durieu E, Chicanne G, Payrastre B, Sbrissa D, Shisheva A, Limanton E, Carreaux F, Bazureau J-P, Meijer L (2014) Cdc-like/dual-specificity tyrosine phosphorylation–regulated kinases inhibitor leucettine L41 induces mTOR-dependent autophagy: implication for Alzheimer’s disease. Mol Pharmacol 85:441–450 31. Miyazaki Y, Maeda Y, Sato H, Nakano M, Mellor GW (2008) Rational design of 4-amino5,6-diaryl-furo[2,3-d]pyrimidines as potent glycogen synthase kinase-3 inhibitors. Bioorg Med Chem Lett 18:1967–1971 32. Wen Y, Planel E, Herman M, Figueroa HY, Wang L, Liu L, Lau L-F, Yu WH, Duff KE (2008) Interplay between cyclin-dependent kinase 5 and glycogen synthase kinase

Computer-Assisted Drug Design: A Toolbox for Novel Tau Kinase Inhibitors. . . 3 mediated by neuregulin signaling leads to differential effects on tau phosphorylation and amyloid precursor protein processing. J Neurosci 28:2624–2632 33. Engmann O (2009) Crosstalk between Cdk5 and GSK3β: implications for Alzheimer’s disease. Front Mol Neurosci 2:2 34. Kimura T, Ishiguro K, Hisanaga S (2014) Physiological and pathological phosphorylation of tau by Cdk5. Front Mol Neurosci 7:65 35. Jankowska A, Satała G, Bojarski AJ, Pawłowski M, Chłon´-Rzepa G (2021) Multifunctional ligands with glycogen synthase kinase 3 inhibitory activity as a new direction in drug research for Alzheimer’s disease. Curr Med Chem 28:1731–1745 36. Chow H-M, Guo D, Zhou J-C, Zhang G-Y, Li H-F, Herrup K, Zhang J (2014) CDK5 activator protein p25 preferentially binds and activates GSK3β. Proc Natl Acad Sci 111:E4887– E4895 37. Khan I, Tantray MA, Alam MS, Hamid H (2017) Natural and synthetic bioactive inhibitors of glycogen synthase kinase. Eur J Med Chem 125:464–477 38. Jain AK, Karthikeyan C, Trivedi P. Unpublished Work 39. Meijer L, Skaltsounis A-L, Magiatis P, Polychronopoulos P, Knockaert M, Leost M, Ryan XP, Vonica CA, Brivanlou A, Dajani R, Crovace C, Tarricone C, Musacchio A, Roe SM, Pearl L, Greengard P (2003) GSK-3-selective inhibitors derived from Tyrian purple indirubins. Chem Biol 10:1255–1266 40. Kaller MR, Zhong W, Henley C, Magal E, Nguyen T, Powers D, Rzasa RM, Wang W, Xiong X, Norman MH (2009) Design and synthesis of 6-oxo-1,6-dihydropyridines as CDK5 inhibitors. Bioorg Med Chem Lett 19: 6591–6594 41. Zhong W, Liu H, Kaller MR, Henley C, Magal E, Nguyen T, Osslund TD, Powers D, Rzasa RM, Wang H-L, Wang W, Xiong X, Zhang J, Norman MH (2007) Design and

185

synthesis of quinolin-2(1H)-one derivatives as potent CDK5 inhibitors. Bioorg Med Chem Lett 17:5384–5389 42. Shiradkar M, Thomas J, Kanase V, Dighe R (2011) Studying synergism of methyl linked cyclohexyl thiophenes with triazole: synthesis and their cdk5/p25 inhibition activity. Eur J Med Chem 46:2066–2074 43. Jain AK, Malla S, Karthikeyan C, Tiwari AK (2020) In search of novel tau kinase inhibitors and their role in Alzheimer’s therapy. Nova Science Publisher. ISBN: 9781536169737 44. Jain AK, Gupta A, Karthikeyan C, Trivedi P, Dutt Konar A (2022) Substituent orchestration in dimethylquinoxaline derivatives: a tool for fishing out appropriate CDK5 inhibitors as potential therapeutics for Alzheimer’s. Chem Biodivers 19:e202200116 45. Jain AK, Gupta A, Karthikeyan C, Trivedi P, Dutt Konar A (2021) Unravelling the selectivity of 6,7-dimethyl quinoxaline analogs for kinase inhibition: an insight towards the development of Alzheimer’s therapeutics. Chem Biodivers 18:e2100364 46. Waiker DK, Karthikeyan C, Poongavanam V, Kongsted J, Lozach O, Meijer L, Trivedi P (2014) Synthesis, biological evaluation and molecular modelling studies of 4-anilinoquinazoline derivatives as protein kinase inhibitors. Bioorg Med Chem 22: 1909–1915 47. Gabr MT, El-Gohary NS, El-Bendary ER, El-Kerdawy MM (2015) Structure-based drug design and biological evaluation of 2-acetamidobenzothiazole derivative as EGFR kinase inhibitor. J Enzyme Inhib Med Chem 30:160–165 48. Schr€ odinger Suite (2012) Schr€ odinger. LLC, New York 49. Madhavi Sastry G, Adzhigirey M, Day T, Annabhimoju R, Sherman W (2013) Protein and ligand preparation: parameters, protocols, and influence on virtual screening enrichments. J Comput Aided Mol Des 27:221–234

Chapter 7 Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease Agents: Case Studies of Phosphodiesterase Inhibitors Vinay Kumar and Kunal Roy Abstract Alzheimer’s disease (AD) is one of the major public health concerns. Phosphodiesterases (PDEs) are a major class of enzymes which hydrolyze two second messengers: cyclic adenosine monophosphate (cAMP) and cyclic guanosine monophosphate (cGMP). Due to the high expression of various PDE subfamilies in the human brain, PDE inhibition has a substantial impact on neurodegenerative diseases by controlling the level of cAMP or cGMP. In this regard, several synthetic or natural compounds that inhibit specific PDE subtypes, for instance, rolipram and roflumilast (PDE4 inhibitors), vinpocetine (PDE1 inhibitor), cilostazol and milrinone (PDE3 inhibitors), sildenafil and tadalafil (PDE5 inhibitors), etc., have been stated as exhibiting excellent results for the treatment of AD. PDEs are currently believed to be a potential target for the treatment of AD since several PDE inhibitors have demonstrated significant cognitive improvement effects in preclinical investigations and more than 33 of them have been subjected to clinical trials. In the search for novel drugs, computational drug design methods are now essential. Computational approaches, whether structure-based (protein structure prediction, molecular docking, MD simulation, pharmacophore modeling, fragment-based de novo design, etc.) or ligand-based (QSAR, chemical read-across, pharmacophore modeling, similarity search), are used in almost every drug discovery project. To investigate new drugs, many drug targets have been researched employing computational techniques. Many researchers across the world have recently focused on the development of more advanced and selective phosphodiesterases as treatments for inflammatory illnesses, CNS disorders (including Alzheimer’s disease), and numerous other diseases. The majority of these groups have used computational tools for drug discovery and design at various stages of their research. The objective of the current chapter is to provide a concise summary of the most relevant and recent research on PDE inhibitors as anti-AD therapeutics with promising results utilizing various computational modeling techniques, which can assist in the further development and identification of new anti-AD agents. In this chapter, we will present relevant and recently published computational studies for the identification or design of potential PDE inhibitors using various computational approaches. Moreover, the chapter will give the audience a broad overview of effective computational drug discovery research in this particular field of applications. Key words Alzheimer’s disease, Phosphodiesterase, QSAR, Pharmacophore modeling, Molecular docking, MD simulation

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_7, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

187

188

1

Vinay Kumar and Kunal Roy

Introduction Alzheimer’s disease (AD), the most prevalent form of dementia, is clinically identified by memory deterioration, cognitive problems, growing reliance, and altered behavior in the elderly [1, 2]. In addition, there are other nonclassical characteristics (not centered on memory), particularly in younger instances, such as visuospatial difficulties, dyspraxia, and anomia [1, 2]. Approximately 55 million people worldwide are affected by this neurological condition in 2021, according to data from the World Alzheimer’s Report 2021 [1, 2]. This population is growing quickly, and by 2030 and 2050, it will be 80 and 152 million, respectively [1, 2]. According to the report, the entire cost of AD care is expected to rise from $1 trillion in 2021 to $2 trillion by 2030 [2, 3]. Along with rising morbidity, there is also a high death rate that must be taken into consideration, such as in the USA, mortality climbed by 145% between 2000 and 2017, especially among persons aged 65 years and older [2, 3]. As a result, AD is one of the world’s most serious health issues, posing a significant burden on individuals, their families, and the community [1–3]. So far, only four medications have been approved by the FDA to treat AD; these are donepezil, rivastigmine, galantamine, and memantine [1–3]. The first three drugs are used in the first 3 months by inhibiting acetylcholinesterase (AChE), while the last one alleviates severe AD by targeting the N-methyl-D-aspartate receptor (NMDAR) [1–3]. Very recently, an anti-amyloid antibody drug named aducanumab (Aduhelm™) developed by Biogen and Eisai got approval (7 June 2021) as a treatment for AD from the US Food and Drug Administration (FDA) [4]. However, the study data provoked a heated debate among experts following Aduhelm’s fast-track clearance [4]. In patients with AD, memory and cognition can be improved, but the disease is incurable. Therefore, finding new targets and therapeutic approaches is urgently needed in AD therapy. Two primary targets, β-amyloid (Aβ) and tau protein, are suggested since senile plaques and neurofibrillary tangles are the key characteristics of AD [1–3]. Several researchers are currently paying close attention to both. Regrettably, phase III trials for many Aβ-targeting therapeutics, including bapineuzumab, solanezumab, and semagacestat (LY450139), have been terminated [1–3]. These two monoclonal antibodies are thought to target inappropriate Aβ isoforms [1– 3]. Some patients’ treatment may be too late if their pathological condition has advanced. Likewise, a tau aggregation inhibitor TRx0237 (LMTX) had disappointing results in a phase III clinical trial in mild to severe AD patients [1–3]. Therefore, it is still extremely difficult to find therapeutic interventions for AD. Several targets have been identified based on pathological studies of AD, including 5-hydroxytryptamine (5-HT) receptors,

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

189

glycogen synthase kinase-3 beta (GSK-3β), the third subtype of histamine (H3) receptor, and phosphodiesterase (PDE) [5, 6]. The primary function of the broad family of enzymes known as PDEs is to hydrolyze the 3′-phosphodiester link in cyclic adenosine monophosphate (cAMP) and cyclic guanosine monophosphate (cGMP), which are involved in signal transduction pathways [5, 6]. Due to their ability to modify cAMP and cGMP levels, they play a significant role in the pathways that are crucial for many pharmacological processes, including cell function [6]. The hippocampus-expressed cAMP response element binding protein (CREB) is regulated by cAMP/cGMP and promotes neuronal survival via pathways related to synapse strengthening and synaptic plasticity [7]. CREB positively regulates memory consolidation and performance by upregulating brain-derived neurotrophic factor (BDNF) [7]. Acute neuroinflammation has been shown to impact learning and memory-related CREB signaling in the hippocampus of mice via tumor necrosis factor (TNF)-dependent processes [6, 7]. Furthermore, Aβ reduces BDNF levels via a mechanism involving transcription factor CREB downregulation. It is widely understood that most of the drugs used against neurodegeneration, including cognition enhancers, target a precise neurotransmitter [5–7]. PDE inhibition contributes to neurodegeneration by enhancing cGMP and/or cAMP intracellular availability [5– 7]. Although PDEs are extensively expressed in the human brain, cAMP and cGMP levels can influence neurodegenerative processes [5–7]. From a broad perspective, PDE inhibitors could enhance cognition by controlling neural transmission by altering presynaptic neurotransmitter release and postsynaptic intracellular processes after extracellular neurotransmitter interaction [5–7]. As a result, researchers have found the PDE family to be an appealing multipotential target for numerous disease pathologies. PDEs are a large enzyme group composed of 11 isoenzyme families (PDE1– PDE11), with distinctions based on substrate specificities and affinities, kinetic properties, tissue and subcellular distributions, regulatory mechanisms, and drug and modulator susceptibility [5– 7]. These isoforms catalyze the hydrolysis of cAMP and cGMP, which are involved in the proliferation, differentiation, apoptosis, gene expression, visual transduction, inflammation, and metabolic pathways [5–7]. These isoenzyme families have been discovered to have over 40 PDE isoforms (encoded by 21 genes), with some specific to cAMP (i.e., PDE4, PDE7, and PDE8) and others unique to cGMP (i.e., PDE5, PDE6, and PDE9), while others can operate on both cAMP and cGMP (i.e., PDE1, PDE2, PDE3, and PDE11) [5–7]. The presence of different isoforms in PDE families was discovered in experimental animals in the 1970s [5–7]. As a result of the abundance of isoforms and subtypes in this enzyme family, isoform-/subtype-specific inhibitors should be created, which is a difficult task in drug research [5–7]. However, a significant effort

190

Vinay Kumar and Kunal Roy

has recently been made to develop novel phosphodiesterase inhibitors (PDEIs) as subtype-selective medicines addressing a variety of disorders using experimental and computational approaches [5–7]. Furthermore, PDEIs have a beneficial impact on cognition improvement through information processing, memory, and executive functioning [5–7]. There are few clinical studies on the effects of PDE inhibitors on cognitive function in AD, and the majority of the available data comes from preclinical animal models of AD, such as transgenic mice or central Aβ insertion [5–7]. Clinical experiments revealed that hippocampus mRNA expression of PDE4D and 8B was altered in age-related memory-impaired participants as well as in patients with mild to moderate AD [5–7]. The findings of behavioral preclinical and clinical studies revealed that PDE7 improved memory function in AD patients and altered PDE7A mRNA expression in the AD brain [5–7]. Heckman et al. [8] proposed that inhibiting PDE4D isoform subtypes is an appropriate target for AD treatments. As a potential new target for AD therapy, PDEIs may prove useful. More than ten medications have been given market approval thus far. The most successful PDE5 inhibitor, sildenafil (Viagra), is used to treat erectile dysfunction (ED), demonstrating the therapeutic potential of PDE targeting [5–7]. Furthermore, the PDE4 inhibitor rolipram has been shown to improve cognitive performance in AD mice, which substantially supports and promotes the development of PDE inhibitors for the treatment of AD. In the current chapter, we have outlined the computational modeling approaches frequently employed in in silico medicinal chemistry for the discovery of novel phosphodiesterase inhibitors as anti-Alzheimer’s disease drugs, and we have also highlighted the recent advancement in the development of PDE inhibitors, thus proposing a viewpoint in the therapeutic potential of targeting PDEs in AD. 1.1 Therapeutic Strategies Targeting PDEs and Existing PDE Inhibitors Clinically Being Used Against AD

A PDE inhibitor (PDEI) is a class of drug that can inhibit one or more PDE subtypes, blocking the inactivation of intracellular second messengers including cAMP and cGMP by the corresponding PDE subtypes [5, 9, 10]. PDEIs may improve cAMP/PKA, cGMP/PKG, and CREB signaling, as well as GSK3-controlled tau phosphorylation, and increase neuron and mitophagy survival, synaptogenesis, mitochondrial biogenesis, ATP generation, antioxidant and detoxifying enzyme expression, and memory formation by enhancing the levels of these secondary messengers [5, 9, 10]. For the time being, many PDE inhibitors have been authorized for use in the treatment of heart failure, pulmonary hypertension, erectile dysfunction, and pulmonary hypertension (Table 1) [5, 9, 10]. All of these would be clinically efficient in preventing AD as well as in slowing and reversing its course. The majority of the data is preclinical, making it difficult to conclude how PDEIs improve cognitive function in AD patients [5, 9, 10]. Rolipram is

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

191

Table 1 PDE inhibitors are approved by FDA for clinical use Compound

Brand name

Target Clinical use

Vinpocetine

Calan, Cavinton

PDE1

Parkinson’s disease, Alzheimer disease

Cilostazol

Pletal, Pletaal

PDE3

Antiplatelet, intermittent claudication

Amrinone lactate

Inocor

PDE3

Heart failure

Milrinone lactate

Primacor

PDE3

Congestive heart failure

Roflumilast

Daxas, Daliresp

PDE4

Chronic obstructive pulmonary disease

Apremilast

Otezla

PDE4

Psoriasis and psoriatic arthritis

Crisaborole

Eucrisa

PDE4

Allergic dermatitis

Sildenafil citrate

Viagra

PDE5

Erectile dysfunction, pulmonary arterial hypertension

Tadalafil

Cialis

PDE5

Erectile dysfunction, pulmonary arterial hypertension

Vardenafil HCL

Vivanza

PDE5

Erectile dysfunction

Avanafil

Zepeed, Stendra

PDE5

Erectile dysfunction

a well-known PDE4I inhibitor that can modify precognition and improve memory impairment induced by injection of the Aβ25–35 or Aβ1–40 peptide in animal models of AD (see Table 2) [5, 9, 10]. PDE4I is thought to potentially have an antidepressant effect while having many adverse effects. Similarly, the well-known PDEI sildenafil improves cognitive functions in a mouse model of AD [5, 9, 10]. Furthermore, Hiramatsu et al. [11] found that cilostazol (PDE3I) significantly lowers Aβ aggregation and improves memory deficits in Aβ-injected mice. PDEIs, on the other hand, show prospective effects in the treatment of learning and memory impairments associated with AD, and it is worth mentioning that PDE families are complicated due to their subtypes, which are encoded by different genes and are connected with the pathophysiology of numerous diseases [5, 9–11]. In addition to preclinical investigations, several PDEIs are currently in clinical trials for cognitive enhancement, particularly related to AD, as stated in Table 3.

2

Materials and Methods This section has addressed the fundamentals of the computational approaches and associated tools that have been employed by many researchers to model PDE inhibitors for the development of antiAlzheimer’s disease agents. There are various challenges in the search for innovative anti-Alzheimer’s disease drugs, due to the unidentified pathogenesis of this disease. Nonetheless, various research groups are constantly working to identify a novel

PDE1 Rats

Morris water maze, 3.0 mg/kg, ip PDE2 Chronic stress-induced Novel object recognition and location task, mice 3.0 mg/kg, ip Rats and C57BL/6J mice Object recognition and social memory Aged rats task, 1.0 mg/kg, po APP/PS1 mice Object recognition task, 0.3 mg/kg, ip Object location test, 0.3 mg/kg, po Y-maze, 0.3 mg/kg, po Forced swim test, 0.3 mg/kg, po Elevated zero maze, 0.3 mg/kg, po

PDE3 Aβ25–35-injected mice

PDE4 APP/PS1 mice Aβ25–35- or Aβ40-infused rats Streptozotocin-injected and aged mice Iron-impaired aged rat

ITI-214

BAY 60–7550

Cilostazol

Rolipram

Improves memory possibly due to the prevention of oxidative damage or by decreasing Aβ accumulation by the reduction of Aβ accumulation and tau phosphorylation Improves memory Improves memory Reduces oxidative-nitrosative stress; upregulates thioredoxin; inhibits iNOS/NO pathway Improves memory probably due to its anticholinesterase, antiamyloid, anti-oxidative, and anti-inflammatory effects Improves memory

Morris water maze, 0.03 mg/kg, sc Radial-arm water maze, 0.03 mg/kg, sc Morris water maze, 0.5 mg/kg, ip Passive avoidance, 0.5 and 1.25 mg/kg, ip Morris water maze, 0.05 mg/kg and 0.1 mg/kg, ip Object recognition, 0.01–0.1 mg/kg, ip

Reverses cognitive impairment via neuroplasticity-related NMDAR-CaMKII-cGMP/cAMP signaling Improves memory, enhanced LTP, reversed MK801-induced deficits Improves cognition and memory through enhancing the nNOS activity Improves memory without anxiety, depressive-like behavior, or hypothalamus–pituitary–adrenal axis regulation No changes on the level of Aβ, pCREB, BDNF, or presynaptic density

Improves acquisition, consolidation, and retrieval memory

Improves memory Reduces oxidative-nitrosative stress Modulates cholinergic functions Prevents neuronal damage

Outcome

Y-maze, 30 mg/kg, po Passive avoidance, 30 mg/kg, po Morris water maze, 10 mg/kg, ip

Object recognition, 0.3–3.0 mg/kg, ip

Morris water maze, 10 mg/kg, icv Passive avoidance, 10 mg/kg, icv

PDE1 Streptozotocin-injected rats

Vinpocetine

Model

Target Subject

Compound

Table 2 List of the PDE inhibitors studied in the animal models of AD

192 Vinay Kumar and Kunal Roy

(continued)

Reverses memory deficits by regulating the Akt/GSK3b and p25/CDK5 pathways Ameliorates age-dependent cognitive impairments, reduces the tau phosphorylation Improves memory by a long-lasting reduction of Aβ levels Restores cognitive deficits by the regulation of PKG/pCREB signaling, anti-inflammatory response, and reduction of Aβ levels Improves memory by reducing APP processing, Aβ levels, and tau hyperphosphorylation

Morris water maze, 15 mg/kg, ip Fear conditioning, 15 mg/kg, ip Morris water maze, 7.5 mg/kg, ip Fear conditioning, 3 mg/kg, ip Reversal Morris water maze, 3 mg/kg, ip Object recognition, 10 mg/kg, ip Passive avoidance, 7.5 mg/kg, ip

Sildenafil

PDE5 Tg2576 mice SAMP8 mice APP/PS1 mice SAMP8 and SAMR1 mice

D159687

Improves memory probably due to stimulation of the cAMP/ PKA/CREB/BDNF pathway and anti-inflammatory effects Little emetic potential

Improves memory at doses without emesis-like behavior

PDE4 Mice

GEBR-7b

Morris water maze, 0.5 mg/kg, po Passive avoidance, 0.5 mg/kg, po Xylazine/ketamine-induced anesthesia

Improves long-term recognition, memory, and cognition

Y-maze, 0.001 mg/kg, iv Object recognition, 0.01 mg/kg, iv Xylazine/ketamine-induced anesthesia

PDE4 APP/PS1 mice

FFPM

Object recognition, 0.01 mg/kg, ip Water maze delayed matching to position test, 0.3 mg/kg, ip

Enhances long-term memory Facilitates expression of CREB-regulated genes in aged hippocampus

PDE4 Scopolamine-injected mice Mice

PDE4 Rats

MK-0952

Fear conditioning, 0.1 mg/kg, ip Morris water maze, 0.15 mg/kg, ip

Improves memory Increased emetic-like properties at a dose 100 times the memory-enhancing dose No improvement on memory when given alone; fully restored memory deficit when given in combination with donepezil (0.1 mg/kg, po)

Improves memory at doses without emesis-like behavior

PDE4 Aged mice

HT-0712

Object location, 0.03 mg/kg, sc Y-maze, 0.1 mg/kg, sc Xylazine/ ketamine-induced anesthesia Object recognition, 0.1 mg/kg, ip

Object location and recognition, 0.003 mg/kg, sc Xylazine/ketamine-induced anesthesia

PDE4 Mice Scopolamine-impaired rats

Roflumilast

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . . 193

PDE5 Aged J20 mice

PDE7 APP/PS1 mice

PDE9 Scopolamine-injected rats Rats Tg2576 mice Aβ25–35-injected mice

Tadalafil

S14

BAY 73–6691 T-maze, 10 mg/kg, po Passive avoidance, 10 mg/kg, po Social recognition, 0.3 mg/kg, po Object recognition, 0.1 mg/kg, po Object location, 5 mg/kg, po Morris water maze, 1 mg/kg, ip

Radial arm water maze, 3 mg/kg, ip

Morris water maze, 15 mg/kg, po

Model

PFPDE9 Scopolamine-injected rats Conditioned avoidance attention, 3 mg/ kg, ip 04447943 Mice Object recognition, 1 mg/kg, po Social recognition, 1 mg/kg, po Y-maze, 1 mg/kg, po

Target Subject

Compound

Table 2 (continued)

Improves learning and memory Improves cognitive performance

Improves long-term and short-term memory Improves both early and late LTP Transforms early into late LTP Improves memory; restores Aβ-induced impairment of longterm potentiation Improves memory

Improves memory

Improves memory, reduces tau phosphorylation but not Aβ

Outcome

194 Vinay Kumar and Kunal Roy

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

195

Table 3 List of the PDE inhibitors in clinical trials for the treatment of AD or associated diseases

Target

Disease/condition

Phase

Completion date/year

NCT04854811 Roflumilast

PDE4

Memory and functional recovery

II (recruiting)

April 1, 2024

NCT04658654 Roflumilast

PDE4

Mild dementia patients

II (recruiting)

October 1, 2023

NCT05297201 CPL500036

PDE10A Parkinson’s disease

II (recruiting)

March 1, 2023

NCT01429740 PF-0999

PDE2

Schizophrenia

I (completed)

2011

NCT01530529 PFPDE2 05180999

Healthy volunteers

I (completed)

2012

NCT02584569 TAK-915

PDE2

Healthy volunteers

I (completed)

2015

NCT01409564 Cilostazol

PDE3

Alzheimer’s disease

IV (completed)

2011

NCT02491268 Cilostazol

PDE3

Mild cognitive impairment

II (recruiting)

2015

NCT01433666 Roflumilast

PDE4

Dementia

II (completed)

2011

NCT02051335 Roflumilast

PDE4

Memory impairment, Alzheimer’s disease

I (completed)

2014

NCT02835716 Roflumilast

PDE4

Alzheimer’s disease

Preclinical (recruiting)

2016

NCT02013310 HT-0712

PDE4

Age-associated memory impairment

II (completed)

2013

NCT00880412 Etazolate

PDE4

Alzheimer’s disease

II (completed)

2009

NCT03030105 BPN14770

PDE4

Alzheimer’s disease

I (not recruiting participants)

2017

NCT02840279 BPN14770

PDE4

Alzheimer’s disease

I (completed)

2016

NCT02648672 BPN14770

PDE4

Alzheimer’s disease

I (completed)

2016

NCT00455715 Sildenafil

PDE5

Schizophrenia

IV (completed)

2007

NCT01941732 Sildenafil

PDE5

Parkinson’s disease

IV (completed)

2013

NCT02450253 Tadalafil

PDE5

Dementia, vascular

II (recruiting)

2017

NCT00930059 PFPDE9 04447943

Alzheimer’s disease

II (completed)

2009

NCT00988598 PFPDE9 04447943

Alzheimer’s disease

I (completed)

2009

NCT01097876 PFPDE9 04447943

Healthy

I (completed)

2010

NCT02240693 BI-409306

Alzheimer’s disease

II (recruiting)

2014

Clinical trial identifier

Druga

PDE9

(continued)

196

Vinay Kumar and Kunal Roy

Table 3 (continued)

Target

Disease/condition

Phase

Completion date/year

PDE9

Alzheimer’s disease

II (recruiting)

2014

NCT02197130 PFPDE10 02545920

Huntington’s disease

II (completed)

2014

NCT02037074 EVP-6308

PDE10

Schizophrenia

I (completed)

2014

NCT01900522 ITI-214

PDE1

Schizophrenia

I (terminated)

2013

NCT00362024 MK-0952

PDE4

Alzheimer’s disease

II (terminated)

2006

NCT01215552 HT-0712

PDE4

Healthy elderly volunteers

I (terminated)

2010

NCT02162979 Sildenafil

PDE5

Parkinson’s disease

II (terminated)

2007

NCT02342548 PFPDE10 02545920

Huntington’s disease

II (terminated)

2015

NCT02477020 TAK-063

PDE10

Schizophrenia

II (terminated)

2015

NCT01952132 OMS643762 PDE10

Schizophrenia

II (terminated)

2013

Clinical trial identifier

Druga

NCT02337907 BI-409306

a

Agents in clinical trials for the treatment of AD and related diseases in 2022 (from clinicaltrials.gov accessed on 05/09/ 2022)

therapeutic for the treatment of AD. To find novel anti-Alzheimer’s disease agents, several researchers are employing computational approaches. In this chapter, we reviewed the important computational modeling research articles published to date in the modeling of PDE inhibitors for the development of anti-Alzheimer’s disease agent. 2.1 Computer-Aided Drug Design (CADD) Approaches

Even though drug research is a time-consuming and costly process, computational biology and bioinformatics approaches have simplified the initial discovery of potential therapeutic compounds. Such computational biology approaches have proven increasingly important, from lead identification to optimization. In this context, computational methodologies such as structure-based drug design (homology modeling, fold recognition and threading methods, ab-initio modeling, protein–protein interactions network analysis, structure-based pharmacophore modeling, fragment-based de novo design (FBDND), molecular docking, and molecular dynamics (MD) simulation) and ligand-based drug design (quantitative structure–activity relationship (QSAR), chemical read-across, ligand-based pharmacophore modeling, similarity search) are playing important roles in the design and discovery of novel compounds with improved therapeutic activity. Chemoinformatics and molecular modeling approaches have been used for decades

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

197

in various fields to identify and optimize novel compounds with improved therapeutic potential. In silico modeling is currently used in the conventional drug discovery process, and such techniques are often used in the search for novel therapies or the optimization of the therapeutic activity of a chemical series during the initial phases of drug development. To combat the predictable and prospective targets for AD, computational techniques have given many possible therapeutic compounds. Several compounds have been revealed to be excellent lead compounds (including flavonoids, carbamates, pyridonepezil, and coumarin derivatives) that can be promulgated against AD [12]. Many inhibitors have entered numerous stages of clinical trials, including MK-8931 (against β-secretase, Merck), TAK-070 (against β-secretase, Takeda Pharmaceuticals), and LMTX (against tau hyperphosphorylation, TauRx Inc.) [12]. 2.1.1 Structure-Based Drug Design Approach Three-Dimensional (3D) Protein Structure Prediction Homology Modeling

Identification of Template and Target Sequence Alignment

Homology modeling is a protein structure prediction approach based on the general finding that proteins with similar sequences have similar structures. Models can be built for a homologous sequence (target) that shares either the template significant sequence (~30% or more) or structural similarity, given an analytically known protein structure (template) [13–15]. The following are the steps involved in the homology modeling of proteins: (1) template identification, (2) alignment of the template and target sequence, (3) model development, and (4) evaluation of the developed model. The first step is to identify the template, which is accomplished by comparing the sequence of unidentified protein structures available in the protein database (PDB) [16]. The identification of template structures is more important in determining evolutionary relationships between proteins and genes [17, 18]. Even when the correct template is selected for comparative modeling, alignment issues are the primary cause of variations [19]. The most widely used online server for the template identification is BLAST (Basic Local Alignment Search Tool) [17–19]. Using BLAST, one can search the database for the best local alignments with the query, which provides a list of recognized protein structures that match the sequence [20]. Sequence similarity with the query sequence must be greater than 30% for BLAST to locate a template; it must be greater than 40% for significant model development [21]. In recent years, an imperative advance has been made in the progress of responsive alignment approaches based on iterative searches, like PSI-BLAST [22], hidden Markov models (HMM), e.g., HMMER [23], or profile–profile alignment such as FFAS03 [24], ProfileScan (available from http://130.88.97.239/bioactivity/newpfscan.html) and HHsearch (available from https://github.com/soedinglab/ hh-suite), and multiple alignment programs like ClustalW [25] and ClustalX [25].

198

Vinay Kumar and Kunal Roy

Model Development and Their Evaluation

The homology modeling process moves on to model development once the target–template alignment is achieved [13–15]. This involves several steps, including backbone building, loop modeling, and side-chain modeling. For model development, tools and servers, such as the BIOVIA Discovery Studio 4.1 client [26], SWISSMODEL web server [19], Modeller 9.23 (available from https:// salilab.org/modeller/9.23/release.html), etc., can be utilized. The obtained models are validated using several techniques because it is likely that the anticipated model has a variety of errors that were introduced at various stages of the homology modeling process [17–20]. The percentage of sequence identity between the template and the target protein is a crucial component for the various errors since, as we all know, errors in the developed model are more likely as sequence similarity decreases [17–21]. Therefore, validating both the established model and the template and comparing the results can be more beneficial in reaching better conclusions regarding the model’s quality [17–21]. The quality of the predicted models can be ensured by using a variety of tools and servers for the evaluation of the developed model, for instance, the MolProbity web server’s Ramachandran plot [27], PROCHECK [28], Verify3D [29], ERRAT [30], etc.

Fold Recognition and Threading Approach

The most fascinating and challenging task is identifying complete protein folds from an unknown amino acid sequence [31, 32]. Fold recognition and threading approaches employ techniques for threading to match sequences with 3D structures in order to select the native fold of a given sequence from a list of potential folds [31, 32]. Once two proteins in a pairwise alignment have less than 25% identity, fold recognition and threading techniques are employed to predict the 3D structure of the proteins [31, 32]. Using energy potentials or similarity score algorithms, it evaluates each target sequence against a database of potential fold templates [31, 32]. Therefore, it is anticipated that the template with the lowest energy score will best match the target protein’s fold [31, 32]. Various servers and tools for fold recognition and threading are available, including the HHpred server (available from http://toolkit.tuebingen.mpg.de/hhpred), FFAS server (available from http://ffas.ljcrf.edu/), THREADER tool (available from http://bioinf.cs.ucl.ac.uk/threader/), Robetta meta-server (available from http://www.robetta.org/), BioInfobank metaserver (available from http://meta.bioinfo.pl/), e-Protein server (available from http://www.e-protein.org/), I-TASSER [33], FUGUE server (available from http://tardis.nibio.go.jp/fugue/), PSIPRED server (http://bioinf.cs.ucl.ac.uk/psipred/), etc.

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

199

Ab-Initio Modeling Approach

The ab-initio approach is commonly used to predict 3D protein structures that have no homologs or have a very low degree of similarity to the protein [34–36]. It is the most challenging and comprehensive approach, in which the query protein is folded randomly [34–36]. The entire approach is established on Anfinsen’s thermodynamic theory proposed in 1962 [34–36]. According to the hypothesis, under a specific set of circumstances, the native structure corresponds to the global free energy minimum [34– 36]. Ab-initio modeling is also known as de novo modeling, physics-based modeling, and free modeling [34–36]. The ab-initio modeling process involves two fundamental steps: first, the prime amino acid sequence is searched for various conformations to estimate the native folds; and second, when the folds have been located and predicted, model validation is carried out to confirm the accuracy of the anticipated structure [34–36]. There are several available ab-initio 3D structure prediction techniques, including Rosetta [37], TOUCHSTONE-II [38], and the most widely known I-TASSER [33]. These methods are based on the Monte Carlo algorithm. For the ab-initio prediction of a 3D protein structure, Rosetta and I-TASSER use the improved technique [33, 37].

Importance of TargetTemplate Modeling

The identification of 3D protein structures from their amino acid sequences using various approaches is a difficult task in computational drug research [39]. Target-template modeling is an unambiguous high-throughput in silico approach that has significant potential as a tool in rational drug discovery [39]. The accuracy of the estimated 3D structure is primarily determined by the target and template alignment [39]. Because no methodology, to our knowledge, ever yields a precise prediction, target-template-based approaches also have certain benefits and drawbacks. Currently, the probability of discovering similar proteins with a known structure for a sequence selected randomly from a genome varies from 30% to 80%, depending on the genome [39]. There is at least one domain in around 70% of all known sequences that may be matched with at least one protein having a known structure [39]. This percentage is significantly higher than the amount of experimentally determined protein structures that have been submitted to the Protein Data Bank (PDB) [16]. In practice, template-based modeling continuously contains evidence that is independent of the template, such as force constraints from general statistical interpretation [39]. The most successful systems frequently explore an increasing amount of template-independent conformational space as a result of better force fields and search algorithms [39]. It is not necessary to select only one template during template selection; hence, the optimal usage of the number of best-score templates improves model accuracy [39]. Combining different template structures can be advantageous in two ways [39]. First, if more than one template structure is aligned with distinct target domains through slight overlap among

200

Vinay Kumar and Kunal Roy

them, the modeling technique can develop a homology-based model of the entire target sequence [39]. Second, the template structures may be aligned with the same target region to identify the best template for developing the model [39]. Another important consideration is that if the sequence identity is less than 40%, a model developed using multiple templates may be more accurate than one developed with a single template [39]. Protein–Protein Interaction Network Approach

Almost all biological functions, such as metabolic cycles, DNA transcription and replication, various signaling cascades, and numerous other activities, are regulated by protein–protein interactions (PPIs), which are of curiosity in biology [40]. It is crucial to understand the specific details of these interactions because proteins work together to perform biological tasks [40]. The need to understand these interactions has led to the creation of many experimental techniques for determining them [40]. The annotation of protein sequences seems to be falling behind, both in terms of quality and quantity, as compared to the exponential growth of genomic sequence information [40]. To close the knowledge gap between the important biochemical and medical data and the raw sequencing data, multipronged, high-throughput functional genomics approaches are required [40]. It follows that finding interactions that are inaccessible to high-throughput approaches requires the use of computational methods [40]. These computational predictions can subsequently be validated using more time-consuming procedures [40]. In recent years, several computational methods for identifying protein interactions have been developed [40]. The feature data that is used to predict protein interactions differ between these approaches [40]. Many studies have shown that knowing the tools and being familiar with the databases is necessary for performing new research in protein–protein interaction analysis [40]. In recent years, many tools and software tools have been developed for the analysis and visualization of biological networks including Cytoscape (available from https://cytoscape.org/down load.html), Medusa [41], and NAViGaTOR (available from https://bio.tools/navigator).

Structure-Based Pharmacophore Modeling

The three-dimensional structure of a macromolecular target or a macromolecule–ligand complex is directly used in structure-based pharmacophore (SBP) modeling [42, 43]. The standard procedure for structure-based pharmacophore modeling is analyzing the complementary chemical properties of the active site and their spatial correlations, followed by the assembly of specific features to produce a pharmacophore [42, 43]. There are two subcategories of structure-based pharmacophore modeling methods: macromolecule–ligand complex-based and macromolecule (without ligand)-based [42, 43]. The macromolecule–ligand complexbased approach is specialized to find the ligand-binding site of the

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

201

macromolecular target and identify the critical areas of ligand interaction [42, 43]. Pharmacophore models are generated for a better molecular understanding of ligand–protein interactions, and they can be used to find new compounds that meet the pharmacophore requirements and have a high probability of being physiologically active [42, 43]. SBPs derive these molecular features by converting protein attributes to reciprocal ligand space. Contrary to ligandbased pharmacophore models, which require templates of ligands in their bioactive configuration, SBPs do not rely on ligand information [42, 43]. The fundamental procedures in the generation of SBPs are (i) protein structure preparation (protonation states and protein conformations), (ii) binding site identification, (iii) pharmacophore feature specification, and (iv) pharmacophore feature selection (energy-based selection, selection based on protein– ligand interaction information, selection based on variation of protein amino acids in the ligand-binding site, training of SBPs with known actives, and shape restraints) [42, 43]. Various software and tools for structure-based pharmacophore modeling are available, including the BIOVIA Discovery Studio [26], Molecular Operating Environment (MOE) (available from https://www. chemcomp.com/Products.htm), LigandScout [44], and Maestro Version 10.7.014 (available from https://sites.google.com/a/ srscicomp.com/maestro/downloads). Fragment-Based De Novo Design (FBDND)

Fragment-based de novo design (FBDND) has been confirmed as a good methodology for the identification of additional chemical features based on the structural details of the target protein as well as scoring the present molecules [45]. In this approach, the de novo fragment-based design of exemplar inhibitors against AD is performed by screening fragment libraries by developing a novel methodology [45]. The fragments are to be retrieved from the fragment libraries (Asinex fragment library (available from http:// www.asinex.com/?page_id=97), FCH Group’s Fragment Libraries (available from http://fchgroup.net/fragment-libraries.php), Life Chemicals 3D-shaped Fragment Library (available from https:// lifechemicals.com/screening-libraries/fragment-libraries), etc.). The retrieved fragments are screened by molecular docking (could be flexible too) to the ligand-binding pocket of the different target proteins of AD, and top fragments are selected based on interacting amino acid residues, interacting forces, and docking score [45]. The ligand is then designed by combining fragments based on the structural attributes at the ligand-binding site of the target proteins and redocking [45]. The basic steps involved in the FBDND method are in the following sequential scheme: (1) collection of target protein (crystal structure from PDB), (2) selection of fragment library, (3) identification of binding pockets, (4) screening of fragment libraries via molecular docking, (5) seed structure mapping, (6) PanDDA analysis to validate the binding interactions,

202

Vinay Kumar and Kunal Roy

(7) fragment linking—using the rules (A) max atom–atom distance 1 Å, (B) minimum fragment centroid distance 2 Å, (C) minimum bond angle deviation 15°, and (D) maximum fragment size of 200 atoms—(8) seed structure extraction, and (9) target ligand efficiency [45]. Various software and tools for FBDND analysis are available, including the BIOVIA Discovery Studio 4.1 [26], Schro¨dinger Glide (available from https://www.schrodinger.com/ products/glide), LigBuilder V3 (available from http://www. pkumdl.cn:8080/ligbuilder3/), PanDDA, etc. (available from https://pandda.bitbucket.io/pandda/index.html). Molecular Docking

To anticipate the best possible identical binding pattern of a ligand to a macromolecular target, one of the key methodologies in structure-based drug discovery (SBDD) is known as molecular docking [46–49]. It entails identifying a range of potential ligand poses inside the protein-binding domain [46–49]. The molecular docking method is divided into two basic steps: the first is to predict the stable conformation and orientation of the ligand, and the second is to compute the binding affinity and binding orientation of the ligand within the active sites of target proteins [46– 49]. Through using data from the docking study, it is possible to make predictions about the binding orientation, binding energy, free energy, interaction energy, and stability of complexes to assist in the search for new pharmaceuticals [46–49]. In drug discovery, molecular docking has a wide range of uses, including lead optimization, chemical mechanism investigations, structure–activity analyses, possible lead identification through virtual screening, the establishment of binding hypotheses for easy prediction, combinatorial library design, etc. [46–49]. The docking approach is divided into three categories based on the degree of flexibility of the compounds: rigid docking, semiflexible docking, and flexible docking [46–49]. There are four important basic steps for molecular docking, which are as follows: (1) target selection and preparation, (2) ligand selection and preparation, (3) molecular docking, and (4) molecular docking analysis [46–49]. There are some software/ tool and web servers accessible for molecular docking, such as AutoDock Vina (available from https://autodock.scripps.edu/ downloads/), Schro¨dinger software (available from https://www. schrodinger.com/), Molegro Virtual Docker software 6.0 (MVD) [50], BIOVIA Discovery Studio [26], idock, “Achilles” Blind Docking server (available from https://bio-hpc.ucam.edu/achil les/), FlexX [51], and smina software (available from https:// sourceforge.net/projects/smina/).

Molecular Dynamics (MD) Simulation

An atomic structure is particularly helpful since it usually provides significant insight into how the biomolecule behaves [52–55]. In a biomolecule, all of the atoms are in constant motion, and the physical movements of the atoms and molecules govern all of the

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

203

functions and intermolecular interactions [52–55]. The skill of molecular biologists to investigate the behavior of these biomolecules, perturb them at the atomic level using various techniques, and then attempt to detect how the atoms respond seems to be what makes this understanding possible. The best method for describing the physical motion of atoms in biomolecules is hence computer simulation [52–55]. Based on an all-inclusive physics model that predicts interatomic interactions, molecular dynamics (MD) simulations estimate the movements of each atom in a protein throughout a certain period [52–55]. Molecular dynamics simulations have developed into an excellent method for understanding macromolecular structure-to-function correlations [52– 55]. Atomic locations and velocities are among the details that molecular dynamics simulations offer at the microscopic level [52–55]. The key concept behind an MD simulation is clear. It details the coordinates of every atom in the biomolecular system (such as a protein surrounded in water and probably a lipid bilayer); one can calculate the force that each atom is subjected to from the interactions of all the other atoms [52–55]. Thus, it is possible to determine the spatial position of each atom as a function of time using Newton’s laws of motion [52–55]. To be more precise, one steps through time, continuously computing the forces on each atom and then using those forces to adjust each atom’s position and velocity [52–55]. The resulting trajectory is essentially a 3D movie that reveals the atomic-level arrangement of the system at every point over the simulated period [52–55]. MD simulations are significant for a variety of reasons [52–55]. First, they confine the placement and movement of every atom at every point in time, which is highly difficult with any experimental method; second, the simulation environment is clearly specified and can be carefully controlled to identify the following: the initial conformation of a protein, which ligands are linked to, whether it has any mutations or posttranslational modifications, and which other molecules are present in the simulation environment [52–55]. One can determine the effects of a wide range of molecular disturbances by relating simulations performed under different settings. An important step in the MD simulation is the estimation of the ligandbinding affinities (binding free energy) with the target macromolecule, which uses approximate scoring functions and struggles with the estimation of binding energies following the experimental values [52–55]. Numerous techniques have been reported for determining the binding free energy of small molecules with biological targets, including thermodynamic integration (TI), linear interaction energy (LIE), free energy perturbation (FEP), molecular mechanics Poisson–Boltzmann surface area (MM-PBSA), and molecular mechanics generalized Born surface area (MM-GBSA) [56, 57]. The most feasible and effective techniques to determine the binding free energy of the ligand with the

204

Vinay Kumar and Kunal Roy

protein target are MM-PBSA and MM-GBSA [56, 57]. Kollman et al. [58] first described the MM-PBSA approach; afterward, it was revised by various research groups. Nevertheless, there is still no agreement on the specific details of the procedure; this may be because the method’s effectiveness varies depending on the system to which it is applied [56–58]. Particularly, the MM-PBSA and MM-GBSA approaches combine molecular mechanics with continuum solvent models to estimate ligand-binding affinities and are orders of magnitude faster than other known techniques [56– 58]. The following procedures are used to compute the binding free energy using MM-PBSA and MM-GBSA in association with MD simulations: (1) An explicit solvent model is employed in an MD simulation of the protein–ligand complex, as implicit solvent simulations have been found to yield less appropriate results; (2) all solvent molecules and charged ions are eliminated from each MD session, and the implicit PBSA or GBSA solvent model is used to evaluate the solvation energy; and (3) the solute conformational entropy change can be determined from a selected set of snapshots [56–58]. After that, the total of these various energy components yields the ultimate binding free energy. It is difficult to adequately summarize the MD simulation process in a few sentences or even a brief paragraph due to its complexity [56–58]. Here, we have outlined the fundamental steps of the MD simulation as follows: (1) generating topology (protein topology, ligand topology) and building the complex, (2) defining box and solvating the system, (3) adding ions, (4) energy minimization, (5) equilibration (equilibration NVT, equilibration NPT), (6) final MD run, and (7) data analysis (protein–ligand interactions and ligand dynamics such as RMSD, radius of gyration, RMSF; protein–ligand interaction energy) [52–55]. The details of the process and basic steps involved in MD simulation are addressed by Karplus et al. [55]. To perform MD simulation, there are some tools and online servers available like NAMD 2.13 [59], MMTK software [60], GROMACS 5.1.2 [61], pmemd.cuda module in AMBER18 [62], VMD [63], etc. 2.1.2 Ligand-Based Drug Design Approach Quantitative Structure– Activity Relationship (QSAR)

Quantitative structure–activity relationship (QSAR) is a statistical approach to finding the consistent relationship between the biological activity (dependent variable) of compounds and their structural arrangements and chemical property (independent variables) [64, 65]. The chemical domain space refers to the chemical information, which is derived in terms of descriptors (independent variables) utilizing various software tools [64, 65]. The dependent variable acquired from an experiment stands as the endpoint and is modeled using the Equation 1 that is given below. There are different regression and pattern recognition techniques (S-MLR, GA, etc.) that can be used for variable selection and QSAR model development [64, 65]. The developed model for the QSAR is used particularly in chemoinformatics and drug discovery and to

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

205

evaluate the biological activity of new chemical compounds; apart from these it is also used for toxicological and ecotoxicological evaluations of specific chemicals within the meaning of risk management [64, 65]. Biological activity=toxicity = f ðStructure, Physicochemical propertiesÞ

ð1Þ

The modeling method may be referred to as quantitative structure–activity relationship (QSAR), quantitative structure–toxicity relationship (QSTR), quantitative structure–property relationship (QSPR), or other terms depending on the type of endpoint modeled [64, 65]. The term “activity” itself may often be employed as additional descriptors, as in the cases of “quantitative structure– activity–activity relationship” (QSAAR), etc. [64, 65]. The selection of data (chemical/biological) is one of the most crucial elements in the accomplishment of any QSAR analysis [64, 65]. Two separate types of data information, namely, biological data (endpoints) and chemical data in terms of molecular descriptors, are needed for the development of QSAR models [64, 65]. Before using any modeling methodology for model development, the feature selection approach is then employed to choose the appropriate number of meaningful and informative descriptors [64, 65]. The model should follow the OECD (Organization for Economic Co-operation and Development) guidelines for QSAR model development [64, 65]. It is difficult to adequately summarize the QSAR modeling process in a few sentences or even a brief paragraph due to its complexity [64, 65]. Here, we have outlined the fundamental steps of the QSAR modeling as follows: (1) dataset collection with defined activity, (2) dataset curation, (3) computation of molecular descriptors, (4) division of the dataset, (5) feature selection, (6) development of QSAR model, (7) validation of QSAR models (internal and external), (8) checking domain of applicability of developed models, and (9) mechanistic interpretation of the QSAR model [64, 65]. The details of the process and basic steps involved in QSAR modeling are discussed by Roy et al. [64]. The freely accessible software tools to perform QSAR modeling are available from the official website (http://teqip.jdvu.ac.in/ QSAR_Tools/) of Jadavpur University (Kolkata, India) developed by Prof. Kunal Roy’s research group (URL: https://sites.google. com/site/kunalroyindia/). Similarity-Based Chemical Read-Across

Read-across prediction is a similarity-based in silico technique that predicts the biological response of unknown compounds based on known activity values [66–69]. The chemical read-across approach is based on machine learning to estimate the activity of the test set chemicals using the modeled descriptors [66–69]. For a successful prediction with Laplacian kernel (LK) similarity-based, and Gaussian kernel (GK) similarity-based estimations, first we need to

206

Vinay Kumar and Kunal Roy

optimize the hyperparameter sigma (σ) and gamma (γ) respectively using validation sets [66–69]. For the optimization, the initial training set will need to be randomly divided into sub-training and sub-test sets in a 3:1 proportion [66–69]. The sub-training and sub-test sets will then be subjected to “Read-Across-v4.1” (available from https://sites.google.com/jadavpuruniversity.in/ dtc-lab-software/home) with different σ and γ values. During optimization, the other tool parameters, including the number of nearby training compounds, the distance threshold, and the similarity threshold, should be kept constant. The optimized setting will be selected by checking the external validation metrics (Q 2F 1 and Q 2F 2 ) [66–69]. Finally, the optimized setting is combined with the original training and test sets to get the final prediction [66–69]. To obtain the best predictions, we need to gradually reduce the number of similar training compounds from ten to two [66–69]. Ligand-Based Pharmacophore Modeling

A pharmacophore model is a combination of steric and electronic properties required for effective supramolecular interactions with a specific biological target and activation or inhibition of its biological response [70–72]. Through various 3D arrangements of conceptual interaction features that take into consideration various kinds of non-covalent interactions, a pharmacophore model illustrates the binding patterns of bioactive compounds with the target binding site [70–72]. These interactions could be in the form of hydrogen bonds, hydrophobic interactions, metal contacts, aromatic contact, charge transfer interactions, etc. [70–72]. The following features are used by numerous algorithms to generate pharmacophore models: hydrogen bond donors (HBD), hydrogen bond acceptors (HBA), positive and negative charge features, hydrophobic features, aromatic rings, steric restrictions features, etc. [70–72]. A pharmacophore model can be developed using either a structure-based technique (using the active site of the protein structure) or a ligand-based approach (using a collection of specified molecules), which involves superimposing a set of active molecules and extracting common chemical characteristics that are essential for their bioactivity [70–72]. The basic steps involved in the development of a potential pharmacophore model are as follows: in the case of ligand-based pharmacophore model generation, (1) dataset collection with defined activity, (2) dataset curation, (3) ligand preparation and conformation search, (4) ligand alignment and features calculation, (5) features selection, (6) model development, and (7) scoring and validation of the developed models (quantitative and qualitative in terms of internal and external) [70–72]. Pharmacophore models have been extensively employed in de novo design, virtual screening, and other applications like lead optimization and multitarget drug design [70– 72]. Several automated pharmacophore generators have been

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

207

developed, including commercially and freely available tools and software such as BIOVIA Discovery Studio [26], Schro¨dinger software (available from https://www.schrodinger.com/), PharmaGist online tool (https://bioinfo3d.cs.tau.ac.il/PharmaGist/), LigandScout [44], etc. Similarity Search

The similarity search approach can be used to start developing a SAR or identify more hit compounds once lead compounds have been identified through experiments [73]. The similarity search strategy is the simplest and fastest. It can find molecules that are similar to the input compound chemically or physiochemically [73]. This method can also be used to validate leads since a molecule with several bioactive analogs from which a SAR can be generated is relevant for further investigation [73]. The basic steps involved in the similarity search are as follows: (1) preparation of query compound, (2) selection of types of the fingerprint to define the compounds in the database, and (3) selection of comparison method and performing the similarity search against an in silico database such as PubChem (available from https://pubchem.ncbi. nlm.nih.gov/), ChEMBL (available from https://www.ebi.ac.uk/ chembl/), etc. To perform similarity search, there are some tools and online servers available like Molecular Operating Environment (MOE) (available from https://www.chemcomp.com/Products. htm), BIOVIA Discovery Studio [26], PubChem similarity search (available from https://pubchem.ncbi.nlm.nih.gov/score_matrix/ score_matrix.cgi), etc.

2.1.3

There are many publicly or commercially available databases that can be used for computational drug discovery applications since high-throughput technologies for biological screening and compound synthesis have recently become available [74]. Virtual screening techniques are increasingly being recognized as the most cost-effective and time-saving approach to introducing new chemical entities into the pharmaceutical market to address the economic pressure on the pharma industry [74]. A ligand-based method and a structure-based approach can both be used as the framework for virtual chemical database screening [74]. Pharmacophore mapping is a ligand-based virtual screening technique that can be used to efficiently identify novel potential lead compounds [74]. A pharmacophore model identifies the key chemical properties underlying the bioactivities of the compounds under investigation [74]. Furthermore, in receptor-based virtual screening, molecular docking is used to identify compounds based on their binding energy and interaction patterns with the target protein [74]. Additionally, one can perform virtual screening simply based on some criteria, such as Lipinski’s rule of five, central nervous system (CNS) drug-like properties, or any other user-defined physicochemical properties [74].

Virtual Screening

208

3

Vinay Kumar and Kunal Roy

Case Studies Due to a dearth of knowledge about the underlying causative mechanisms and pathways, defects in brain activities linked to aging and neurodegenerative disorders gain only marginally from current treatments [1, 2]. AD is a generally incurable, age-related neurological degeneration for which only symptomatic treatment is available, and the therapeutic potential of modifying disease development remains far from being realized [1, 2]. Targeting cyclic nucleotide signaling in AD has become a new idea as a result of altered phosphodiesterase (PDE) expression and unregulated cyclic nucleotide signaling in AD [3–6]. PDEs are intracellular enzymes that degrade the cyclic nucleotide adenosine and/or guanosine monophosphate to modify cellular signaling through the cAMP/ cGMP pathways [3–6]. For several biological processes in the brain, and controlling them significantly, targeting cyclic nucleotides as an intracellular messenger appears to be a promising option. However, cyclic nucleotide synthesis, execution, and/or degradation have been directly connected to cognitive impairments [3–6]. About cognition, the cyclic nucleotides (cAMP and cGMP) have an imperative execution in several phases of memory, such as gene transcription, neurogenesis, neural circuitry, synaptic plasticity, and neuronal survival, among others [3–6]. A key role for cAMP/ cGMP signaling in AD populations is suggested by the abnormalities of these fundamental cognitive processes that are observed in AD individuals [3–6]. As the only group of enzymes that can allow the hydrolysis and destruction of cAMP and cGMP while maintaining their optimal levels, phosphodiesterase inhibitors make an intriguing research target. Various scientists all over the world are attempting to identify and design novel PDE enzyme inhibitors using a variety of computational methodologies, including ligandbased and structure-based approaches. In this chapter, we have covered the most important recent studies on computational modeling of PDE inhibitors that have been done in an effort to find new anti-AD drugs targeting the PDE enzyme.

3.1 In Silico Modeling of PDE Inhibitors Against Alzheimer’s Disease: Case Studies 3.1.1 Repurposing of PDE2 Inhibitors Against AD

Zhu, J. et al. [75] revealed the composite crystal structure of PDE2 and the highly selective inhibitor BAY 60-7550 (see Fig. 1) (protein number: 4HTX at 1.9 Å resolution; https://www.rcsb.org/ structure/4HTX) in 2013, which laid the groundwork for the rational design of PDE2-targeted small-molecule ligands. The crystal structures demonstrate that the inhibitor binds to the PDE2 active site via a binding-induced, hydrophobic pocket, in addition to the conserved glutamine-switch mechanism for substrate binding. In silico affinity profiling using molecular docking shows that the inhibitor binding to this pocket greatly adds to the binding affinity and hence increases inhibitor selectivity for PDE2. The findings highlight a structure-based design technique that takes

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

209

Fig. 1 Chemical structures of top selected compounds (PDE2 and PDE4 inhibitors against AD) from the listed studies

advantage of possible binding-induced pockets to obtain greater selectivity in the development of PDE inhibitors. However, there are currently no PDE2 medicines on the market. As a result, more research is urgently required to develop PDE2 inhibitors to address clinical needs. In 2021, Tang, L. et al. [76] selected urolithins (hydroxylated 6H-benzo[c]chromen-6-ones) as the leading compound, and a series of 44 alkoxylated 6H-benzo[c]chromen-6-one derivatives were designed and synthesized to identify the neuroprotective agent as PDE2 inhibitors. The potential neuroprotective benefits of the resulting compounds were initially tested using a PDE2 inhibition assay. The compounds with the highest PDE2 inhibition activity were tested further in hippocampus HT-22 cells using corticosterone-induced cytotoxicity. The authors have used the PDE2 inhibitor BAY 60-7550 (see Fig. 1) as the reference compound with an IC50 of 8.4 nM. After the detailed analysis, the authors found the compound 1 (see Fig. 1) to have the optimal inhibitory potential (IC50: 3.67 ± 0.47 μM) against the PDE2 enzyme. Very recently, Zhou, Y. et al. [77] have designed and synthesized 28 dihydropyranopyrazole derivatives as PDE2 inhibitors. The in vitro inhibitory potencies of the hit compound and its synthesized derivatives against PDE2 were tested using the compound erythro-9-(2-hydroxy-3-nonyl)adenine (EHNA; see Fig. 1) (IC50 = 2460 nM), which was acquired from SIGMA and served as the reference compound. The majority of the target

210

Vinay Kumar and Kunal Roy

compounds exhibited moderate to strong inhibitory efficacy against PDE2. With an IC50 value of 41.5 nM, compound 2 (see Fig. 1) stood out among them as the most effective PDE2 inhibitor. The authors have also compared the selectivity profile of compound 2 with other PDE families and found that inhibition of PDE4D2 and PDE8A1 was very weak (IC50 > 10,000 nM), while IC50 value against PDE1C was 289-fold higher than that against PDE2A. In summary, compound 2 exhibits strong selectivity toward PDE2 when compared to other PDEs. The authors have also performed the molecular docking study of selected compound 2 with the crystal structure of PDE2A (PDB ID: 4HTX, https:// www.rcsb.org/structure/4HTX). The authors have used the compound (R)-LZ77 (see Fig. 1) as a hit compound with moderate PDE2 inhibitory activity from their previous study to define the active site of PDE2A and to test the docking [78, 79]. Accelrys Discovery Studio 2.5.5 software [26] was used for molecular docking study through the CDOCKER method. The molecular docking of PDE2 with compound 2 demonstrates that the compound’s 4-(trifluoromethyl)benzyl)oxyl side chain enters the H-pocket and forms strong hydrophobic contacts with LEU770, LEU809, and PHE862 amino acids, improving inhibitory efficacy. Finally, selected poses were subjected to MD simulation and MM-PBSA binding free energy calculations. 3.1.2 Repurposing of PDE4 Inhibitors Against AD

Bruno and colleagues [80] used the method of replacing imino ether with five-membered heterocycles with the identical alkyl chains to learn more about pharmacophore properties. A crucial stage in the creation of memories is the breakdown of cAMP, which is catalyzed by phosphodiesterase type 4D (PDE4D). Inhibiting PDE4 has been shown to improve recognition in degenerative disorders like Alzheimer’s. The imino ether moieties had a key role in forming a wide network of stable, direct, and watermediated H-bond contacts with the catalytic region of PDE4D in docking and molecular dynamics simulations of lead compound GEBR-7b-35 and the crystal structure of PDE4D. The molecular flexibility was decreased and fewer conformers were produced by heterocycle insertion. The substance GEBR-7b-36 was swiftly absorbed and enters the brain, according to the pharmacokinetic analyses. The BBB penetration of compound GEBR-7b-36 (see Fig. 2) was higher than that of compound GEBR-7b-35 (see Fig. 2). Inhibiting phosphodiesterase (PDE) enzymes, which are responsible for the hydrolysis of the intracellular second messenger cAMP, is a new strategy for treating AD [81]. PDE4D is a subtype of PDEs. Rolipram and roflumilast have been identified as PDE4 inhibitors [81]. Hu, J. et al. [82] synthesized the clioquinol-rolipram/roflumilast derivatives and examined their PDE4D and Aβ aggregation inhibiting effects in light of these findings. When

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

211

Fig. 2 Chemical structures of top selected compounds (PDE4 and PDE5 inhibitors against AD) from the listed studies

compared to the reference drugs rolipram and roflumilast, the synthesized compounds demonstrated excellent inhibitory efficacy against PDE4D. Among the investigated compounds, compound 3 displayed inhibitory action (IC50, PDE4D2: 0.399 μM) comparable to and even better than the reference drugs (IC50, PDE4D2 values are 0.621 and 0.480 μM for rolipram and roflumilast, respectively). The most effective inhibitors had either an unsubstituted 8-hydroxyquinoline moiety or iodine at the 7-position on 8-hydroxyquinoline. Other substitutions or changes in substituent position have not shown positive results. Finally, docking analyses with PDE4D were performed on compound 3. The hydrophobic amino acid residues PHE372 and ILE336 form a sandwich outer shell around compound 3 (see Fig. 1). The molecular docking investigations revealed that compound 3 has a good hydrogen bond interaction with the amide nitrogen of the amino acid GLN369 via the cycloproylmethyl and difluoromethyl oxygen of derivative 3. Using structure-based drug design and fragment identification, Liao, Y. et al. [83] designed and synthesized a series of 1-phenyl3,4-dihydroisoquinoline-3-carboxamide analogs as selective PDE4B inhibitors. All of the newly synthesized 3,4-dihydroisoquinoline-3-carboxamide derivatives were initially evaluated in vitro for PDE4B inhibitory efficacy. Rolipram, a commercial agent, was

212

Vinay Kumar and Kunal Roy

employed as a positive control for comparison [81]. Following an investigation, every substance exhibited inhibitory action against PDE4B with potencies ranging from moderate (IC50 > 5 μM) to good (IC50 < 5 μM, which is comparable to the IC50 of a positive rolipram). After detailed analysis, the authors found that compound 4 (see Fig. 2) was showing the best inhibitory activity against the PDE4B, with an IC50 of 0.22 μM and good selectivity toward PDE4B over PDE4D, making it a viable lead for studying the biological effects of PDE4 inhibition. Preliminary SAR analysis revealed that the addition of methyl glycinate or anilines with different substituents at the ortho-position to the C-3-COOH group was typically favorable in increasing the inhibitory action against PDE4B. A current study revealed that attaching a methoxy group in the ortho-position of the phenyl ring was the best way to increase inhibition against PDE4B. A new set of arylbenzylamine derivatives were designed and synthesized by Tang, L. et al. [84] using the well-known PDE4 inhibitors FCPR16 [85] and FCPR03 [86, 87] as lead compounds. As a result, aryl benzylamine derivatives were initially tested for binding efficacy against full-length human PDE4B1 and PDE4D7 using rolipram and FCPR16 as positive controls. The most potent compounds were further investigated for their IC50 values. An array of promising compounds having morpholine or pyridin-3-amine side chains and strong inhibitory effects were found during preliminary screening of this class of compounds against human PDE4B1 and PDE4D7 isoforms. After a detailed analysis, the authors found that compound 5 (see Fig. 2) was showing the best inhibitory activity against the PDE4B1, with an IC50 of 0.34 μM and good selectivity toward PDE4B1 over PDE4D7, making it a viable lead for studying the biological effects of PDE4 inhibition. The SAR of the entire set of analogs revealed that substituting a difluoromethoxy group in the B-ring for a methoxy group was harmful to inhibitory action. In contrast, increasing the aryl ring (C-ring) hydrophilicity boosted its activity toward PDE4B1 and PDE4D7. Furthermore, molecular docking findings demonstrated that compound 5 has the best interactions with UCR2, which could explain the partial inhibition of PDE4. Additionally, compound 5 was nontoxic to SH-SY5Y cells and demonstrated neuroprotection against MPP+-induced apoptosis in SH-SY5Y cells. Therefore, compound 5 outperformed the lead compound FCPR03 in terms of oral bioavailability. Using compound 5 as a lead molecule, future efforts will focus on the identification of novel PDE4 inhibitors with better bioavailability and inhibitory activity. Recently, Liu, J. and coworkers [88] designed and synthesized 28 novel 2,3-dihydro-1H-inden-1-one derivatives as catechol ether-based dual PDE4 and AChE inhibitors to treat AD. All of the newly synthesized derivatives were initially evaluated in vitro for

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

213

PDE4D and AChE inhibitory efficacy. Rolipram and donepezil, commercial agents, were employed as positive controls for comparison. After detailed analysis, the authors found that compound 6 (see Fig. 2) having a 2-(piperidin-1-yl)ethoxy group at the 6-position of indanone ring was showing the best inhibitory activity and selectivity against AChE (IC50 = 0.28 μM) and PDE4D (IC50 = 1.88 μM), making it a viable lead for studying the biological effects of PDE4/ACh inhibition. Molecular docking studies revealed that catechol ether of compound 6 occupied the pocket of the catechol moiety in nature-ligand roflumilast, creating one hydrogen connection with GLN535 and one interaction with PHE538. A hydrophobic connection was formed with PHE506, MET523, and PRO522 by the 2-(piperidin-1-yl)ethoxy group at the 6-position of the indanone ring, which had extended into the long and narrow hydrophobic pocket. As ligands for magnesium ions, nitrogen atoms in the pyridine ring formed one intermolecular hydrogen bond with one molecule of water. According to the molecular docking findings, compound 6 in the S configuration may exhibit greater inhibitory efficacy against AChE and PDE4D than 6 in the R configuration. These findings show that compound 6 is a promising multifunctional drug for the treatment of AD. Y. Lin et al. [89] developed and synthesized furan and oxazole carboxylic acid derivatives after prior research [90, 91] indicating that oxazole and furan-containing compounds possessed strong PDE4B inhibiting action. Evaluation of the compounds revealed good to moderate anti-PDE4B activity. 4-Methoxyphenyl analogs 7 and 8 (see Fig. 2) had the highest anti-PDE4B activity, with IC50 values of 2.8 M and 1.4 M, respectively. The anti-PDE4B activity of oxazole analogs was shown to be stronger than that of equivalent furan derivatives, according to SAR. The presence of a methoxy group at the 4-position of the phenyl ring, in particular, resulted in increased activity. According to the docking data, the title compounds interacted favorably with the PDE4B protein through intermolecular hydrogen bonds (-CO-N, GLN443), hydrophobic interactions (thiazolidine, PHE 446), and metal coordination in the ligand–receptor complex (p-OMe, Zn2+, and Mg2+). 3.1.3 Repurposing of PDE5 Inhibitors Against AD

Mao, F. et al. [92] have rationally designed and synthesized a series of novel tadalafil derivatives for the identification of dual-target AChE/PDE5 inhibitors for the treatment of AD. The modified Ellman method was used to test the inhibition of these tadalafil derivatives against AChE and butyrylcholinesterase (BuChE) using Hup A and donepezil as positive references, and an IMAP-FP (immobilized metal ion affinity-based fluorescence polarization) assay was used to determine PDE5 inhibition using tadalafil as a reference. The SAR investigations demonstrated that stereo configuration at position 6 of the tadalafil parent nucleus was essential for exhibiting strong AChE inhibitory activity, with 6R being the

214

Vinay Kumar and Kunal Roy

preferred configuration, while stereo configuration at both positions 6 and 12 of the tadalafil parent nucleus, particularly position 6, was essential for exhibiting strong PDE5A1 inhibitory activity, with 6R and 12R being the preferred configuration. According to research, compounds 10 (see Fig. 2) (AChE, IC50 = 0.036 μM; PDE5A, IC50 = 0.153 μM) and 11 (AChE, IC50 = 0.032 μM; PDE5A, IC50 = 1.530 μM) are excellent selective dual-target AChE/PDE5 inhibitors with increased BBB penetrability. Interestingly, at a dose of 10 mg/kg, compound 11 (see Fig. 2) displayed comparable efficacy to donepezil in improving cognitive abilities in the AD mice model, as well as a great effect on boosting the amount of CREB phosphorylation in vivo, a critical element in alleviating cognitive impairment and improving synaptic function in AD. Finally, molecular docking simulations of compound 11 with hAChE and hPDE5A validated the rationale of the planned strategy. Zhou, L. et al. [93] have designed and synthesized 2-substituted-6-amino-4 (3H) quinazolinone, quinazoline-2,4-(1H,3H)dione, isoindoline-1,3-dione, and indoline-2,3-dione derivatives for the identification dual-target AChE/PDE5 inhibitors for the treatment of AD. All the synthesized compounds were evaluated for the AChE inhibitory activities in vitro by applying the assay of the Ellman method using donepezil, used as the positive control, and human PDE5A level kit assay was used to determine PDE5 inhibition using sildenafil as a reference. After detailed analysis, the biological test revealed that compound 9 (see Fig. 2) displays good biological activities against AChE with IC50 values of about 79.43 nM and PDE5A with IC50 values of about 50 μM. Finally, molecular docking studies demonstrated that compound 9 interacts with the AChE amino acid residues TRP86, TYR337, TYR341, and TRP286, as well as the PDE5A amino acid residues PHE820, ASP654, ASP764, and TYR612. Finally, the authors conclude that since the carbonyl of indole-2,3-dione is important in enzyme interactions, ring-opening derivatives and indolin-2-one derivatives may be challenging to provide high inhibitory activity to AChE and PDE5A simultaneously. Rabal, O and their colleagues in 2018 [94] used structure- and knowledge-based techniques to build first-in-class sildenafil and vardenafil-based analogs with the required target compound profile: dual PDE5 and HDAC6-selective inhibitors. All the synthesized compounds were subjected to the HDAC6 and PDE5 inhibitory activities in vitro by applying the assay of the specific fluorescence-labeled substrate and HTRF cGMP assay kit, respectively. After detailed in vitro analysis, the authors have chosen compound 12 (PDE5, IC50 = 11 nM; HDAC6, IC50 = 15 nM) for further in vivo testing based on its promising in vitro activity and brain accessibility. The pCREB-SER133 phosphorylation mark

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

215

Fig. 3 Chemical structures of top selected compounds (PDE7 and PDE9 inhibitors against AD) from the listed studies

was measured in the hippocampus of WT mice 30 min after receiving an intraperitoneal (ip) dosage of 40 mg/kg. When compared to untreated mice, there was a 48% rise in the levels of this mark. Thus, at the measured dose, compound 12 (see Fig. 2) penetrates the BBB and achieves its functional response in the hippocampus. Finally, molecular docking analysis of compound 12 with PDE5 and HDAC6 validated the rationale of the planned approach. 3.1.4 Repurposing of PDE7 Inhibitors Against AD

Purine-based agents have shown significant promise as PDE7 inhibitors. Pitts et al. [95] identified a potent purine-based PDE7 (compound 13; see Fig. 3) inhibitor (IC50 = 0.150 μM) through screening the Bristol-Myers Squibb chemical library. The addition of an aryl sulfonamide to the structure (compound 14 (see Fig. 3), IC50 = 0.010 μM) surrounding the purine core at C-6 revealed that the potency and PDE selectivity might be increased over the first lead. However, physicochemical data were also considered by using their PAMPA (parallel artificial membrane permeation assay) and solubility data to design a series of fused pyrimidine-based inhibitors to enable the in vivo evaluation. Lorthiois et al. [96] discovered the spiroquinazolinone (compound 15; see Fig. 3) by high-throughput screening, which inhibited PDE7 with an IC50 of 0.17 μM. This main lead chemical was optimized in search of structures of a powerful PDE7 inhibitor

216

Vinay Kumar and Kunal Roy

with an IC50 range of 0.07–0.014 μM. Selectivity against PDE7A1, 3A, 4D, and 5 was also demonstrated by some of the best compounds. As a new and promising class of drugs for the treatment of neurological disorders, a neural network model has been developed by Redondo, M. et al. [97] to predict the inhibitory potential of any chemical structure to be a phosphodiesterase 7 (PDE7) inhibitor. The CODES program was used to achieve the numerical definition of the structures. The validation of this neural network model revealed a novel family of 5-imino-1,2,4-thiadiazoles (ITDZs) as a PDE7 inhibitor. Extensive biological research has shown that ITDZs can inhibit PDE7 and enhance intracellular cAMP levels. Among them, compound 16 (IC50 = 1.59 μM) demonstrated great in vitro efficacy with a favorable pharmacokinetic profile (safe genotoxicity and blood–brain barrier penetration). The potential of ITDZ derivative, particularly compound 16 (see Fig. 3), for the efficient treatment of multiple sclerosis is demonstrated by the considerable attenuation of clinical symptoms following the administration of ITDZ derivative compound 16 in an experimental autoimmune encephalomyelitis (EAE) mouse model. Using a structure-based drug design methodology, a library of new anilide and benzylamide derivatives of ɷ-(4-(2-methoxyphenyl)piperazin-1-yl)alkanoic acids was developed by Jankowska, A. et al. [98] as a combination of 5-HT1A/ 5-HT7 receptor ligands and phosphodiesterase PDE4B/PDE7A inhibitors. In vitro investigations of 33 newly synthesized compounds allowed authors to identify compound 17 (see Fig. 3) as the most promising multifunctional 5-HT1A/5-HT7 receptor antagonist (5-HT1A Ki = 8 nM, Kb = 0.04 nM; 5-HT7 Ki = 451 nM, Kb = 460 nM) with PDE4B/PDE7A inhibitory activity (PDE4B IC50 = 80.4 μM; PDE7A IC50 = 151.3 μM). Compound 17 demonstrated excellent passive membrane penetration as well as strong metabolic stability in vitro. Additionally, the pharmacological analysis of compound 17 demonstrated its procognitive and antidepressant characteristics in rat behavioral tests. Compound 17 at 3 mg/kg (ip) considerably restored MK-801-induced episodic memory deficits in the novel object recognition test while at 10 mg/kg (ip) greatly reduced animal immobility time (by approximately 34%) in the forced swimming test. Compound 17 had an antidepressant-like effect, although it was more potent than escitalopram, the reference drug. 3.1.5 Repurposing of PDE9 Inhibitors Against AD

Combining structure-based design and computational docking, Meng, F. et al. [99] have identified a novel class of phosphodiesterase-9 (PDE9) inhibitors with a scaffold of 6-amino-pyrazolopyrimidinone. This approach is useful for discovering inhibitors since it considerably reduces the workload of chemical synthesis. The best compound, 18 (see Fig. 3), exhibits a selectivity against PDE families of nearly three orders of magnitude

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

217

with an IC50 for PDE9 and PDE5 of 21 nM and 3.3 μM, respectively. The crystal structure of the PDE9 catalytic domain in association with 28 has been established, and it reveals a hydrogen bond between compound 18 and TYR424. The 860-fold selectivity of compound 18 against PDE1B, as opposed to the approximately 30-fold selectivity of BAY73-6691, may be explained by this hydrogen bond. Thus, the authors claimed that TYR424, a particular residue of PDE8 and PDE9, is a possible target for increasing the selectivity of PDE9 inhibitors. A combinatorial strategy involving pharmacophores, molecular docking, molecular dynamics simulations, binding free energy calculations, and bioassay was utilized by Li, Z. et al. [100] to identify novel PDE9A inhibitors with new scaffolds rather than pyrazolopyrimidinones from the SPECS database of around 200,000 chemicals. As a consequence, 15 hits out of 29 compounds with 5 unique scaffolds were found to be PDE9A inhibitors with inhibitory affinities no more than 50 mM to enhance the structural diversity, differing from the pyrazolopyrimidinone-derived family. The virtual screening method’s high hit ratio of 52% showed that the combinatorial approach provides a good trade-off between computational expense and precision. According to binding pattern analysis, hits with scaffolds other than pyrazolopyrimidinones can bind the same active site pocket of PDE9A as conventional PDE9A inhibitors. Additionally, a novel molecule, 19 (see Fig. 3), was identified as a result of structurally altering compound AG-690/40135604 (see Fig. 3) (IC50 = 8.0 μM), which had an expectedly higher inhibitory affinity of 2.1 μM. A variety of new pyrazolopyrimidinone derivatives coupled with the pharmacophore of antioxidants such as ferulic and lipoic acids have been designed by Zhang, C. et al. [101] through the use of molecular docking and dynamics simulations to find multifunctional anti-AD medicines capable of inhibiting PDE9 and having antioxidant activity. In the ORAC analysis, 12 of the 14 synthesized compounds inhibited PDE9A with IC50 values less than 200 nM. The most effective multifunctional anti-AD compound, compound 20 (see Fig. 3), exhibited an IC50 of 56 nM against PDE9A and showed strong antioxidant activity (Trolox of ORAC = 3.3). Furthermore, compound 20 did not exhibit any cytotoxicity toward human neuroblastoma SH-SY5Y cells. Structure–activity relationship (SAR) and binding mode analyses of the compounds may shed light on potential future modifications. A series of PDE9 inhibitors (new pyrimidinone derivatives) with the rosiglitazone pharmacophore has been rationally designed, synthesized, and evaluated by Wu, X. N. et al. [102]. The human neuroblastoma cell line SH-SY5Y was used to test the cytotoxic effects of four of the proposed compounds, 21, 22, 23, and 24 (see Fig. 4), whose IC50 against PDE9 was less than 5 nmol/L. These tests revealed low toxicity to SH-SY5Y cells. The most effective molecule, 21, has an IC50 of 1.1 nmol/L against PDE9, which is

218

Vinay Kumar and Kunal Roy

Fig. 4 Chemical structures of top selected compounds (PDE9 inhibitors against AD) from the listed studies

much better than the reference compounds PF-04447943 and BAY73-6691. The following residues with binding energies of -53.89 kcal/mol were proposed by molecular docking research to be significant for the inhibitory actions of compound 21 on PDE9: PHE456, GLN453, LEU420, ILE403, TYR424, MET365, VAL460, PHE441, and PHE459. According to the binding patterns, all of the ligands formed π–π stacking with Phe456 and hydrogen bonding with GLN453 and PHE456, which made up the majority of these residues’ contributions to the overall binding free energy. PHE456, PHE459, and VAL460 in the hydrophobic pocket contributed the most energy. To develop effective anti-AD agents, many dual-target AChE/ PDE9A inhibitors were rationally designed, synthesized, and evaluated by Hu, J. et al. [103]. Among these targets, compounds 25 (see Fig. 4) (AChE, IC50 = 0.048 μM; PDE9A, IC50 = 0.530 μM) and 26 (AChE, IC50 = 0.223 μM; PDE9A, IC50 = 0.285 μM) displayed outstanding and balanced dual-target AChE/PDE9A inhibitory activity. These two substances have low neurotoxicity and good blood–brain barrier (BBB) penetrability. Additionally, in the Morris water maze test, compound 25 was able to enhance both cognitive and spatial memory in mice with Aβ25–35-induced cognitive deficits. In a molecular docking investigation, the authors found that compound 25 formed two hydrogen bonds with the important residue GLN453 and π–π stacking with PHE456, presumably explaining its potent inhibitory effect against PDE9A. Due

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

219

to the hydrophobic linker, compound 26 (see Fig. 4) was unable to make hydrogen bonds with GLN453 and also lost its π–π stacking connection with PHE456. The binding mode of compound 25 with AChE was demonstrated to be strong π–π stacking of the phenyl ring of the benzylpiperidine group with the residue TRP86 in AChE. The pyrazolopyrimidinone moiety of compound 25 formed π–π stacking interactions with TRP286. So, based on the information gathered from the aforementioned investigation, compound 25 was suggested as a promising candidate medicine for the treatment of AD. To highlight the differences in their interaction patterns in the presence of various metal systems, such as Zn/Mg, Mg/Mg, and Zn/Zn, Sivakumar, D. et al. [104] have collected highly potent previously reported selective PDE9 inhibitors BAY73-6691, 27, 28, PF-0447943, PF-4181366, and 29 (see Fig. 4). Molecular docking was employed to establish the initial complexes, which were then subjected to molecular dynamics simulations for 100 ns in triplicate for each system to understand the stability of the interactions before being subjected to MM/PBSA binding free energy calculations. The data were thoroughly examined by the authors, concentrating on the nonbonded interactions of the ligands with PDE9 in various metal complexes. Since all the compounds have demonstrated efficacy and selectivity and some are still in various stages of clinical trials, the goal of this study is not to declare one of the compounds to be the best. The relevance of the PDE9 metal system and its influence on the inhibitory function of the protein is only demonstrated by the authors here. To identify the novel potent PDE9 inhibitors, Swetha, R. et al. [105] have performed a structure-based virtual screening. Pharmacophore models were established using the PDB structure of the phosphodiesterase 9 enzyme (PDB ID: 6A3N, available from https://www.rcsb.org/structure/6A3N) with 9Q9 as a co-crystallized ligand, and later validated model was used to screen the ZINC15 library comprising 13,190,317 compounds to search for virtual hits. After the virtual screening, the leads were further subjected to molecular docking, ADMET predictions, and molecular dynamics simulation studies. The top hits, compounds ZINC000001305675 and ZINC000000377099 (see Fig. 5), had good anticipated ADMET scores as well as excellent docking scores of -10.90 and -10.30 kcal/mol. Finally, the hits were subjected to molecular dynamics (MD) investigations, which revealed that they formed stable complexes with PDE9 protein and had acceptable ligand RMSD values. 3.1.6 Repurposing of PDE10 Inhibitors Against AD

A combined molecular docking and pharmacophore framework approach was used by Zago´rska, A. et al. [106] to rationally design the novel series of multitarget compounds against PDE10A and 5-HT1AR targets. The majority of selective PDE10A inhibitors have been designed based on the results of co-crystallographic

220

Vinay Kumar and Kunal Roy

X-ray analyses involving the interaction of PDE10A catalytic domain bound with AMP. Following that, the novel compounds selected from the in silico investigations were synthesized and evaluated for their dual inhibitory activity against PDE10A and 5-HT1AR targets. Finally, compounds 30 (PDE10A IC50 = 5.15 μM, HT1AR IC50 = 6.90 μM) and 31 (PDE10A IC50 = 5.77 μM, HT1AR IC50 = 5.32 μM) were selected for additional research on their pharmacokinetic and brain distribution characteristics. Compounds 30 and 31, which are partial agonists of the 5-HT1AR and PDE10A inhibitors, respectively, showed the highest blood–brain barrier permeability. Compounds 30 and 31 (see Fig. 5) demonstrated antipsychotic and antidepressant-like effects in rats, as well as the restoration of recognition memory impairments. The overall efficacy, pharmacokinetics, and bioavailability studies suggest that the reported dual-acting compounds have therapeutic-like promise as a technique for treating NPS in dementia. To identify the novel potent PDE10A inhibitors, Al-Nema, M. et al. [107] have performed a structure-based virtual screening. Based on the alignment of the binding sites, the PDE10A structure (PDB ID: 5UWF, available from https://www.rcsb.org/structure/ 5UWF) in complex with the inhibitor (16d) was selected for the pharmacophore model development, in which the alignment of four protein complexes (PDB ID: 3HQW (available from https:// www.rcsb.org/structure/3HQW), 4DDL (available from https:// www.rcsb.org/structure/4DDL), 4P0N (available from https://

Fig. 5 Chemical structures of top selected compounds (PDE9 and PDE10 inhibitors against AD) from the listed studies

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

221

www.rcsb.org/structure/4P0N), and 4HF4 (available from https://www.rcsb.org/structure/4HF4)) on 5UWF demonstrated the complete integration of the four ligands into the active region of 5UWF. The well-validated model was used to screen the ZINC database comprising six million lead-like compounds for the search for virtual hits. Based on the fit value, the authors have selected the 14 top hits and subjected them to the molecular docking study in which 2 compounds displayed the highest affinity to PDE10A, thereby selected for MD simulations. The overall modeling studies demonstrated that compound ZINC42657360 formed three hydrogen bonds with ASN226, THR187, and ASP228 and two aromatic interactions with TYR78 and PHE283, besides the common interactions with the P-clamp residues PHE283 and ILE246. Finally, compound ZINC42657360 (see Fig. 5) was evaluated in a PDE-Glo Phosphodiesterase Assay and shown to have a considerable inhibitory activity of 1.60 μM against PDE10A. According to the aforementioned results, compound ZINC42657360 may be further investigated to develop new PDE10A inhibitors. A combined strategy involving pharmacophores and molecular docking was utilized by Fan, H. et al. [108] to identify novel PDE10A inhibitors with new scaffolds from the database of Sophora flavescens, a well-known Chinese herb that comprises around 78 compounds. A diverse dataset of 30 experimentally identified PDE10A inhibitors (retrieved from the published literature) was used to develop the pharmacophore models; afterward, the validated model was used to screen the database. The top compounds obtained from the virtual careening were subjected to molecular docking investigation. Compounds kosamol Q (total score = 8.82) and kosamol A (total score = 8.39) (see Fig. 5) were chosen as the top two compounds based on the docking score and may be potential PDE10A inhibitors since they have pharmacological effects, including CNS protectant properties. In this investigation, Czopek, Anna et al. [109] synthesized a library of novel 4-methoxy-2,3-dihydro-1H-isoindole-1,3-dione derivatives with different aminoalkyl moieties as possible PDE10A inhibitors and serotonin receptor ligands. After a detailed analysis, 1H-benzimidazole derivatives were found to be effective PDE10A inhibitors among the synthesized and evaluated compounds. Compound 32 (see Fig. 5), the most effective PDE10 inhibitor (IC50 = 0.886 ± 0.017 μM) of the group exhibiting drug-like properties according to Lipinski’s rule of five, was selected for further pharmacological study. The binding mechanism of compound 32 was identified to explain its inhibitory effect against PDE10A. Molecular modeling studies revealed that interactions with PDE10A depend on H-bonds formed by the 4-methoxy group in the phthalimide moiety as well as the optimum length of the carbon linker. Preliminary in vitro neurotoxicity and hepatotoxicity experiments demonstrated that compound 32 is safe and

222

Vinay Kumar and Kunal Roy

has no cytotoxicity. According to the authors, compound 32 exhibited notable antipsychotic effects in the d-AMPH-induced hyper locomotor activity test in mice, although additional pharmacological research is required to understand the mechanism. 3.1.7 Activity Data Sources, Chemical Databases, and Freely Available CADD Software and Tools

4

One of the main elements needed to perform the various tasks involved in in silico studies is a software tool. A publicly available software tool allows any researcher to execute, copy, and share the tool with the scientific community. Table 4 contains a collection of freely available databases and in silico software applications.

Conclusions and Prospects There are numerous other cases of the effective use of in silico approaches in the identification and design of PDE inhibitors aside from those studies covered in the previous section. The applications are extremely diverse, spanning from lead generation to optimization. Molecular docking, quantitative structure–activity relationship (QSAR) predictive models, pharmacophore approaches, and MD simulations have all been effectively employed to identify novel therapeutics through virtual screening techniques, or to optimize chemical derivatives and introduce new potential PDE inhibitors as anti-Alzheimer’s disease agents. Detailed research and clinical studies have demonstrated that therapeutic targets implicated in PDE inhibition can be considered potential therapeutic approaches in cognitive disorders and dementia. PDEs are a wide class of enzymes whose primary functions involve hydrolyzing cyclic nucleotides into monophosphate isoforms. It has been discovered that cyclic nucleotides are second messengers, playing important roles in hormone and neurotransmitter signaling. PDE dysfunction can lead to dysregulation of several cellular pathways, including the immune system, transduction and transcription signaling pathways, and the inflammatory response, all of which play important roles in neurodegenerative conditions. Unfortunately, the main issue with many PDE subtype inhibitors, whether developed or discovered, is a lack of selectivity. Despite the recent design of inhibitors with notable selectivity for the PDE subtype using computational approaches, none of them have yet been able to reach the market. To prevent and cure AD, new therapeutic strategies are required. It is important to note that many PDE-specific therapies have additional pharmacological effects that may interpret their efficacy. Furthermore, some of these drugs’ potential additional pharmacological properties have not been extensively investigated. Other pharmacological effects of the drugs, including neurotransmitter receptor activity or some downstream effects, may also be noteworthy in addition to issues like the protective benefits of such drugs, which are directly attributable to PDE inhibition. PDEs are becoming more useful as targets for

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

223

Table 4 The Weblinks of tools and servers in alphabetical order Tools/Servers

Weblinks

Achilles Blind Docking server https://bio-hpc.ucam.edu/achilles/ ACD/ChemSketch

https://www.acdlabs.com/resources/freeware/chemsketch/download. php

AMBER18

https://ambermd.org/GetAmber.php

AutoDock Vina

http://vina.scripps.edu/

Avogadro version 1.2

https://avogadro.cc/news/avogadro-1-2-0-released/

BIOVIA Discovery Studio

https://discover.3ds.com/discovery-studio-visualizer-download

Binding Database

https://www.bindingdb.org/bind/index.jsp

CHARMM-GUI Glycolipid Modeler

http://www.charmm-gui.org/?doc=input/glycan

Chiron online server

https://dokhlab.med.psu.edu/chiron/login.php

ChemDraw Ultra12v

https://en.freedownloadmanager.org/users-choice/Chemdraw_Ultra_ 7.0_Free_Download.html

ClustalW

https://www.genome.jp/tools-bin/clustalw

ClustalX

http://www.clustal.org/clustal2/

DISPHOS 1.3 server

https://dabi.temple.edu/disphos/pred/predict

ERRAT

https://servicesn.mbi.ucla.edu/ERRAT/

ChEMBL database

https://www.ebi.ac.uk/chembl/

FlexX

https://www.biosolveit.de/FlexX/

GaussView program

https://gaussian.com/gaussview6/

GROMACS 5.1.2

http://manual.gromacs.org/documentation/5.1.2/download.html

GlycoEP server

https://bio.tools/glycoep

HMMER tool

http://hmmer.org/

HHsearch tool

https://github.com/soedinglab/hh-suite

HyperChem

http://www.hyper.com/

idock tool

https://github.com/HongjianLi/idock

I-TASSER web server

https://zhanglab.ccmb.med.umich.edu/I-TASSER/

LigandScout tool

https://ligandscout.software.informer.com/

Modeller 9.23

https://salilab.org/modeller/9.23/release.html

Molegro Molecular Viewer

http://molexus.io/molegro-molecular-viewer/

Molegro Virtual Docker software

http://molexus.io/molegro-virtual-docker/

MolProbity web server

http://molprobity.biochem.duke.edu/

MMTK software

http://dirac.cnrs-orleans.fr/MMTK.html (continued)

224

Vinay Kumar and Kunal Roy

Table 4 (continued) Tools/Servers

Weblinks

NAMD 2.13

https://www.ks.uiuc.edu/Development/Download/download.cgi? PackageName=NAMD

NCBI database

https://www.ncbi.nlm.nih.gov/

NetPhos 3.1 server

http://www.cbs.dtu.dk/services/NetPhos/

NetSurfP-2.0 server

http://www.cbs.dtu.dk/services/NetSurfP/

OGTSite server

http://csb.cse.yzu.edu.tw/OGTSite/

PaleAle 5.0 server

http://distilldeep.ucd.ie/paleale/quickhelp.html

PeptideMass online server

https://web.expasy.org/peptide_mass/

PEP-FOLD3 server

https://bioserv.rpbs.univ-paris-diderot.fr/services/PEP-FOLD/

PharmaGist online tool

https://bioinfo3d.cs.tau.ac.il/PharmaGist/

PileUp tool

http://www.dbbm.fiocruz.br/cgc/pileup.html

pmemd.cuda module in AMBER18 tool

https://ambermd.org/GPUHowTo.php

PROCHECK

https://www.ebi.ac.uk/thornton-srv/software/PROCHECK/

ProfileScan

http://130.88.97.239/bioactivity/newpfscan.html

Profile-profile alignment tool http://ffas.godziklab.org/ffas-cgi/cgi/ffas.pl FFAS03 PubChem

https://pubchem.ncbi.nlm.nih.gov/

PyMOL tool

https://pymol.org/dsc/

PyRx 0.8

https://pyrx.sourceforge.io/

QSAR model

http://teqip.jdvu.ac.in/QSAR_Tools/

SAM tool

https://compbio.soe.ucsc.edu/sam.html

Schro¨dinger software

https://www.schrodinger.com/

smina software

https://sourceforge.net/projects/smina/

SANCDB

https://sancdb.rubi.ru.ac.za/

SwissDock server

http://www.swissdock.ch/

SWISS-MODEL web server

https://swissmodel.expasy.org/

UCSF Chimera

https://www.cgl.ucsf.edu/chimera/

Verify3D

https://servicesn.mbi.ucla.edu/Verify3D/

Verification Server

https://servicesn.mbi.ucla.edu/SAVES/

vROCS (OpenEye)

https://docs.eyesopen.com/applications/rocs/vrocs/vrocs.html

YASARA server

http://www.yasara.org/minimizationserver.htm

YinOYang 1.2 server

http://www.cbs.dtu.dk/services/YinOYang/

ZINC15 database

https://zinc15.docking.org/

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . .

225

investigating potential pharmacotherapeutic drugs to treat AD. The progress of computational modeling approaches will provide a significant boost in the future for tackling the selectivity problem. These developments will enable more in-depth research on a smaller time scale, enabling computational researchers to identify the interactions and structural features necessary for excellent selectivity. These results give reason to believe that the unexplored potential of diverse PDE inhibitors will be achieved in the near future. At last, we believe that the information provided here will serve as a guide for researchers working to develop safe, efficient PDE inhibitors from synthetic or naturally derived compounds as the antiAlzheimer’s disease agent.

Acknowledgments VK thanks the Indian Council of Medical Research (ICMR), New Delhi (File No: BMI/11(03)/2022, IRIS Cell No.: 2021-8243, dated: 13/05/2022), for financial support in the form of a Research Associateship (RA). References 1. Alzheimer’s Association (2021) 2021 Alzheimer’s disease facts and figures. Alzheimers Dement 17(3):327–406 2. Gauthier S, Rosa-Neto P, Morais JA, Webster C (2021) World Alzheimer Report 2021: journey through the diagnosis of dementia. Alzheimer’s Disease International, London 3. Tang Y, Zhang D, Gong X, Zheng J (2022) A mechanistic survey of Alzheimer’s disease. Biophys Chem 281:106735. https://doi. org/10.1016/j.bpc.2021.106735 4. Esang M, Gupta M (2021) Aducanumab as a novel treatment for Alzheimer’s disease: a decade of hope, controversies, and the future. Cureus 13(8):e17591. https://doi.org/10. 7759/cureus.17591 5. Dutta BJ, Singh S, Seksaria S, Gupta GD, Bodakhe SH, Singh A (2022) Potential role of IP3/Ca2+ signaling and phosphodiesterases: relevance to neurodegeneration in Alzheimer’s disease and possible therapeutic strategies. Biochem Pharmacol 201:115071. https://doi.org/10.1016/j.bcp.2022. 115071 6. Nabavi SM, Talarek S, Listos J, Nabavi SF, Devi KP, de Oliveira MR et al (2019) Phosphodiesterase inhibitors say NO to Alzheimer’s disease. Food Chem Toxicol 134: 110822. https://doi.org/10.1016/j.fct. 2019.110822

7. Xi M, Sun T, Chai S, Xie M, Chen S, Deng L et al (2022) Therapeutic potential of phosphodiesterase inhibitors for cognitive amelioration in Alzheimer’s disease. Eur J Med Chem 232:114170. https://doi.org/10. 1016/j.ejmech.2022.114170 8. Heckman PR, Blokland A, Prickaerts J (2017) From age-related cognitive decline to Alzheimer’s disease: a translational overview of the potential role for phosphodiesterases. In: Phosphodiesterases: CNS functions and diseases. Springer, Cham, pp 135–168 9. Wu Y, Li Z, Huang YY, Wu D, Luo HB (2018) Novel phosphodiesterase inhibitors for cognitive improvement in Alzheimer’s disease: miniperspective. J Med Chem 61(13): 5467–5483. https://doi.org/10.1021/acs. jmedchem.7b01370 10. Prickaerts J, Heckman PR, Blokland A (2017) Investigational phosphodiesterase inhibitors in phase I and phase II clinical trials for Alzheimer’s disease. Expert Opin Investig Drugs 26(9):1033–1048. https://doi.org/10. 1080/13543784.2017.1364360 11. Hiramatsu M, Takiguchi O, Nishiyama A, Mori H (2010) Cilostazol prevents amyloid β peptide25-35-induced memory impairment and oxidative stress in mice. Br J Pharmacol 161(8):1899–1912. https://doi.org/10. 1111/j.1476-5381.2010.01014.x

226

Vinay Kumar and Kunal Roy

12. Kumar A, Nisha CM, Silakari C, Sharma I, Anusha K, Gupta N, Kumar A (2016) Current and novel therapeutic molecules and targets in Alzheimer’s disease. J Formos Med Assoc 115(1):3–10. https://doi.org/10.1016/j. jfma.2015.04.001 13. Cavasotto CN, Phatak SS (2009) Homology modeling in drug discovery: current trends and applications. Drug Discov Today 14(13–14):676–683. https://doi.org/10. 1016/j.drudis.2009.04.006 14. Saxena A, Sangwan RS, Mishra S (2013) Fundamentals of homology modeling steps and comparison among important bioinformatics tools: an overview. Sci Int 1(7):237–252 15. Hasani HJ, Barakat K (2017) Homology modeling: an overview of fundamentals and tools. Int Rev Model Simul 10(2):1–14. https://doi.org/10.15866/iremos.v10i2. 11412 16. Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig H et al (2000) The protein data bank. Nucleic Acids Res 28(1):235–242. https://doi.org/10.1093/ nar/28.1.235. Available from: https://www. rcsb.org/ 17. Franc¸a TCC (2015) Homology modeling: an important tool for the drug discovery. J Biomol Struct Dyn 33(8):1780–1793. https:// doi.org/10.1080/07391102.2014.971429 18. Muhammed MT, Aki-Yalcin E (2019) Homology modeling in drug discovery: overview, current applications, and future perspectives. Chem Biol Drug Des 93(1):12–20. https://doi.org/10.1111/cbdd.13388 19. Schwede T, Kopp J, Guex N, Peitsch MC (2003) SWISS-MODEL: an automated protein homology-modeling server. Nucleic Acids Res 31(13):3381–3385. https://doi. org/10.1093/nar/gkg520 20. Vyas VK, Ukawala RD, Ghate M, Chintha C (2012) Homology modeling a fast tool for drug discovery: current perspectives. Indian J Pharm Sci 74(1):1. https://doi.org/10. 4103/0250-474X.102537 21. Pitman MR, Menz RI (2006) Methods for protein homology modelling. Appl Microbiol Biotechnol 6:37–59. https://doi.org/10. 1016/S1874-5334(06)80005-5 22. Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ (1997) Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res 25:3389–3402. https://doi.org/10.1093/nar/25.17.3389 23. Potter SC, Luciani A, Eddy SR, Park Y, Lopez R, Finn RD (2018) HMMER web server: 2018 update. Nucleic Acids Res 46

(W1):W200–W204. https://doi.org/10. 1093/nar/gky448 24. Jaroszewski L, Rychlewski L, Li Z, Li W, Godzik A (2005) FFAS03: a server for profile– profile sequence alignments. Nucleic Acids Res 33(suppl_2):W284–W288. https://doi. org/10.1093/nar/gki418 25. Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, McWilliam H et al (2007) Clustal W and Clustal X version 2.0. Bioinformatics 23(21):2947–2948. https://doi.org/10.1093/bioinformatics/ btm404 26. Syste`mes D (2016) Biovia, discovery studio modeling environment. Dassault Syste`mes Biovia, San Diego 27. Davis IW, Murray LW, Richardson JS, Richardson DC (2004) MOLPROBITY: structure validation and all-atom contact analysis for nucleic acids and their complexes. Nucleic Acids Res 32(suppl_2):W615– W619. https://doi.org/10.1093/nar/ gkh398 28. Laskowski RA, MacArthur MW, Moss DS, Thornton JM (1993) PROCHECK: a program to check the stereochemical quality of protein structures. J Appl Crystallogr 26(2): 283–291. https://doi.org/10.1107/ S0021889892009944 29. Eisenberg D, Lu¨thy R, Bowie JU (1997) VERIFY3D: assessment of protein models with three-dimensional profiles. In: Methods in enzymology, vol 277. Academic Press, New York, pp 396–404. https://doi.org/10. 1016/S0076-6879(97)77022-8 30. Colovos C, Yeates TO (1993) ERRAT: an empirical atom-based method for validating protein structures. Protein Sci 2(9): 1511–1519 31. McGuffin LJ (2008) Protein fold recognition and threading. In: Computational structural biology. World Scientific, London, pp 37–60 32. Xu Y, Liu Z, Cai L, Xu D (2007) Protein structure prediction by protein threading. In: Computational methods for protein structure prediction and modeling. Springer, New York, pp 1–42 33. Zhang Y (2008) I-TASSER server for protein 3D structure prediction. BMC Bioinformatics 9(1):40. https://doi.org/10.1186/14712105-9-40 34. Bernasconi A, Segre AM (2000) Ab initio methods for protein structure prediction: a new technique based on Ramachandran plots. ERCIM News 43:13–14 35. Bonneau R, Baker D (2001) Ab initio protein structure prediction: progress and prospects. Annu Rev Biophys Biomol Struct 30(1):

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . . 173–189. https://doi.org/10.1146/ annurev.biophys.30.1.173 36. Simons KT, Bonneau R, Ruczinski I, Baker D (1999) Ab initio protein structure prediction of CASP III targets using ROSETTA. Proteins 37(S3):171–176. https://doi.org/10. 1002/(SICI)1097-0134(1999)37:3 37. Schaap MG, Leij FJ, Van Genuchten MT (2001) Rosetta: a computer program for estimating soil hydraulic parameters with hierarchical pedotransfer functions. J Hydrol 251(3–4):163–176. https://doi.org/10. 1016/S0076-6879(04)83004-0 38. Zhang Y, Kolinski A, Skolnick J (2003) TOUCHSTONE II: a new approach to ab initio protein structure prediction. Biophys J 85(2):1145–1164. https://doi.org/10. 1016/S0006-3495(03)74551-2 39. Fiser A (2010) Template-based protein structure modeling. In: Computational biology. Humana Press, Totowa, pp 73–94 40. Canzar S, Ringeling FR (2020) Proteinprotein Interaction Networks: Methods and Protocols. Humana Press, New York, eBook ISBN 978-1-4939-9873-9, Springer Science +Business Media, LLC, part of Springer Nature 2020. https://doi.org/10.1007/ 978-1-4939-9873-9 41. Pavlopoulos GA, Hooper SD, Sifrim A, Schneider R, Aerts J (2011) Medusa: a tool for exploring and clustering biological networks. BMC Res Notes 4(1):1–6. https:// doi.org/10.1186/1756-0500-4-384 42. Sanders MP, McGuire R, Roumen L, de Esch IJ, de Vlieg J, Klomp JP, de Graaf C (2012) From the protein’s perspective: the benefits and challenges of protein structure-based pharmacophore modeling. Med Chem Commun 3(1):28–38. https://doi.org/10.1039/ C1MD00210D 43. Gaurav A, Gautam V, Pereira S, Alvarez-LeiteJ, Vetri F, Choudhury M et al (2014) Structure-based three-dimensional pharmacophores as an alternative to traditional methodologies. J Recept Ligand Channel Res 7: 27–38. https://doi.org/10.2147/JRLCR. S46845 44. Wolber G, Langer T (2005) LigandScout: 3-D pharmacophores derived from proteinbound ligands and their use as virtual screening filters. J Chem Inf Model 45(1):160–169. https://doi.org/10.1021/ci049885e 45. Park H, Shin Y, Kim J, Hong S (2016) Application of fragment-based de novo design to the discovery of selective picomolar inhibitors of glycogen synthase kinase-3 beta. J Med Chem 59(19):9018–9034. https://doi.org/ 10.1021/acs.jmedchem.6b00944

227

46. Dar AM, Mir S (2017) Molecular docking: approaches, types, applications and basic challenges. J Anal Bioanal Tech 8(2):1–3. https:// doi.org/10.4172/2155-9872.1000356 47. Morris GM, Lim-Wilby M (2008) Molecular docking. In: Molecular modeling of proteins. Humana Press, Totowa, pp 365–382 48. Meng XY, Zhang HX, Mezei M, Cui M (2011) Molecular docking: a powerful approach for structure-based drug discovery. Curr Comput Aided Drug Des 7(2): 146–157. https://doi.org/10.2174/ 157340911795677602 49. Salmaso V, Moro S (2018) Bridging molecular docking to molecular dynamics in exploring ligand-protein recognition process: an overview. Front Pharmacol 9:923. https:// doi.org/10.3389/fphar.2018.00923 50. Bitencourt-Ferreira G, de Azevedo WF (2019) Molegro virtual docker for docking. In: Docking screens for drug discovery. Humana Press, New York, pp 149–167. https://doi.org/10.1007/978-1-49399752-7_10 51. Schellhammer I, Rarey M (2004) FlexX-scan: fast, structure-based virtual screening. Proteins 57(3):504–517. https://doi.org/10. 1002/(SICI)1097-0134(19991101)37:2 52. Hollingsworth SA, Dror RO (2018) Molecular dynamics simulation for all. Neuron 99(6): 1129–1143. https://doi.org/10.1016/j.neu ron.2018.08.011 53. Allen MP (2004) Introduction to molecular dynamics simulation. In: Computational soft matter: from synthetic polymers to proteins, NIC series, vol 23. John von Neumann Institute for Computing, pp 1–28 54. Zheng L, Alhossary AA, Kwoh CK, Mu Y (2019) Molecular dynamics and simulation. Elsevier Inc. Nanyang Technological University, Singapore 55. Karplus M, Petsko GA (1990) Molecular dynamics simulations in biology. Nature 347(6294):631–639. https://doi.org/10. 1038/347631a0 56. Rastelli G, Rio AD, Degliesposti G, Sgobba M (2010) Fast and accurate predictions of binding free energies using MM-PBSA and MM-GBSA. J Comput Chem 31(4): 797–810. https://doi.org/10.1002/jcc. 21372 57. Homeyer N, Gohlke H (2012) Free energy calculations by the molecular mechanics Poisson–Boltzmann surface area method. Mol Inform 31(2):114–122. https://doi. org/10.1002/minf.201100135 58. Kollman PA, Massova I, Reyes C, Kuhn B, Huo S, Chong L et al (2000) Calculating

228

Vinay Kumar and Kunal Roy

structures and free energies of complex molecules: combining molecular mechanics and continuum models. Acc Chem Res 33(12): 889–897. https://doi.org/10.1021/ ar000033j 59. Phillips JC, Zheng G, Kumar S, Kale´ LV (2002) NAMD: biomolecular simulation on thousands of processors. In: SC’02: Proceedings of the 2002 ACM/IEEE conference on supercomputing. IEEE, Washington, DC, pp 36–36. https://doi.org/10.1109/SC.2002. 10019 60. Bishop KP, Constable S, Faruk NF, Roy PN (2015) OpenMM accelerated MMTK. Comput Phys Commun 191:203–208. https:// doi.org/10.1016/j.cpc.2015.01.025 61. Van Der Spoel D, Lindahl E, Hess B, Groenhof G, Mark AE, Berendsen HJ (2005) GROMACS: fast, flexible, and free. J Comput Chem 26(16):1701–1718. https:// doi.org/10.1002/jcc.20291 62. Lee TS, Cerutti DS, Mermelstein D, Lin C, LeGrand S, Giese TJ et al (2018) GPU-accelerated molecular dynamics and free energy methods in Amber18: performance enhancements and new features. J Chem Inf Model 58(10):2043–2050. https://doi.org/10.1021/acs.jcim.8b00462 63. Humphrey W, Dalke A, Schulten K (1996) VMD: visual molecular dynamics. J Mol Graph 14(1):33–38 64. Roy K, Kar S, Das RN (2015) Understanding the basics of QSAR for applications in pharmaceutical sciences and risk assessment. Academic Press. ISBN 0128016337, 9780128016336 65. Roy K, Kar S, Das RN (2015) A primer on QSAR/QSPR modeling: fundamental concepts. Springer. ISBN:3319172816, 9783319172811 66. De P, Kumar V, Kar S, Roy K, Leszczynsk J (2022) Repurposing FDA approved drugs as possible anti-SARS-CoV-2 medications using ligand-based computational approaches: sum of ranking difference-based model selection. Struct Chem 33:1741–1753. https://doi. org/10.1007/s11224-022-01975-3 67. Chatterjee M, Roy K (2022) Application of cross-validation strategies to avoid overestimation of performance of 2D-QSAR models for the prediction of aquatic toxicity of chemical mixtures. SAR QSAR Environ Res 33(6): 463–484. https://doi.org/10.1080/ 1062936X.2022.2081255 68. Banerjee A, Chatterjee M, De P, Roy K (2022) Quantitative predictions from chemical read-across and their confidence measures. Chemom Intell Lab Syst 227:104613.

https://doi.org/10.1016/j.chemolab.2022. 104613 69. Banerjee A, Roy K (2022) First report of q-RASAR modeling toward an approach of easy interpretability and efficient transferability. Mol Divers 26(5):2847–2862. https:// doi.org/10.1007/s11030-022-10478-6 70. Choudhury C, Sastry GN (2019) Pharmacophore modelling and screening: concepts, recent developments and applications in rational drug design. In: Structural bioinformatics: applications in preclinical drug discovery process. Springer, Cham, pp 25–53. https://doi. org/10.1007/978-3-030-05282-9_2 71. Yang SY (2010) Pharmacophore modeling and applications in drug discovery: challenges and recent advances. Drug Discov Today 15(11–12):444–450. https://doi.org/10. 1016/j.drudis.2010.03.013 72. Schaller D, Dora Sˇ, Theresa N, Lihua D, Trung NN, Szymon P, David M, Marcel B, Gerhard W (2020) Next generation 3D pharmacophore modeling. Wiley Interdiscip Rev Comput Mol Sci 10:e1468. https://doi.org/ 10.1002/wcms.1468 73. Yu W, MacKerell AD (2017) Computer-aided drug design methods. In: Sass P (ed) Antibiotics. Methods in molecular biology, vol 1520. Humana Press, New York. https://doi.org/10.1007/978-1-49396634-9_5 74. Moro S, Bacilieri M, Deflorian F (2007) Combining ligand-based and structure-based drug design in the virtual screening arena. Expert Opin Drug Discov 2(1):37–49. https://doi.org/10.1517/17460441.2.1.37 75. Zhu J, Yang Q, Dai D, Huang Q (2013) X-ray crystal structure of phosphodiesterase 2 in complex with a highly selective, nanomolar inhibitor reveals a binding-induced pocket important for selectivity. J Am Chem Soc 135(32):11708–11711. https://doi.org/10. 1021/ja404449g 76. Tang L, Jiang J, Song G, Wang Y, Zhuang Z, Tan Y et al (2021) Design, synthesis, and biological evaluation of novel 6H-benzo[c] chromen-6-one derivatives as potential phosphodiesterase II inhibitors. Int J Mol Sci 22(11):5680. https://doi.org/10.3390/ ijms22115680 77. Zhou Y, Li J, Yuan H, Su R, Huang Y, Huang Y et al (2021) Design, synthesis, and evaluation of dihydropyranopyrazole derivatives as novel PDE2 inhibitors for the treatment of Alzheimer’s disease. Molecules 26(10):3034. h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / molecules26103034 78. Qvortrup K, Jensen JF, Sørensen MS, Kouskoumvekaki I, Petersen RK, Taboureau

Computational Modeling Approaches in Search of Anti-Alzheimer’s Disease. . . O et al (2017) Synthesis and biological evaluation of dihydropyrano-[2,3-c]pyrazoles as a new class of PPARγ partial agonists. PLoS One 12(2):e0162642. https://doi.org/10. 1371/journal.pone.0162642 79. Li Z, Huang Y, Wu Y, Chen J, Wu D, Zhan CG, Luo HB (2019) Absolute binding free energy calculation and design of a subnanomolar inhibitor of phosphodiesterase-10. J Med Chem 62(4):2099–2111. https://doi. org/10.1021/acs.jmedchem.8b01763 80. Bruno O, Fedele E, Prickaerts J, Parker LA, Canepa E, Brullo C et al (2011) GEBR-7b, a novel PDE4D selective inhibitor that improves memory in rodents at non-emetic doses. Br J Pharmacol 164(8):2054–2063. https://doi.org/10.1111/j.1476-5381. 2011.01524.x 81. Gong B, Vitolo OV, Trinchese F, Liu S, Shelanski M, Arancio O (2004) Persistent improvement in synaptic and cognitive functions in an Alzheimer mouse model after rolipram treatment. J Clin Investig 114(11): 1624–1634. https://doi.org/10.1172/ JCI22831 82. Hu J, Pan T, An B, Li Z, Li X, Huang L (2019) Synthesis and evaluation of clioquinol-rolipram/roflumilast hybrids as multitarget-directed ligands for the treatment of Alzheimer’s disease. Eur J Med Chem 163: 512–526. https://doi.org/10.1016/j. ejmech.2018.12.013 83. Liao Y, Jia X, Tang Y, Li S, Zang Y, Wang L et al (2019) Discovery of novel inhibitors of phosphodiesterase 4 with 1-phenyl-3, 4-dihydroisoquinoline scaffold: structurebased drug design and fragment identification. Bioorg Med Chem Lett 29(22): 126720. https://doi.org/10.1016/j.bmcl. 2019.126720 84. Tang L, Huang C, Zhong J, He J, Guo J, Liu M et al (2019) Discovery of arylbenzylamines as PDE4 inhibitors with potential neuroprotective effect. Eur J Med Chem 168:221–231. https://doi.org/10.1016/j.ejmech.2019. 02.026 85. Zhou ZZ, Ge BC, Zhong QP, Huang C, Cheng YF, Yang XM et al (2016) Development of highly potent phosphodiesterase 4 inhibitors with anti-neuroinflammation potential: design, synthesis, and structureactivity relationship study of catecholamides bearing aromatic rings. Eur J Med Chem 124:372–379. https://doi.org/10.1016/j. ejmech.2016.08.052 86. Zhou ZZ, Cheng YF, Zou ZQ, Ge BC, Yu H, Huang C et al (2017) Discovery of N-alkyl catecholamides as selective phosphodiesterase-4 inhibitors with anti-

229

neuroinflammation potential exhibiting antidepressant-like effects at non-emetic doses. ACS Chem Neurosci 8(1):135–146. https://doi.org/10.1021/acschemneuro. 6b00271 87. Zou ZQ, Chen JJ, Feng HF, Cheng YF, Wang HT, Zhou ZZ et al (2017) Novel phosphodiesterase 4 inhibitor FCPR03 alleviates lipopolysaccharide-induced neuroinflammation by regulation of the cAMP/PKA/ CREB signaling pathway and NF-κB inhibition. J Pharmacol Exp Ther 362(1):67–77. https://doi.org/10.1124/jpet.116.239608 88. Liu J, Liu L, Zheng L, Feng KW, Wang HT, Xu JP, Zhou ZZ (2022) Discovery of novel 2, 3-dihydro-1H-inden-1-ones as dual PDE4/AChE inhibitors with more potency against neuroinflammation for the treatment of Alzheimer’s disease. Eur J Med Chem 238: 114503. https://doi.org/10.1016/j.ejmech. 2022.114503 89. Lin Y, Ahmed W, He M, Xiang X, Tang R, Cui ZN (2020) Synthesis and bioactivity of phenyl substituted furan and oxazole carboxylic acid derivatives as potential PDE4 inhibitors. Eur J Med Chem 207:112795. https://doi.org/ 10.1016/j.ejmech.2020.112795 90. Li YS, Hu DK, Zhao DS, Liu XY, Jin HW, Song GP et al (2017) Design, synthesis and biological evaluation of 2,4-disubstituted oxazole derivatives as potential PDE4 inhibitors. Bioorg Med Chem 25(6):1852–1859. https://doi.org/10.1016/j.bmc.2017. 01.047 91. Hu DK, Zhao DS, He M, Jin HW, Tang YM, Zhang LH et al (2018) Synthesis and bioactivity of 3,5-dimethylpyrazole derivatives as potential PDE4 inhibitors. Bioorg Med Chem Lett 28(19):3276–3280. https://doi. org/10.1016/j.bmcl.2018.03.031 92. Mao F, Wang H, Ni W, Zheng X, Wang M, Bao K et al (2018) Design, synthesis, and biological evaluation of orally available firstgeneration dual-target selective inhibitors of acetylcholinesterase (AChE) and phosphodiesterase 5 (PDE5) for the treatment of Alzheimer’s disease. ACS Chem Neurosci 9(2): 328–345. https://doi.org/10.1021/ acschemneuro.7b00345 93. Zhou LY, Zhu Y, Jiang YR, Zhao XJ, Guo D (2017) Design, synthesis and biological evaluation of dual acetylcholinesterase and phosphodiesterase 5A inhibitors in treatment for Alzheimer’s disease. Bioorg Med Chem 27(17):4180–4184. https://doi.org/10. 1016/j.bmcl.2017.07.013 94. Rabal O, Sa´nchez-Arias JA, CuadradoTejedor M, de Miguel I, Pe´rez-Gonza´lez M, Garcı´a-Barroso C et al (2018) Design,

230

Vinay Kumar and Kunal Roy

synthesis, biological evaluation and in vivo testing of dual phosphodiesterase 5 (PDE5) and histone deacetylase 6 (HDAC6)-selective inhibitors for the treatment of Alzheimer’s disease. Eur J Med Chem 150:506–524. https://doi.org/10.1016/j.ejmech.2018. 03.005 95. Pitts WJ, Vaccaro W, Huynh T, Leftheris K, Roberge JY, Barbosa J et al (2004) Identification of purine inhibitors of phosphodiesterase 7 (PDE7). Bioorg Med Chem Lett 14(11): 2955–2958. https://doi.org/10.1016/j. bmcl.2004.03.021 96. Bernardelli P, Lorthiois E, Vergne F, Oliveira C, Mafroud AK, Proust E et al (2004) Spiroquinazolinones as novel, potent, and selective PDE7 inhibitors. Part 2: Optimization of 5, 8-disubstituted derivatives. Bioorg Med Chem Lett 14(18):4627–4631 97. Redondo M, Palomo V, Brea J, Pe´rez DI, ´ lvarez R, Pe´rez C et al (2012) IdenMartı´n-A tification in silico and experimental validation of novel phosphodiesterase 7 inhibitors with efficacy in experimental autoimmune encephalomyelitis mice. ACS Chem Neurosci 3(10): 793–803. https://doi.org/10.1021/ cn300105c 98. Jankowska A, Satała G, Kołaczkowski M, Bucki A, Głuch-Lutwin M, S´wierczek A et al (2020) Novel anilide and benzylamide derivatives of arylpiperazinylalkanoic acids as 5-HT1A/5-HT7 receptor antagonists and phosphodiesterase 4/7 inhibitors with procognitive and antidepressant activity. Eur J Med Chem 201:112437. https://doi.org/ 10.1016/j.ejmech.2020.112437 99. Meng F, Hou J, Shao YX, Wu PY, Huang M, Zhu X et al (2012) Structure-based discovery of highly selective phosphodiesterase-9A inhibitors and implications for inhibitor design. J Med Chem 55(19):8549–8558. https://doi. org/10.1021/jm301189c 100. Li Z, Lu X, Feng LJ, Gu Y, Li X, Wu Y, Luo HB (2015) Molecular dynamics-based discovery of novel phosphodiesterase-9A inhibitors with non-pyrazolopyrimidinone scaffolds. Mol BioSyst 11(1):115–125. https://doi.org/10.1039/C4MB00389F 101. Zhang C, Zhou Q, Wu XN, Huang YD, Zhou J, Lai Z et al (2018) Discovery of novel PDE9A inhibitors with antioxidant activities for treatment of Alzheimer’s disease. J Enzyme Inhib Med Chem 33(1):260–270. https://doi.org/10.1080/14756366.2017. 1412315 102. Wu XN, Huang YD, Li JX, Yu YF, Qian Z, Zhang C et al (2018) Structure-based design, synthesis, and biological evaluation of novel

pyrimidinone derivatives as PDE9 inhibitors. Acta Pharm Sin B 8(4):615–628. https://doi. org/10.1016/j.apsb.2017.12.007 103. Hu J, Huang YD, Pan T, Zhang T, Su T, Li X et al (2018) Design, synthesis, and biological evaluation of dual-target inhibitors of acetylcholinesterase (AChE) and phosphodiesterase 9A (PDE9A) for the treatment of Alzheimer’s disease. ACS Chem Neurosci 10(1):537–551. https://doi.org/10.1021/acschemneuro. 8b00376 104. Sivakumar D, Mudedla S, Jang S, Kim H, Park H, Choi Y et al (2021) Computational study on selective PDE9 inhibitors on PDE9Mg/Mg, PDE9-Zn/Mg, and PDE9-Zn/Zn systems. Biomol Ther 11(5):709. https:// doi.org/10.3390/biom11050709 105. Swetha R, Sharma A, Singh R, Ganeshpurkar A, Kumar D, Kumar A, Singh SK (2022) Combined ligand-based and structure-based design of PDE 9A inhibitors against Alzheimer’s disease. Mol Divers 26(5):2877–2892. https://doi.org/10. 1007/s11030-022-10504-7 106. Zago´rska A, Bucki A, Partyka A, Jastrze˛bskaWie˛sek M, Siwek A, Głuch-Lutwin M et al (2022) Design, synthesis, and behavioral evaluation of dual-acting compounds as phosphodiesterase type 10A (PDE10A) inhibitors and serotonin ligands targeting neuropsychiatric symptoms in dementia. Eur J Med Chem 233:114218. https://doi.org/10.1016/j. ejmech.2022.114218 107. Al-Nema M, Gaurav A, Lee VS, Gunasekaran B, Lee MT, Okechukwu P, Nimmanpipug P (2022) Structure-based discovery and bio-evaluation of a cyclopenta [4, 5] thieno [2, 3-d] pyrimidin-4-one as a phosphodiesterase 10A inhibitor. RSC Adv 12(3): 1576–1591. https://doi.org/10.1039/ D1RA07649C 108. Fan H, Guo J, Zhang Y, Gu Y, Ning Z, Qiao Y, Wang X (2018) The rational search for PDE10A inhibitors from Sophora flavescens roots using pharmacophore and docking-based virtual screening. Mol Med Rep 17:388–393. https://doi.org/10. 3892/mmr.2017.7871 109. Czopek A, Partyka A, Bucki A, Pawłowski M, Kołaczkowski M, Siwek A, Głuch-Lutwin M, Koczurkiewicz P, Pe˛kala E, Jaromin A, Tyliszczak B, Wesołowska A, Zago´rska A (2020) Impact of N-alkylamino substituents on serotonin receptor (5-HTR) affinity and phosphodiesterase 10A (PDE10A) inhibition of isoindole-1,3-dione derivatives. Molecules 25(17):3868. https://doi.org/10.3390/ molecules25173868

Chapter 8 Recent Advances in Computational Modeling of Multi-targeting Inhibitors as Anti-Alzheimer Agents Khac-Minh Thai , Thai-Son Tran , The-Huan Tran , Thi-Cam-Nhung Cao, Hoang-Nhan Ho, Phuong Nguyen Hoai Huynh, Tan Thanh Mai , Thanh-Dao Tran , Minh-Tri Le, and Van-Thanh Tran Abstract Alzheimer’s disease (AD) is a neurodegenerative illness that affects the brain and is linked to cognitive decline, memory problems, and behavioral changes. It is highly prevalent in the elderly, with a constantly growing number of new cases worldwide. In affluent nations with aging populations, AD has been a major source of economic and social problems. As a result, the discovery of novel treatment methods for this disease is now crucial. With advances in research on the pathological mechanisms of AD, many new drug targets have been proposed and focused on in-depth investigations. AD has now been identified as a multifactorial disease. Therefore, the goal of therapeutic drug development has largely been directed at acting on multiple therapeutic targets of the disease at the same time. Computational modeling is a potent and robust method in the discovery and development of pharmacological drugs. Recently, this approach has played an increasingly important role in the search for new medications to treat AD. Computational modeling helps conserve experimental resources and dramatically accelerates advances in drug research. In this chapter, various computational modeling methods utilized in designing multi-targeting inhibitors as anti-Alzheimer agents would be described. Key words Alzheimer’s disease (AD), Multi-targeting inhibitors, Amyloid beta (Aβ), β-Secretase (BACE-1), γ-Secretase, Acetylcholinesterase (AChE), Butyrylcholinesterase (BuChE), Tau, Computer-aided drug design (CADD), Homology modeling, Molecular docking, Molecular dynamics (MDs) simulation, Pharmacophore modeling, Quantitative structure–activity relationship (QSAR)

1

Introduction Alzheimer’s disease (AD) is one of today’s most serious medical problems. The condition exhibits intellectual impairment, memory loss, language difficulties, and the capacity to utilize motions as well as cognition, causing major consequences for the patient’s occupational employment and social communication. Because caring for an Alzheimer’s patient is typically difficult and expensive, it has a significant impact on the patients’ relatives. AD worsens with time

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_8, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

231

232

Khac-Minh Thai et al.

and finally becomes a cause of death [1]. The current global population of patients with AD-related dementia is believed to be over 50 million (2018), with a projected increase to 152 million by 2050, also with the expense of global health care for this condition reaching billions of US dollars [2]. Risk factors of AD include age, genetic factors, lifestyle, diet, alcohol, tobacco, cardiovascular diseases, diabetes, depression, and obesity. Furthermore, factors that increase cognitive reserves (education, intelligence, occupation, and social activity) also contribute to a decreased risk for AD [3]. Pathologically, AD is defined as a disease with multiple pathogeneses including (i) extracellular deposition of amyloid beta (Aβ) plaques; (ii) intracellular appearance of neurofibrillary tangles containing over-phosphorylated Tau proteins [1, 4]; (iii) disorders of cholinergic neurotransmission [5]; (iv) consequences of prolonged inflammatory and immune responses [6]; and (v) many other factors [7]. As for the treatment, aducanumab, an anti-Aβ monoclonal antibody licensed by the US Food and Drug Administration (FDA) in June 2021, is the most recently approved therapy for AD. In addition, acetylcholinesterase inhibitors, namely, donepezil, rivastigmine, and galantamine, are by far the most widely used, all of which are reversible inhibitors of acetylcholinesterase (AChE). The therapeutic effects, as well as the undesirable effects of this group of drugs, are almost equivalent, and the choice of which drug to use in treatment is mainly based on the cost, patient’s tolerability, and doctor’s experience [8]. Memantine is an N-methyl-D-aspartate receptor antagonist licensed in 2003 by the FDA for the treatment of moderate to severe AD. Memantine can be used by itself or in combination with acetylcholinesterase inhibitors [9]. In addition to licensed drugs, there have been many drugs being clinically tested for AD treatment in different stages such as (i) anti-amyloid therapies including BACE-1 inhibitors and γ-secretase inhibitors or modulators; (ii) immunization therapies such as vaccines or immunoglobulins; (iii) monoclonal antibodies; (iv) therapies that act on the Tau protein such as Tau phosphorylation inhibitors and microtubule stabilizers; and (v) many others [8]. Currently, there is no real multi-targeting agent for AD that is being tested in clinical trials. Table 1 lists several multi-targeting therapies currently in phase 3 clinical trials for the treatment of AD [10]. With the goal of developing medicines that actually cure AD, a multifactorial disease, it is vital to hunt for compounds that serve as true multi-targeting inhibitors for the disease.

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

233

Table 1 Several multi-targeting therapies are currently in phase 3 clinical trials for the treatment of AD No Agent

Mechanism of action

1

ALZT-OP1 (cromolyn + ibuprofen)

Mast cell stabilizer (cromolyn), anti-inflammatory (ibuprofen)

2

ANAVEX2-73 (blarcamesine)

Sigma-1 receptor agonist, M2 autoreceptor antagonist

3

AVP-786

Sigma-1 receptor agonist; NMDA receptor antagonist

4

AXS-05

Sigma-1 receptor agonist; NMDA receptor antagonist (dextromethorphan); dopamine–norepinephrine reuptake inhibitor (bupropion)

5

Azeliragon

Amyloid, inflammation

6

Losartan and amlodipine and atorvastatin + exercise

Angiotensin II receptor blocker (losartan), calcium channel blocker (amlodipine), cholesterol agent (atorvastatin)

7

CT1812

Sigma-2 receptor antagonist; competes with oligomeric Aβ binding

2

Materials

2.1 Drug Targets in Alzheimer’s Disease

Alzheimer’s disease has been identified as a multifactorial disease. The hallmarks of AD are the existence of extracellular amyloid plaques and intracellular neurofibrillary tangles made up of overphosphorylated Tau proteins, the dysfunction in cholinergic neurotransmission, the disorder in intracellular signaling pathways, and other factors. Currently, many therapeutic targets of AD have been found to be involved in disease initiation and progression, such as Aβ peptides and enzymes involved in its formation including β- and γ-secretase, acetylcholinesterase, and over-phosphorylated Tau protein and enzymes involved in its formation, aggregation, and clearance, such as glycogen synthase kinase-3β (GSK-3β), cyclindependent kinase 5 (Cdk5), casein kinase 1 (CK1), protein kinase A (PKA), protein phosphatase 2A (PP2A), caspase, calpain, cathepsin, etc. In different cellular signaling pathways, the targets can interact with each other to create a disease network. Therefore, the drugs currently used clinically to treat AD are not able to completely cure the disease because they only act on one single therapeutic target of AD. Hence, there is an urgent need to find new therapies that affect multiple therapeutic targets of AD at the same time, in which the design of substances that act simultaneously on multiple targets is a research area that attracts a lot of interest from scientists [11]. Over the past century, many theories have been proposed to explain this multifactorial disease [12]. The amyloid cascade hypothesis has been the dominant theory for the past two decades. This theory indicates that the amyloid precursor protein (APP) is

234

Khac-Minh Thai et al.

normally cleaved by α-secretase and abnormal action of β- and γ-secretase disrupts the balance between Aβ peptide production and clearance [13]. As a result, the Aβ peptides will spontaneously aggregate into soluble oligomers which are then linked together to form insoluble filaments, and eventually, these filaments will be deposited in dispersed senescence plaques [14]. Aβ peptides (e.g., Aβ1–40 and Aβ1–42) are cleaved from APP by three enzymes, α-, β-, and γ-secretase [15]. Both α- and γ-secretase are involved in APP cleavage, whereas β-secretase (BACE-1) may be involved only in Aβ formation. Most genetic mutations promote Aβ formation by cleavage of APP by β- or γ-secretase. In addition, some mutations also promote the self-aggregation of Aβ into amyloid filaments. To modify AD progression, many therapeutic strategies have been developed, including drugs acting on β-amyloid and β- or γ-secretase. The structures of target proteins also play a very important role in the development of new anti-AD drugs as well as in the study of biological disease mechanisms [16]. To date, no X-ray crystal structure of Aβ plaques has been published. Techniques, including X-ray diffraction, solid-phase nuclear magnetic resonance, and electron microscopy, have shown that the internal structure of Aβ fibers is a sequence of β-sheet structures arranged in a structural pattern, cross-β architecture [17]. The Aβ arrays have solved different degrees of structure: (1) the Aβ1–40 and the Aβ1–42 in first-order structure; (2) β-plate; (3) protofibril, the junction between the plates; and (4) threedimensional arrangement of fiber bundles. Structures of aggregates GNNQQNY and NNQQNY were obtained from synthetic peptides. The β chains within each parallel plate of the GNNQQNY aggregate have a conformational extension and are linked together by hydrogen bonds [18]. The side branches of each chain in sideby-side panels can form a network linking with each other. BACE-1 is an aspartic protease that has the highest enzymatic activity at acidic pH [19] and is found mostly in nerve cells. This enzyme exhibits greater activity in the Golgi apparatus, secretory vesicles, and endocytosis [20]. Many BACE-1 crystal structures have been determined, containing or without an inhibitor at the active site. BACE-1 contains a protease active region at the N-terminus, a linkage chain, a transmembrane region, and a cytoplasmic region [21]. The structure includes a folded aspartic protease region, but its site of action is much more extensive than that of pepsin. The substrate-binding region is positioned between the C-terminal and N-terminal lobes, which include the Asp32/ Asp228 catalytic dyad [22]. A ring with a hairpin structure termed a “flap” (amino acids 67–75) covers the binding pocket. This ring (flap) is positioned in the N-terminal lobe and regulates substrate entrance by conformational change. The site of action of BACE-1 is large and consists of several sub-sites called pockets. The S1 and S3 pockets are located close to each other and are composed of

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

235

hydrophobic amino acids. Pockets S2 and S4 are solvent-sensitive and are composed of hydrophilic amino acids. In contact with solvents, there are also S3′ and S4′ pockets. Next to the S4′ pocket is the S2′ one containing mostly hydrophobic and amphoteric amino acids. The catalytic dyad is housed in the S1′ pocket, which is placed in the heart of the active site. The γ-secretase enzyme is a high molecular weight complex made up of four subunits, namely, presenilin (PS), nicastrin (NCT), anterior pharynx-defective 1 (APH-1), and presenilin enhancer 2 (PEN-2) [23]. The presenilin catalytic subunit of γ-secretase is a complex of two forms, PS1 and PS2 [24]. PS1 lysis during γ-secretase complex formation results in a C-terminal fragment (CTF) and an N-terminal fragment (NTF) [25]. PEN-2 is required for the maturation of the complex, and APH-1 is required for the enzyme complex to be stabilized [26]. Finally, NCT was identified as having an important role in APP cohesion [27]. In the co-crystallized structure of γ-secretase and DAPT, this inhibitor is capable of interacting with crucial amino acids that also form a binding site containing the catalytic dyad [28]. The cholinergic hypothesis stresses the presence of cholinergic synapses throughout the human central nervous system, with high concentrations in many regions of the brain. This suggests that cholinergic neurotransmission is important for memory, learning, focus, and other sophisticated brain activities. Thus, the cerebral cholinergic system occupies a central role in the research on AD and other forms of age-related dementia [29]. Acetylcholinesterase (AChE) is a protein that regulates the hydrolysis of acetylcholine (ACh) as well as other neurotransmitter choline esters. AChE is typically present at nerve terminals, where it quickly degrades ACh to yield choline and acetate. Thus, AChE plays an essential role in cholinergic neurotransmission [30]. Acetylcholinesterase inhibitors raise synaptic cleft acetylcholine levels and are one of the few clinically validated medicines beneficial in the treatment of AD. This validates the cholinergic system as a key treatment target for this condition [5]. In many neurodegenerative illnesses, including AD, soluble Tau monomers cluster within the cell and trigger the production of oligomers, which self-assemble to produce aggregates of which Tau has a more complicated shape [31]. The aggregated Tau β-sheet structures produce straight filaments (SFs) and paired helical filaments (PHFs), which then reassemble into intracellular fibrous depositions that induce neurofibrillation [32]. Phosphorylation [33], acetylation [34], glycosylation [35], and cleavage [36] are posttranscriptional alterations that distinguish pathological forms of Tau from the normal form present in the human brain. The phosphorylation sites contain Ser-Pro and Thr-Pro amino acid chain types and are expected to affect Tau conformation and biological activity [37]. The most extensively researched kinases

236

Khac-Minh Thai et al.

involved in Tau phosphorylation include Cdk5, CK1, GSK-3β, and PKA [38]. It should also be noted that the activity of PP2A, the enzyme accountable for the dephosphorylation of more than 70% of the phosphorylation sites of Tau, is dysfunctional in the brains of AD sufferers [39]. Amino acid residues including Thr123, Ser208, Ser400, Ser409, Ser412, and Ser413 are among the Tau O-glycosylation sites that have been experimentally verified [40]. The amino acids Lys274 and Lys280 were implicated as acetylation sites involved in Tau disease [41]. Tau acetylation was also discovered in the microtubule binding region suggesting that this acetylation is critical for Tau aggregation and clearance [42]. C-terminal cleavage has also been linked to AD etiology [36]. The major enzymes involved in Tau cleavage include caspase, calpain, cathepsin, and a thrombin-like protease. 2.2 Computational Modeling Methods in Drug Discovery

Bringing a new drug from research into clinical use is a process that is money-consuming and time-demanding. On average, it takes about 10–15 years to discover/develop a new drug with a cost of about 800 million USD [43, 44]. And it is not surprising that pharmaceutical companies have always focused on how to shorten the time and budget to develop drugs without compromising the quality of medicine. In the 1990s, many drug development projects were carried out using large-scale screening and synthesis techniques that accelerated the drug discovery process [44, 45]. These methods have been widely accepted because they make it possible to rapidly perform chemical synthesis and screen very large libraries of substances. Unfortunately, however, no significant success has been achieved, and very few projects have come forward to bring new drugs to market during this time [46, 47]. The combination of modern computing technologies, biological sciences, and chemical synthesis has been used to facilitate drug discovery, and it is computational modeling methods that enhance the scale in drug discovery. Thus, the term computer-aided drug design (CADD) was coined and applied to the use of computers and assistive programs in the process of drug discovery [47, 48]. Modern computer software programs have proven to be effective tools and many projects have had remarkable success with these methods. CADD is a distinct field in which different computational approaches are used to model interactions between receptors and pharmacological compounds in order to assess binding affinity [49]. However, this approach is not restricted to chemical interaction investigations and binding prediction; it has a wide range of applications, from developing compounds with desired physicochemical qualities to managing digital chemical compound archives. CADD is divided into three types: ligand-based, structure-based, and de novo drug designs. Figure 1 illustrates some of the different computational methods used in CADD.

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

237

Fig. 1 Some methods used in rational drug design

Virtual screening (VS) is a computational approach used in in silico research to screen vast databases of chemical compounds and has been effectively utilized with high-throughput screening (HTS) in drug discovery [50–52]. The main purpose of virtual screening is to enable the quick and low-cost examination of massive virtual databases of chemical substances to search for biologically active leads that set the stage for further studies of chemical synthesis and subsequent evaluations [53]. Virtual screening uses a number of computer approaches to screen massive libraries of chemicals for those that are likely to bind to a desired molecular target [54, 55]. Furthermore, because VS may employ enormous libraries of already synthesized compounds, it helps to lessen the difficulty of drug production. 2.2.1 Structure-Based Drug Design

Structure-based drug design (SBDD) uses structural data in three dimensions (3D) of protein molecules to design potential pharmacological active compounds [56]. Therefore, identifying a target molecule that is active and its structure is a fundamental first step of SBDD [54, 57]. The discovered target might be an enzyme linked to a condition of interest. Potential drugs would be molecules that

238

Khac-Minh Thai et al.

inhibit the target and so lower its activity based on the binding affinity findings. SBDD searches for prospective therapeutic medications using information about a biological target. SBDD has advanced computational algorithms used in biophysics, pharmacology, statistics, biochemistry, and other domains significantly [58]. Scientific research has resulted in a plethora of strategies for predicting the structure of proteins. Many proteins’ structures may now be determined using nuclear magnetic resonance (NMR) and cryo-electron microscopy (cryo-EM) approaches, as well as X-ray crystallography and computational techniques such as homology modeling and molecular dynamics simulations [59]. Homology Modeling

Structure determination of a therapeutic target is the first step before searching for a specific drug molecule on that target [60]. Although many advanced techniques are currently being applied, a huge number of protein structures have not been established. In this scenario, homology modeling has shown to be effective since it may be used to design protein structures based on information about comparable proteins [29]. Structural data of an effector target has been identified as a prerequisite in SBDD; however, the structure of several therapeutic targets of medications used to treat neurodegenerative illnesses is unknown [61, 62]. A wide range of studies has been conducted utilizing homology modeling to construct structures of a chosen target molecule. In order to have a better knowledge of protein action, structural information is also required. Dhanavade et al. (2013) [63] have solved the structure of cysteine protease, which is an enzyme capable of breaking down peptide β-amyloid—an important pathogen in Alzheimer’s disease. Several in silico studies employing structural simulations of cysteine proteases have been conducted to study the characterization of this enzyme’s binding site. Many other therapeutic targets of AD have also been modeled homologously, from which further in silico studies have been performed [64, 65]. The steps for building a homology model are outlined in Fig. 2. Template Selection The first stage in a homology modeling methodology is to identify homologous sequences in the protein structure database and utilize them as templates for modeling. It typically depends on serial pairwise sequence alignments supported by database search techniques like BLAST and FASTA, but it may also use PSI-BLAST, protein threading, and other approaches. Target and Template Sequence Alignment To get optimal alignment, the full-length sequences of the template and target proteins must be realigned using enhanced alignment algorithms once the structure with the highest sequence similarity

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

239

Fig. 2 Steps in the homology model-building process

is determined as a template. For this aim, the finest multiple alignment algorithms available, such as Praline and T-Coffee, should be utilized, followed by manual adjustment of the alignment to increase alignment quality. Target Protein Backbone Structure Construction Once optimum alignment is obtained, the coordinates of the template proteins’ corresponding residues may be easily duplicated onto the target protein. If the coordinates of the two aligned residues are equal, the coordinates of the side chain atoms are copied together with the coordinates of the main chain atoms. Only the backbone atoms can be replicated if the two residues vary. Side Chain Atom and Loop Addition and Optimization There are frequent sections in the sequence alignment for modeling that are generated by insertions and deletions, resulting in gaps in the sequence alignment. The gaps cannot be represented directly, resulting in “holes” in the model. Closing the gaps necessitates the use of loop modeling, which is a tough challenge. After the main chain atoms are constructed, the locations of unmodeled side chains must be established. It is critical to model side chain geometry when analyzing protein–ligand interactions at active sites and protein–protein interactions at the contact interface. The side chain refinement function is included in the majority of modeling programs.

240

Khac-Minh Thai et al.

Model Refinement and Optimization Structure abnormalities such as unfavorable bond angles, bond lengths, or tight atomic contacts are removed from the complete raw homology model. If structural anomalies are discovered, the entire model can be repaired using the energy reduction process. Model Validation To ensure that the structural elements of the model are compatible with the physicochemical rules, the final homology model must be validated. Alignment and model construction is repeated as needed until the desired result is attained. When there is no obvious homology, fold recognition and threading algorithms may be employed to assign tertiary structures to protein sequences. They are relatively quick and affordable methods of constructing a close approximation of a structure from a sequence without the time and expense of experimental processes. The continued development of such approaches has had a huge influence on structural biology, giving us a greater capacity to properly simulate 3D protein structures utilizing highly evolutionarily distant fold templates. The goal of fold recognition and threading approaches is to assign folds to target sequences that have very little sequence similarity to known structures. Finally, if structural templates are unavailable, ab initio procedures are applied. Ab initio structure prediction is the process of predicting a protein’s structure using just its amino acid sequence (ab initio means “from the beginning” in Latin). Although many alternative techniques for ab initio protein construction have been developed over the years, they all solve a similar challenge. Several applications and servers, both commercial and free, are available for protein structure prediction, with the goal of creating a full model using query sequences. The list of these software and servers is presented in Table 2. Molecular Docking

Molecular docking is a modeling method widely used for predicting the binding behavior and affinity of small molecules to their target (usually proteins) in a short amount of time [84, 85]. This computational approach has become tremendously useful in the field of drug discovery [52, 86, 87]. Molecular docking has grown in popularity over the last two decades and is now widely used in CADD and structural biology, where it has proved to be more productive than traditional drug discovery approaches. Because of the fast expansion of computational power and the huge database of small molecules and proteins, molecular docking is likewise getting more powerful. Recent advances in computational approaches and accessibility to 3D structural data of biological targets have boosted the usefulness of this technology for large-scale applications in protein–ligand binding interaction investigations. Docking may be divided into three types: rigid docking (both the ligand and the target are rigid

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

241

Table 2 Useful servers and programs for protein structure prediction No

Name

Program/server

Commercial

Ref

Homology modeling 1

SWISS-MODEL

Server

[66]

2

WHATIF4

Server

[67]

3

SCWRL

Server

[68]

4

PROCHECK

Server

[69]

5

BLAST

Server

[70]

6

Clustal Omega

Server

[71]

7

MODELLER

Program

[72]

8

ICM Homology

Program

X

[73]

9

Schro¨dinger Prime

Program

X

[74]

10

MOE

Program

X

[75]

11

GOLD

Program

X

[76]

Fold recognition and threading 12

ORION

Server

[77]

13

PHYRE2

Server

[78]

14

PSIPRED

Server

[79]

15

IntFOLD

Server

[80]

16

I-TASSER

Server

[81]

Ab initio prediction 17

QUARK

Server

[82]

18

Rosetta@home

Program

[83]

structures), flexible docking (both the ligand and the target are flexible molecules), and semiflexible docking (ligand is considered flexible, and the target is rigid molecule) [88]. Many computer programs for molecular docking have been developed in recent times, including AutoDock [89], DOCK [90], Glide [85], FlexX [91], LigandFit [92], GOLD [93], Surflex [94], and ICM [95], and have been utilized successfully in several computer-based drug discovery studies. Table 3 lists the popular molecular docking tools used in practice. Fundamentally, the primary purpose of molecular docking is to find the best binding ligands in the protein’s effector region and to determine which binding position is most energetically favorable. A binding pose is a word for an empirically validated

242

Khac-Minh Thai et al.

Table 3 Some popular molecular docking programs No

Program

1

AutoDock

2

GOLD

3

Commercial

Searching algorithm

Ref

Genetic algorithms/Monte Carlo

[89]

X

Genetic algorithms

[93]

Glide

X

Monte Carlo

[85]

4

FlexX

X

Incremental construction algorithm

[91]

5

Dock

Shape fitting

[90]

6

LigandFit

Monte Carlo

[92]

7

FRED

Shape fitting

[96]

8

ICM

X

Monte Carlo

[95]

9

eHiTS

X

Incremental construction algorithm

[97]

10

Surflex-Dock

X

Incremental construction algorithm

[94]

X

conformation of the ligand within the binding site of the target protein or a theoretical conformation calculated by a computer. Two critical components for determining protein–ligand interactions are search algorithms and scoring functions [98]. The searching algorithm is responsible for finding different poses and conformations of ligands within a given target protein, and the scoring function is responsible for estimating the binding affinity of generated poses, ranking them, and determining the most favorable binding patterns between ligands and proteins [98, 99]. An ideal searching algorithm would make the search fast and efficient, and an ideal scoring function should be able to determine the physicochemical properties of the molecule and thermodynamic interactions. Various studies are being conducted to identify how the ligands bind and to pick the most energetically favorable positions. To do this, molecular docking techniques are used to produce a series of alternative binding poses for the ligand, and a scoring function is used to quantify the binding affinity of these poses to identify the optimal one. The energy shift generated by the creation of the ligand/protein complex is given as Gibbs free energy (ΔG) and binding constant (Kd) [100, 101]. A complex’s binding energy is predicted by analyzing physicochemical factors associated with protein–ligand interactions such as desolvation, intermolecular interactions, and entropy effects [102].

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

243

Fig. 3 The molecular dynamics simulation process Molecular Dynamics Simulation

Protein mobility is vital in ligand binding. Expensive and complicated studies have prompted a quest for computer modeling approaches that can anticipate protein mobility. However, even with the finest supercomputers, the computations necessary to simulate the quantum mechanical motion and chemical reactions of huge molecular systems are too complicated, and simulations are time-consuming. To circumvent these restrictions, molecular dynamics (MD) simulations were created in the late 1970s. MD can lessen the strain on the computer by simulating atomic motion using basic approximations based on Newtonian physics [103]. Figure 3 depicts the general procedure of dynamics simulation. First, the NMR structure, X-ray crystal, or homologous model is used to create a topological model of the molecular system. Based on the interactions between bound and unbound atoms, the forces exerted on each atom of the system will be estimated. The unbound forces are constituted by van der Waals interactions, which are simulated using Lennard–Jones potentials, and electrostatic interactions simulated using Coulomb’s law. The bonding force is based on three factors: the length of the chemical bond, the bond angle, and the rotation of the bond. Chemical bonds and bond angles are simulated as simple springs, and the dihedral angle (or bond rotation) is simulated using the sine function to approximate the energy difference between the obscured structures and staggered shapes. The abovementioned energy factors will be tuned to match quantum mechanical calculations and experimental data in order to imitate the real molecule in motion as precisely as feasible. This parameterization includes determining the stiffness and length of the chemical bond simulation spring, as well as the bond angle; calculating the atomic charge used to approximate the electrostatic interaction energy; correctly

244

Khac-Minh Thai et al.

determining the van der Waals interaction radius; and many other factors. These factors are referred to together as force fields. AMBER, CHARMM, and GROMOS are three popular simulation force fields. The difference between the force fields is in how they are parameterized, yet the outcomes are frequently comparable when applied [103]. After estimating the force exerted on each atom in the system, the atoms will be moved using Newton’s equations of motion. The simulation time is frequently within 1–2 femtoseconds. Then, the process of estimating the molecular energy acting on each atom is repeated, and the simulation continues until the end of the set time [103]. Many researches have been conducted to validate this simulation approach by comparing the findings of dynamics simulations with experimental data. Many ligand–receptor binding conformations predicted by molecular dynamics may be utilized to predict NMR features such as spin recovery, making NMR data helpful in this comparative investigation. Indeed, these studies have shown a high degree of consistency between in-machine and experimental macromolecule simulations [103]. To perform a dynamic simulation, the following elements are required: • Initial conditions include the composition of the system, the coordinates of the atoms, and the initial velocity. • Topological geometry of molecules and formulas for potential energy describe interactions between atoms. • Conditions in the simulation environment: algorithms used to measure different types of interactions, motion limitations of groups of atoms, external forces, time of description, time of each step, and the frequency at which the data will be saved. • Computer to perform simulation and device to store data and results. • Analyzed results. The results of molecular dynamics simulation are evaluated through the values of RMSD (root-mean-square deviation), RMSF (root-mean-square fluctuation), percent hydrogen bond (percentage occupancy of hydrogen bond), and potential energy. • RMSD in molecular dynamics simulation is a measure of the average distance between atoms (usually the atoms in the main chain) of a protein structure. RMSD is also computationally applied to nonprotein molecules such as small organic molecules. RMSD plots are often analyzed in protein dynamics simulations. Typically, these graphs are very steep during the first nanoseconds and oscillate around a mean point for the rest of the dynamic simulation. The RMSD values fluctuate widely

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

245

around the mean which means that the simulated complex has undergone conformational changes. The increasing RMSD indicates that there are changes in the structure of the simulated substance. The RMSD value is calculated according to Eq. 1. RMSDðt 1 ,t 2 Þ =

1 N

N i=1

½x i ðt 1 Þ - x i ðt 2 Þ2

ð1Þ

where N is the atomic number and xi(t) is the position of atom i at time t [104]. • The RMSF provides information about the fluctuations of each amino acid during the simulation period. A high RMSF value indicates high mobility of amino acids. Thus, each atom will have an RMSF. The RMSF value is calculated according to Eq. 2. RMSFv =

1 T

T i=1

ðvi - vÞ2

ð2Þ

where T is the number of frames, v is the average coordinate of the atom, and vi is the coordinate of that atom at each frame [105]. • Percentage of hydrogen bonding is defined as the ratio between the time in which the hydrogen bond is present and the total time considered [106]. • Potential Energy. The binding ability between proteins and ligands is not only due to binding forces but also to nonbinding forces, including van der Waals interactions and electrostatic interactions. Thus, the protein–ligand interaction is divided into two parts: (1) electrostatics modeled by the Poisson–Boltzmann function and (2) van der Waals interaction, represented by the Lennard–Jones potential [107]. 2.2.2 Ligand-Based Drug Design

Ligand-based drug design (LBDD) is a drug design method aimed at elucidating the relationship between the structural and physicochemical properties of a compound/ligand molecule and its biological activity. This approach is useful when there is no knowledge about the target protein’s 3D structure. In this method, accessible information on ligands and their biological activity is leveraged to produce prospective novel therapeutic candidates. LBDD is commonly employed in pharmaceutical research, with more than half of the targeted medications working as membrane proteins (of which their 3D structure is frequently lacking, such as G-protein coupled receptors (GPCRs)) being licensed. LBDD is a strategy based on the premise that compounds with similar structural features share biological activity and interactions/inhibitions with common therapeutic target molecules [87, 108].

246

Khac-Minh Thai et al.

Table 4 Some common types of molecular descriptors used in ligand-based drug design Type of descriptor Based on

Example

Theoretical descriptor 0D

Molecular formula

Molecular mass, atomic number, bond number

1D

Structural formula

Number of molecular fragments, number of functional groups

2D

Topological structure

Balaban index, BCUTS, Randic index, Weiner index

3D

Geometric structure 3D-MORSE, autocorrelation GETAWAY, WHIM

4D

Chemical conformation

GRID, Raptor, Volsurf

Experimental descriptor Hydrophobic constant

Hydrophobicity

Distribution coefficient (logP), hydrophobic substituent constant (π)

Electronic constant Electronic properties

Hammett’s constant, acid dissociation constant

Stereoscopic constant

Charton’s constant, Taft’s constant

Stereoscopic properties

The LBDD technique is based on the entire information describing the ligand molecule. Molecular descriptors are numerical numbers that are used to express a molecule’s structural and physical features. The area of molecular descriptors is multidisciplinary and involves several hypotheses. Active compounds are defined by zero- to four-dimensional molecular descriptors [109]. 0D descriptors are the counting and covariance parameters. SMILES and SLNs are 1D parameters that are chemical fingerprints or lists of molecular fragments. 2D descriptors are graph constants or invariants in which atoms are represented by points and bonds are represented by edges. 3D descriptors include geometric quantities, WHIM, and others. Finally, 4D molecular descriptors are values derived from comparative molecular field analysis (CoMFA) or DRID computations [110]. Table 4 summarizes some common types of molecular descriptors used in LBDD [111] LBDD relies heavily on similarity searches. This method finds comparable chemicals using a known active ingredient as a query and then ranks the found compounds in a database. This assumption implies that structurally identical compounds will have comparable biological activity and physical qualities. To assess the degree of equivalence, numerical descriptors are used, and the similarity factor is calculated. In similarity searches, the measure of similarity based on fingerprints or 2D similarity is often utilized. Many

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

247

coefficients are used in similarity searches against various molecular fingerprint databases (Cosine, Forbes, Euclidean distance, Tanimoto coefficient, etc.). Despite the well-documented bias, the Tanimoto coefficient is the most frequent and commonly recognized similarity metric for binary variables. This coefficient can be expressed by Eq. 3. Tc=

OV ab OV ab - OV aa - OV bb

ð3Þ

where Tc is the Tanimoto coefficient and OVab, OVaa, and OVbb are the overlapping volumes of the molecule a with molecule b, molecule a, and molecule b, respectively. The value of Tc ranges from 0 (not similar) to 1 (completely similar). LBDD methods are generally classified into pharmacophore modeling and quantitative structure–activity relationship (QSAR) studies. Pharmacophore Modeling

A pharmacophore is a 3D combination of the “stereoscopic” and “electronic” features necessary for effective macromolecular interaction with a certain biological target conformation in order to promote/inhibit this biological target’s biological response [112]. The technique of developing ligand-based pharmacophore models utilizing knowledge of compounds’/ligands’ biological activity. A pharmacophore is a hypothetical depiction of the fundamental molecular features of interactions between ligands and macromolecular molecules rather than a molecule/ligand or real links between functional groups. Pharmacophore modeling is commonly used to find possible lead compounds quickly. Many prospective and widely acknowledged therapeutic targets have been found in the recent period of drug creation; nevertheless, the geometry of their active sites is unclear. The study of pharmacophore models allows for the efficient and rapid screening of large databases of chemicals. Common pharmacophores are often constructed by matching conformational models and active chemicals in 3D. The method used to calculate the degree of overlap will stack the compounds in the training set in 3D space based on their chemical properties’ comparable position/arrangement. The pharmacophore features are defined in such a way that all/most substances have a similar chemical property. Information regarding physiologically inactive chemicals can be integrated during the model-building process to refine this generic pharmacophore score. Many computer software and tools for pharmacophore modeling have been created, including Catalyst/Discovery Studio, LigandScout, MOE, and Phase [113].

248

Khac-Minh Thai et al.

Quantitative Structure– Activity Relationship

In the process of finding new drugs, the number of compounds that need to be synthesized and tested for biological activity is extremely large, so the search for new active ingredients requires a great deal of human resources and costs. Along with the development of informatics, the systematization of structural information of molecules and data analysis has introduced a new method called QSAR to predict the activity of substances. QSAR research methods are based on the assumption that molecular structure is directly related to biological activity and that changes in structure will alter biological activity. QSAR is defined as a process that involves building computational and mathematical models using computational chemistry (chemometric) technologies to determine meaningful correlations between structures and functions [114]. For QSAR, the main theory is that “compounds that are similar in structure or physicochemical properties should exhibit similar biological activity.” Since the structural features of compounds are often expressed as molecular descriptors, QSARs are also defined as mathematical relationships between molecular descriptors and the biological activity of known substances tested in vitro. And so, the result obtained will be in the form of a mathematical model used to predict the biological activity value of compounds with similar structures to known substances. A library of lead compounds with the requisite biological activity is created in order to find possible lead compounds. A model was then created to predict the quantitative link between these chemicals’ structural and physicochemical attributes and their biological activity. These correlations will be utilized to develop statistical models, which will subsequently be used to statistically optimize the biological features of the substances and maximize the related biological activities. QSAR is commonly utilized in drug discovery to improve the activity of current medications and is used to change existing chemicals and improve their activity.

2.2.3 De Novo Drug Design

This is a molecular fragment-based drug design method. De novo design begins with small building blocks (molecular fragments), initial building blocks with desired characteristics that are brought together either through direct bonding or with each other through an intermediate group (linker). This process can be repeated many times until one (or more) molecule(s) with the desired characteristics is (are) obtained. De novo design also includes structurebased and ligand-based methods. The structure-based de novo design searches for new ligand molecules using the target protein’s 3D information, and the new ligand is built directly inside the target protein’s binding site and evaluated by calculating the interaction energy between the target protein and that ligand. For ligand-based de novo design, information about the target protein is often not known, so new ligand molecules will be proposed based on the homologous structures of previously known molecules [115].

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

3

249

Method

3.1 Studies on Multitargeting Inhibitors for Alzheimer’s Disease

To date, most therapy for AD has focused on the use of acetylcholinesterase inhibitors (AChEIs) such as donepezil, rivastigmine, and galantamine to enhance cholinergic transmission. These agents can only improve symptoms in the majority of patients but do not treat the etiology of AD [116]. Therefore, the search for a more effective therapeutic strategy to combat this disease is urgently needed. Because of the complex multifactorial nature of AD, molecules that modulate the function of only one target protein are unlikely to significantly alter disease progression. In contrast, molecules showing multiple biological effects that complement each other would represent an important improvement in the treatment of this disease [117–119]. Currently, single targeted therapy mainly focuses on acetylcholinesterase inhibition. In the design of multi-targeted drug molecules, acetylcholinesterase has been extensively studied in combination with other therapeutic targets. The first dual-acting inhibitor for AChE is based on the theory that the interaction of Aβ with the peripheral anionic site (PAS) of AChE will promote some conformational changes of Aβ leading to the formation of a plate structure and cause aggregation to take place [120]. Therefore, donepezil-derived inhibitors have been designed to interact with the catalytic site and the PAS region of AChE. This idea has opened up a new era of research in which inhibitors of known therapeutic targets are combined with different chemical groups to form a heterodimer or homodimer. Bis(7)-tacrine is the first homodimer reported to inhibit AChE and BChE simultaneously. Years later, it was also determined that bis(7)-tacrine also interacts with the PAS and Aβ regions, as well as inhibiting BACE-1 [121]. Molecular modeling and docking studies were used to design the AP2238 inhibitor that binds to both the catalytic site and the PAS region of human AChE. At the same time, this substance also inhibits Aβ aggregation [122]. AP2238 inhibitor contains a benzylamino group and a coumarin moiety. In addition to the simultaneous inhibition of AChE and Aβ aggregation, many scientists have designed lipocrin-like molecules that add the ability to control oxidative stress (an important aspect of AD) [123]. Lipocrin was the first to inhibit AChE and Aβ aggregation induced by AChE and protect against the effects of reactive oxygen species (ROS) [123]. To get an effect against ROS, two structural components, tacrine and melatonin, were combined. This heterodimer molecule is reported to have antioxidant effects along with the inhibitory activity of AChE and BChE [124]. Their various derivatives were further evaluated for their potential against Aβ toxicity. In one study, a lipocrin- and memoquin-based hybrid compound was generated by combining a benzoquinone fragment

250

Khac-Minh Thai et al.

with a lipoyl functional group that was selective with potent inhibitory activity against AChE and decreased Aβ self-aggregation as well as decreased production of ROS [124]. For the effect on oxidative stress, researchers have attempted to create substances that inhibit the enzyme monoamine oxidase (MAO), which is the enzyme that releases ROS in the deamination of neurotransmitters. MAO inhibitors reduce ROS and oxidative stress in patients with AD. Therefore, several structural groups such as ladostigil [124] and propargylamine (ASS234) [125] have been reported to inhibit MAO and AChE simultaneously. Ladostigil is currently in phase 2 clinical trials. A small number of substances have also been reported to have calcium channel inhibitory activity along with AChE based on previous studies suggesting that disturbances in Ca2+ homeostasis are involved in AD pathogenesis [126]. Tacripyrine is a simultaneous inhibitor of Ca2+ channels, AChE, BChE, and Aβ aggregation; in addition, it also crosses the blood–brain barrier [127]. In addition, several other recent studies have also been carried out. Li, Y. et al. (2014) in a study conducted the design, synthesis, and bioactivity testing of a series of 2-methoxy-phenyl dimethyl-carbamate derivatives with inhibitory activities against AChE, BChE, and amyloid aggregation; free radical neutralization; and chelation with metal ions [128]. In another study, Scipioni et al. (2019) synthesized new vanillin derivatives with antioxidant, anti-AChE, inhibitory of amyloid aggregation, and neuroblast protection effects [129]. Umar et al. (2018) in his study also synthesized derivatives of N-(4-((7-chloroquinolin-4-yl)oxy)-3ethoxybenzyl)amine that inhibit amyloid aggregation, with antioxidant properties, and can chelate with metal ions [130]. BACE-1 is another important therapeutic target in the AD drug discovery process, but there are very few candidate substances in clinical trials. The discovery of BACE-1 inhibitors is fraught with challenges due to the enzyme’s large site of action and its very versatile nature [22]. Very few BACE-1 inhibitors have been reported to have additional effects on AChE, Aβ, α-secretase, and GSK-3β. Bis(7)-tacrine was found to be an α-secretase activator and BACE-1 inhibitor [131]. Both BACE-1 and AChE enzymes are involved in Aβ formation and aggregation, so designing dual-acting inhibitors of these two enzymes is a very interesting approach. However, to date only very few constructs have been reported to have inhibitory activity on both of these enzymes. The first dualacting inhibitor was designed by combining the AP2238 molecule (AChE inhibitor) with a dihalophenyl acid group (BACE-1 inhibitor). This substance has been suggested to be able to bind to the S1 and S1’ pockets of BACE-1 through molecular docking modeling studies [132]. Likewise, fragment-based drug design strategies have been used to design a simultaneous inhibitor of BACE-1 and AChE by importing the isophthalamide group on the GRL-8234 (BACE-1 inhibitor) into donepezil to form a new molecule

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

251

[133]. Multiple quinoxaline-based ligands [134] and coumarin derivatives [132] have also been reported to inhibit AChE and BACE-1 simultaneously. In addition, novel memoquin derivatives have also been designed to simultaneously inhibit AChE, BACE-1, and Aβ self-aggregation, as well as act on oxidative stress [135]. An association between BACE-1 inhibition and metal ion chelation has also been reported. In this study, a database of 1,3-diphenylurea derivatives was constructed by combining the structure of LR-90 with BACE-1 inhibitors and screening using pharmacophore models on BACE-1 [136]. Another important pathophysiological feature of AD is the formation of neurofibrillary tangles (NFTs) induced by hyperphosphorylation of the protein Tau, which plays an important role in several protein kinases including GSK-3β. GSK-3β is another wellstudied target for AD as well as cancer. Recently, the first inhibitor with dual action on BACE-1 and GSK-3β was reported [137]. In this study, scientists synthesized and performed a SAR survey on triazone derivatives (with either the guanidino group or cyclic amide group attached). The obtained substances (34 substances) showed the ability to inhibit both BACE-1 and GSK-3β. Murata K. et al. (2015) in a screening study for the AChE and BACE-1 inhibitory effects of naturally occurring substances identified curcumin derivatives in turmeric extract including curcumin, demethoxycurcumin, and bisdemethoxycurcumin as the active substances with strong inhibitory effects simultaneously on AChE and BACE-1 [138]. In addition, the curcumin structural framework has also been reported to inhibit BACE-1 and GSK-3β simultaneously [139]. Yan J. et al. (2017), in their study, designed, synthesized, and tested the bioactivity of hybrid compounds between donepezil and curcumin acting on AChE, inhibiting amyloid aggregation and anti-oxidation [140]. In recent years, many research groups have carried out studies on groups of flavonoid derivatives with effects on many therapeutic targets of AD including AChE, BChE, inhibition of Aβ aggregation, antioxidant chemotactic, neuroprotective, BACE-1, metal ion chelation, and MAO. Many studies have been performed on AChE in combination with other therapeutic targets [141–143]. Several other research groups have also conducted studies on BACE-1 in combination with other therapeutic targets [144]. Some other research groups have performed combination studies between other therapeutic targets [145, 146]. Several studies on flavonoids have led to the inclusion of derivatives in clinical trials. Kim H. et al. (2005) developed flavonoids that inhibit Aβ aggregation and have no toxicity, one of which has been in phase 1 clinical trials [147]. Porat Y. et al. (2006) have developed flavonoids that have effects on multiple therapeutic targets of AD including inhibition of Aβ aggregation, anti-oxidation, inhibition of BACE-1, and chelation with metal ions. Two of these derivatives have been in phase 2 and phase 3 clinical trials [148].

252

Khac-Minh Thai et al.

Many studies have been performed to discover flavonoid derivatives with dual effects on AChE and BACE-1. Xu QX. et al. (2018) in one of the studies isolated four derivatives bavachin, bavachinin, bavachalcone, and isobavachalcone from Psoralea fructus species that have effects on many therapeutic targets of AD, notably AChE and BACE-1 [143]. In another study, Parasuraman P. et al. (2015) used an in silico method to design condensates between flavones and thiazole rings, including one that was predicted to have a very potent inhibitory effect on AChE and BACE1 (predicted of IC50 100 nM and 2.1 μM, respectively). This study guides the chemical synthesis and biological activity test in vitro as well as further studies in vivo to find new drugs in the treatment of AD [149]. In addition, Wang SN et al. (2016) have identified xanthone derivatives that have effects on multiple therapeutic targets of AD, in which garcinone D derivative exhibits a simultaneous inhibitory effect on AChE and BACE-1 with inhibition percentage of 12.9% and 62.7%, respectively [150]. Ferna´ndez-Bachiller M.I. (2012) in a study using a fragment-based drug design method designed a group of hybrid tacrine-4-oxo-4H-chromene derivatives. The tacrine molecule was selected for its inhibitory effect on acetylcholinesterase; the flavonoid 4-oxo-4H-chromene framework was selected for its BACE-1 inhibitory effect. Of the derivatives obtained, 6-hydroxy-4-oxo-N-{10-[(1,2,3,4-tetrahydroacridine9-yl)amino]decyl}-4H-chromene-2-carboxamide showed strong inhibition of AChE and BACE-1 with IC50 values of 75 nM and 2.8 μM, respectively [151]. More recent studies using multiple chemoinformatics servers and tools to identify new multi-targeting anti-AD compounds have been conducted. Kumar et al. (2020) [152] developed and validated 2D-QSAR models for AChE and BuChE dual inhibitors utilizing multiple data sets and internal and external validation metrics. Furthermore, molecular docking was used in this work to discover the binding pattern between ligand and target enzyme by employing the most and least active molecules from the data sets. In another study, Stern et al. (2022) [153] utilized their ISE technique to model each of the AChE peripheral site inhibitors and BACE-1 inhibitors using available data and built classification models for each. Both models were then used to search through vast molecular databases. The top-scoring compounds were docked into the crystal structures of AChE and BACE-1, and the 36 molecules with the highest weighted scores (based on ISE indices and docking findings) were sent for inhibition experiments on the 2 enzymes. Two of them inhibited both AChE (IC50 ranging from 4 to 7 μM) and BACE-1 (IC50 ranging from 50 to 65 μM). Two more compounds inhibited solely AChE, while two more inhibited only BACE-1. In preliminary testing on APPswe/ PS1dE9 transgenic mice, F681-0222 presented a decrease in soluble Aβ42 in brain tissue. In a work conducted by Khan et al. (2023)

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

253

[154] which included a combination of in silico and in vitro studies, the authors have discovered a multi-targeting compound potentially for use in the treatment of AD. This substance showed promising inhibitory activities against four enzymes, i.e., AChE, BChE, MAO-A, and MAO-B. The in silico methods used in this work were density functional theory (DFT) studies, molecular docking, and molecular dynamic simulations. A class of new fused tricyclic compounds based on a pyrazino[1,2-a]indol-1(2H)-one ring scaffold with a C3 phenyl substituent were designed, synthesized, and tested for the bioactivities against various AD targets indicating potential inhibitory effects on amyloid aggregation and cholinesterase enzymes (AChE and BuChE) in combination with antioxidant activity. In silico investigations revealed that the fused tricyclic pyrazino[1,2-a]indol-1(2H)-one scaffold should bind to Aβ, as well as cholinesterase enzymes, and that a C3 3,4-diOMephenyl is a new Aβ-binding pharmacophore that can be employed in the creation of anti-AD medicines [155]. Machine learning models using linear method, genetic function approximation (GFA) and nonlinear approaches, support vector machine (SVM), and artificial neural network (ANN) were developed by Dhamodharan et al. (2022) to predict the activity of AChE and BACE-1 dual inhibitors [156]. Several classes of molecular descriptors utilized in the construction of these models were found to be critical in defining the inhibitory activity of AChE and BACE-1 enzymes and should be exploited further in the rational design of multi-targeting anti-AD lead compounds. In search of multitargeting inhibitors for AD, drug repurposing using in silico methods is a usable approach. Shrivastava et al. (2022) [157] developed various computational models to screen a library of 4199 FDA-approved drugs on 2 AD targets, AChE and BACE-1. After structural modifications, chemosynthesis, and bioactivity assays, compound 27 showed promising inhibitory activity on both AChE and BACE-1 (with IC50 at submicromolar levels). In vivo behavioral investigations of this compound revealed a considerable reduction in cognitive dysfunctions compared to scopolamineinduced amnesia animal models. 3.2 Computational Modeling Approaches to Design Multitargeting Inhibitors

Computational modeling techniques have been used effectively to screen, design, and generate leads against specific AD targets [158]. To discover leads, molecular docking is employed, and molecular dynamics simulation is utilized to comprehend the amyloid-destabilizing mechanism [159–161]. Virtual screening, pharmacophore mapping, and molecular dynamic simulations have also been effectively used to discover natural inhibitors of AD drug targets from chemical databases [162, 163]. These computational methods were coupled to find multi-targeting inhibitors of AD. Due to the complicated and multifaceted character of AD, the same strategies have been proposed to develop better

254

Khac-Minh Thai et al.

medications for many established therapeutic targets [164]. In silico methods have been widely used to design selective singletarget inhibitors in the drug discovery process [162], and other computational methods [165] combined with the docking, molecular dynamics, and free binding energy calculation are providing very promising solutions for identifying new leads with multiple activities and binding mechanism. 3.2.1 Docking-Based Methods Preparation of Compounds and Proteins

To perform molecular docking successfully, it is necessary to have the appropriate protein binding sites and protonated ligands, as well as create the appropriate tautomeric forms of the ligand. If this step is skipped, the docking procedure may fail due to steric or electromagnetic problems. When 3D crystallographic structures of the therapeutic targets are known, and 3D structural information is required for both therapeutic targets in the case of dual inhibitor design, molecular docking-based approaches are beneficial. An active site should be properly defined after downloading the 3D structure. If the target’s 3D X-ray structure included an inhibitor or substrate linked to the active site, molecular docking would be more reliable. The binding site of the bound ligand or substrate should be used for molecular docking. If the target was newly identified and the active site is unknown, blind docking should be done, in which a grid is established to cover the complete protein and then molecular docking is performed to identify probable locations of binding for ligands [166]. If the protein is big enough or the specific binding site is unknown, the grid should be applied to the entire protein. In this scenario, two approaches should be taken: (ii) First, a huge grid spacing of 1 Å (in AutoDock) may be utilized, allowing the full target to fit into the map space. However, this may produce accuracy issues. (ii) Second, a regular small-size grid should be arranged overlappingly to cover the complete protein surface, and then molecular docking would be done to determine the location where the ligand binds securely in comparison to the other sites. In this instance, various active site prediction tools and/or a web server should be of assistance. Furthermore, because most protein active site residues are conserved within the same protein class or functionally related proteins, sequence alignment to identify conserved residues inside the cavity or putative binding site should be performed. Following active site validation, molecular docking is used to separate binders from nonbinders. Top hits are then docked with a robust technique into the active site of both targets in order to evaluate the binding energy or binding affinity in terms of docking scores provided by docking software. Each docking program has a unique set of algorithms and scoring criteria. As a result, validating the procedure (algorithm and scoring function) for the provided list of proteins/drug targets is a recommended practice [167].

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

255

Molecular Docking Evaluation

Various methods of molecular docking evaluation and scoring functions have been reported [168]. The most commonly used method is the “pose selection” method, in which a docking program is used to re-dock to the active site a molecule with a previously known conformation and orientation, usually from a co-crystallization structure. Docking programs capable of reproducing poses with an RMSD value of no more than 1.5 Å (or 2.0 Å depending on ligand size) from a known configuration are considered successful. A low RMSD suggests that the docking program or algorithm can anticipate the right pose for the given protein–ligand combination. The RMSD value is also affected by the number of rotatable bonds. The RMSD tends to increase as the number of rotatable bonds increases. After the pose selection process, the pose grading and grading process will be conducted to find out which of the existing scoring functions will most accurately rank the poses corresponding to their RMSD values [169]. Other than redocking, the cross-docking method should be leveraged. The target 3D structure is acquired from a PDB file in which another ligand has been removed and utilized for docking of small molecules in crossdocking. Another method of evaluating molecular docking is called the decoy set method. A decoy set is a set of inactive or assumed inactive substances created from substances with known activity for the target of action under study. After classifying the substances in the decoy set based on the docking score, the enrichment factor will be calculated, and the receiver operating characteristic (ROC) curve will be established. The ROC curve is a graph showing the sensitivity (Se) of a docking/scoring function in addition to its specificity (Sp), and the area under the curve (AUC) can be calculated to compare different methods. The ROC curve can be independent of the number of active substances in the decoy set [169, 170].

Searching Algorithm

Docking a ligand into a binding site accounts for some degree of bond freedom. There are six relative rotational and translational degrees of freedom between entities, as well as conformational degrees of freedom between ligands and proteins. In rigid docking, the algorithms consider all six degrees of freedom of small-molecule organic compounds, i.e., in terms of translation and rotation, where both the ligands and the proteins are considered rigid entities. The current docking standard is semiflexible docking, in which the conformational flexibility of the ligand molecule is considered but the protein molecule remains a rigid entity. A systematic search for all conceivable rotational bonds of a drug-like molecule is computationally inefficient due to the fact that the number of possible combinations between spin conformations grows exponentially as the number of spinning links increases. Searching algorithms will solve this issue and aim to investigate all

256

Khac-Minh Thai et al.

the solutions (called the conformation space) of the ligand within the protein’s active site quickly and efficiently. In systematic searchbased methods, the results are often exactly repeatable and the number of conformations is reduced and slightly simplified. An example of a searching algorithm of this type is the incremental construction algorithm included in FlexX software [171], where the ligand’s configuration space is obtained based on a discrete model and a tree search is used to position the ligand molecule in such a way that this ligand molecule develops inside the site of action. More common algorithms include genetic algorithms (random search as in GOLD [172]), or tabu search algorithms, which are capable of avoiding repetitive searches of previously defined poses [173] (as in MOE-Dock software [174]). Molecular mechanical force field simulation approaches, such as molecular dynamics and energy minimization, are frequently integrated with searching algorithms (as in Glide [85]). Other approaches rely on pharmacophore modeling. The concept of pharmacophore has already been discussed in depth. Ph4Dock software performs a docking process based on mapping the pharmacophore points of the ligand corresponding to the pharmacophore points of the protein active site [175]. A more recently developed algorithm is FLAP (Fingerprints for Ligands and Proteins). This algorithm uses 4-point pharmacophore fingerprints for both the ligand and the effector site of the protein [176]. The molecular interaction fields (MIFs) of the target binding pocket were calculated using GRID software [177], and the “hotspots” were converted into pharmacophore points. The ligands are described by either atom-based pharmacophores (using GRID atoms) or MIFs (as in ligand–ligand similarity search or pharmacophore interpretation, etc.). Four-point pharmacophores that are able to select potential ligand-binding controls are filtered through a molecular cavity conformational model generated from the target protein (3-point pharmacophores will give less filtering power). The FLAP algorithm can be implemented automatically in processes such as protein access, GRID parameter calculation, cavity detection, and cavity virtual search of similarity with a binding cavity database. Flexible docking is a method using algorithms in which the conformational space of a protein is also partially explored. In an effort to minimize computation, usually, only conformations close to those determined experimentally are evaluated. One possibility that requires less computational work is to add the flexibility of side chain amino acids to the library of spin isomers. Flexibility at the target level of impact is also considered when integrating docking with molecular dynamics as in AMBER [178] or GROMACS [179] software for post-processing of docking results. The flexibility of the molecular framework in these docking processes is also taken into account but may result in inaccurate results due to molecular mechanical force fields.

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . . Scoring and Ranking Function

257

Energy scoring functions are mathematical functions utilized to estimate the free energy of binding of small molecules to the active site of the target protein. Scoring functions, which are routinely employed in docking software, have greatly simplified thermodynamic events and coherence. The major components of the protein–ligand interaction are taken into account when rapidly estimating the most important energetic contributions. The quantities of electrostatic and stereochemical energies are often combined together with the evaluation of hydrogen interactions. While entropy and desolvation effects are ignored, some internal energy can be incorporated. Scoring can also be a multistep process that begins with a quick analysis step followed by re-scoring phases that require more precision and take more computing power. Scoring functions can be classified into three groups: molecular mechanics, empirical, and knowledge-based force fields. • In molecular mechanics scoring functions, the energy portion will include both intramolecular and intermolecular contributions. Molecules will be represented by atomic types and are specifically bound by force fields with partial charges located at the center of the atom. Binding energy is the sum of bond stretching energies, bond angle, torsion angle, and forced torsion angular energy. The Coulomb equation is used to calculate electromagnetic energy, while the Lennard–Jones energy quantity is used to calculate van der Waals forces. The OPLS [180] and AMBER [178] force fields are well computed for protein and ligand molecules; however, they have the disadvantage of taking longer to compute than empirical or knowledge-based scoring functions. • The empirical scoring functions will estimate the binding energy as the sum of the unadjusted energy quantities. The binding energies to the impact target and the accessible complex X-ray crystal structures for a group of ligands with known experimental values will be used to determine the coefficients. These scoring functions have the role of compensating for possible errors in the energy quantities used. Some examples of this type of scoring function include ChemScore [62], Piecewise Linear Potential (PLP) [181], and X-Score [182]. Their precision is determined by how effectively the ligands and proteins are represented in the training set. They could well be tuned to do specific tasks like anticipating binding patterns, categorizing a group of inhibitors, or researching a specific target of action. • Knowledge-based scoring functions use many weighted properties of the molecule that are related to protein–ligand binding patterns. These properties are usually the atom-to-atom distance between the protein and the ligand in the complex, the number of intermolecular hydrogen bonds, or the atom–atom contact energies. Many X-ray crystals of protein–ligand complexes are used as the basis for knowledge-based scoring. A predicted

258

Khac-Minh Thai et al.

protein–ligand complex can be evaluated based on the similarity of its properties to known experimental samples. The contributions used in the evaluation are the sum of all pairs of atoms in the complex, and the resulting score is converted into an energy pseudofunction to estimate the binding affinity. Attribute coefficients can be calculated through linear regression analysis or nonlinear statistical methods, such as Bayesian models, or machine learning techniques such as random forest analysis. Examples of docking programs that use this scoring function include PMF [183], DrugScore [184], LUDI [185], and RF-Score [186]. The disadvantages of scoring functions of this type are the difficulties in assessing the numerical–physical significance of the scores and the risk of error when trying to predict ligands that are not in the training set. Applications of Molecular Docking

Molecular docking is currently a common technique in pharmaceutical businesses. Docking strategies have been used in the creation of several medications at various stages [187–189]. Docking is commonly used for three purposes: (1) predicting binding patterns, (2) virtual screening, and (3) predicting binding free energy. The performance of molecular docking is now thoroughly evaluated and confirmed, particularly for the first two purposes. Binding Pattern Prediction Many published studies have shown that docking can retrieve the binding pattern of co-crystallized ligands within a suitable computational time frame [190]. Low-complexity small-molecule organic compounds are correctly predicted in more than 50% of the cases and can reach as high as 90% in some studies [191]. The inaccuracy increases dramatically when the number of rotatable bonds is considerable (>8) or the binding site is large, and the docked conformations having RMSD values ≤2 Å with the co-crystallized ligand are only achieved in about 30% of cases. The prediction’s success is determined by the target type and the scoring mechanism applied. Systems having little hydrophobic interactions between ligands and proteins are very difficult. On the other hand, scoring functions are not reliable in identifying poses more similar to the existing pose in the co-crystallized crystal, even in the case of the pose being sampled by searching algorithms. In summary, it can generally be admitted that docking programs successfully addressed the problem of conformational search, but the scoring function remains a pitfall of the docking algorithm. Virtual Screening To find prospective novel ligands, molecular docking is frequently used to screen very large libraries of virtual compounds (105–106) to discover potential new leads. In this scenario, the algorithm parameters are improved to boost calculation speed. Each molecule will have one pose chosen and will be assessed by the scoring

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . .

259

algorithms. Molecular docking has recently been employed for the virtual screening of molecular fragments prior to testing [192, 193]. Molecular fragment-based drug discovery focuses on low molecular weight compounds ( 0:6 [209– 212]. QSAR methods are quite effective in predicting single-target activity. Furthermore, multi-target QSAR models and chemoinformatics techniques have been used effectively to find compounds capable of hitting several targets in neurological illnesses [213– 216]. To generate excellent models, the data collection of known inhibitors must be large enough to encompass a wider range of inhibitor activity for specified targets. Furthermore, the molecular size of the testing medications must be within a specified range for proper computation of multi-target-dependent or speciesdependent molecular descriptors, which may influence its capacity to create multi-target QSAR models in some circumstances [217, 218]. A good QSAR model can be built if concepts’ compatibility, assay data representativeness, the impact of data outliers, the selection of molecular descriptors, the fitness of developed quantitative relationships, the occurrence of chance correlations, and starting geometry are all carefully considered. The use of QSAR to identify anti-AD leads has been considered [219].

4

Notes AD is the leading cause of dementia. The disease is highly prevalent in the elderly and entails a wide range of economic and social impacts. The disease also places enormous burdens on the health systems of many countries. Pathophysiologically, AD is a multifactorial disease, so the approach in therapy as well as in research to find new drugs for the treatment of this disease is currently aimed at multiple therapeutic targets at the same time. Computational modeling is a method that has been widely used in the field of drug

268

Khac-Minh Thai et al.

discovery and has played a key role in the discovery of many effective therapeutic agents. Using this approach, scientists can provide valuable insights to guide the experimental stages of drug development. Recent studies have shown that computational modeling methods have been successfully applied in identifying molecular structures with the potential to inhibit multiple therapeutic targets of AD. Computational modeling strategies have been used effectively to screen, design, and generate leads against specific AD targets. To discover leads, molecular docking has been employed, and molecular dynamics modeling has been used to understand deeper the mechanism of molecular interactions. To find inhibitors using chemical databases, virtual screening, pharmacophore mapping, and molecular dynamic simulations have all been used effectively. These computational methods have been integrated together with each other to develop multi-targeting inhibitors of AD. Because AD is complicated and multifaceted, the same strategies have been proposed to develop better medications for many validated therapeutic targets. In the drug development process, in silico approaches are frequently utilized to create selective singletarget inhibitors, while additional computational methods combined with docking, molecular dynamics, and free binding energy calculation are giving extremely promising solutions for discovering novel leads with various activities and binding mechanisms. The success of computational modeling methods makes the prospect of finding new substances in AD treatment increasingly open and promising in the following years.

Acknowledgments This work was supported by University of Medicine and Pharmacy at Ho Chi Minh city (Grant number: 224/2022/HÐ-ÐHYD to Khac-Minh Thai) and Hue University (Grant number: DHH202204-166 to Thai-Son Tran). References 1. Masters CL, Bateman R, Blennow K et al (2015) Alzheimer’s disease. Nat Rev Dis Primers 1(1):15056 2. Jain P, Jadhav HR (2013) Quantitative structure activity relationship analysis of aminoimidazoles as BACE-I inhibitors. Med Chem Res 22(4):1740–1746 3. Richard AA (2019) Risk factors for Alzheimer’s disease. Folia Neuropathol 57(2): 87–105 4. Vogel JW, Iturria-Medina Y, Strandberg OT et al (2020) Spread of pathological tau proteins through communicating neurons in

human Alzheimer’s disease. Nat Commun 11(1):2612 5. Hampel H, Mesulam MM, Cuello AC et al (2018) The cholinergic system in the pathophysiology and treatment of Alzheimer’s disease. Brain 141(7):1917–1933 6. Kinney JW, Bemiller SM, Murtishaw AS et al (2018) Inflammation as a central mechanism in Alzheimer’s disease. Alzheimers Dement (N Y) 4(1):575–590 7. Cheng X, Zhang L, Lian Y-J (2015) Molecular targets in Alzheimer’s disease: from pathogenesis to therapeutics. Biomed Res Int 2015: 760758

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . . 8. Athar T, Al Balushi K, Khan SA (2021) Recent advances on drug development and emerging therapeutic agents for Alzheimer’s disease. Mol Biol Rep 48(7):5629–5645 9. Kabir MT, Sufian MA, Uddin MS et al (2019) NMDA receptor antagonists: repositioning of memantine as a multitargeting agent for Alzheimer’s therapy. Curr Pharm Des 25(33): 3506–3518 10. Cummings J, Lee G, Ritter A et al (2020) Alzheimer’s disease drug development pipeline: 2020. Alzheimers Dement (N Y) 6(1): e12050 11. Zhang P, Xu S, Zhu Z et al (2019) Multitarget design strategies for the improved treatment of Alzheimer’s disease. Eur J Med Chem 176:228–247 12. Kurz A, Perneczky R (2011) Novel insights for the treatment of Alzheimer’s disease. Prog Neuro-Psychopharmacol Biol Psychiatry 35(2):373–379 13. Salomone S, Caraci F, Leggio GM et al (2012) New pharmacological strategies for treatment of Alzheimer’s disease: focus on disease modifying drugs. Br J Clin Pharmacol 73(4):504–517 14. Hardy J (2009) The amyloid hypothesis for Alzheimer’s disease: a critical reappraisal. J Neurochem 110(4):1129–1134 15. Savage MJ, Gingrich DE (2009) Advances in the development of kinase inhibitor therapeutics for Alzheimer’s disease. Drug Dev Res 70(2):125–144 16. Long JM, Holtzman DM (2019) Alzheimer disease: an update on pathobiology and treatment strategies. Cell 179(2):312–339 17. Stromer T, Serpell LC (2005) Structure and morphology of the Alzheimer’s amyloid fibril. Microsc Res Tech 67(3–4):210–217 18. Nelson R, Sawaya MR, Balbirnie M et al (2005) Structure of the cross-β spine of amyloid-like fibrils. Nature 435(7043): 773–778 19. Dislich B, Lichtenthaler SF (2012) The membrane-bound aspartyl protease BACE1: molecular and functional properties in Alzheimer’s disease and beyond. Front Physiol 3:8 20. Vassar R, Bennett BD, Babu-Khan S et al (1999) Beta-secretase cleavage of Alzheimer’s amyloid precursor protein by the transmembrane aspartic protease BACE. Science (New York, NY) 286(5440):735–741 21. Hong L, Koelsch G, Lin X et al (2000) Structure of the protease domain of memapsin 2 (beta-secretase) complexed with inhibitor. Science (New York, NY) 290(5489):150–153 22. Xu Y, Li MJ, Greenblatt H et al (2012) Flexibility of the flap in the active site of BACE1 as

269

revealed by crystal structures and molecular dynamics simulations. Acta Crystallogr D Biol Crystallogr 68(Pt 1):13–25 23. Kimberly WT, LaVoie MJ, Ostaszewski BL et al (2003) Gamma-secretase is a membrane protein complex comprised of presenilin, nicastrin, Aph-1, and Pen-2. Proc Natl Acad Sci U S A 100(11):6382–6387 24. Wolfe MS, Xia W, Ostaszewski BL et al (1999) Two transmembrane aspartates in presenilin1 required for presenilin endoproteolysis and gamma-secretase activity. Nature 398(6727): 513–517 25. Thinakaran G, Borchelt DR, Lee MK et al (1996) Endoproteolysis of presenilin 1 and accumulation of processed derivatives in vivo. Neuron 17(1):181–190 26. Takasugi N, Tomita T, Hayashi I et al (2003) The role of presenilin cofactors in the gammasecretase complex. Nature 422(6930): 438–441 27. Shah S, Lee SF, Tabuchi K et al (2005) Nicastrin functions as a gamma-secretase-substrate receptor. Cell 122(3):435–447 28. X-c B, Rajendra E, Yang G et al (2015) Sampling the conformational space of the catalytic subunit of human γ-secretase. eLife 4:e11182 29. Vyas VK, Ukawala RD, Ghate M et al (2012) Homology modeling a fast tool for drug discovery: current perspectives. Indian J Pharm Sci 74(1):1–17 30. Colletier J-P, Fournier D, Greenblatt HM et al (2006) Structural insights into substrate traffic and inhibition in acetylcholinesterase. EMBO J 25(12):2746–2756 31. Medina M (2018) An overview on the clinical development of tau-based therapeutics. Int J Mol Sci 19(4):1160 32. Fichou Y, Al-Hilaly YK, Devred F et al (2019) The elusive tau molecular structures: can we translate the recent breakthroughs into new targets for intervention? Acta Neuropathol Commun 7(1):31 33. Yoshida H, Goedert M (2012) Phosphorylation of microtubule-associated protein tau by AMPK-related kinases. J Neurochem 120(1): 165–176 34. Cohen TJ, Guo JL, Hurtado DE et al (2011) The acetylation of tau inhibits its function and promotes pathological tau aggregation. Nat Commun 2:252 35. Schedin-Weiss S, Winblad B, Tjernberg LO (2014) The role of protein glycosylation in Alzheimer disease. FEBS J 281(1):46–62 36. Garcı´a-Sierra F, Mondrago´n-Rodrı´guez S, Basurto-Islas G (2008) Truncation of tau protein and its pathological significance in

270

Khac-Minh Thai et al.

Alzheimer’s disease. J Alzheimers Dis 14(4): 401–409 37. Bretteville A, Ando K, Ghestem A et al (2009) Two-dimensional electrophoresis of tau mutants reveals specific phosphorylation pattern likely linked to early tau conformational changes. PLoS One 4(3):e4843 38. Mucke L (2009) Alzheimer’s disease. Nature 461(7266):895–897 39. Liu F, Grundke-Iqbal I, Iqbal K et al (2005) Contributions of protein phosphatases PP1, PP2A, PP2B and PP5 to the regulation of tau phosphorylation. Eur J Neurosci 22(8): 1942–1950 40. Zhu Y, Shan X, Yuzwa SA et al (2014) The emerging link between O-GlcNAc and Alzheimer disease. J Biol Chem 289(50): 34472–34481 41. Grinberg LT, Wang X, Wang C et al (2013) Argyrophilic grain disease differs from other tauopathies by lacking tau acetylation. Acta Neuropathol 125(4):581–593 42. Cook C, Carlomagno Y, Gendron TF et al (2014) Acetylation of the KXGS motifs in tau is a critical determinant in modulation of tau aggregation and clearance. Hum Mol Genet 23(1):104–116 43. Pan SY, Zhou SF, Gao SH et al (2013) New perspectives on how to discover drugs from herbal medicines: CAM’s outstanding contribution to modern therapeutics. Evid Based Complement Alternat Med 2013:627375 44. Szyman´ski P, Markowicz M, Mikiciuk-Olasik E (2012) Adaptation of high-throughput screening in drug discovery-toxicological screening tests. Int J Mol Sci 13(1):427–452 45. Clark RL, Johnston BF, Mackay SP et al (2010) The drug discovery portal: a resource to enhance drug discovery from academia. Drug Discov Today 15(15–16):679–683 46. Lahana R (1999) How many leads from HTS? Drug Discov Today 4(10):447–448 47. Song CM, Lim SJ, Tong JC (2009) Recent advances in computer-aided drug design. Brief Bioinform 10(5):579–591 48. Veselovsky AV, Zharkova MS, Poroikov VV et al (2014) Computer-aided design and discovery of protein-protein interaction inhibitors as agents for anti-HIV therapy. SAR QSAR Environ Res 25(6):457–471 49. Paˆrvu L (2003) QSAR – a piece of drug design. J Cell Mol Med 7(3):333–335 50. Baig MH, Ahmad K, Roy S et al (2016) Computer aided drug design: success and limitations. Curr Pharm Des 22(5):572–581 51. Kim KH, Kim ND, Seong BL (2010) Pharmacophore-based virtual screening: a

review of recent applications. Expert Opin Drug Discovery 5(3):205–222 52. Sousa SF, Cerqueira NM, Fernandes PA et al (2010) Virtual screening in drug design and development. Comb Chem High Throughput Screen 13(5):442–453 53. Waszkowycz B, Perkins TDJ, Sykes RA et al (2001) Large-scale virtual screening for discovering leads in the postgenomic era. IBM Syst J 40(2):360–376 54. Lionta E, Spyrou G, Vassilatis DK et al (2014) Structure-based virtual screening for drug discovery: principles, applications and recent advances. Curr Top Med Chem 14(16): 1923–1938 55. Shoichet BK (2004) Virtual screening of chemical libraries. Nature 432(7019): 862–865 56. Lounnas V, Ritschel T, Kelder J et al (2013) Current progress in structure-based rational drug design marks a new mindset in drug discovery. Comput Struct Biotechnol J 5: e201302011 57. Anderson AC (2012) Structure-based functional design of drugs: from target to lead compound. Methods Mol Biol 823:359–366 58. Andricopulo AD, Salum LB, Abraham DJ (2009) Structure-based drug design strategies in medicinal chemistry. Curr Top Med Chem 9(9):771–790 59. Goh BC, Hadden JA, Bernardi RC et al (2016) Computational methodologies for real-space structural refinement of large macromolecular complexes. Annu Rev Biophys 45:253–278 60. Fang Y (2015) Combining label-free cell phenotypic profiling with computational approaches for novel drug discovery. Expert Opin Drug Discovery 10(4):331–343 61. Cavasotto CN (2011) Homology models in docking and high-throughput docking. Curr Top Med Chem 11(12):1528–1534 62. Eldridge MD, Murray CW, Auton TR et al (1997) Empirical scoring functions: I. The development of a fast empirical scoring function to estimate the binding affinity of ligands in receptor complexes. J Comput Aided Mol Des 11(5):425–445 63. Dhanavade MJ, Jalkute CB, Barage SH et al (2013) Homology modeling, molecular docking and MD simulation studies to investigate role of cysteine protease from Xanthomonas campestris in degradation of Aβ peptide. Comput Biol Med 43(12): 2063–2070 64. Khare N, Maheshwari SK, Rizvi SMD et al (2022) Homology modelling, molecular docking and molecular dynamics simulation

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . . studies of CALMH1 against secondary metabolites of Bauhinia variegata to treat Alzheimer’s disease. Brain Sci 12(6):770 65. Mahendran SR, Jeyabaskar DS, Francis A et al (2017) Homology modeling and in silico docking analysis of BDNF in the treatment of Alzheimer’s disease. Res J Pharm Technol 10:2899–2906 66. Waterhouse A, Bertoni M, Bienert S et al (2018) SWISS-MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res 46(W1):W296–W303 67. Vriend G (1990) WHAT IF: a molecular modeling and drug design program. J Mol Graph 8(1):52–56, 29 68. Krivov GG, Shapovalov MV, Dunbrack RL Jr (2009) Improved prediction of protein sidechain conformations with SCWRL4. Proteins Struct Funct Bioinform 77(4):778–795 69. Laskowski RA, MacArthur MW, Moss DS et al (1993) PROCHECK: a program to check the stereochemical quality of protein structures. J Appl Crystallogr 26(2):283–291 70. Altschul SF, Gish W, Miller W et al (1990) Basic local alignment search tool. J Mol Biol 215(3):403–410 71. Sievers F, Higgins DG (2018) Clustal omega for making accurate alignments of many protein sequences. Protein Sci 27(1):135–145 72. Webb B, Sali A (2016) Comparative protein structure modeling using MODELLER. Curr Protoc Bioinformatics 54:5.6.1–5.6.37 73. Cardozo T, Totrov M, Abagyan R (1995) Homology modeling by the ICM method. Proteins 23(3):403–414 74. Jacobson MP, Pincus DL, Rapp CS et al (2004) A hierarchical approach to all-atom protein loop prediction. Proteins Struct Funct Bioinform 55(2):351–367 75. Molecular Operating Environment (MOE), 2022.02 Chemical Computing Group ULC, 1010 Sherbooke St. West S et al. 76. Jones G, Willett P, Glen RC et al (1997) Development and validation of a genetic algorithm for flexible docking. J Mol Biol 267(3): 727–748 77. Ghouzam Y, Postic G, Guerin PE et al (2016) ORION: a web server for protein fold recognition and structure prediction using evolutionary hybrid profiles. Sci Rep 6:28268 78. Kelley LA, Mezulis S, Yates CM et al (2015) The Phyre2 web portal for protein modeling, prediction and analysis. Nat Protoc 10(6): 845–858 79. McGuffin LJ, Bryson K, Jones DT (2000) The PSIPRED protein structure prediction server. Bioinformatics 16(4):404–405

271

80. McGuffin LJ, Adiyaman R, Maghrabi AHA et al (2019) IntFOLD: an integrated web resource for high performance protein structure and function prediction. Nucleic Acids Res 47(W1):W408–W413 81. Yang J, Zhang Y (2015) I-TASSER server: new development for protein structure and function predictions. Nucleic Acids Res 43 (W1):W174–W181 82. Mortuza SM, Zheng W, Zhang C et al (2021) Improving fragment-based ab initio protein structure assembly using low-accuracy contactmap predictions. Nat Commun 12(1):5011 83. Simons KT, Bonneau R, Ruczinski I et al (1999) Ab initio protein structure prediction of CASP III targets using ROSETTA. Proteins Suppl 3:171–176 84. Meng XY, Zhang HX, Mezei M et al (2011) Molecular docking: a powerful approach for structure-based drug discovery. Curr Comput Aided Drug Des 7(2):146–157 85. Friesner RA, Banks JL, Murphy RB et al (2004) Glide: a new approach for rapid, accurate docking and scoring. 1. Method and assessment of docking accuracy. J Med Chem 47(7):1739–1749 86. Huang SY, Zou X (2010) Advances and challenges in protein-ligand docking. Int J Mol Sci 11(8):3016–3034 87. Sousa SF, Ribeiro AJ, Coimbra JT et al (2013) Protein-ligand docking in the new millennium--a retrospective of 10 years in the field. Curr Med Chem 20(18):2296–2314 88. Mohan V, Gibbs AC, Cummings MD et al (2005) Docking: successes and challenges. Curr Pharm Des 11(3):323–333 89. Morris GM, Goodsell DS, Huey R et al (1996) Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4. J Comput Aided Mol Des 10(4):293–304 90. Ewing TJ, Makino S, Skillman AG et al (2001) DOCK 4.0: search strategies for automated molecular docking of flexible molecule databases. J Comput Aided Mol Des 15(5): 411–428 91. Kramer B, Rarey M, Lengauer T (1999) Evaluation of the FLEXX incremental construction algorithm for protein-ligand docking. Proteins 37(2):228–241 92. Venkatachalam CM, Jiang X, Oldfield T et al (2003) LigandFit: a novel method for the shape-directed rapid docking of ligands to protein active sites. J Mol Graph Model 21(4):289–307 93. Verdonk ML, Cole JC, Hartshorn MJ et al (2003) Improved protein-ligand docking using GOLD. Proteins 52(4):609–623

272

Khac-Minh Thai et al.

94. Jain AN (2003) Surflex: fully automatic flexible molecular docking using a molecular similarity-based search engine. J Med Chem 46(4):499–511 95. Neves MA, Totrov M, Abagyan R (2012) Docking and scoring with ICM: the benchmarking results and strategies for improvement. J Comput Aided Mol Des 26(6): 675–686 96. McGann M (2012) FRED and HYBRID docking performance on standardized datasets. J Comput Aided Mol Des 26(8): 897–906 97. Ravitz O, Zsoldos Z, Simon A (2011) Improving molecular docking through eHiTS’ tunable scoring function. J Comput Aided Mol Des 25(11):1033–1051 98. Du X, Li Y, Xia YL et al (2016) Insights into protein-ligand interactions: mechanisms, models, and methods. Int J Mol Sci 17(2): 144 99. Sousa SF, Fernandes PA, Ramos MJ (2006) Protein-ligand docking: current status and future challenges. Proteins 65(1):15–26 100. Ferreira LG, Dos Santos RN, Oliva G et al (2015) Molecular docking and structurebased drug design strategies. Molecules (Basel, Switzerland) 20(7):13384–13421 101. Foloppe N, Hubbard R (2006) Towards predictive ligand design with free-energy based computational methods? Curr Med Chem 13(29):3583–3608 102. Jain AN (2006) Scoring functions for proteinligand docking. Curr Protein Pept Sci 7(5): 407–420 103. Durrant JD, McCammon JA (2011) Molecular dynamics simulations and drug discovery. BMC Biol 9:71 104. Schreiner W, Karch R, Knapp B et al (2012) Relaxation estimation of RMSD in molecular dynamics immunosimulations. Comput Math Methods Med 2012:173521 105. Blessy JJ, Sharmila DJ (2015) Molecular modeling of methyl-α-Neu5Ac analogues docked against cholera toxin--a molecular dynamics study. Glycoconj J 32(1–2):49–67 106. Kieseritzky G, Morra G, Knapp EW (2006) Stability and fluctuations of amide hydrogen bonds in a bacterial cytochrome c: a molecular dynamics study. J Biol Inorg Chem 11(1): 26–40 107. Pacholczyk M, Kimmel M (2011) Exploring the landscape of protein-ligand interaction energy using probabilistic approach. J Comput Biol 18(6):843–850 108. Manly CJ, Chandrasekhar J, Ochterski JW et al (2008) Strategies and tactics for

optimizing the Hit-to-Lead process and beyond--a computational chemistry perspective. Drug Discov Today 13(3–4):99–109 109. Andrade CH, Pasqualoto KF, Ferreira EI et al (2010) 4D-QSAR: perspectives in drug design. Molecules (Basel, Switzerland) 15(5):3281–3294 110. Myint KZ, Xie XQ (2010) Recent advances in fragment-based QSAR and multidimensional QSAR methods. Int J Mol Sci 11(10):3846–3866 111. Lo Y-C, Rensi SE, Torng W et al (2018) Machine learning in chemoinformatics and drug discovery. Drug Discov Today 23(8): 1538–1546 112. Kaserer T, Beck KR, Akram M et al (2015) Pharmacophore models and pharmacophorebased virtual screening: concepts and applications exemplified on hydroxysteroid dehydrogenases. Molecules (Basel, Switzerland) 20(12):22799–22832 113. Liao C, Sitzmann M, Pugliese A et al (2011) Software and resources for computational medicinal chemistry. Future Med Chem 3(8):1057–1085 114. Karelson M, Lobanov VS, Katritzky AR (1996) Quantum-chemical descriptors in QSAR/QSPR studies. Chem Rev 96(3): 1027–1044 115. Nicolaou CA, Kannas C, Loizidou E (2012) Multi-objective optimization methods in de novo drug design. Mini Rev Med Chem 12(10):979–987 116. Chan HH, Leong YQ, Voon SM et al (2021) Effects of amyloid precursor protein overexpression on NF-κB, rho-GTPase and pro-apoptosis Bcl-2 pathways in neuronal cells. Rep Biochem Mol Biol 9(4):417–425 117. Kumar A, Singh A, Ekavali (2015) A review on Alzheimer’s disease pathophysiology and its management: an update. Pharmacol Rep 67(2):195–203 118. Yiannopoulou KG, Papageorgiou SG (2013) Current and future treatments for Alzheimer’s disease. Ther Adv Neurol Disord 6(1): 19–33 119. Wang TT, Chen Q, Zhou D (2016) Alzheimer’s disease therapeutics: current and future therapies. Minerva Med 107(2):108–113 120. De Ferrari GV, Canales MA, Shin I et al (2001) A structural motif of acetylcholinesterase that promotes amyloid beta-peptide fibril formation. Biochemistry 40(35): 10447–10457 121. Leo´n R, Garcia AG, Marco-Contelles J (2013) Recent advances in the multitargetdirected ligands approach for the treatment

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . . of Alzheimer’s disease. Med Res Rev 33(1): 139–189 122. Piazzi L, Rampa A, Bisi A et al (2003) 3-(4-[[Benzyl(methyl)amino]methyl]phenyl)-6,7-dimethoxy-2H-2-chromenone (AP2238) inhibits both acetylcholinesterase and acetylcholinesterase-induced beta-amyloid aggregation: a dual function lead for Alzheimer’s disease therapy. J Med Chem 46(12):2279–2282 123. Rosini M, Andrisano V, Bartolini M et al (2005) Rational approach to discover multipotent anti-Alzheimer drugs. J Med Chem 48(2):360–363 124. Rodrı´guez-Franco MI, Ferna´ndez-Bachiller MI, Pe´rez C et al (2006) Novel tacrinemelatonin hybrids as dual-acting drugs for Alzheimer disease, with improved acetylcholinesterase inhibitory and antioxidant properties. J Med Chem 49(2):459–462 125. Marco-Contelles J, Unzeta M, Bolea I et al (2016) ASS234, as a new multi-target directed propargylamine for Alzheimer’s disease therapy. Front Neurosci 10:294 126. Reddy PH, Tripathi R, Troung Q et al (2012) Abnormal mitochondrial dynamics and synaptic degeneration as early events in Alzheimer’s disease: implications to mitochondriatargeted antioxidant therapeutics. Biochim Biophys Acta 1822(5):639–649 127. Bartolini M, Marco-Contelles J (2019) Tacrines as therapeutic agents for Alzheimer’s disease. IV. The tacripyrines and related annulated tacrines. Chem Rec 19(5):927–937 128. Li Y, Peng P, Tang L et al (2014) Design, synthesis and evaluation of rivastigmine and curcumin hybrids as site-activated multitargetdirected ligands for Alzheimer’s disease therapy. Bioorg Med Chem 22(17):4717–4725 129. Scipioni M, Kay G, Megson IL et al (2019) Synthesis of novel vanillin derivatives: novel multi-targeted scaffold ligands against Alzheimer’s disease. Medchemcomm 10(5): 764–777 130. Umar T, Shalini S, Raza MK et al (2018) New amyloid beta-disaggregating agents: synthesis, pharmacological evaluation, crystal structure and molecular docking of N-(4-((7-chloroquinolin-4-yl)oxy)-3-ethoxybenzyl)amines. Medchemcomm 9(11): 1891–1904 131. Fu H, Li W, Luo J et al (2008) Promising anti-Alzheimer’s dimer bis(7)-tacrine reduces beta-amyloid generation by directly inhibiting BACE-1 activity. Biochem Biophys Res Commun 366(3):631–636 132. Piazzi L, Cavalli A, Colizzi F et al (2008) Multi-target-directed coumarin derivatives:

273

hAChE and BACE1 inhibitors as potential anti-Alzheimer compounds. Bioorg Med Chem Lett 18(1):423–426 133. Zhu Y, Xiao K, Ma L et al (2009) Design, synthesis and biological evaluation of novel dual inhibitors of acetylcholinesterase and beta-secretase. Bioorg Med Chem 17(4): 1600–1613 134. Huang W, Tang L, Shi Y et al (2011) Searching for the multi-target-directed ligands against Alzheimer’s disease: discovery of quinoxaline-based hybrid compounds with AChE, H3R and BACE 1 inhibitory activities. Bioorg Med Chem 19(23):7158–7167 135. Cavalli A, Bolognesi ML, Capsoni S et al (2007) A small molecule targeting the multifactorial nature of Alzheimer’s disease. Angew Chem Int Ed Engl 46(20):3689–3692 136. Huang W, Lv D, Yu H et al (2010) Dualtarget-directed 1,3-diphenylurea derivatives: BACE 1 inhibitor and metal chelator against Alzheimer’s disease. Bioorg Med Chem 18(15):5610–5615 137. Prati F, De Simone A, Armirotti A et al 3,4-Dihydro-1,3,5-triazin-2(1H)(2015) ones as the first dual BACE-1/GSK-3β fragment hits against Alzheimer’s disease. ACS Chem Neurosci 6(10):1665–1682 138. Murata K, Matsumura S, Yoshioka Y et al (2015) Screening of β-secretase and acetylcholinesterase inhibitors from plant resources. J Nat Med 69(1):123–129 139. Di Martino RMC, De Simone A, Andrisano V et al (2016) Versatility of the curcumin scaffold: discovery of potent and balanced dual BACE-1 and GSK-3β inhibitors. J Med Chem 59(2):531–544 140. Yan J, Hu J, Liu A et al (2017) Design, synthesis, and evaluation of multitarget-directed ligands against Alzheimer’s disease based on the fusion of donepezil and curcumin. Bioorg Med Chem 25(12):2946–2955 141. Sang Z-p, Qiang X-m, Li Y et al (2015) Design, synthesis, and biological evaluation of scutellarein carbamate derivatives as potential multifunctional agents for the treatment of Alzheimer’s disease. Chem Biol Drug Des 86(5):1168–1177 142. Singh M, Silakari O (2016) Design, synthesis and biological evaluation of novel 2-phenyl1-benzopyran-4-one derivatives as potential poly-functional anti-Alzheimer’s agents. RSC Adv 6(110):108411–108422 143. Xu QX, Hu Y, Li GY et al (2018) Multi-target anti-Alzheimer activities of four prenylated compounds from Psoralea fructus. Molecules (Basel, Switzerland) 23(3):614

274

Khac-Minh Thai et al.

144. Chakraborty S, Basu S (2017) Multifunctional activities of citrus flavonoid narirutin in Alzheimer’s disease therapeutics: an integrated screening approach and in vitro validation. Int J Biol Macromol 103:733– 743 145. Ahmad A, Ali T, Park HY et al (2017) Neuroprotective effect of fisetin against amyloidbeta-induced cognitive/synaptic dysfunction, neuroinflammation, and neurodegeneration in adult mice. Mol Neurobiol 54(3): 2269–2285 146. Liang Z, Zhang B, Su WW et al (2016) C-glycosylflavones alleviate tau phosphorylation and amyloid neurotoxicity through GSK3β inhibition. ACS Chem Neurosci 7(7):912–923 147. Kim H, Park B-S, Lee K-G et al (2005) Effects of naturally occurring compounds on fibril formation and oxidative stress of β-amyloid. J Agric Food Chem 53(22): 8537–8541 148. Porat Y, Abramowitz A, Gazit E (2006) Inhibition of amyloid fibril formation by polyphenols: structural similarity and aromatic interactions as a common inhibition mechanism. Chem Biol Drug Des 67(1):27–37 149. Pavadai P, Swaminathan S (2015) Design and insilico molecular prediction of flavonefusedthiazole analogues as Acetyl Cholinesterase and β-Secretase inhibitor in the treatment of Alzheimer’s disease. Int J Pharmtech Res 7:125–131 150. Wang SN, Li Q, Jing MH et al (2016) Natural xanthones from Garcinia mangostana with multifunctional activities for the therapy of Alzheimer’s disease. Neurochem Res 41(7): 1806–1817 151. Ferna´ndez-Bachiller MI, Pe´rez C, Monjas L et al (2012) New tacrine-4-Oxo-4H-chromene hybrids as multifunctional agents for the treatment of Alzheimer’s disease, with cholinergic, antioxidant, and β-amyloidreducing properties. J Med Chem 55(3): 1303–1317 152. Kumar V, Saha A, Roy K (2020) In silico modeling for dual inhibition of acetylcholinesterase (AChE) and butyrylcholinesterase (BuChE) enzymes in Alzheimer’s disease. Comput Biol Chem 88:107355 153. Stern N, Gacs A, Ta´trai E et al (2022) Dual inhibitors of AChE and BACE-1 for reducing Aβ in Alzheimer’s disease: from in Silico to in vivo. Int J Mol Sci 23(21):13098 154. Khan BA, Hamdani SS, Alsfouk BA et al (2023) Synthesis, biological evaluation and

computational investigations of S-benzyl dithiocarbamates as the cholinesterase and monoamine oxidase inhibitors. J Mol Struct 1271:134138 155. Gujral SS, Shakeri A, Hejazi L et al (2022) Design, synthesis and structure-activity relationship studies of 3-phenylpyrazino[1,2-a] indol-1(2H)-ones as amyloid aggregation and cholinesterase inhibitors with antioxidant activity. Eur J Med Chem Rep 6:100075 156. Dhamodharan G, Mohan CG (2022) Machine learning models for predicting the activity of AChE and BACE1 dual inhibitors for the treatment of Alzheimer’s disease. Mol Divers 26(3):1501–1517 157. Shrivastava SK, Nivrutti AA, Bhardwaj B et al (2022) Drug reposition-based design, synthesis, and biological evaluation of dual inhibitors of acetylcholinesterase and β-Secretase for treatment of Alzheimer’s disease. J Mol Struct 1262:132979 158. Zeng H, Wu X (2016) Alzheimer’s disease drug development based on ComputerAided Drug Design. Eur J Med Chem 121: 851–863 159. Kumar A, Srivastava S, Tripathi S et al (2016) Molecular insight into amyloid oligomer destabilizing mechanism of flavonoid derivative 2-(4’ benzyloxyphenyl)-3-hydroxy-chromen-4-one through docking and molecular dynamics simulations. J Biomol Struct Dyn 34(6):1252–1263 160. Verma A, Kumar A, Debnath M (2016) Molecular docking and simulation studies to give insight of surfactin amyloid interaction for destabilizing Alzheimer’s Aβ42 protofibrils. Med Chem Res 25(8):1616–1622 161. Singh SK, Sinha P, Mishra L et al (2013) Neuroprotective role of a novel copper chelator against Aβ 42 induced neurotoxicity. Int J Alzheimers Dis 2013:567128 162. Kumar A, Roy S, Tripathi S et al (2016) Molecular docking based virtual screening of natural compounds as potential BACE1 inhibitors: 3D QSAR pharmacophore mapping and molecular dynamics analysis. J Biomol Struct Dyn 34(2):239–249 163. Roy S, Kumar A, Baig MH et al (2015) Virtual screening, ADMET profiling, molecular docking and dynamics approaches to search for potent selective natural molecules based inhibitors against metallothionein-III to study Alzheimer’s disease. Methods 83:105– 110 164. Iqbal K, Grundke-Iqbal I (2010) Alzheimer’s disease, a multifactorial disorder seeking

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . . multitherapies. Alzheimers Dement 6(5): 420–424 165. Arooj M, Sakkiah S, Cao G et al (2013) An innovative strategy for dual inhibitor design and its application in dual inhibition of human thymidylate synthase and dihydrofolate reductase enzymes. PLoS One 8(4):e60470 166. Cosconati S, Forli S, Perryman AL et al (2010) Virtual screening with AutoDock: theory and practice. Expert Opin Drug Discovery 5(6):597–607 167. Kumar A, Sharma A (2018) Computational modeling of multi-target-directed inhibitors against Alzheimer’s disease. In: Roy K (ed) Computational modeling of drugs against Alzheimer’s disease. Springer, New York, pp 533–571. https://doi.org/10. 1007/978-1-4939-7404-7_19 168. Cole JC, Murray CW, Nissink JW et al (2005) Comparing protein-ligand docking programs is difficult. Proteins 60(3):325–332 169. Hevener KE, Zhao W, Ball DM et al (2009) Validation of molecular docking programs for virtual screening against dihydropteroate synthase. J Chem Inf Model 49(2):444–460 170. Triballeau N, Acher F, Brabet I et al (2005) Virtual screening workflow development guided by the “receiver operating characteristic” curve approach. Application to highthroughput docking on metabotropic glutamate receptor subtype 4. J Med Chem 48(7): 2534–2547 171. Rarey M, Kramer B, Lengauer T et al (1996) A fast flexible docking method using an incremental construction algorithm. J Mol Biol 261(3):470–489 172. Jones G, Willett P, Glen RC (1995) Molecular recognition of receptor sites using a genetic algorithm with a description of desolvation. J Mol Biol 245(1):43–53 173. Baxter CA, Murray CW, Clark DE et al (1998) Flexible docking using Tabu search and an empirical estimate of binding affinity. Proteins 33(3):367–382 174. Hall SB, Venkitaraman AR, Whitsett JA et al (1992) Importance of hydrophobic apoproteins as constituents of clinical exogenous surfactants. Am Rev Respir Dis 145(1):24–30 175. Goto J, Kataoka R, Hirayama N (2004) Ph4Dock: pharmacophore-based proteinligand docking. J Med Chem 47(27): 6804–6811 176. Baroni M, Cruciani G, Sciabola S et al (2007) A common reference framework for analyzing/comparing proteins and ligands. Fingerprints for Ligands and Proteins (FLAP):

275

theory and application. J Chem Inf Model 47(2):279–294 177. Park K, Kim D (2006) A method to detect important residues using protein binding site comparison. Genome Inform 17(2): 216–225 178. Case DA, Cheatham TE 3rd, Darden T et al (2005) The Amber biomolecular simulation programs. J Comput Chem 26(16): 1668–1688 179. Congreve M, Chessari G, Tisi D et al (2008) Recent developments in fragment-based drug discovery. J Med Chem 51(13):3661–3680 180. Jorgensen W, Maxwell D, Tirado-Rives J (1996) Development and testing of the OPLS all-atom force field on conformational energetics and properties of organic liquids. J Am Chem Soc 118:11225–11236 181. Verkhivker GM (2004) Computational analysis of ligand binding dynamics at the intermolecular hot spots with the aid of simulated tempering and binding free energy calculations. J Mol Graph Model 22(5):335–348 182. Wang R, Lai L, Wang S (2002) Further development and validation of empirical scoring functions for structure-based binding affinity prediction. J Comput Aided Mol Des 16(1): 11–26 183. Muegge I (2006) PMF scoring revisited. J Med Chem 49(20):5895–5902 184. Velec HF, Gohlke H, Klebe G (2005) DrugScore(CSD)-knowledge-based scoring function derived from small molecule crystal data with superior recognition rate of near-native ligand poses and better affinity prediction. J Med Chem 48(20):6296–6303 185. Bo¨hm HJ (1994) The development of a simple empirical scoring function to estimate the binding constant for a protein-ligand complex of known three-dimensional structure. J Comput Aided Mol Des 8(3):243–256 186. Ballesteros JA, Weinstein H (1995) [19] Integrated methods for the construction of three-dimensional models and computational probing of structure-function relations in G protein-coupled receptors. In: Sealfon SC (ed) Methods in neurosciences, vol 25. Academic Press, pp 366–428. https://doi. org/10.1016/S1043-9471(05)80049-7 187. Garman E, Laver G (2004) Controlling influenza by inhibiting the virus’s neuraminidase. Curr Drug Targets 5(2):119–136 188. Kaldor SW, Kalish VJ, Davies JF 2nd et al (1997) Viracept (nelfinavir mesylate, AG1343): a potent, orally bioavailable inhibitor of HIV-1 protease. J Med Chem 40(24):3979–3985

276

Khac-Minh Thai et al.

189. von Itzstein M, Wu WY, Kok GB et al (1993) Rational design of potent sialidase-based inhibitors of influenza virus replication. Nature 363(6428):418–423 190. Chen H, Lyne PD, Giordanetto F et al (2006) On evaluating molecular-docking methods for pose prediction and enrichment factors. J Chem Inf Model 46(1):401–415 191. Warren GL, Andrews CW, Capelli AM et al (2006) A critical assessment of docking programs and scoring functions. J Med Chem 49(20):5912–5931 192. Huang JW, Zhang Z, Wu B et al (2008) Fragment-based design of small molecule X-linked inhibitor of apoptosis protein inhibitors. J Med Chem 51(22):7111–7118 193. Murray CW, Callaghan O, Chessari G et al (2007) Application of fragment screening by X-ray crystallography to beta-secretase. J Med Chem 50(6):1116–1123 194. Fink T, Bruggesser H, Reymond JL (2005) Virtual exploration of the small-molecule chemical universe below 160 Daltons. Angew Chem Int Ed Engl 44(10):1504–1508 195. Bohacek RS, McMartin C, Guida WC (1996) The art and practice of structure-based drug design: a molecular modeling perspective. Med Res Rev 16(1):3–50 196. Huey R, Morris GM, Olson AJ et al (2007) A semiempirical free energy force field with charge-based desolvation. J Comput Chem 28(6):1145–1152 197. Xie H, Wen H, Zhang D et al (2017) Designing of dual inhibitors for GSK-3β and CDK5: virtual screening and in vitro biological activities study. Oncotarget 8(11):18118–18128 198. Tran T-S, Le M-T, Tran T-D et al (2020) Design of curcumin and flavonoid derivatives with acetylcholinesterase and betasecretase inhibitory activities using in silico approaches. Molecules (Basel, Switzerland) 25(16):3644 199. Duan S, Guan X, Lin R et al (2015) Silibinin inhibits acetylcholinesterase activity and amyloid β peptide aggregation: a dual-target drug for the treatment of Alzheimer’s disease. Neurobiol Aging 36(5):1792–1807 200. Yang SY (2010) Pharmacophore modeling and applications in drug discovery: challenges and recent advances. Drug Discov Today 15(11–12):444–450 201. Sliwoski G, Kothiwale S, Meiler J et al (2014) Computational methods in drug discovery. Pharmacol Rev 66(1):334–395

202. Caporuscio F, Tafi A (2011) Pharmacophore modelling: a forty year old approach and its modern synergies. Curr Med Chem 18(17): 2543–2553 203. Fei J, Zhou L, Liu T et al (2013) Pharmacophore modeling, virtual screening, and molecular docking studies for discovery of novel Akt2 inhibitors. Int J Med Sci 10(3): 265–275 204. Goyal M, Dhanjal JK, Goyal S et al (2014) Development of dual inhibitors against Alzheimer’s disease using fragment-based QSAR and molecular docking. Biomed Res Int 2014:979606 205. Tetko IV, Gasteiger J, Todeschini R et al (2005) Virtual computational chemistry laboratory--design and description. J Comput Aided Mol Des 19(6):453–463 206. MOE. 2008.10 edition. Chemical Computing Group Inc. SSW, Suite 910, Montreal, Quebec, Canada H3A 2R7. https://www. chemcomp.com/. Accessed 20 May 2021 207. Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional neural networks. Adv Neural Inf Process Syst 25:1097–1105 208. Ngo T-D, Tran T-D, Le M-T et al (2016) Computational predictive models for P-glycoprotein inhibition of in-house chalcone derivatives and drug-bank compounds. Mol Divers 20(4):945–961 209. Consonni V, Ballabio D, Todeschini R (2009) Comments on the definition of the Q2 parameter for QSAR validation. J Chem Inf Model 49(7):1669–1678 210. Todeschini R, Ballabio D, Grisoni F (2016) Beware of unreliable Q(2)! A comparative study of regression metrics for predictivity assessment of QSAR models. J Chem Inf Model 56(10):1905–1913 211. Chirico N, Gramatica P (2011) Real external predictivity of QSAR models: how to evaluate it? Comparison of different validation criteria and proposal of using the concordance correlation coefficient. J Chem Inf Model 51(9):2320–2335 212. Thai K-M, Bui Q-H, Tran T-D et al (2012) QSAR modeling on benzo[c]phenanthridine analogues as topoisomerase I inhibitors and anti-cancer agents. Molecules (Basel, Switzerland) 17(5):5690–5712 213. Youdim MB, Buccafusco JJ (2005) Multifunctional drugs for various CNS targets in the treatment of neurodegenerative disorders. Trends Pharmacol Sci 26(1):27–35

Recent Advances in Computational Modeling of Multi-targeting Inhibitors as. . . 214. McGaughey GB, Colussi D, Graham SL et al (2007) Beta-secretase (BACE-1) inhibitors: accounting for 10s loop flexibility using rigid active sites. Bioorg Med Chem Lett 17(4): 1117–1121 215. Kumalo HM, Bhakat S, Soliman ME (2016) Investigation of flap flexibility of β-secretase using molecular dynamic simulations. J Biomol Struct Dyn 34(5):1008–1019 216. Berhanu WM, Masunov AE (2015) Atomistic mechanism of polyphenol amyloid aggregation inhibitors: molecular dynamics study of Curcumin, Exifone, and Myricetin interaction with the segment of tau peptide oligomer. J Biomol Struct Dyn 33(7):1399–1411

277

217. Ma XH, Shi Z, Tan C et al (2010) In-silico approaches to multi-target drug discovery: computer aided multi-target drug design, multi-target virtual screening. Pharm Res 27(5):739–749 218. Gonza´lez-Dı´az H, Prado-Prado FJ, Santana L et al (2006) Unify QSAR approach to antimicrobials. Part 1: predicting antifungal activity against different species. Bioorg Med Chem 14(17):5973–5980 219. Ambure P, Roy K (2014) Advances in quantitative structure-activity relationship models of anti-Alzheimer’s agents. Expert Opin Drug Discovery 9(6):697–723

Chapter 9 Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for Alzheimer’s Disease Priyanka De and Kunal Roy Abstract The use of biomarkers in the detection of early and preclinical Alzheimer’s disease (AD) has become of central importance. The use of in vivo amyloid and tau imaging agents can detect early AD pathological processes and subsequent neurodegeneration. Positron emission tomography (PET) and single photon emission computed tomography (SPECT) are efficient imaging techniques for the in vivo detection of Alzheimer’s disease pathologies. In the current chapter, the authors have reviewed different computational modeling studies recently performed on the PET and SPECT diagnostic imaging agents for AD. The information obtained from these studies involves the detection of chemical scaffolds and their interactions with AD-related pathologies which help in the development of new PET and SPECT imaging agents with clinical applicability. Key words Alzheimer’s disease (AD), Positron emission tomography (PET), Single photon emission computed tomography (SPECT), In silico methods, Molecular docking, Molecular dynamics (MD) simulation

1

Introduction An array of syndromes resulting in the loss and degradation of cells of the nervous system give rise to various insidious but lethal neuropathies like Alzheimer’s disease (AD) and dementia. AD is a complex neurodegenerative disease afflicting millions of individuals worldwide. It is the most common form of dementia characterized by neuronal impairment, irreversible cognitive dysfunction, and memory loss [1]. The main pathological feature of AD is the formation of neuritic plaques which are composed of amyloid-β peptide (Aβ) fibrils, neurofibrillary tangles (NFTs), and neurotransmitter shortfalls. The abnormal protein deposition is associated with other biochemical processes like oxidative damage, inflammation, and lysosomal dysfunction [2, 3]. Diagnosis of early AD proceeds with the symptomatic characterization which is not always precise, giving rise to a demand for biomarkers that efficiently and

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_9, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

279

280

Priyanka De and Kunal Roy

accurately work in the AD diagnosis. Molecular imaging modalities are frequently used for biomarker quantification; and thus, this field actively assists in biomarker identification for AD. Modern clinical scientists apply molecular imaging technology in studying the basis/cause of the disease from the molecular abnormalities found in the cells. This method of analysis also accelerates the achievement of other important clinical goals of (a) early disease detection, (b) therapy optimization for important molecular targets, (c) forecasting and monitoring response to therapy, and (d) disease recurrence monitoring. Radionuclide molecular imaging is one of the earliest and most mature methods of imaging technique which is efficient in detecting any harboring infection. Positron emission tomography (PET) and single photon emission computed tomography (SPECT) imaging were the first molecular imaging modalities used clinically and are used now also due to their advantages of noninvasive localization, high sensitivity, and quantifiability [4]. PET and SPECT imaging modalities use radiopharmaceuticals, commonly referred to as radioactive tracers or probes, or radiolabeled diagnostic agents injected intravenously into individuals for the identification of diseased conditions. AD pathology possesses a complex group of symptoms requiring a battery of neuropsychological tests for the diagnosis of dementia and differentiating from other dementias (vascular, frontotemporal, and Lewy body dementia). With an improved understanding of AD and other dementia, the use of molecular imaging as a biomarker helps in refined disease detection. Contemporary research approaches provide ample evidence of PET and SPECT biomarkers’ applications. These imaging agents were found to detect glucose metabolism and perfusion in the brain as evidenced by different tests in clinical diagnosis of AD [5]. Early observations report that the imbalance in the whole brain blood flow in dementia patients was detected by metabolic tracer [18F]fluorodeoxyglucose or FDG followed by identification of regional changes in blood flow and oxygen metabolism and glucose metabolism [6, 7]. Comparative studies reported [18F]FDG PET having better sensitivity (88–94%) and specificity (63–73%) than common other neuropathological examinations [8]. [99mTc]hexamethylpropyleneamine oxime ([99mTc]HMPAO) brain perfusion SPECT was more specific in distinguishing AD compared to other types of dementia [8]. Due to the scarce availability of experimental data for these imaging/diagnostic agents, the use of in silico methods has gained significant momentum. The main benefit of the in silico drug design is cost-effectiveness and less time consumption in research and development. Furthermore, a large number of available tools provide a stronger basis for the design of these imaging agents with preferred specificity. In the present book chapter, we have elaborately explained different PET and SPECT imaging agents, their principles of action, and their applications in the diagnosis of AD.

Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for. . .

281

Furthermore, we have extensively studied the application of different computational approaches like quantitative structure–activity relationship (QSAR), pharmacophore mapping, molecular docking [9], molecular dynamics simulation [10], and machine learning studies [11] researched worldwide in the last few years with reference to the diagnostic agents for Alzheimer’s disease. These computational methods are elaborately explained in Chapter 7 of this book.

2

PET Imaging PET imaging technology involves the administration of a radioactive, positron-emitting nuclide, which labels a biomolecule specific to the physiologic process under investigation by PET. PET allows the three-dimensional mapping of administered positron-emitting radiopharmaceuticals and also enables the study of biological function in both healthy and diseased conditions [12]. Radiopharmaceuticals, labeled with positron-emitting isotopes like 11C and 18F, are administered for disease diagnosis. The positron-emitting decay process is identified by the alteration of a proton into a neutron along with the emission of a positron (positively charged antiparticle of an electron) and a neutrino (chargeless particle): A Z XN

þ →A Z - 1 Y N þ1 þ e þ v

After emission, the positron travels a short distance known as the positron range before it annihilates by combining with an electron. During an annihilation event, when a positron unites with an electron nearby, its mass is converted into energy producing two 511 keV γ-rays which travel simultaneously in equal and nearly opposite directions (Fig. 1). This pair of photons is detected by the PET scanners, equipped with coincidence γ detectors that hit the detectors almost at the same time [13]. The resolving time between the two coincidence detectors is about 4–5 ns, and this

Fig. 1 The main principle of the PET imaging technique

282

Priyanka De and Kunal Roy

Table 1 Some commonly used PET radionuclides Radionuclide

Half-life (t1/2)

Emax (MeV)

β+ branching fraction

11

Carbon

20.3 min

0.96

1.00

13

Nitrogen

9.97 min

1.20

1.00

15

Oxygen

2.1 min

1.73

1.00

18

Fluorine

110 min

0.63

0.97

22

Sodium

2.60 y

0.55

0.90

62

Copper

9.74 min

2.93

0.97

64

Copper

12.7 h

0.65

0.29

68

Gallium

67.6 min

1.89

0.89

76

Bromine

16.2 h

Various

0.56

82

Rubidium

1.25 min

2.60, 3.38

0.96

4.17 d

1.53, 2.14

0.23

124

Iodine

time interval is called the coincidence time window. However, in the newly developed PET scanners, the time-of-flight coincidence detectors have a time resolution close to 500 ps [14]. As soon as the opposite detectors detect the two released photons, within the coincidence window, a coincidence event is logged, and the positron annihilation is expected to have commenced somewhere along the line of response (LOR) connecting the two detectors. In some annihilation events, a detectable coinciding event is not generated either because one of the two γ-rays is absorbed or because it is simply not detected. However, about 97% of the emitting photons are detected. Many such events are summed which help in the quantification of line integrals through the isotope distribution. The rationality of this calculation depends on the number of counts collected [15]. A PET study commences with the injection or inhalational administration of a radiopharmaceutical. The scan is initiated after a time lag extending from seconds to minutes to allow for transport and uptake by the organ of interest. Several PET radionuclides with their respective half-life (t1/2) are listed in Table 1. The majority of the PET radiopharmaceuticals used by medical and clinical researchers for the AD diagnosis are labeled with four common PET radionuclides 15O, 13N, 11C, and 18F.

3

SPECT Imaging Single photon emission computed tomography (SPECT) is a nuclear imaging modality used frequently in diagnostic medicines. It gives a three-dimensional nuclear image with combined

Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for. . .

283

Table 2 Commonly used radionuclides for SPECT imaging Nuclide

Half-life/h

Principal photon emission energies/MeV

Type of emission

123

13.2

0.16

Electron capture

Tc

6

0.14

Isomeric transition

In

67.9

0.17/0.25

Electron capture

Ga

78.3

0.09/0.19/0.30

Electron capture

73.1

0.17

Electron capture

I

99m 111 67

201

Tl

Fig. 2 Principles of SPECT imaging

knowledge obtained from scintigraphy with that of computed tomography. This allows a three-dimensional display offering better detail, contrast, and spatial information. SPECT imaging uses radionuclides that directly emit gamma (γ) rays such as technetium-99 m (99mTc) and iodine-123 (123I). Generally, the half-lives of SPECT radiotracers are longer than those used in PET imaging (Table 2). This makes them more accessible for imaging and longer radiosynthesis times make them more viable. SPECT machines combine an array of gamma cameras (ranging from one to four cameras) that rotate around the patient [16]. Radionuclide distribution within tissues can be determined spatially using specially designed gamma cameras rotating around the patient. The use of multiple gamma cameras increases detector efficiency and spatial resolution. Three-dimensional images are then constructed from the projection data obtained from the cameras [16, 17]. Figure 2 shows the basic principles on how SPECT imaging works.

284

Priyanka De and Kunal Roy

Fig. 3 Examples of 99mTc radiopharmaceuticals

SPECT radiopharmaceuticals are used to diagnose neurodegenerative diseases, cancer, and infections by emitting gamma (γ) radiation. The longer half-life of these chemicals has the advantage of enabling SPECT imaging studies to be conducted over longer periods. Figure 3 shows some commonly used technetium (99mTc)labeled SPECT radiopharmaceuticals.

4

Imaging Agents for Alzheimer’s Disease PET imaging has aided in the investigations of the underlying pathophysiology of different neurological conditions. It has been employed to investigate metabolism, receptor binding, and alterations in regional blood flow. One of the major applications is in the favor of elucidating complex neurological disorders such as dementia or Alzheimer’s disease (AD). The molecular sensitivity in the central nervous system (CNS) allows the PET radiotracers the quantification of target–ligand interactions with good selectivity in humans giving information about disease pathology. Widely accepted radiopharmaceuticals for brain imaging involve [18F]FDOPA tracers for dopamine synthesis in PD and schizophrenia, [18F]-FDG analogs for imaging the glucose metabolism alterations, and translocator protein detection in AD and/or PD [18, 19]. Additionally, [11C]-PIB compounds are used for tracking the amyloid-β plaque accumulation in AD [20]. A PET tracer should have the potential to cross the blood–brain barrier (BBB), while the tracer’s selectivity ultimately impacts its usefulness and applicability. Therefore, they should follow essential criteria: (a) molecular weight should be less than 500 kDa; (b) lipophilic coefficient should be between 1 and 5; and (c) topological polar surface area should be below 90 Å2 [21, 22]. Benzothiazole and benzoxazole derivatives like [18F]-flutemetamol, [18F]-florbetapir, and [18F]-florbetaben are used for the detection of pathological amyloid depositions within the brain tissue. Tracers like [18F]-AV-1451 and [18F]THK help in the detection of aggregation rates of tau proteins [23]. At present, FDA approves only three PET imaging agents including [18F]-flutemetamol (Vizamyl), [18F]-florbetapir (Amyvid), and [18F]-florbetaben (Neuraceq) against amyloid plaques

Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for. . .

285

[24–26]. [18F]THK523 (2-(4-aminophenyl)-6-(2-([18F]fluoroethoxy))quinoline), a derivative of quinolone, showed 12-fold higher affinity toward tau than Aβ protein in the transgenic mice model and high uptake in orbitofrontal, parietal, hippocampal, and temporal cortices of the human brain [27]. [18F]THK5105 and [18F]THK5117 are two other radiolabeled 2-arylquinoline SPECT derivatives which exhibit high retention in the mesial temporal lobes associated with the pathology of AD. Neuroinflammation imaging in AD is performed with [11C]N-butan-2-yl-1-(2-chlorophenyl)-N-methylisoquinoline-3-carboxamide or [11C]PK11195 which binds to the translocator protein (TSPO) or peripheral benzodiazepine receptor (PBR). Commonly used SPECT imaging agents for neurological disorders (mainly for Parkinson’s disease) are [99mTc(V)]-HMPAO, [123I]-ioflupane, and [99mTc(I)]-TRODAT-1 [28, 29]. 123I-based imidazopyridine compounds are used for amyloid beta imaging [30]. A thioflavin derivative, 6-iodo-2-(4′-dimethylamino)-phenyl-imidazo[1,2-a]pyridine (IMPY), radiolabeled with 125I/123I showed good binding with amyloid plaque labeled in a transgenic mouse model of AD [31]. SPECT imaging with 99mTc-hexamethylpropyleneamine (99mTc-HMPAO) and 99mTc-ethylcysteine dimer efficiently distinguished AD from vascular dementia (VD), Lewy body dementia (LBD), and normal controls [32]. It was found that radiolabeled DRM106 (123/125I-DRM106 [6-iodo-2[4-(1H-3-pyrazolyl)phenyl]imidazo[1,2-a]pyridine]) showed higher sensitivity than a promising SPECT probe [125I]IMPY in a transgenic mice amyloid beta model [30]. [123I]ABC577, a radioiodinated imidazopyridine derivative, featured alluring characteristics and high binding affinity for Aβ in the frontal, temporal, and posterior cortices in AD patients with minimum retention in the white matter [33]. Iodinated PK11195 and CLINDE ([123I] PK11195 and [123I]CLINDE) are promising SPECT agents for neuroinflammation imaging in AD diagnosis [34, 35].

5 Computational Modeling of PET and SPECT Imaging Agents Against Alzheimer’s Disease The development of biomarkers is an emerging field in the research of AD diagnosis facilitating quantitative analysis of imaging agents’ behavior toward AD pathologies. Here we present the case studies of computational modeling of PET and SPECT imaging agents for Alzheimer’s disease reported in the last 4 years. Ambure and Roy [36] researched the exploration of structural features of imaging agents against amyloid beta (Aβ) plaques in AD. In their study, they used a congeneric series of 44 imaging agents, which included 17 PET and 27 SPECT imaging agents. The study involved the development of 2D-quantitative structure–activity relationship

286

Priyanka De and Kunal Roy

Fig. 4 Highly active PET and SPECT compounds showing the contribution of the S_aasC descriptor [36]

(2D-QSAR) and group-based QSAR (G-QSAR) models using the genetic function approximation (GFA) method of feature selection. Various kinds of statistical metrics were calculated for the internal and external validation of the models. The 2D-QSAR model consisted of three descriptors including Atype_Unknown, S_aasC, and B05[N-N] descriptors, and the model showed significant predictive ability (R2 = 0:812, Q 2 = 0:764, R2pred = 0:784 ). The Atype_Unknown descriptor is associated with rhenium complexed with SN3, SN2O, and SN2S chelators, the presence of which lowers the binding affinity of the imaging agents. The S_aasC descriptor is related to the presence/absence of functional groups at the sixth position of the benzothiazole ring and the 4′ position of the phenyl ring. The presence of a bulky group with electronegative atoms shoots up the binding affinity as seen in PET compound P1 and SPECT compound S1 (Fig. 4). The B05[N-N] descriptor representing the topological distance between two nitrogen atoms has a negative influence on the binding affinity. It was noted that the presence of SN2O and SN2S chelators (in the absence of N-N fragment at distance 5) possesses good binding affinity than in SN3, SN2O, and SN2S chelators (in the presence of N-N fragment at distance 5). The G-QSAR was a four descriptor model with R2 = 0.862, Q2 = 0.817, and R2pred = 0:765. From the descriptor analysis, it was inferred that the –OCH3 group at the R position, bulky substituent like dimethyl, substituted amine at the 4′ position and less electronegative groups like Br at the R position increase the binding affinity toward amyloid beta. They have also designed a few PET and SPECT imaging agents which were found to be similar or more active than compounds in the original dataset. The information obtained from this research is helpful to design new novel PET and SPECT imaging agents efficiently binding to Aβ plaques in the diagnosis of AD.

Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for. . .

287

Table 3 Statistical metrics for the QSAR models developed by De et al. [37]

Target

Number of compounds in train and test sets

R2

Q 2LOO

Q 2F1

PET imaging of amyloid beta (Aβ)

Ntrain = 29, Ntest = 9

0.766

0.600

0.534

SPECT imaging of amyloid beta (Aβ)

Ntrain = 55, Ntest = 18

0.771

0.700

0.739

PET and SPECT imaging of tau protein

Ntrain = 22, Ntest = 9

0.910

0.839

0.865

De et al. [37] applied a multilayered variable selection strategy to develop QSAR models for PET and SPECT imaging agents targeted against amyloid beta (Aβ) and tau fibrils for AD diagnosis. Following the strict Organization for Economic Co-operation and Development (OECD) guidelines, they have strived to select significant descriptors from the large initial pool of descriptors using a multilayered variable selection strategy using the double crossvalidation (DCV) method followed by the best subset selection (BSS) method prior to the development of the final PLS models. The developed models showed significant statistical performance and reliability (Table 3). For PET binding toward Aβ protein, the important features observed were (a) total polar surface area contributed by N, O, S, and P atoms; (b) topological distances between oxygen and sulfur; (c) topological distance between two carbon atoms (C–C); (d) the number of halogens (X) on the aromatic ring; and (e) the number of donor atoms for H bonds. In the case of SPECT binding to Aβ fibrils, the features like the surface area of acceptor atoms, the topological distance between two carbon atoms, presence or absence of fluorine and iodine, and presence of ten-membered rings like 4H-1-benzopyran ring were found to be crucial. Important parameters for PET and SPECT binding to tau fibril include (a) detour distance of the ring system; (b) –CH groups in benzene nucleus; and (c) presence or absence of N-F at the topological distance 8. These observations were further corroborated by the molecular docking binding interactions. Furthermore, the authors have designed new potential PET and SPECT imaging agents with better predicted binding affinities toward Aβ plaques and tau fibrils. Thus, this amalgamation of QSAR and molecular docking techniques are efficient tools useful for the design of potential imaging agents for the diagnosis of AD. Kuang et al. [38] studied a second-generation tau PET tracer, PI2620, as a potential diagnostic agent for AD. A multiscale simulation algorithm including molecular docking, molecular dynamics (MD) simulation, metadynamics simulation, and binding free energy calculations were executed. Metadynamics simulation is a modification of MD simulation in which harmonic restraints are

288

Priyanka De and Kunal Roy

Fig. 5 Concave sites of tau fibril for PI2620 binding as explained by Kuang et al. [38]

enforced on selected collective coordinates of the system along with a history-dependent potential [39]. Molecular docking revealed that this compound tends to bind at the concave sites of the fibril which can be segregated into three subsites (V1, V2, and V3) (Fig. 5). At the top of the concave site, V1 is located which contains polar amino acids such as Arg349 and Gln351. Below V1, V2 is located which is formed by Gln351 and Lys353 polar residues. PI2620 binding is stronger with V2 site than V1. This is due to that fact that V1 being exposed tends to suffer solvent molecule disturbance. The V3 concave binding site is located at the bottom which is formed by Ile360 and His362. Here also, PI2620 gets solvent exposed by binding at the top of the Ile360 and His362 residues. From the metadynamics simulation analysis, it was found that the ligand binds at the core C1 site of the protein with the lowest energy on the free energy surface (FES). It also has a very high binding affinity (-16.38 ± 0.71 kcal/mol) at the C1 site, much higher than at the concave sites V1 and V2. The C1 site is formed by nonpolar amino acid residues Val339, Ile344, Phe346, and Ile354. Metadynamics simulation also identified an entry portal for PI2620 which is a high-affinity binding site (E1) formed by the β-sheets at N- and C-terminals of the tau fibril. Possessing a purely hydrophobic microenvironment, this E1 site is lined by nonpolar residues Ile308, Leu375, and Phe378 which shield the PET tracer from solvent exposure. Several other surface binding sites (S1–S7) were found where PI2620 could bind with a significant binding energy profile. The most important surface binding site S4 was formed by Asn327 and His329 residues, and they interact with van der Waals forces and π–π interactions, respectively. Besides the C1 site, two other core sites C2 and C3 were identified. It was understood that the tracer does not bind at any single spot but may bind at multiple sites. This work gives an idea of the

Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for. . .

289

binding mechanism of the PET tracer with tau fibril and enlightens about tau fibril-specific binding properties. The knowledge would help in the design of novel compounds with better binding and tau selectivity properties. Marondedze et al. [40] reported a pharmacophore-based model and a QSAR model constructed from a dataset of 166 amyloid beta diagnostic compounds targeted against AD with the aim of finding important structural features influencing and predicting bioactivity. The pharmacophore-based virtual screening and QSAR study were used to identify new diagnostic agents against amyloid beta. The highest-ranked pharmacophoric feature (AHRR) contained one hydrogen bond acceptor, one hydrophobic feature, and two ring features. A partial least squares (PLS)-based 3D-QSAR model was generated which showed good statistical metric values (R2 = 0.76, Q2 = 0.72, and Pearson r = 0.76). Pharmacophore-based and molecular docking-based virtual screening of 1,073,715 molecules procured from the ZINC15 database and the ChemBridge CNS-Set were performed to filter out seven hit compounds (Hit 1–7). Molecular docking analysis identified important interactions including hydrogen bond interaction with GLY33, π–π stacking with HIS14, and other hydrophobic interactions with VAL12, LEU17, ILE32, and LEU34. Four of the hit compounds (Hit 1, Hit 2, Hit 5, and Hit 6) passed the ADME analysis satisfactorily. Further, MD simulation studies were carried out for 100 ns, and the root-mean-square deviation was found to be less than 1 Å. The total virtual screening process identified four hit compounds (ZINC IDs: ZINC00164885540, ZINC000283123570, ZINC000279331543, and ZINC000437028032) with predicted activities of 7.126, 7.040, 7.013, and 7.094, respectively. These research findings, thus, can be extremely helpful for the design and identification of new diagnostic molecules targeting Alzheimer’s disease amyloid beta. Murugan et al. [41] reported that there are a minimum of three different high-affinity binding sites for different PET tracers in the tau fibril through binding energy analysis. Integration of modeling techniques including molecular docking, molecular dynamics simulation, and binding free energy calculations helped in the identification of binding sites. The different PET tracers used for the analysis fall under a variety of chemical groups including naphthylethylidene derivative (FDDNP), pyridinyl-butadienyl-benzothiazole derivative (PBB3), benzimidazole-pyrimidine derivatives (T807 and T808), arylquinoline derivatives (THK5105, THK523, THK5351, and THK5117), pyrrolo-pyridine-isoquinolineamine (MK6240), and other miscellaneous compounds (like JNJ311 and R06955). The main aim of the work was to explore the binding mechanisms of these tracer molecules with tau fibrils. Four binding sites were initially identified: B1, B3, and B4 being the core site and B2 being the surface site. It was found that except

290

Priyanka De and Kunal Roy

Table 4 Site preference for different tracers as observed from molecular docking and MD analysis

Compound

Binding site preference in molecular docking

Binding site preference in MD simulation

FDDNP

B1

B1, B3

PBB3

B2

B1, B4

T807

B1, B2

B1

T808

B2

–

THK523

B2

B4

THK5105

B1

B3

THK5317

B1, B2

B3

THK5351

B1

B3

RO6955

B1

B3

MK6240

B1

B1

JNC311

B1, B2

B2

for the T808 tracer, all of the other tracers have a substantial binding affinity toward the surface site. Tracers like FDDNP, THK5105, THK5351, RO6955, and MK6240 preferably bind to the core site more strongly than to the surface site, whereas PBB3, T808, and THK523 have an affinity toward the surface site more. T807, THK5317, and JNJ311 tracers have similar affinities toward more than one site. Since molecular docking analysis considers the target as a rigid structure and the solvent information remains unknown, the authors calculated the free energies of binding for tracers using molecular dynamics (MD) and molecular mechanicsgeneralized Born surface area (MMGBSA) approach. A significant change in the preference of binding sites was observed in the MD analysis, occurring due to the flexibility and dynamic nature of the binding site. Initially, from docking, sites B1 and B2 were found to be preferable; however, MD studies showed a significant preference for B3 and B4 also. Table 4 shows a comparison of the binding site with a preference as observed in molecular docking as well as in MD studies of different tracers. The tracer PBB3 was observed to possess a stronger binding affinity to the tau fibril with a preference to sites 1 and 4. Further, the authors [41] have studied a decomposition analysis to understand the residue-wise contribution to the binding energy for all four sites. This study is a noteworthy example proving that in silico methods can be effectively used to evaluate different imaging tracers before they are used for clinical diagnostic applications.

Computational Modeling of PET and SPECT Imaging Agents as Diagnostics for. . .

6

291

Conclusions The rapid development of different markers suitable for the visualization of AD pathologies during the past few years has led to significant applications of computational methods. Molecular tracers in combination with PET and SPECT imaging agents are widely established as noninvasive in vivo technology. The present book chapter emphasizes on various in silico techniques and gives detailed information about their application in the design and development of novel PET and SPECT imaging agents targeted against AD. The chapter covered computational modeling of PET or SPECT agents reported recently for amyloid beta plaques and tauopathies. Various computational methods involving QSAR, molecular docking, molecular dynamics simulation, and pharmacophore modeling have been studied in these reports to identify novel tracer molecules for the diagnosis of AD.

Acknowledgments PD thanks the Indian Council of Medical Research for Research Associateship (File No: BMI/11(35)/2022). References 1. Breijyeh Z, Karaman R (2020) Comprehensive review on Alzheimer’s disease: causes and treatment. Molecules 25:5789 2. Blennow K, de Leon MJ, Zetterberg H (2006) Alzheimer’s disease. Lancet 368:387–403 3. Alzheimer’s Association (2010) 2010 Alzheimer’s disease facts and figures. Alzheimers Dement 6:158–194 4. Anderson CJ, Ferdani R (2009) Copper-64 radiopharmaceuticals for PET imaging of cancer: advances in preclinical and clinical research. Cancer Biother Radiopharm 24:379–393 5. Jagust W (2004) Molecular neuroimaging in Alzheimer’s disease. NeuroRx 1:206–212 6. Friedland RP, Budinger TF, Ganz E et al (1983) Regional cerebral metabolic alterations in dementia of the Alzheimer type: positron emission tomography with [18f]fluorodeoxyglucose. J Comput Assist Tomogr 7:590–598 7. Benson DF, Kuhl DE, Hawkins RA et al (1983) The fluorodeoxyglucose 18F scan in Alzheimer’s disease and multi-infarct dementia. Arch Neurol 40:711–714 8. Zipursky RB, Meyer JH, Verhoeff P (2007) PET and SPECT imaging in psychiatric disorders. Can J Psychiatr 52:146–157

9. Morris GM, Lim-Wilby M (2008) Molecular docking. In: Methods in molecular biology. Humana Press, Clifton, pp 365–382 10. Hollingsworth SA, Dror RO (2018) Molecular dynamics simulation for all. Neuron 99:1129– 1143 11. Batta M (2019) Machine learning algorithms-a review. Int J Sci Res 9:381–386 12. Townsend DW (2004) Physical principles and technology of clinical PET imaging {. Ann Med Singapore 33:133–145 13. Saha GB (2015) Basics of PET imaging: physics, chemistry, and regulations. Springer, Cham 14. Spanoudaki VC, Levin CS (2010) Photodetectors for time of flight positron emission tomography (ToF-PET). Sensors 10:10484– 10505 15. Ollinger JM, Fessler JA (1997) Positronemission tomography. IEEE Signal Process Mag 14:43–55 16. Lee HW, Hong SB, Tae WS (2000) Opposite ictal perfusion patterns of subtracted SPECT. Hyperperfusion and hypoperfusion. Brain 123: 2150–2159 17. Van Paesschen W, Dupont P, Van Driel G et al (2003) SPECT perfusion changes during

292

Priyanka De and Kunal Roy

complex partial seizures in patients with hippocampal sclerosis. Brain 126:1103–1111 18. Che´telat G, Arbizu J, Barthel H et al (2020) Amyloid-PET and 18F-FDG-PET in the diagnostic investigation of Alzheimer’s disease and other dementias. Lancet Neurol 19:951–962 19. Minoshima S, Mosci K, Cross D, Thientunyakit T (2021) Brain [F-18]FDG PET for clinical dementia workup: differential diagnosis of Alzheimer’s disease and other types of dementing disorders. Semin Nucl Med 51:230–240 20. Blazhenets G, Frings L, Ma Y et al (2021) Validation of the Alzheimer disease dementia conversion-related pattern as an ATN biomarker of neurodegeneration. Neurology 96: e1358–e1368 21. Minoshima S, Drzezga AE, Barthel H et al (2016) SNMMI procedure standard/EANM practice guideline for amyloid PET imaging of the brain. J Nucl Med 57:1316–1322 22. Yap SY, Frias B, Wren MC et al (2021) Discriminatory ability of next-generation tau PET tracers for Alzheimer’s disease. Brain 144: 2284–2290 23. Harada R, Okamura N, Furumoto S, Yanai K (2018) Imaging protein misfolding in the brain using β-sheet ligands. Front Neurosci 12:585 24. FDA (2013) US prescribing information for Amyvid. accessdata.fda.gov/drugsatfda_docs/ label/2012/202008s000lbl.pdf. Accessed 2023 25. FDA (2014) US prescribing information for Neuraceq. https://www.accessdata. fda.gov/ drugsatfda_docs/label/2014/204677s000 lbl.pdf. Accessed 2023 26. FDA (2016) US prescribing information for Vizamyl. https://www.accessdata. fda.gov/ drugsatfda_docs/label/2016/203137s00 5lbl.pdf. Accessed 2023 27. Fodero-Tavoletti MT, Okamura N, Furumoto S et al (2011) 18F-THK523: a novel in vivo tau imaging ligand for Alzheimer’s disease. Brain 134:1089–1100 28. Adak S, Bhalla R, Vijaya Raj KK et al (2012) Radiotracers for SPECT imaging: current scenario and future prospects. Radiochim Acta 100:95–107 29. Valotassiou V, Malamitsi J, Papatriantafyllou J et al (2018) SPECT and PET imaging in Alzheimer’s disease. Ann Nucl Med 329(32): 583–593 30. Chen CJ, Bando K, Ashino H et al (2015) In vivo SPECT imaging of amyloid-β deposition with radioiodinated imidazo[1,2-a]pyridine derivative DRM106 in a mouse model of Alzheimer’s disease. J Nucl Med 56:120–126

31. Kung MP, Hou C, Zhuang ZP et al (2004) Characterization of IMPY as a potential imaging agent for β-amyloid plaques in double transgenic PSAPP mice. Eur J Nucl Med Mol Imaging 31:1136–1145 32. Yeo JM, Lim X, Khan Z, Pal S (2013) Systematic review of the diagnostic utility of SPECT imaging in dementia. Eur Arch Psychiatry Clin Neurosci 263:539–552 33. Maya Y, Okumura Y, Kobayashi R et al (2016) Preclinical properties and human in vivo assessment of 123 I-ABC577 as a novel SPECT agent for imaging amyloid-β. Brain 139:193– 203 34. Versijpt JJ, Dumont F, Van Laere KJ et al (2003) Assessment of neuroinflammation and microglial activation in Alzheimer’s disease with radiolabelled PK11195 and single photon emission computed tomography. Eur Neurol 50:39–47 35. Arlicot N, Katsifis A, Garreau L et al (2008) Evaluation of CLINDE as potent translocator protein (18 kDa) SPECT radiotracer reflecting the degree of neuroinflammation in a rat model of microglial activation. Eur J Nucl Med Mol Imaging 35:2203–2211 36. Ambure P, Roy K (2015) Exploring structural requirements of imaging agents against Aβ plaques in Alzheimer’s disease: a QSAR approach. Comb Chem High Throughput Screen 18: 411–419 37. De P, Bhattacharyya D, Roy K (2019) Application of multilayered strategy for variable selection in QSAR modeling of PET and SPECT imaging agents as diagnostic agents for Alzheimer’s disease. Struct Chem 30:2429–2445 38. Kuang G, Murugan NA, Zhou Y et al (2020) Computational insight into the binding profile of the second-generation PET tracer PI2620 with tau fibrils. ACS Chem Neurosci 11:900– 908 39. Barducci A, Chelli R, Procacci P et al (2006) Metadynamics simulation of prion protein: β-structure stability and the early stages of misfolding. J Am Chem Soc 128:2705–2710 40. Marondedze EF, Govender KK, Govender PP (2020) Ligand-based pharmacophore modelling and virtual screening for the identification of amyloid-beta diagnostic molecules. J Mol Graph Model 101:107711 41. Murugan NA, Nordberg A, Ågren H (2018) Different positron emission tomography tau tracers bind to multiple binding sites on the tau fibril: insight from computational modeling. ACS Chem Neurosci 9:1757–1767

Part III Computational Modeling of Anti-Alzheimer Drugs Against Newer Targets

Chapter 10 Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents Eva Serrano-Candelas, Laureano E. Carpio, and Rafael Gozalbes Abstract Dual-specificity tyrosine phosphorylation-regulated kinase 1A (DYRK1A) is a promising target for the treatment of different neurodegenerative diseases, especially Alzheimer’s disease (AD). In this chapter, different ligand-based and structure-based computational approaches were explored to develop a workflow for the selection of potential candidates as inhibitors of DYRK1A. The NuBBE database—comprising different compounds from the Brazilian biodiversity landscape—was screened with the designed workflow allowing to search for potential inhibitors. Five different compounds from this database were identified as candidates, and one of them presented not only a good interaction profile with the ATP binding site of DYRK1A but also a great synthesizable accessibility score and an optimal predicted toxicological profile. These results show the capability of the developed in silico workflow to screen large databases to find hit compounds from natural sources, therefore representing a good starting point for future further studies. Key words Alzheimer’s disease, DYRK1A, Computational chemistry, (Q)SAR, Molecular docking

1

Introduction One of the main features of Alzheimer’s disease (AD) is the intraneuronal presence of neurofibrillary tangles (NFTs), which are abnormal insoluble deposits, mainly composed of hyperphosphorylated Tau (pTau) protein [1]. These NFTs contribute to mitochondrial dysfunction, synaptic deficits, and memory impairment [2]. Moreover, as Tau acts as a stabilizer of microtubules (polymeric proteins that support neuronal architecture and projections), Tau dissociation from microtubules induces their destabilization, leading to morpho-functional connectivity impairment and contributing to mental and cognitive symptoms [3].

Supplementary Information The online version contains supplementary material available at https://doi.org/ 10.1007/978-1-0716-3311-3_10. Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_10, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

295

296

Eva Serrano-Candelas et al.

Fig. 1 DYRK1A structure. NLS, nuclear localization signal; DH, DH-box (DYRK homology domain); PEST, region rich in proline (P), glutamic acid (E), serine (S), and threonine (T) residues; His region, polyhistidine region; S/T region, a region rich in serine and threonine residues

Among the kinases involved in in vivo Tau phosphorylation, one of the most relevant is the dual-specificity tyrosine phosphorylation-regulated kinase 1A (DYRK1A), which phosphorylates at least 11 serine and threonine residues on Tau protein [4]. DYRK1A is codified between the 21q22.13 and 21q22.2 regions of chromosome 21, inside the known Down syndrome (DS) critical area [5]. As expected, this extra copy of DYRK1A contributes to the increased susceptibility to AD in DS patients [6]. DYRK1A autophosphorylates on tyrosine and phosphorylates substrates on serine/threonine residues [7]. It contains a central catalytic domain with a nuclear localization signal (NLS), preceded by a secondary NLS and a DH-box (DYRK homology domain, a domain conserved among all DYRK subfamily members) and followed by three crucial regions: the PEST motif (rich in proline (P), glutamic acid (E), serine (S), and threonine (T) residues, involved in the regulation of kinase activity), a polyhistidine tract (composed of 13 histidine residues) that allows DYRK1A accumulation in nuclear speckle-like compartments, and the S/T region (17 subsequent serine/threonine residues of unknown function) [7, 8] (Fig. 1). Considering the importance of the Tau protein in AD and the function of DYRK1A in Tau phosphorylation, finding DYRK1A inhibitors is a promising strategy to ameliorate the AD phenotype. DYRK1A inhibitors are classified into three different types: Type I refers to ATP-competitive inhibitors that block the DYRK1A through binding at its ATP-binding domain, presenting a series of key residues such as PHE238, GLU239, LEU241, LYS188, PHE170, SER242, ASN29, and ASP307 amino acids [9– 13]. Type II and Type III are non-ATP-competitive inhibitors, and both act through allosteric mechanisms but with different modes of binding to the pockets [14]. Due to the lack of reported DYRK1A inhibitors advancing to and beyond clinical trials, there remains a need to develop new high-quality compounds that can be used for detailed biological study [15]. Among the different procedures to identify novel DYRK1A inhibitors, computational (in silico) approaches could represent essential savings in time, resources, and money. Moreover, computational approaches represent valuable alternative methods to mitigate the use of animals [16].

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents

297

Fig. 2 Computational approaches reviewed in this work. SB structure-based, LB ligand-based

In this chapter, we show a case study of the application of different ligand-based (LB) and structure-based (SB) computational approaches for type I DYRK1A inhibitor discovery (Fig. 2). Their advantages and limitations in screening compounds based on their inhibitory profile predictions are also discussed. Moreover, these computational methods were integrated in a workflow applied to screen the NuBBE database [17] to find novel potential DYRK1A inhibitors. Finally, the most promising compounds were analyzed in terms of their ease to be synthesized and their toxicological profile, which are essential aspects to be considered for the final selection of best candidates.

2

Overview of in Silico Methods

2.1 Ligand-Based (LB) Approaches

The LB computational approaches described in the present work are known as structure–activity relationships (SARs) and quantitative SARs (QSARs), collectively referred to as (Q)SARs. Both terms substantiate the idea that there is a correlation between a chemical structure and its physicochemical, biological, or toxicological effects [18].

298

Eva Serrano-Candelas et al.

SARs are based on the presence of molecular substructures linked to a particular property or activity and are widely used for the toxicological characterization of molecules (e.g., it is well known that the presence of a nitroso group in a molecule is closely related to its carcinogenicity potential) [19]. SARs are usually called “expert rules” because they rely on expert judgment, and the presence of those well-characterized substituents is often also considered as “structural alert” (SA). Several computational tools may be used to help in the analysis, such as the SARpy software [20, 21], which identifies the relevant substructures from an input composed of a dataset of molecules with clearly defined chemical structures and identified bioactivity values. The main advantage of SAR analyses is that they are very fast to apply to new molecules. Moreover, as they are based on functional groups, they can provide ideas and hypotheses about the underlying mechanisms of action (MoAs). The main limitation of SARs is that negative predictions are inconclusive, that is, the absence of alerts implies the absence of effects, especially when evaluating toxicity. QSARs are statistical models that use molecular descriptors as predictor variables for a parameter of interest (either physicochemical, toxicological, or biological). Usually, they involve the use of machine learning (ML) algorithms to develop such models. The general workflow of QSAR models has been widely described [22– 25], and as in SAR analysis, QSARs require a dataset of molecules of defined structure and known activity. The characteristics of the molecules are numerically codified as molecular descriptors (or molecular features). These descriptors include, among others, count of specific atoms (e.g., number of heteroatoms), counts of fragments (e.g., number of benzene groups), topological properties (that describe the planar structure of the molecule), or 3D descriptors (which take into account the tridimensional structure of the molecules). QSAR models require higher technical skills than SAR models to be developed, but their application to predict the properties of new molecules is equally fast. The main risks on QSAR model development are (a) possibility of “chance correlations”; (b) possibility of overtraining; and (c) possibility of weak reproducibility of statistical quality of a suggested approach [26]. These risks can be overcome by following different strategies: using two or three independent datasets and the use of additional statistical tests, as y-randomization test (i.e., to perform the model with random dependent variables) or the k-fold cross-validation test (i.e., dividing the data into folds and ensuring that each fold is used as a testing set at some point). On the other side, molecular descriptors are able to better describe the chemical space than the structural characteristics used by SAR analysis, but the underlying MoA of the property/activity cannot usually be inferred from the molecular descriptors used in the model.

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents

299

Table 1 Metrics for evaluation of classification (Q)SAR models Metrics

Equation

Accuracy (ACC)

TP ðTP þ TN þ FP þ FNÞ

Precision (PPV)

TP ðTP þ FPÞ

Sensitivity (TPR)

TP ðTP þ TNÞ

Specificity (TNR)

TN ðFP þ TNÞ

F1-score

2  TP 2  TP þ FP þ FN

NPV

TN ðTN þ FNÞ

FNR

FN ðFN þ TPÞ

FPR

FP ðFP þ TNÞ

FDR

FP ðFP þ TPÞ

FOR

FN ðFN þ TNÞ

MCC

ðTP  TNÞ - ðFP  FNÞ ðTP þ FPÞ  ðTP þ FNÞ  ðTN þ FPÞ  ðTN þ FNÞ

TP a test result that correctly indicates the presence of a condition or characteristic, FP a test result that wrongly indicates that a particular condition or attribute is present, TN a test result that correctly indicates the absence of a condition or characteristic, FN a test result which wrongly indicates that a particular condition or attribute is absent

The use of (Q)SARs has three main advantages over other methodologies: (i) once a model has been developed, the prediction of the property/activity of a compound can be made from the simple knowledge of its chemical structure, (ii) models can be easily automated, thus providing an extremely rapid means for evaluating a large number of chemical structures (thousands or even millions), and (iii) (Q)SARs are completely valid from a regulatory point of view, and EU instances such as the OECD (https://www.oecd. org/) or the ECHA (https://echa.europa.eu/) encourage their use. These kinds of predictive models can be evaluated by different metrics that represent the goodness of fit of the developed models; some of these metrics are shown in Table 1.

300

Eva Serrano-Candelas et al.

2.2 Structure-Based (SB) Approaches

3

SB methods are highly dependent on the available data of threedimensional structure(s) of the target of interest [27]. One of the most accepted and successful in silico practices in this field is molecular docking. This approach aims to predict the interaction between a small molecule (ligand) and a protein at the atomic level by identifying ligand-molecular interactions. Docking procedures allow to establish a ranking of the molecules from databases of chemicals according to ligands’ electronic and structural complementarity to a given target [27]. Molecular docking is a very helpful technique and nowadays is vastly employed to assist in different tasks for drug discovery processes, such as the identification of novel scaffolds within huge libraries of compounds, performing profiling for drug repositioning, or predicting adverse effects, among others [28]. In contrast, the involvement of 3D structures is highly time-consuming, resource-expensive, and not recognized at a regulatory level. Moreover, it requires information about the 3D structure of the protein (obtained from crystallization or nuclear magnetic resonance (NMR) data), which is not always available, and the analysis does not usually discriminate between agonists and antagonists. Three-dimensional structural studies allow us to obtain valuable structural data from 3D information about macromolecules. The Protein Data Bank (PDB, https://www.rcsb.org/) is the dataset of reference, since it includes around 200,000 experimentally determined 3D biomolecular structures [29]. The PDB archive was initially established to serve the structural biology community. However, the ample use of PDB structures across various fields proves the significance of structural studies and how threedimensional data archives support interdisciplinary research [30]. Biomolecular structures have helped us understand the fundamental mechanisms that keep cells alive. For example, with an atomic structure, it has been possible to explore the detailed mechanism of proteins and understand how they stabilize. In addition, atomic structures have revealed the complex motions of motor proteins and the basis of immune system function, as they have shown how antibodies recognize foreign molecules. Each new 3D entry adds a new piece to the puzzle of how life works [31]. Depending on the available 3D information, different results could be obtained depending on the type of crystallized structures. For example, if the macromolecule is co-crystallized with a ligand, the information about the ligand–target interaction can be obtained.

Case Study: DYRK1A Computational Models In order to demonstrate the efficacy of computational tools to the discovery of novel DYRK1A inhibitors, we designed an integrated workflow through the sequential application of the computational

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents

301

methods described in the previous section. A graphical representation of this workflow is depicted in Fig. 3. 3.1 Database Collection

A dataset of 1291 molecules was collected from the ChEMBL database [32], by compiling molecules presenting some inhibitory activity (IC50) against human DYRK1A (ChEMBL ID: CHEMBL2292). This raw dataset was processed to check duplicated and incorrect molecules. Moreover, only molecules from a BioAssay Ontology label (BAO label) equal to “single protein assay” were retained. After curation, the final dataset included 354 molecules with IC50 values between 0.5 nM and 100 μM. A threshold of pIC50 ≥ 6 was used as a positivity criterion (i.e., chemicals with pIC50 ≥ 6 were considered as “positives” and the rest as “negatives”), following the criterion described in [33].

Fig. 3 Graphical representation of the computational workflow designed for the discovery of potential DYRK1A inhibitors by computational methods. The workflow applies SAR, QSAR, molecular docking, and final refinement by means of 3D data to identify hit compounds

302

Eva Serrano-Candelas et al.

Table 2 IDs from structures retrieved from PDB Activity

PDB IDs

Positive ligands

3ANQ, 3ANR, 4AZE, 4MQ1, 4MQ2, 4NCT, 4YLJ, 4YLK, 4YLL, 4YU2, 5A3X, 5A4E, 5A54, 6A1F, 6A1G, 6EIL, 6EIQ, 6EIR, 6EIS, 6EJ4, 6T6A, 6UIP, 6UWY, 7A4R, 7A4S, 7A4W, 7A4Z, 7A51, 7A52, 7A53, 7A5D, 7A5L, 7AJ2, 7AJ4, 7AJ5, 7AJ7, 7AJ8, 7AJM, 7AJS, 7AJV, 7AJW, 7AKA, 7AKB, 7AKE, 7AKL, 7O7K, 7AKH, 7AKF

Negative ligands

5A4L, 5A4Q, 5A4T, 5AIK, 6EIF, 6EIJ, 6EIP, 6EIV, 6LN1, 6QU2, 6S11, 6S14, 6S17, 6S1B, 6S1H, 6S1I, 6S1J, 6YF8, 7A4O, 7A55, 7A5B, 7A5N, 7AJA, 7AJY, 7AK2, 7FHT

Chemicals with pIC50 ≥ 6 were considered as “positives” and the rest as “negatives”

Regarding the SB approach, 74 crystal structures of DYRK1A inhibitors bound to this protein were retrieved from the PDB, and selected PDB IDs were collected (Table 2). 3.2

LB Models

3.2.1

SAR

3.2.2

QSAR

In order to characterize significant molecular substructures common in DYRK1A inhibitors, the curated dataset from ChEMBL was subjected to SAR analysis by SARpy software (v1.0), and molecular representations were obtained with RDKit library [34]. A group representing about 75% of the molecules (composed of 116 active molecules and 148 inactive ones) was reserved to extract the rules (“training set” (TS)), whereas the remaining 25% of the molecules (50 actives and 40 inactive) were reserved to verify and validate those rules (“validation set” (VS)). Only positive substructures having between 4 and 20 atoms and present in at least 3 molecules were included for rules generation. As a result of SAR analysis, a total of 14 SAs for active molecules were found (Table 3). The confusion matrices for the TS and VS are shown in Tables 4 and 5, respectively. As it can be seen, the SAs were able to efficiently detect positive molecules inside the different datasets (accuracy going from 86% (TS) to 76% (VS)). The collected dataset was used to generate QSAR models for exploring the inhibitory profile of molecules on DYRK1A. The same threshold set before as a positivity criterion (pIC50 ≥ 6) was maintained. The models were developed using an in-house set of scripts, following the steps below: I. Descriptor calculation: We have calculated a total of 4676 molecular descriptors for each chemical, using in-house software called WOTAN, which uses the RDKit [34] and Mordred [35] descriptors, as well as other descriptors, implemented as described in [36].

17/1

29/4

O=C(Nc1cccc(O)c1)c1cnc2[nH]cc (c3ccc(O)cc3)c2c1

COc1ccc2c3ccnc(C)c3n(CCCCN(C)C (=O)C)c2c1

c2c[nH]c3ncccc23

CCCCn1c2cc (OC)ccc2c2ccnc (C)c12

1

Found in TS (+/-)

0

Depiction

Example SMILES

No. SMARTS

Table 3 Rule set identified by SARpy analysis

(continued)

5/3

2/1

Found in VS (+/-)

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents 303

CN(C)c1ncnc2ccccc12 Cc1ccsc1CN(C)c1ncnc2ccc(c3ccc (CO)o3)cc12

c1nc(NCC)c2ccccc2n1 c1nc(NCCc2c[nH]cn2)c2cc(c3ccc4c(c3) OCO4)ccc2n1

3

Example SMILES

2

No. SMARTS

Table 3 (continued)

Depiction

11/0

9/0

Found in TS (+/-)

3/0

3/1

Found in VS (+/-)

304 Eva Serrano-Candelas et al.

C(CCC)c4ccccc4

c5cnn(C)c5

4

5

Cn1cc(c2cccc(c3coc4ccc(c5cnn(C)c5) nc34)c2)cn1

COc1ccc2c(c1)[nH]c1c (N3CCCC3c3ccccc3)nccc21

5/0

5/0

(continued)

2/0

0/0

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents 305

COC(=N)c1nc2ccc3ncnc(Nc4ccc(C(C) (C)C)cc4)c3c2s1

COCc1nc2ccc3ncnc (N)c3c2s1

c1nc(N)c2cc(c3ccc4c (c3)OCO4)ccc2n1

6

7

Fc1c(c2ccc3ncnc(NCc4cccs4)c3c2) ccc2c1OCO2

Example SMILES

No. SMARTS

Table 3 (continued)

Depiction

19/1

32/1

Found in TS (+/-)

6/0

11/1

Found in VS (+/-)

306 Eva Serrano-Candelas et al.

Cc1cccc(N)c1

COc1ccc2c3ccnc (C)c3n(CC)c2c1

8

9

COc1ccc2c3ccnc(C)c3n(CCCN)c2c1

COc1ncc2cc(C(=O)Nc3cc(C(=O) NCc4ccsc4)ccc3Cl)c(=O)[nH]c2n1

5/1

14/1

(continued)

7/1

23/4

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents 307

Example SMILES

COc1ccc2[nH]c(c3cccc(F)c3)c (c3ccncc3)c2n1

NS(=O)(=O)c1ccc(Nc2nc (OCC3CCCCC3)c3nc[nH]c3n2)cc1

No. SMARTS

10 COc1ccc2[nH]c (c3ccccc3)cc2n1

11 CCCO

Table 3 (continued)

Depiction

15/3

4/1

Found in TS (+/-)

3/2

0/1

Found in VS (+/-)

308 Eva Serrano-Candelas et al.

CCOC(=O)c1c(C(=CN)C#N)c2ccc (Cl)c(Cl)c2n1C

13 C(=CN)C

+/- means the ratio between the number of occurrences of the structural alert found in positive and negative molecules inside the TS and the VS

COc1ccc2sc(NC(=O)C)nc2c1

12 c1nc2cc(O)ccc2s1

5/0

2/0

0/0

0/0

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents 309

310

Eva Serrano-Candelas et al.

Table 4 Confusion matrix for the SAR study (TS) Predictions Inactive

Active

Experimental values Inactive 106 10 91.4% (TNR) Active 27 121 81.8% (TPR) 79.7% (NPV) 92.4% (PPV) 86% (ACC)

Table 5 Confusion matrix for the SAR study (VS) Predictions Inactive

Active

Experimental values Inactive 34 6 85% (TNR) Active 15 35 70% (TPR) 69.4% (NPV) 85.4% (PPV) 76.7% (ACC)

II. Feature elimination and imputation: First, an unsupervised feature elimination step was performed to eliminate (i) constant, infinite, and non-calculated values and (ii) highly correlated values (when some descriptors presented a correlation >90%, only one was kept). Those descriptors with empty values in more than 15% of the molecules were eliminated, and the remaining empty values were assigned using k-nearest neighbors. III. Training set (TS)/validation set (VS) split and feature selection: The initial datasets were clustered by k-means method, and each cluster was randomly split, in order to maintain about 25% of the chemicals for the purposes of validation (VS). Each part of each cluster was joined to construct the final train and test datasets. The TS was used for final feature selection by employing an internal software to select the most relevant features for the model. IV. Models’ development: Predictive models were generated using the TS following a ProtoQSAR’s designed workflow that explores different ML algorithms available in scikit-learn Python package [36]. V. Models’ evaluation: Different metrics were obtained for the TS and VS and by a tenfold cross-validation (CV) step, in order to evaluate the robustness and the predictivity of the models. Confusion matrices for the TS and VS are shown in Tables 4 and 5 respectively, and CV metrics are shown in Table 6.

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents

311

Table 6 Definitions of the chemical descriptors selected for QSAR model Chemical descriptor

Definition

SlogP_VSA2

MOE logP VSA Descriptor 2

R3e+

R maximal autocorrelation of lag 3/weighted by Sanderson electronegativity

AATS6d

Averaged Broto–Moreau autocorrelation of lag 6 (log function) weighted by valence electrons

F06[C-O]

Frequency of C-O at topological distance 6

Mor22s

3D-MoRSE-signal 22, I-state-weighted

AATSC6dv

Averaged centered Broto–Moreau autocorrelation of lag 6 (log function) weighted by valence electrons

PMI1

Principal moment of inertia of order 1

NPR1

Normalized principal moment of order 1

ATS0s

Broto–Moreau autocorrelation of lag 0 (log function) weighted by I-state

SIC2

Structural information content index (neighborhood symmetry of 2-order)

MATS1s

Moran autocorrelation of lag 1 (log function) weighted by I-state

Table 7 Confusion matrix for the QSAR study (TS) Predictions

Exp. values

Inactive Active

Inactive

Active

109 4 96.4% (NPV)

7 144 95.4% (PPV)

94.6% (TNR) 97.2% (TPR) 95.8% (ACC)

QSAR models for DYRK1A ability were developed following the procedure described above, and the best one was retained. In this model, after feature selection by light gradient boosting machine method, 11 descriptors were used to develop a gradient boosting classifier QSAR model (Table 6). The performance metrics of the final model are reported in Tables 7 (TS) and Table 8 (VS). Moreover, the model was also evaluated by applying a tenfold cross-validation, and the results confirmed the capability of the model for detecting potential inhibitors of DYRK1A (metrics shown in Table 9).

312

Eva Serrano-Candelas et al.

Table 8 Confusion matrix for QSAR study (VS) Predictions

Exp. values

Inactive Active

Inactive

Active

28 8 77.8% (NPV)

12 42 77.8% (PPV)

70.0% (TNR) 84.0% (TPR) 77.8% (ACC)

Table 9 Model classification tenfold CV metrics for TS and VS

3.3 3.3.1

SB Models Docking

Metrics

TS

VS

Accuracy (ACC)

0.94 ± 0.01

0.77 ± 0.07

Precision (PPV)

0.94 ± 0.01

0.78 ± 0.07

Sensitivity (TPR)

0.96 ± 0.01

0.83 ± 0.07

Specificity (TNR)

0.92 ± 0.02

0.71 ± 0.10

F1

0.95 ± 0.01

0.80 ± 0.06

NPV

0.95 ± 0.01

0.77 ± 0.08

FNR

0.04 ± 0.01

0.17 ± 0.07

FPR

0.08 ± 0.02

0.29 ± 0.10

FDR

0.06 ± 0.01

0.22 ± 0.07

FOR

0.05 ± 0.01

0.23 ± 0.08

MCC

0.89 ± 0.02

0.54 ± 0.14

To generate a molecular docking approach for searching for new DYRK1A inhibitors, we took advantage of this protein’s co-crystallized structures with ligands (74 PDB files). Among these only active (considering the same positivity criterion described above, shown in Table S1) were retained (a total of 48 structures) to perform following analysis. In order to validate the docking algorithm to be employed, the first step was to perform a redocking of selected structures employing YASARA [37] software with default settings and a simulation cell of 2 Armstrong (Å) in each coordinate axis around the ligands. Previous to the docking, selected structures were refined with YASARA by adding missing hydrogens and correcting clashes with standard energy minimization. Afterward, they were docked into the ATP-binding pocket of DYRK1A with the AutoDock Vina module implemented in YASARA. The docked ligand’s top poses based on the obtained scores in the DYRK1A ATP site were aligned

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents

313

with the co-crystallized structures to validate the protocol. In this step, the mean heavy atom RMSD calculated was found to be 0.562 ± 0.239 Å. A molecular docking protocol that predicts poses’ distances lower than a preselected value (generally 1.5–2 Å) is generally considered to have been implemented effectively [38]. Moreover, in order to extract valuable information from the crystallized data, all the scores computed after the redocking validation step were retrieved to obtain the range of values of the inhibitors. Scores of these experiments are shown in Fig. 5, resulting in a mean value of 9.65 ± 1.58 Kcal/mol. This result allowed us to establish a cutoff score to discern the best outputs for later docking virtual screening campaigns. 3.3.2

3D Structural Study

For this approach, we used the aforementioned selected PDB structures (Table S1) to study the different interactions taking place with the chemical candidates. This study was done employing ProLIF [39], a Python library that encodes the molecular interactions between the structures by using standard parameters and allowing the generation of interaction diagrams. The retrieved crystallized structures were employed to obtain an interaction profile of the binding. Figure 4 shows different types of interactions that appeared while analyzing these structures. The residues involved in more than half of the structures showing those interactions were selected as relevant. As shown in Fig. 5, hydrophobic and pi-stacking interactions were the two main types of interactions appearing in the crystallized structures. Among them, the pi-stacking interaction involving PHE238 was the one highlighted for this type, while for the hydrophobic ones, other residues were involved, such as LYS188, PHE170, VAL306, or LEU241. Those residues that are part of the ATP-binding domain seem to play a relevant role in the binding of inhibitors.

Fig. 4 Redocking score of the structures retrieved from the PDB. Mean value represented as a dark green line. Standard deviation shown in purple dotted line

314

Eva Serrano-Candelas et al.

Fig. 5 Interactions observed in the crystalized structures from the PDB. (a) Hydrophobic interactions, (b) pi-stacking interactions. *, residues presenting the interactions in more than half of the crystallized structures 3.4 Chemical Profile Predictions 3.4.1 Synthetic Accessibility Prediction (SAP)

One of the most important aspects of drug discovery is the production/synthesis of the active principle. It is desirable to develop compounds that are cheap and easy to synthesize or obtain from natural extracts. Synthetic accessibility prediction (SAP) can be defined as a measurement of the ease/difficulty to synthesize a chemical. It is calculated in the form of a score based on four molecular descriptors that include stereochemical, molecular, fused, and bridged complexities. The SAP score is presented in a range from 0 to 100, in which a compound is more easily synthesizable as closer it is to 100. In our study, the SyntheticAccessibiliyCli (Ambit-SA) software tool [40] was used for SAP evaluation.

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents 3.4.2 Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) Profile

4

315

A first evaluation of the toxicity profile of hit compounds is of great importance in drug discovery processes. In silico toxicity prediction can help to reduce time, cost, and animal experimentation in the first steps of drug discovery. Two specific modules of the technological platform ProtoPRED [41], namely ProtoTOX and ProtoADME, were utilized for predicting a complete panel of toxicological endpoints of the selected compounds, such as mutagenicity (Ames test), genotoxicity (micronucleus assay), chromosomal aberration, acute oral toxicity, and developmental toxicity.

Application: Screening of the NuBBE Database The workflow described here was used to screen the NuBBE database to identify promising hit candidates that could act as DYRK1A inhibitors. The NuBBE database is comprised of 2201 unique Brazilian natural compounds and includes validated multidisciplinary information such as species sources, geographic locations, and spectroscopic data (NMR). The database comprises 78% compounds present in plants, 15% from semi-synthesis, 8% from microorganisms, 2% from biotransformation, and less than 1% from marine species. Since its launch in 2013, the NuBBE database has proven to be an important resource for new drug design [17].

4.1

LB Screening

As depicted in Fig. 3, the first step was to screen the database employing the SAR approach related to positive molecules found in the previous section. For this approach, all the molecules from the NuBBE database were treated calculating the molecular descriptors that our LB models used. Then, in the first step, the database was screened with the SAR developed model. From this analysis, 1444 molecules presented at least 1 of the substituents identified by the SAR and were considered as potential active molecules. Subsequently, these molecules were screened by the QSAR model, and 12 of them were classified as actives, allowing to reduce the initial database to a number of compounds approachable for the SB screening. These 12 compounds, predicted as active inhibitors of DYRK1A by both LB models, were retained to perform further structural studies.

4.2

SB Screening

The 12 compounds predicted as potential actives by SAR and QSAR methodologies were subsequently studied by molecular docking employing YASARA. For this purpose, an optimal resolution DYRK1A structure (1.40 Å) was retrieved from the Protein Data Bank (PDB ID: 4YLL) to be used as the receptor structure. With a simulation cell of 2 Å around the ATP-binding pocket and default parameters, the 12 selected compounds were screened

316

Eva Serrano-Candelas et al.

Table 10 Molecular docking screening results Compound ID

Binding Energy Score (Kcal/mol)

L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L11 L12

9.19 7.792 7.494 7.444 7.229 7.349 8.287 8.656 7.628 7.646 10.68 8.944

The best hit is shown in bold. Colored in green: hits with docking score > mean redocking value – std. Colored in red: hits with docking score < mean redocking value – std

against DYRK1A. Binding energy scores of this molecular docking are shown in Table 10. The docking step resulted in five poses above the mean – std. limit established from the crystallized structure redocking analysis. Compound L11 presented a docking score above the mean value, but not very big differences with the other four compounds. This step allowed us to discriminate the screened compounds further and reduce the number of compounds for next studies. In the next step, the interactions of the best five compounds from the molecular docking with the DYRK1A ATP binding site were studied and compared with the relevant ones found from all the available structural data of DYRK1A in previous actions. The predicted interactions of these five compounds are shown in Table 11. Comparing these interactions with the ones obtained previously from the crystallized structures, we have found that the same hydrophobic interactions appeared in both cases (e.g., this is the case of LYS188, PHE170, LEU241, or VAL306 residues). Therefore, the relevance of these residues for the binding of DYRK1A inhibitors is highlighted. 4.3

Profiling

Considering that future development of these potential hits may require their chemical synthesis, we performed a SAP showing that three of the five selected compounds possessed great scores (>50), with compound L7 presenting the higher one (Table 12).

Table 11 2D interaction diagrams of five best compounds ID

SMILES

L1

C=C[C@]1(C)CC2=C(C(=O)[C@H] 1O)[C@@]1(C)CCCC(C)(C)[C@@H] 1CC2=O

L7

COC1=CC(=O)O[C@@H] (CCc2ccccc2)C1

L8

COC1=CC(=O)O[C@@H](CCc2ccc3c (c2)OCO3)C1

Interactions

(continued)

Table 11 (continued) ID

SMILES

L11 CC(=O)C[C@@H]1c2cc3c (cc2-c2ccc4cc5c(cc4c2N1C)OCO5) OCO3

L12 C=C(C)[C@@H]1Cc2cc3ccc(=O) oc3cc2O1

Interactions

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents

319

Table 12 Synthetic accessibility prediction (SAP) score of the selected compounds, predicted by Ambit-SA software tool SA score

Natural source

C=C[C@]1(C)CC2=C(C(=O)[C@H]1O)[C@@]1(C)CCCC(C) (C)[C@@H]1CC2=O

42.73

Vellozia compacta

L7

COC1=CC(=O)O[C@@H](CCc2ccccc2)C1

79.95

Aniba gigantifolia

L8

COC1=CC(=O)O[C@@H](CCc2ccc3c(c2)OCO3)C1

70.95

Aniba gigantifolia

L11 CC(=O)C[C@@H]1c2cc3c(cc2-c2ccc4cc5c(cc4c2N1C)OCO5) OCO3

49.66

Zanthoxylum riedelianum

L12 C=C(C)[C@@H]1Cc2cc3ccc(=O)oc3cc2O1

66.57

Rutaceae

ID

SMILES

L1

The natural source in which the compounds are found is shown as well

4.4 ADMET Predictions

5

The computational ProtoTOX and ProtoADME modules of the ProtoPRED web server were utilized to predict five different toxicity-related endpoints (Table 13). In the case of ADME predictions, different endpoints were obtained, showing optimal results for human intestinal absorption, desired for developing oral drugs, good bioavailability, and permeation for the blood–brain barrier (BBB), important for the DYRK1A target. Moreover, different rules related to drug development were also calculated, i.e., Lipinski, Ghose, and drug-like (QED), obtaining optimal results for the three of them, not observing significative violations in any of these rules. On the other hand, very different toxicity profiles were obtained for the selected compounds. Only compound L8 presented an optimal ADMET profile, being predicted as nontoxic at all for any of the measured endpoints, which makes this compound the best candidate studied from the NuBBE database.

Conclusion/Discussion The discovery of small molecules for treating AD by targeting different proteins (such as the γ-secretase and β-secretase) has not been as successful as expected [42]. This is the main reason why studying other targets for AD treatment is still a relevant matter. Among the possible targets to work on, DYRK1A stands out. The progressive improvement in computational approaches has speeded up the application of these techniques in drug discovery, not only allowing to reduce costs and accelerate the first steps of the process but also helping to reduce animal experimentation [43]. In this study, different in silico methods were implemented to work together in an optimal computational workflow, with the

Non-genotoxic

Genotoxic

Non-mutagenic

L8

Genotoxic

L12 Non-mutagenic

Non-mutagenic

L7

Non-genotoxic

Non-genotoxic

Non-mutagenic

L1

Genotoxicity Micronucleous

L11 Mutagenic

Mutagenicity (AMES test)

ID

Positive

Negative

Negative

Positive

Positive

Chromosomal aberration

Toxic

Non-toxic

Non-toxic

Non-toxic

Non-toxic

Acute Oral toxicity

Non-toxic

Toxic

Non-toxic

Non-toxic

Toxic

Developmental toxicity

Table 13 Toxicity prediction results obtained with ProtoTOX and ProtoADME

BBB +

BBB +

BBB +

BBB +

BBB +

Blood Brain Barrier permeation

Positive

Positive

Positive

Positive

Positive

Human intestinal absorption

Positive

Positive

Positive

Positive

Positive

Bioavailability (30%)

0/4

0/4

0/4

0/4

0/4

Lipinski rules violations

1/4

0/4

0/4

1/4

0/4

Ghose Rules violations

0/6

0/6

0/6

0/6

0/6

QED rules violations

320 Eva Serrano-Candelas et al.

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents

321

purpose of screening large databases of compounds and finding potential candidates to inhibit DYRK1A. The first type of computational techniques applied was the LB-based ones consisting of SAR and QSAR predictive models. SAR approaches are widely used in AD-related drug discovery processes [44–47], as they serve to identify relevant substructures related to biological activity. This identification can be helpful to elucidate the mechanism of action (MoA) of a compound but sometimes can be a bit simple if the studied dataset presents a high grade of diversity, making it difficult to establish a relationship between the detected substructures and the activity. In contrast, QSAR models, which are also widely employed in anti-Alzheimer drug discovery [48–50], can overcome this limitation due to the use of different and complex molecular descriptors to encode the molecules. However, the significance of these descriptors is sometimes difficult to interpret, making it difficult to elucidate the MoA. In the second step, these LB approaches were complemented with SB techniques, which computationally are more costly but have the advantage to go deeper in the study of the molecular interactions between the structures. Molecular docking techniques have been widely employed in drug discovery processes, allowing to obtain relevant structural information of the binding [51– 53]. However, the need for computational resources makes it difficult to apply SB techniques to large datasets of thousands of molecules. In this study, different computational approaches were applied in a sequential manner, in order to increase the chance to obtain good candidates and to progressively reduce the number of molecules for the identification of potential inhibitors of DYRK1A. The predicted inhibitory hit compounds by SAR and QSAR models were further screened by molecular docking technique, and at the end five different compounds were identified by applying this workflow to the NuBBE database. These compounds were studied in depth by obtaining interactions with the DYRK1A ATP binding site, essential for the binding of inhibitors. This conserved binding site consists of key residues such as PHE238, GLU239, LEU241, LYS188, PHE170, SER242, ASN29, and ASP307 amino acids [9– 13]. Comparing these key residues with those obtained from the identified hit compounds, we found molecules L8 and L11 as having a higher number of interactions with residues LYS188, PHE170, PHE238, LEU241, SER242, and ASP307. It is also noticeable that compound L12 is the only one that presents a hydrophobic interaction with residue ASN292. Other compounds, L1 and L7, also presented hydrophobic interactions with different residues from the ATP binding site. Finally, among the hit compounds, L8 stands out not only for the SAR, QSAR, and docking results but also for presenting an optimal toxicity profile (no predicted toxicity for any of the five endpoints calculated),

322

Eva Serrano-Candelas et al.

making this candidate an interesting starting point for further studies. These discoveries show the usefulness of the application of different computational techniques working sequentially to obtain compounds with desired properties/activities in the first steps of the drug discovery process. Even so, before moving on to in vitro and in vivo tests, more computational predictions are necessary to ensure the reliability of our choices. Molecular dynamics (MD) simulations of the compounds bound to the binding site and subsequent MMGBSA energy calculations would enrich the compound selection process and provide more security in the selection of compounds. In addition, in order to be able to computationally study the selectivity of the proposed compounds, it would be interesting in future works to examine other proteins similar to DYRK1A and obtain the predicted binding of the inhibitors’ candidates with those targets. As a conclusion, in the present work, a computational workflow integrating sequentially different computational techniques, SAR, QSAR, and molecular docking, to identify potential inhibitors of DYRK1A has been successfully developed. This workflow has been applied effectively, thus managing to screen the NuBBE database, finding 5 promising compounds from the initial set of more than 2000 chemicals. The five selected compounds were further studied, predicting their synthesizable applicability and toxicological profile, obtaining relevant results for future works with a promising target as it is DYRK1A for the treatment of AD. References 1. Selkoe DJ (1991) The molecular pathology of Alzheimer’s disease. Neuron 6:487–498. https://doi.org/10.1016/0896-6273(91) 90052-2 2. Zhang H, Cao Y, Ma L et al (2021) Possible mechanisms of tau spread and toxicity in Alzheimer’s disease. Front Cell Dev Biol 9:2064. https://doi.org/10.3389/FCELL.2021. 707268/BIBTEX 3. Verstraelen P, Detrez JR, Verschuuren M et al (2017) Dysregulation of microtubule stability impairs morphofunctional connectivity in primary neuronal networks. Front Cell Neurosci 11:173. https://doi.org/10.3389/FNCEL. 2017.00173/BIBTEX 4. Stotani S, Giordanetto F, Medda F (2016) DYRK1A inhibition as potential treatment for Alzheimer’s disease. Future Med Chem 8:681– 6 9 6 . h t t p s : // d o i . o r g / 1 0 . 4 1 5 5 / f m c 2016-0013 5. de Lagra´n MM, Bortolozzi A, Gispert J et al (2008) Ageing in down syndrome: DYRK1A as a candidate gene for cognitive decline. Int Med

Rev Down Syndr 12:34–40. https://doi.org/ 10.1016/S2171-9748(08)70039-4 6. Wegiel J, Gong CX, Hwang YW (2011) The role of DYRK1A in neurodegenerative diseases. FEBS J 278:236. https://doi.org/10. 1111/J.1742-4658.2010.07955.X 7. Becker W, Joost HG (1998) Structural and functional characteristics of Dyrk, a novel subfamily of protein kinases with dual specificity. Prog Nucleic Acid Res Mol Biol 62:1–17. https://doi.org/10.1016/S0079-6603(08) 60503-6 8. Arbones ML, Thomazeau A, NakanoKobayashi A et al (2019) DYRK1A and cognition: a lifelong relationship. Pharmacol Ther 194:199–221. https://doi.org/10.1016/j. pharmthera.2018.09.010 9. Falke H, Chaikuad A, Becker A et al (2015) 10-iodo-11H-indolo[3,2-c]quinoline-6-carboxylic acids are selective inhibitors of DYRK1A. J Med Chem 58:3131–3143. https://doi.org/10.1021/JM501994D/ SUPPL_FILE/JM501994D_SI_001.PDF

Computational Modeling of DYRK1A Inhibitors as Potential Anti-Alzheimer Agents 10. Rothweiler U, Stensen W, Brandsdal BO et al (2016) Probing the ATP-binding pocket of protein kinase DYRK1A with benzothiazole fragment molecules. J Med Chem 59:9814– 9824. https://doi.org/10.1021/ACS. JMEDCHEM.6B01086 11. Soundararajan M, Roos AK, Savitsky P et al (2013) Structures of down syndrome kinases, DYRKs, reveal mechanisms of kinase activation and substrate recognition. Structure 21:986– 996. https://doi.org/10.1016/j.str.2013. 03.012 12. Tahtouh T, Elkins JM, Filippakopoulos P et al (2012) Selectivity, cocrystal structures, and neuroprotective properties of leucettines, a family of protein kinase inhibitors derived from the marine sponge alkaloid leucettamine B. J Med Chem 55:9312–9330. https://doi. org/10.1021/JM301034U/SUPPL_FILE/ JM301034U_SI_001.PDF 13. Ogawa Y, Nonaka Y, Goto T et al (2010) Development of a novel selective inhibitor of the Down syndrome-related kinase Dyrk1A. Nat Commun 1(1 1):1–9. https://doi.org/ 10.1038/ncomms1090 14. Liu T, Wang Y, Wang J et al (2022) DYRK1A inhibitors for disease therapy: current status and perspectives. Eur J Med Chem 229: 1 1 40 6 2 . h t t p s : // do i . o r g / 1 0 . 1 01 6 / J . EJMECH.2021.114062 15. Henderson SH, Sorrell F, Bennett J et al (2020) Mining public domain data to develop selective DYRK1A inhibitors. ACS Med Chem Lett 11:1620–1626. https://doi.org/10. 1021/ACSMEDCHEMLETT.0C00279/ SUPPL_FILE/ML0C00279_SI_002.XLSX 16. Taylor K, Stengel W, Casalegno C, Andrew D (2014) Experiences of the REACH testing proposals system to reduce animal testing. ALTEX 31:107–128. https://doi.org/10. 14573/ALTEX.1311151 17. Pilon AC, Valli M, Dametto AC et al NuBBE DB: an updated database to uncover chemical and biological information from Brazilian biodiversity. 7. https://doi.org/10. 1038/s41598-017-07451-x 18. Jones RL, Madinaveitia J, Metcalfe TP, Sexton WA (1950) The relationship between the constitution and the effect of chemical. Biochem J 47:110–114. https://doi.org/10.1042/ bj0470110 19. Ashby J, Paton D (1993) The influence of chemical structure on the extent and sites of carcinogenesis for 522 rodent carcinogens and 55 different human carcinogen exposures. Mutat Res 286:3–74. https://doi.org/10. 1016/0027-5107(93)90003-X 20. Ferrari T, Cattaneo D, Gini G, et al (2013) Automatic knowledge extraction from

323

chemical structures: the case of mutagenicity prediction. 101080/1062936X2013773376 24:365–383. https://doi.org/10.1080/ 1062936X.2013.773376 21. SARpy for SAR analysis. http://sarpy. sourceforge.net/. Accessed 22 Aug 2022 22. Carpio LE, Sanz Y, Gozalbes R, Barigye SJ (2021) Computational strategies for the discovery of biological functions of health foods, nutraceuticals and cosmeceuticals: a review. Mol Divers 25:1425–1438. https://doi.org/ 10.1007/S11030-021-10277-5/TABLES/3 23. Go´mez-Ganau S, de Julia´n-Ortiz JV, Gozalbes R (2018) Recent advances in computational approaches for designing potential anti-Alzheimer’s agents. NeuroMethods 132:25–59. https://doi.org/10.1007/978-1-4939-74047_2/FIGURES/3 24. Tropsha A, Golbraikh A (2007) Predictive QSAR modeling workflow, model applicability domains, and virtual screening. Curr Pharm Des 13:3494–3504. https://doi.org/10. 2174/138161207782794257 25. Tropsha A (2010) Best practices for QSAR model development, validation, and exploitation. Mol Inform 29:476–488. https://doi. org/10.1002/MINF.201000061 26. Toropov AA, Toropova AP (2020) QSPR/ QSAR: state-of-art, weirdness, the future. Molecules 25:1292. https://doi.org/10. 3390/MOLECULES25061292 27. Kitchen DB, Decornez H, Furr JR, Bajorath J (2004) Docking and scoring in virtual screening for drug discovery: methods and applications. Nat Rev Drug Discov 3(11):935–949. https://doi.org/10.1038/nrd1549 28. Pinzi L, Rastelli G (2019) Molecular docking: shifting paradigms in drug discovery. Int J Mol Sci 20:4331. https://doi.org/10.3390/ IJMS20184331 29. Berman HM, Battistuz T, Bhat TN et al (2002) The protein data bank. Acta Crystallogr D Biol Crystallogr 58:899–907. https://doi.org/10. 1107/S0907444902003451 30. Feng Z, Verdiguel N, di Costanzo L et al (2020) Impact of the protein data bank across scientific disciplines. Data Sci J 19:1–14. https://doi.org/10.5334/DSJ-2020-025/ METRICS/ 31. The Protein Data Bank: Protein structure | Learn science at Scitable. https://www.nature. com/scitable/topicpage/the-protein-databank-exploring-biomolecular-structure-141 99109/. Accessed 29 Aug 2022 32. Mendez D, Gaulton A, Patrı´cia Bento A et al (2019) ChEMBL: towards direct deposition of bioassay data. Nucleic Acids Res 47:D930. https://doi.org/10.1093/nar/gky1075

324

Eva Serrano-Candelas et al.

33. Lin TE, Chao MW, HuangFu WC et al (2022) Identification and analysis of a selective DYRK1A inhibitor. Biomed Pharmacother 146:112580. https://doi.org/10.1016/j.bio pha.2021.112580 34. RDKit. RDKit: Open-Source Cheminformatics Software News 35. Moriwaki H, Tian YS, Kawashita N, Takagi T (2018) Mordred: a molecular descriptor calculator. J Cheminform 10:1–14. https://doi. org/10.1186/S13321-018-0258-Y/ FIGURES/6 36. Pedregosa Fabianpedregosa F, Michel V, Grisel Oliviergrisel O et al (2011) Scikit-learn: Machine Learning in Python Gae¨l Varoquaux Bertrand Thirion Vincent Dubourg Alexandre Passos PEDREGOSA, VAROQUAUX, GRAMFORT ET AL. Matthieu Perrot. J Mach Learn Res 12:2825–2830. https:// onlinelibrary.wiley.com/doi/book/10.1002/ 9783527628766 37. Krieger E, Vriend G (2014) YASARA view— molecular graphics for all devices—from smartphones to workstations. Bioinformatics 30: 2981–2982. https://doi.org/10.1093/bioin formatics/btu426 38. Hevener KE, Zhao W, Ball DM et al (2009) Validation of molecular docking programs for virtual screening against dihydropteroate synthase. J Chem Inf Model 49:444–460. https://doi.org/10.1021/CI800293N/ SUPPL_FILE/CI800293N_SI_001.PDF 39. Bouysset C, Fiorucci S (2021) ProLIF: a library to encode molecular interactions as fingerprints. J Cheminform 13:1–9. https://doi. org/10.1186/S13321-021-00548-6/ FIGURES/4 40. Simulation of chemical reactions and synthetic accessibility | AMBIT2. https://ambit. sourceforge.net/reactor.html. Accessed 4 Oct 2022 41. ProtoPRED | ProtoQSAR. https://protoqsar. com/en/protopred-en/. Accessed 4 Oct 2022 42. Extance A (2010) Alzheimer’s failure raises questions about disease-modifying strategies. Nat Rev Drug Discov 9(10):749–750. https://doi.org/10.1038/nrd3288 43. Pathak RK, Singh B, Sagar M et al (2020) Computational approaches in drug discovery and design. In: Computer-aided drug design, pp 1–21. https://doi.org/10.1007/978-98115-6815-2_1 44. Ferreira JPS, Albuquerque HMT, Cardoso SM et al (2021) Dual-target compounds for Alzheimer’s disease: natural and synthetic AChE and BACE-1 dual-inhibitors and their

structure-activity relationship (SAR). Eur J Med Chem 221:113492. https://doi.org/10. 1016/J.EJMECH.2021.113492 45. Li X, Hong L, Coughlan K et al (2013) Structure-activity relationship of memapsin 2: implications on physiological functions and Alzheimer’s disease. Acta Biochim Biophys Sin Shanghai 45:613–621. https://doi.org/ 10.1093/ABBS/GMT050 46. Li J, Sun M, Cui X, Li C (2022) Protective effects of flavonoids against Alzheimer’s disease: pathological hypothesis, potential targets, and structure-activity relationship. Int J Mol S c i 2 3 . h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / IJMS231710020 47. Malafaia D, Albuquerque HMT, Silva AMS (2021) Amyloid-β and tau aggregation dualinhibitors: a synthetic and structure-activity relationship focused review. Eur J Med Chem 214:113209. https://doi.org/10.1016/J. EJMECH.2021.113209 48. Wong KY, Duchowicz PR, Mercader AG, Castro EA (2012) QSAR applications during last decade on inhibitors of acetylcholinesterase in Alzheimer’s disease. Mini Rev Med Chem 12: 9 3 6 – 9 4 6 . h t t p s : // d o i . o r g / 1 0 . 2 1 7 4 / 138955712802762365 49. Ambure P, Roy K (2014) Advances in quantitative structure-activity relationship models of anti-Alzheimer’s agents. Expert Opin Drug Discov 9:697–723. https://doi.org/10. 1517/17460441.2014.909404 50. de Moura E´P, Fernandes ND, Monteiro AFM et al (2021) Machine learning, molecular modeling, and QSAR studies on natural products against Alzheimer’s disease. Curr Med Chem 28:7808–7829. https://doi.org/10.2174/ 0929867328666210603104749 51. Meng X-Y, Zhang H-X, Mezei M, Cui M (2011) Molecular docking: a powerful approach for structure-based drug discovery. Curr Comput Aided Drug Des 7:146. h t t p s : // d o i . o r g / 1 0 . 2 1 7 4 / 157340911795677602 52. Vlachakis D (2018) Introductory chapter: molecular docking - overview, background, application and what the future holds. In: Molecular docking. https://doi.org/10. 5772/INTECHOPEN.78266 53. Shahroz MM, Sharma HK, Altamimi ASA et al (2022) Novel and potential small molecule scaffolds as DYRK1A inhibitors by integrated molecular docking-based virtual screening and dynamics simulation study. Molecules 27. h t t p s : // d o i . o r g / 1 0 . 3 3 9 0 / MOLECULES27041159

Chapter 11 Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents Gurmeet Kaur, Deepti Goyal, and Bhupesh Goyal Abstract The incidence of neuropsychiatric diseases like Alzheimer’s disease (AD), depression, Parkinson’s disease (PD), etc. has increased rapidly, imposing a huge economic as well as societal burden. The markers of oxidative stress have been noted early in AD, which highlight the likely participation of reactive oxygen species (ROS) in the cascade of events that lead to neuronal dysfunction. The mitochondrial enzyme monoamine oxidase (MAO) helps in the metabolism of neurotransmitters such as adrenaline, dopamine, noradrenaline, serotonin, and other amines. The oxidative deamination by MAOs generates hydrogen peroxide (H2O2), a delegate of oxidative stress, and neurotoxic ammonia (NH3), which have prompted the development of MAO inhibitors. Among several natural and synthesized inhibitors of MAOs, only a few have displayed potent inhibitory activity against MAOs. A complete understanding of the key binding interactions between MAOs and inhibitors will greatly benefit the design and development of highly efficacious and selective MAO inhibitors. In the present chapter, recent computational studies on the natural and synthetic inhibitors of MAOs have been compiled. Key words Alzheimer’s disease (AD), Molecular docking, Molecular dynamics simulation, Monoamine oxidase (MAO), MAO inhibitors

1

Introduction In the present scenario, the prevalence of neuropsychiatric disorders, for example, Alzheimer’s disease (AD), depression, and Parkinson’s disease, has increased tremendously, imposing global healthcare pressure. AD is the most persistent cause of dementia mainly influencing elderly people. It is a progressive neurodegenerative condition and is identified by substantial cognitive decline and memory damage, which advances to failure in the use of language, decreases in daily performance, increases in neuropsychiatric symptoms, and finally death [1]. The World Alzheimer Report 2022, highlighted that dementia affects nearly 55 million people globally, and this number will reach 139.0 million by 2050 [2].

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_11, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

325

326

Gurmeet Kaur et al.

Pathologically, AD comprises amyloid-β (Aβ) deposits (present in the extracellular senile plaques), neurofibrillary tangles (τ aggregation), metal ion dyshomeostasis, excessive production of reactive oxygen species (ROS), oxidative stress, and loss of cholinergic transmission [3]. The generation of excessive ROS, leading to oxidative stress and neuronal cell death, constitutes one of the key hallmark attributes of AD [4]. The abnormal expression of a mitochondrial enzyme monoamine oxidase (MAO) [5] in the AD brain leads to the production of excessive monoamine metabolites such as hydrogen peroxide (H2O2), a delegate of oxidative stress [6]. 1.1 General Physiology of MAO

MAOs are flavin adenine dinucleotide (FAD) binding proteases intricated in the metabolism of dopamine (DA), norepinephrine (NE), 2-phenylethylamine (PEA), tyramine, serotonin or 5-hydroxytryptamine (5-HT), and exogenous amines like 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) [7]. MAO exists in two isoforms (MAO-A and MAO-B), encoded by two different genes on the X chromosome (Xp11.23-11.4), and has separate substrate specificity. MAO-A is mainly located in the intestines, heart, and placenta, while MAO-B is restricted to brain glial cells, platelets, and liver cells. MAO-A has specificity for large intrinsic amines like epinephrine, 5-HT, and NE, while MAO-B positively deaminates small amines like β-phenylethylamine and benzylamine. Both MAO-A and MAO-B catalyze the deamination of some common amines such as tyramine and DA [8]. Biogenic amines like 5-HT, DA, and NE are continuously metabolized by MAO, which ensure proper synaptic neurotransmission and regulate some of the emotional activities and other brain functioning [9]. The MAO-catalyzed deamination of amines yields the respective aldehydes, which were later oxidized to the corresponding acids in the presence of aldehyde dehydrogenase (ALDH), or alcohols or glycols by aldehyde reductase (ALR) (Fig. 1). In addition, various neurotoxic by-products such as H2O2 and ammonia (NH3) are produced during the deamination of amines. Particularly, H2O2 plays a key role in the production of ROS [10], which further leads to mitochondrial dysfunction and neuronal death.

1.2 Structures of MAO-A and MAO-B

MAO-A and MAO-B consist of 527 and 520 amino acids and have molecular weights of 59,700 and 58,800, respectively. The two isoforms possess 70% similarity in the amino acid sequence [11]. The Fourier Transform Attenuated Total Reflection Spectroscopy (FTIR-ATR) studies of human MAO-A and bovine MAO-B highlighted that the two MAO subtypes have significant folding and molecular specificity [12] and possess four highly conserved regions. These are, namely, (1) adenosine diphosphate (ADP) binding region (residues 6–43); (2) substrate binding domain (residues 178–221); (3) FAD binding site (residues 350–458); and (4) C-terminus (residues 491–511), projected to form a

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

327

Fig. 1 Abnormal expression of monoamine oxidase (MAO), a mitochondrial enzyme, in the AD brain leads to the production of excessive monoamine metabolites such as hydrogen peroxide (H2O2), a delegate of oxidative stress, and neurotoxic ammonia (NH3). Abbreviations: ALDH, aldehyde dehydrogenase; O2, singlet oxygen; NAD+, oxidized form of nicotinamide adenine dinucleotide (NAD); NADH, reduced form of NAD

Fig. 2 Human monoamine oxidase A (MAO-A, PDB ID, 2Z5X) is shown in the cartoon representation with N- and C-termini labelled. The active site cavityshaping loop (Val210–Glu216) and the entrance domain (Phe108–Pro118) of MAO-A are depicted as red coil encircled in black and black coil, respectively. The FAD-binding domain (Met13–Thr88, Val220–Leu294, and Glu400–Gly462) is in salmon, and the substrate-binding domain (Tyr89–Phe219, Ile295–Glu399) is in cyan. The C-terminal membrane region (Leu463–Leu524) of MAO-A is in blue

transmembrane-associated α-helix [13]. The crystallographic examination of human MAO-A and human MAO-B highlighted that they crystallize as monomer and dimer, respectively [14]. The structural comparison of two MAO isoforms exhibited a noteworthy conformational difference in the cavity-shaping loops [MAO-A (210–216) and MAO-B (201–207)], which bring about substrate specificity [15]. Human MAO-A and MAO-B have a cavity volume of 400 Å3 and 700 Å3, respectively. The cavity of MAO-B is

328

Gurmeet Kaur et al.

Fig. 3 Human monoamine oxidase B (MAO-B, PDB ID, 6RKB) is depicted in the cartoon representation. The active site cavity-shaping loop (Thr201–Glu207) of MAO-B is depicted as a red coil encircled in black. The flexible loop (Phe99– Tyr112), which serves as a “gating switch” for the substrate access to the entrance cavity, is depicted as a black coil. The FAD-binding domain (Lys4– Thr79, Val211–Met285, and Glu391–Ala453) is in light pink, and the substratebinding domain (Tyr80–Phe210, Ile286–Glu390, and Met454–Ser488) is in green. The N- and C-termini are labelled, and C-terminal membrane binding region (Val489–Ile501) of MAO-B is in pale yellow

bipartite, and it comprises substrate binding space with the FAD cofactor (400 Å3) and entrance hydrophobic region (300 Å3) [14b, 16], parted by Ile199 and Tyr326. The structures of MAO-A and MAO-B (PDB code: 2Z5X and 6RKB) are shown in Figs. 2 and 3, respectively. The Ile199 in MAO-B is the structural determinant of substrate specificity, and its conformation regulates the plasticity of the catalytic site. The shape of the gating residue Ile199 depends on the nature of the binding ligand, adopting closed and open conformations. This further determines the catalytic domain of MAO-B as a large single cavity or bipartite [17]. In human MAO-A, Phe208 (a cavity entrance residue) does not play a role as gating residue. In human MAO-B, the residue Tyr326 did not truly partition the two cavities; however, it creates a restriction that is more pronounced as compared to human MAO-A, and the Ile335 occupied the equivalent position [9a]. The residues present in active site Tyr (Tyr407 and Tyr 444 in MAO-A; Tyr 398 in MAO-B) and Lys (Lys305 in MAO-A and Lys296 in MAO-B) are conserved in both isoforms of MAO. The residues Phe208 and

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

329

Ile335 of MAO-A and Ile199 and Tyr326 of MAO-B determine substrate specificity exhibited by the two isoforms. 1.3 Involvement of MAO in Increasing Oxidative Stress in AD

By-products like NH3 and H2O2 were produced during the MAO-assisted oxidative deamination of biogenic amines such as 5-HT, DA, and NE. Particularly, H2O2 is responsible for the production of ROS, which plays a role in the neurodegeneration processes as in AD. MAO-A and MAO-B are primarily located in the catecholaminergic neuronal cells and glia and serotonergic neurons, respectively. The biochemical investigations of the postmortem brain indicated that the variations in the two MAO isoforms arise in the initial phase of AD and are sustained throughout the disease [18]. Further, the immunostaining studies highlighted that MAO-B activity was considerably inclined in the cortical and hippocampal part of the brain of patients affected with AD, representing neuronal loss and noteworthy gliosis. On the other hand, MAO-A was elevated in the hypothalamus and frontal lobe. MAO-A activity was decreased in the locus coeruleus in an AD-affected brain and accounts for 80% neuronal reduction, suggesting that reactive MAO-A in neuronal cells is associated with the pathophysiology of AD [19]. The abnormal expression of MAO in the AD brain in conjugation with the Fenton’s and Haber-Weise reactions resulted in the overproduction of ROS, which ultimately leads to neuroinflammation (Fig. 4). Neuroinflammation leads to cognitive impairment and oxidative stress, which is firmly associated with AD [20].

1.4 Involvement of MAO in Aberrant Amyloid Aggregation in AD

The activation of MAO-B contributes to the generation of Aβ aggregates. Oxidative stress is known to play an important role in the pathophysiology of AD, which further catalyzes the amyloid aggregation process. A higher level of MAO-B in the astrocytes of an AD-affected brain is considered an indicator of oxidative stress. The enhanced MAO activity produces more H2O2 and free radicals during the imprudent oxidative deamination of monoamines. This further leads to the overproduction of ROS and, thus, oxidative stress during disease progression [21]. Overactive MAO catalyzes the cleavage of amyloid precursor protein (APP) by directly activating two distinct enzymes, i.e., β-secretase and γ-secretase, thus resulting in aberrant amyloid aggregation (Fig. 5) [22]. The first step toward the production of Aβ involves the cleavage of APP between residues Met671 and Asp672 by β-secretase, resulting in a cell membrane-bound insoluble segment and an extracellular segment. Further, the cell membrane-bound segment was cleaved by γ-secretase generating Aβ (an intracellular fragment of APP). The Aβ undergoes aggregation to generate toxic soluble and insoluble aggregated species, a hallmark of AD. Further, in AD-affected brain, APP produces noxious Aβ fibrils under the effect of 5-HT through activation of various serotoninergic receptors [23].

330

Gurmeet Kaur et al.

Fig. 4 Reactive oxygen species formed by overactive monoamine oxidase in combination with Fenton’s and Haber–Weise reactions leads to the oxidative stress associated in AD

In addition, the acid monoamine metabolites are related to the cerebrospinal levels of Aβ, whereas higher MAO-B expression in amyloid plaques was related to astrocytes, a clinical feature of AD progression [24]. 1.5 Targeting MAO as a Therapeutic Approach in AD Treatment

Considering the presumed pharmacological effects of MAO in AD, the inhibition of overactive MAO expression could result in decreased oxidative stress and neurodegeneration. Furthermore, inhibition of MAO-B could regulate the neuro-modulatory amine levels that will be useful for intellectual indications. The MAO inhibitors induce a considerable neuroprotective effect in AD due to (1) improvement in cognition, (2) modulation of Aβ and APP gene expression, (3) improved metal chelation (iron), and reduced oxidative stress, thus averting hyperphosphorylation of tau [25]. Taking into account the pharmacological effect of MAO in AD, MAO inhibitors persevere to be under considerable attention and are worthy of further exploration. In the present chapter, computational studies on the recent inhibitors of MAO activity were compiled. The inhibitors discussed in the chapter are divided into two categories: (i) natural inhibitors of MAOs (Fig. 6) and (ii) synthetic inhibitors of MAOs (Fig. 7). The studies covered in the chapter are listed in Tables 1 and 2 and are arranged in chronological order. Several inhibitors targeting the substrate binding pocket and active site of MAOs are reported in

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

331

Fig. 5 The overactive MAO catalyzes amyloid precursor protein (APP) cleavage by activating β-secretase and γ-secretase, which leads to aberrant amyloid-β aggregation in AD

the literature, and they can be found elsewhere [6, 9a, 26]. The focus of the present chapter is to review the key interactions of the synthetic and natural inhibitors with the substrate binding pocket and active site of MAOs, whereas several other studies [27–33] highlighted the quantitative structure-activity relationship (QSAR) approach to identify the key structural fragments of the inhibitors that blocked the activity of MAO-A and MAO-B.

2

Insights into the Binding Mechanism of Natural Inhibitors with MAOs In 2022, Fadaka et al. evaluated the inhibitory effects of phytochemicals [leaf extracts of Cannabis sativa (Cannabis)] 5H-naphtho[2,3c]carbazole, 8a-methyl-5-methylene,

332

Gurmeet Kaur et al.

Fig. 6 Two-dimensional structures of natural inhibitors of MAOs

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

333

Fig. 6 (continued)

cannabicoumaronone, cyclopropaneoctanal, cannabichromene, cannabinol, delta-9- tetrahydrocannabinol, linoelaidic acid, morphinan-6-one, dronabinol, octadecadienoic acid, and sterigmatocystin against acetylcholinesterase (AChE), dopa decarboxylase (DDC), MAO-B, and serotonin receptor 2C (HTR2C) [34]. The absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of all twelve compounds were examined using QikProp. All Cannabis sativa derivatives are in the satisfactory range of druggability that indicates the potential of these compounds as therapeutic candidates. The binding-free energy calculations using molecular mechanics for the Generalized Born Model and Solvent Accessibility (MM-GBSA) method highlighted cannabinol, cannabichromene, linoelaidic acid, and morphinan-6one as potential inhibitors against AChE, DDC, MAO-B, and HTR2C activity, respectively. The linoelaidic acid binds strongly to MAO-B (PDB ID: 2V5Z) with a binding free energy of 89.38 kcal/mol. An identical low value of root-mean-square fluctuations (RMSFs) was observed for MAO-B–safinamide and MAO-B–linoelaidic acid complexes, which highlighted the stability of MAO-B–inhibitor complexes during simulation. Next, Jalal et al. performed a virtual screening of 29,000 compounds of Natural Product Atlas [NpATLAS (v2020_06)] from microbial sources using computational techniques [35]. The molecular docking, pharmacological, physiological, and ADMET properties identified 20 novel potent compounds, and further six compounds, i.e., NPA000487 (20α-methyltetrahymanol), NPA012149 (Citreamicin δ), NPA001637 (C35-terpenes), NPA015149 (Ganocin A), NPA007129 (MS-347a), and NPA018844 (Preussomerin EG1), were shortlisted with an ability to cross the blood-brain-barrier (BBB) and target MAO-A (PDB ID: 2BXS) activity. NPA012149 displayed the highest binding affinity to the binding pocket of MAO-A with a binding energy of -14.6 kcal/mol. NPA012149 participated in the hydrogen bond interactions with Tyr58 and Glu205 of MAO-A. The pyran ring of NPA012149 displayed π–π stacking interaction with Tyr396. The MD simulations highlighted that NPA007129,

334

Gurmeet Kaur et al. S

O N R2

R1

O

MK5 R1= NO2, R2= Cl MK6 R1= H, R2= Br MK12 R1= OCH3, R2= F MK14 R1= OCH3, R2= H X1=

R1 N R4

X

O

1 R= p-CF3 2 R= m-CF3

O

HO

X2=

O

N N N

N

NH2

4K5

3 R1 = H, R2= CH2OH, R3= CH3, R4= (CH2)3CF2H, X=X2 4 R1 = R2= CH3, R3= C2H5, R4= 3-FBn, X=CH2 5 R1 = R2= CH3, R3= iPr, R4= 3-FBn, X=CH2 6 R1 = H, R2= CHF2, R3= CH3, R4= Bn, X=X2 7 R1 = R2= R3= CH3, R4= (CH2)3CF2H, X=X2 8 R1 = R2= R3= CH3, R4= (CH2)3CF2H, X=X1 O N H

O N

O

O

N H O

O O

O

O

O

11

12 Br

O

O

O

O

O

O

O

NO

O

O

13

14

15

R

O

N

O O

O

HN

O

N

O N

O

O

O

HN

O

O

O 9

O

10

O

O VB1 R= NH2 VB3 R= CH3 R

16 N

N

N

O

O

N

O

N

N

O

O

VB8

VP1 R= NH2 VP2 R= CH3

O

O N

R

HO

R2 R3

S

N

O

R

N

17 R= 5-NH; 18 R= 6-NH 19 R= 5-O; 20 R= 6-O

N

R

N H

21 R= 5-NH; 22 R= 6-NH 23 R= 5-O; 24 R= 6-O

Fig. 7 Two-dimensional structures of synthetic inhibitors of MAOs

N H

O N

O Ladostigil

HN

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

335

Table 1 Key findings of the recent computational studies on the natural inhibitors of MAOs S. no. Inhibitors

Key findings

References

1.

Linoelaidic acid

The linoelaidic acid binds strongly to MAO-B with a binding free energy of -89.38 kcal/mol. Identical low values of RMSFs were observed for MAO-B– safinamide and MAO-B–linoelaidic acid complexes that highlight the stability of MAO-B–inhibitor complexes during simulation

Fadaka et al. [34]

2.

NPA007129, NPA0012149, and NPA0015149

Jalal et al. NPA012149 exhibited highest binding affinity to [35] MAO-A with a binding score of -14.6 kcal/mol and displayed hydrogen bond interactions with Tyr58 and Glu205 of MAO-A. MD simulations highlighted that NPA007129, NPA0012149, and NPA0015149 form stable complexes with MAO-A and thus can be used as potential MAO-A inhibitors. MD simulations predicted lower stability of complexes of shortlisted compounds with MAO-B as compared to MAO-A

3.

Soy-B (soyasapogenol B) Soy-B displayed no toxicity, low TPSA (60.69), and high Iqbal et al. [36] absorption (82.6%) and satisfied the Lipinski rule. Soy-B effectively binds to AChE, BuChE, GSK3β, MAO-A, MAO-B, and NMDA. Soy-B displayed a strong interaction with Cys323, Thr336, and Phe352 of MOA-A, whereas Leu171, Ile198, and Tyr398 of MOA-B participated in the interaction with Soy-B. Molecular docking and molecular dynamics analyses highlight Soy-B as a potential multitarget therapeutic agent against AD

4.

Glycitein

Glycitein displayed significant inhibitory activity against Prajapati MAO-A (IC50 = 8.30 ± 0.77 μM). Docking studies et al. [46] highlighted that glycitein interacted with major catalytic sites, including Tyr 407 and Tyr444, along with cofactor FAD, of MAO-A, which is similar to the interactions displayed by MAO-A inhibitor harmine. Glycitein interacted with Leu171, Leu199, Ile316, Tyr326, and Phe343 by hydrophobic contacts and with Cys172 and FAD600 through hydrogen bonds in the active pocket of MAO-B. Enzyme kinetics and molecular docking analysis highlighted that glycitein inhibited MAO-A more efficiently as compared to MAO-B

5.

Isoliquiritigenin (ILG)

ILG exhibited competitive inhibition of MAO-A Prajapati (IC50 = 0.68 μM) and mixed inhibition of MAO-B et al. [49] (IC50 = 0.33 μM). ILG binds strongly (-7.44 kcal/ mol) to MAO-A as compared to harmine (6.46 kcal/mol). ILG interacted with allosteric site residues of MAO-B with a binding energy of 8.03 kcal/mol. ILG displayed hydrogen bond interactions with Pro104, His115, Asn116, and Asp123 residues, and hydrophobic contacts with Val106, Trp119, and Arg120 at the allosteric site

336

Gurmeet Kaur et al.

Table 2 Computational scrutiny of the key interactions of the synthetic inhibitors with MAOs S. no. Inhibitors

Key findings

References

1.

Halogen-bearing multiconjugated dienones, MK1–MK15

MK6 (IC50 = 2.82 nM) and MK12 Mathew et al. (IC50 = 3.22 nM) exhibited significant inhibitory [50] activity against MAO-B. MD simulations highlighted that lead compounds, MK5, MK6, MK12, and MK14, inhibit MAO-B activity by distorting the conformational dynamics of MAO-B. Reversibility and kinetic studies highlighted MK6 and MK12 as reversible and competitive inhibitors of MAO-B

2.

S-benzyl dithiocarbamates 1 and 2

S-benzyl dithiocarbamate 1 with p-CF3 Khan et al. [51] (IC50 = 16.02 ± 5.135 μg/mL, 1.284 ± 0.66 μg/ mL) displayed higher inhibitory activity than mCF3 S-benzyl dithiocarbamate 2 (IC50 = 18.37 ± 4.030 μM, 3.164 ± 0.456 μM) against MAO-A and MAO-B, respectively. Molecular docking predicted binding energy of -8.4438 kcal/mol for compound 1 and -8.3308 kcal/mol for compound 2 with MAO-A

3.

Coumarin derivative 3 with -CF2H motif

Coumarin derivative 3 possessing a -CF2H motif was reported as a potent dual inhibitor of AChE (IC50 = 550 nM) and MAO-B (IC50 = 8.2 nM, B/A selectivity >1200). Molecular docking predicted favorable binding of 3 with MAO-B with a binding energy of -10.88 kcal/mol

4.

Pyridoxine derivative, 4k5

Jia et al. 4k5 displayed potent inhibitory activity against [53] AChE (IC50 = 0.0816 ± 0.075 μM) and MAO-B (IC50 = 0.039 ± 0.003 μM). Molecular docking analysis depicted that phenyl ring of 4k5 fits into the enzymatic “aromatic cage” formed by FAD, Tyr398, and Tyr435 and displays π–π stacking interactions with Tyr398 and Tyr435 of MAO-B. In the case of MAO-A, the phenyl ring of 4k5 is situated below “aromatic cage” formed by FAD, Tyr407, and Tyr444. Notably, 4k5 didn’t display π–π stacking interactions with MAO-A

5.

3,7-substituted coumarin derivatives 4 and 5

3,7-substituted coumarin derivatives 4 and Mzezewa 5 displayed good potential as MTDLs due to the et al. inclusion of the propargylamine functionality in [54] their structures. Molecular docking conformations depicted that compounds 4 and 5 lack the flexibility to fit and interact with the residues in the cavity’s active site, which is consistent with the observed weak inhibition of MAO-A in the presence of 4 and 5 (IC50 value of 3.86 μM for 4 and 20.80 μM for 5). 3-propargylamine derivatives 4 and 5 oriented properly in the active site of MAO-B due to its elongated nature

Rullo et al. [52]

(continued)

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

337

Table 2 (continued) S. no. Inhibitors

Key findings

References

6.

Piperine derivatives

Dhiman Compounds 6 (IC50 = 15.38 ± 0.071 μM) and et al. 7 (IC50 = 16.11 ± 0.091 μM) were reported as [58] potent inhibitors of MAO-A. They displayed docking score of -9.72 and -7.98, respectively, which are consistent with the in vitro results. Compound 9 displayed favorable binding with the active site residues MAO-B with docking score of -11.76. Compound 10 displayed π–π stacking interactions with Tyr398 of MAO-B. The 2-methoxyphenyl unit of compound 10 was situated close to FAD domain and the “aromatic cage” formed by Phe103, Trp119, Phe168, and the aromatic ring

7.

Tau PET tracers

PET tracers bind to MAO-B in the same binding site Murugan et al. with similar binding affinity as that of safinamide. [59] Recently developed tau tracers JNJ-311, MK-6240, and PI-2620 displayed lowest relative affinity for MAO-B

8.

4,6-diphenylpyrimidine derivatives, VB1-VB15

VB1 inhibited MAO-A, BuChE, and AChE activity Kumar with IC50 values of 18.34 ± 0.38 nM, 0.666 et al. [60] ± 0.03 μM, and 30.46 ± 0.23 nM, respectively. The analysis of the docking poses indicated differences in the conformation of the propargyl group in VB8 as compared to VB1 and VB3, which is consistent with the observed lower activity of VB8 against MAO-A. MD simulations depicted no major structural changes in MAO-A on the incorporation of VB1, and the orientation of pyrimidine moiety in the active site of MAO-A was identical to that obtained in the docking studies

9.

N-propargyl-substituted diphenylpyrimidine derivatives

Kumar VP1 exhibited good inhibitory activity against et al. AChE (IC50 = 7.21 ± 0.15 μM) and MAO-B [61] (IC50 = 0.04 ± 0.002 μM) and displayed selectivity over MAO-A. The docked complexes of MAO-B–VP1 and MAO-B–V2 depicted interactions of the pyrimidine moiety of VP1 and VP2 with gate residue, Ile199 of MAO-B, which resulted in the open-gate conformation of MAO-B. Molecular docking analysis highlighted that both VP1 and VP2 blocked substrate as well as entrance cavity of MAO-B. Further, MD simulations predicted the structural stability of the MAO-B–VP1 and MAO-B–VP2 complexes (continued)

338

Gurmeet Kaur et al.

Table 2 (continued) S. no. Inhibitors

Key findings

10.

Denya Molecular docking results predicted compounds et al. 15–18 lacked interactions with active site residues [62] of MAO-A. The indole ring of 15–18 was positioned toward FAD in the substrate cavity along with diethylcarbamoyl/urea moiety accommodated in the entrance cavity. The docked poses of compound 11–14 with MAO-A depict interactions with active site residues along with the orientation of propargylamine moiety in close proximity to FAD. The enhanced stability and good ADMET qualities of 12 along with a IC50 value of 4.31 μM and 2.62 μM for MAO-A and MAO-B, respectively, render compound 12 as a potential therapeutic candidate against MAOs

Indole derivatives

References

NPA0012149, and NPA0015149 form stable complexes with MAO-A and thus can be used as potential MAO-A inhibitors. To check the binding specificity of the shortlisted compounds, i.e., NPA000487 (20α-methyltetrahymanol), NPA012149 (Citreamicin δ), NPA001637 (C35-terpenes), NPA015149 (Ganocin A), NPA007129 (MS-347a), and NPA018844 (Preussomerin EG1), the binding potential of these compounds to MAO-B (PDB IF: 1GOS) was evaluated using molecular docking and molecular dynamics (MD) simulations. The molecular docking analyses highlighted favorable binding of NPA015149 (-6.97 kcal/mol), NPA018844 (-8.06 kcal/mol), and NPA007129 (-8.7 kcal/ mol) with MAO-B, whereas NPA000487, NPA001637, and NPA012149 displayed unfavorable binding to MAO-B with binding energies of +12.2, +2.10, and +8.07 kcal/mol, respectively. The MD simulations predicted lower stability of complexes of shortlisted compounds with MAO-B as compared to MAO-A. The study depicted the identification of potential MAO inhibitors from microbial sources and the selectivity of the shortlisted compounds toward MAO-A using computational techniques. In another study, Iqbal et al. screened a total of 44 natural brain-derived neurotrophic factor (BDNF) or abrineurin inducer compounds by the assessment of toxicity and physicochemical properties [36]. Among screened compounds [Alpinetin, Calycosin, Huperzine A, Hydroxytyrosol, and Soyasapogenol B (Soy-B)], a pentacyclic triterpenoid, i.e., Soy-B, displayed potent inhibitory activity against various targets [AChE, butyrylcholinesterase (BuChE), MAO-A, MAO-B, GSK3β, and NMDA] of AD. Soy-B relates to soyasaponins that is isolated from Glycine max and exhibited anti-inflammatory [37], phytooestrogenic [38], and memory

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

339

enhancement properties [39, 40]. Soy-B displayed no Lipinski rule violation, high absorption (82.6%), low topological polar surface area (TPSA) (60.69), and no toxicity. Soy-B effectively binds to all targeted proteins [AChE, BuChE, glycogen synthase kinase 3 beta (GSK3β), MAO-A, MAO-B, and N-methyl-D-aspartate (NMDA)]. Notably, Soy-B displayed binding with the residues of the active site of the targeted proteins identical to the native ligand inhibitor. The MD simulations depicted stable complexes of Soy-B with all targeted proteins. The RMSF and secondary structure analysis highlighted the structural stability of Soy-B–protein complexes. Soy-B displayed a strong interaction with Cys323, Thr336, and Phe352 of MAO-A (PDB ID: 2Z5X), whereas Leu171, Ile198, and Tyr398 of MAO-B (PDB ID: 2V5Z) participated in the interaction with Soy-B. Notably, Soy-B displayed interaction with Cys323 of MAO-A, which is consistent with studies that highlight the interaction of the thiol group of Cys323 with piperine and β-carboline analogs [41, 42]. Remarkably, Soy-B interacted with key residues (Leu171, Ile198, and Tyr398) of MAO-B. Previous studies reported the interaction of Desmodeleganine (a natural alkaloid) with Leu171 and Ile198 of MAO-B [43] and derivatives of natural compounds with the aromatic cage of MAO-B through Tyr398 [44, 45]. The molecular docking and MD analyses highlight Soy-B as a potential multitarget therapeutic agent against AD. In 2021, Prajapati et al. investigated the inhibition potential of glycitin and glycitein against MAO, β-site amyloid precursor protein cleaving enzyme 1 (BACE1, β-secretase), AChE, and BuChE using in vitro enzyme assays [46]. Glycitin, a major isoflavone glucoside extracted from Pueraria lobata (P. lobata), can be converted to bioactive glycitein by human intestinal bacteria [47, 48]. Glycitein exhibited potent inhibitory activity against MAO-A (IC50 = 8.30 ± 0.77 μM). Further, glycitein inhibited AChE, BACE1, and BuChE with IC50 values of 142.98 ± 10.02, 59.46 ± 3.94, and 69.40 ± 3.24 μM, respectively. The 6-methoxy group of chromen-4-one moiety of glycitein displayed hydrophobic contacts with key catalytic site residues Tyr407, Tyr444, and FAD600 and exhibited interaction with Phe208 and Ile335 by nonionic interactions with the 3-phenyl ring. The docking studies highlighted that glycitein interacted with major catalytic sites, including Tyr 407 and Tyr444, along with cofactor FAD, of MAO-A (PDB ID: 2Z5X), which is similar to the interactions displayed by selective and reversible MAO-A inhibitor harmine. Glycitein interacted with MAO-B (PDB ID: 2BYB) catalytic site residues Leu171, Leu199, Ile316, Tyr326, and Phe343 by hydrophobic contacts and with Cys172, FAD600 through hydrogen bonds. The chromen-4-one moiety interacted with Leu171, Cys172, Leu199, and Tyr326 situated at the entrance cavity of MAO-B, thereby preventing the substrate access to the active

340

Gurmeet Kaur et al.

substrate cavity [14a]. Enzyme kinetics and molecular docking analysis highlighted that glycitein inhibited MAO-A more effectively as compared to MAO-B. The study identified glycitein as a potent MAO-A inhibitor along with the inhibition potential against BACE1 and cholinesterase enzymes. In another study, Prajapati et al. reexamined the inhibitory potential of isoliquiritigenin (= 4,2′,4′-trihydroxychalcone) (ILG) against MAO in vitro and elucidated the inhibitory mechanism using molecular docking analysis and kinetics study [49]. ILG exhibited competitive inhibition of MAO-A (IC50 = 0.68 μM) and mixed inhibition of MAO-B (IC50 = 0.33 μM). ILG binds strongly (-7.44 kcal/mol) with the catalytic site residues of MAO-A (PDB ID: 2BXR) as compared to harmine (-6.46 kcal/ mol). The 2′,4′-dihydroxyphenyl moiety of ILG displayed a π–π T-shaped hydrophobic contact with Tyr407 and hydrogen bonds with Tyr444 and FAD of MAO-A. A carbonyl group at C1 displayed hydrogen bond interaction with FAD. Importantly, the aromatic ring of 4″-hydrophenyl was involved in hydrophobic contacts with active site residues Phe208, Ile335, and Leu337, through π–σ and π–π stack bonds. A hydrogen bond between 4″-OH and Thr336 was observed in the MAO-A-ILG complex. The molecular docking analysis predicted favorable binding between MAO-B (PDB ID: 2V60) and ILG (-8.69 kcal/mol). The 4″-hydroxyphenyl moiety of ILG displayed interactions with Tyr398 and FAD, while 2′,4′-dihydroxyphenyl moiety showed interactions with the gating residues Leu171 and Ile199. The 2′,4′-dihydroxyphenyl moiety and 1-CO group of ILG are involved in π–sulfur and hydrogen bond interactions with Cys172 present in the hydrophobic cavity. Additionally, ILG interacted with allosteric site residues of MAO-B with a binding energy of -8.03 kcal/mol. ILG displayed hydrogen bond interactions with Pro104, His115, Asn116, and Asp123 and hydrophobic contacts with Val106, Trp119, and Arg120 at the allosteric site. The multi-target nature along with the appropriate pharmacokinetics and toxicity profile of ILG renders this flavonoid a potential candidate for various neurodegenerative diseases.

3 Computational Insights into the Key Interactions of Synthetic Inhibitors with MAOs In 2022, Mathew et al. synthesized and evaluated 15 halogenbearing multi-conjugated dienones against MAO-A, MAO-B, AChE, BuChE, and BACE1 enzymes [50]. Among the synthesized compounds, MK6 (IC50 = 2.82 nM) and MK12 (IC50 = 3.22 nM) exhibited significant inhibitory activity against MAO-B. MK5, MK6, MK12, and MK14 displayed low toxicities on Vero cells,

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

341

with IC50 values of 218.4, 149.1, 99.96, and 162.3 μg/mL, respectively. Moreover, MK5, MK6, MK12, and MK14 lowered H2O2-associated cell damage by scavenging ROS. The molecular mechanics Poisson-Boltzmann surface area (MM-PBSA) calculations depicted favorable binding of MK5, MK6, MK12, and MK14 to MAO-B (PDB ID: 6RKB) with binding free energies of -40.86, -43.47, -41.69, and -38.92 kcal/ mol, respectively. MK5 displayed interactions with Leu171, Ile198, Ile199, Tyr398, and Tyr435 of MAO-B, whereas Leu171, Tyr188, Ile199, Tyr398, and Tyr435 contributed significantly in the binding of MK6 with MAO-B. MK12 displayed π-sulfur interaction, halogen interaction, and π-π stacked-T-shaped interactions with residues of MAO-B. MK14 bind strongly to Cys172, Tyr188, Ile198, Gln206, Tyr398, and Tyr435 of MAO-B. The binding free energy analysis suggested strong interactions of MK5, MK6, MK12, and MK14 with specific residues of MAO-B, which resulted in the stabilization of MAO-B–inhibitor complexes and led to the inhibition of MAO-B activity. MD simulations highlighted that lead compounds, MK5, MK6, MK12, and MK14, inhibit MAO-B activity by distorting the conformational dynamics of MAO-B. The kinetic and reversibility studies highlighted MK6 and MK12 as reversible and competitive inhibitors of MAO-B. The study depicted MK6 and MK12 as potential therapeutic candidates against AD and PD. In another study, Khan et al. synthesized new S-benzyl dithiocarbamates (1–2) possessing a trifluoromethyl group and evaluated their inhibition potential against MAO-A and MAO-B [51]. Compound 1 with p-CF3 (IC50 = 16.02 ± 5.135 μg/mL, 1.284 ± 0.66 μg/mL) displayed higher inhibitory activity than compound 2 (IC50 = 18.37 ± 4.030 μM, 3.164 ± 0.456 μM) against MAO-A and MAO-B, respectively. Molecular docking predicted binding energy of -8.4438 kcal/mol for compound 1 and -8.3308 kcal/mol for compound 2 with MAO-A (PDB ID: 2Z5X). The molecular docking analysis displayed π–sulfur interactions between tyrosine (Tyr69 and Tyr407) residues of MAO-A and the carbonothioyl group of compound 1. Halogen interactions were observed between the CF3 group in compound 1 and Arg51, Val65, and Gly66 of MAO-A. Compound 2 displayed π–sulfur interactions with Tyr69 and Phe352, π–π interaction with Tyr407, and halogen interactions with Gly443 of MAO-A. Compounds 1 (-9.9463 kcal/mol) and 2 (-9.4588 kcal/mol) bind with more affinity to MAO-B (PDB ID: 2V5Z) as compared to MAO-A. Compound 1 displayed π–sulfur interaction with Cys172, hydrogen bond interaction with Ser59, π–π stacking interactions with Tyr326, and π–sigma type of interaction with Tyr435 of MAO-B. Compound 2 displayed π–sulfur interactions with Tyr60, Tyr398, and Tyr435, hydrogen bond interaction with Gln206, and halogen interaction with Ile198 of MAO-B. The

342

Gurmeet Kaur et al.

results of docking studies display consistency with in vitro results, which depicted higher inhibitory activity of compound 1 against MAO-A and MAO-B as compared to 2. In 2022, Rullo et al. highlighted the coumarin derivative 3 as a water-soluble, orally bioavailable central nervous system (CNS) permeant potent dual inhibitor of AChE (IC50 = 550 nM) and MAO-B (IC50 = 8.2 nM, B/A selectivity >1200) [52]. The ethylsubstituted 4 displayed higher MAO-B inhibitory activity as compared to iPr-derivative 5 (IC50 = 73 and 350 nM, respectively); however, both derivatives were noted to be less active than parent racemate (±)-3. The strongly lipophilic derivative 6 inhibited MAO-B, AChE, and BChE with IC50 values of 132, 561, and 430 nM, respectively, and exhibited good B/A selectivity [selectivity index (SI) > 73]. The para-substituted derivative 7 displayed a better B/A selectivity than 8 due to lower inhibitory activity against MAO-A and higher potency toward MAO-B. Restoring -CH2OH at coumarin C4 yielded the most active MAO-B inhibitor (3, IC50 = 8.2 nM) with notable B/A selectivity (SI > 1250). The molecular docking predicted favorable binding of compound 3 with MAO-B (PDB ID: 2V5Z) with a binding energy of 10.88 kcal/mol. The binding pose of compound 3 with MAO-B depicted that the inhibitor is fully buried inside the enzymatic cavity. Compound 3 displayed π-π interaction with Tyr398 and was placed near FAD. Additionally, compound 3 displayed interaction with Tyr188 via a bidentate hydrogen bonding with phenolic OH and the carbonyl group of Cys172. The steric clashes with gating residue Ile199 were avoided due to the flipping of the xylyl linker upon interacting with Tyr326. In another study, Jia et al. synthesized functionalized pyridoxines by click chemistry and evaluated their inhibition potential against AChE, BuChE, MAO-A, and MAO-B [53]. Among the synthesized compounds, 4k5 displayed potent inhibitory activity against AChE (IC50 = 0.0816 ± 0.075 μM) and MAO-B (IC50 = 0.039 ± 0.003 μM). The molecular docking poses depicted that the phenyl ring of 4k5 fits into the enzymatic “aromatic cage” formed by FAD, Tyr398, and Tyr435 and displays π–π stacking interactions with Tyr398 and Tyr435 of MAO-B (PDB ID: 2BYB). Importantly, 4k5 displayed hydrophobic contacts with Leu167, Leu171, Ile199, Gln206, and Ile316. In the case of MAO-A (PDB ID: 2BXR), the phenyl ring of 4k5 was situated below the “aromatic cage” formed by FAD, Tyr407, and Tyr444. Notably, 4k5 didn’t display any π–π stacking interaction with MAO-A. In 2021, Mzezewa et al. synthesized a series of 3,7-substituted coumarin derivatives and evaluated their inhibitory activity against MAOs and cholinesterase enzymes [54]. Among the synthesized derivatives, compounds 9 and 10 displayed good potential as multi-target directed ligands (MTDLs) due to the presence of the

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

343

propargylamine moiety in their structures. The benzyloxy moiety has been reported for the enhanced potency and selectivity of the coumarin scaffold against MAO-B [55–57]. The initial substitutions to position 7 resulted in higher inhibitory activity and MAOB selectivity of the coumarin scaffold. In comparison to the carbamate moiety, the phenylethyloxy moiety afforded higher inhibition and selectivity. The incorporation of the propargylamine moiety led to the increased potency and selectivity of the derivatives against MAO-B. 3-propargylamine derivatives 9 and 10 exhibited 130 and 206 times higher selectivity to MAO-B as compared to MAO-A. Together, these results show that whereas compound 11 was more active in the inhibition of MAO-B, compounds 9 and 10 showed better selectivity, and this can be attributed to the addition of the propargylamine moiety. The inclusion of the propargylamine moiety yielded compounds that are unable to fit into MAO-A (PDB ID: 2BXS) active site. The molecular docking analysis predicted that compounds 9 and 10 lack the flexibility to fit and interact with the residues in the cavity’s active site, which is consistent with the relatively poor MAO-A inhibitory potential (IC50 value of 3.86 μM for 9 and 20.80 μM for 10). In contrast to MAO-A, the elongated nature of MAO-B (PDB ID: 2V5Z) active site permits for improved accommodation of the 3-propargylamine derivatives 9 and 10. Both compounds interacted with the FAD cofactor due to the addition of propargylamine moiety. The FAD cofactor is required for substrate catalysis, and thus designed inhibitors that are in close proximity or bind to FAD will inhibit MAO activity [16a]. Compound 9 displayed π–H bonding with Cys172, and the propargylamine moiety lie close to FAD. The propargylamine moiety in compound 10 displayed π–H bonds with FAD. The low nanomolar IC50 (29 nM for 9 and 101 nM for 10) and high SI values displayed 9 and 10, highlighting the critical role of propargylamine moiety for its ability to yield highly selective and potent inhibitors of MAO-B. In 2020, Dhiman et al. synthesized compounds similar to piperine and evaluated their inhibitory activity against MAO-A and MAO-B and assessed their free radical scavenging activity [58]. Compounds 12 (IC50 = 15.38 ± 0.071 μM) and 7 (IC50 = 16.11 ± 0.091 μM) displayed potent inhibitory activity against MAO-A (PDB ID: 2Z5X). The docking scores of compounds 12 (-9.72) and 13 (-7.98) are consistent with in vitro studies. The docking pose of MAO-A–13 depicted that the methoxyphenyl unit of 13 was situated in the “aromatic cage” outlined by Tyr197, Tyr407, and Tyr444. In the case of MAO-B (PDB ID: 2V5Z), compounds 14 and 15 were noted to be potent MAO-B inhibitors along with good selectivity in comparison to piperine. Compound 15 displayed favorable binding with the active site residues MAO-B with a

344

Gurmeet Kaur et al.

docking score of -11.76. The nitrosonaphthalen-2-yl ring displayed π–π stacking interactions with Tyr326, whereas the ester carbonyl group formed a hydrogen bond with Tyr326. Compound 14 displayed hydrogen bond interactions through the 2,4-dienamide group of 14. A hydrogen bond was observed between NH of 14 and Pro102 (OH), and another hydrogen bond interaction was noted between the carbonyl oxygen and Tyr326 (C=O) of MAO-B. Compound 16 displayed π–π stacking interactions with Tyr398 of MAO-B. The 2-methoxyphenyl unit of compound 16 was situated close to the FAD domain and the “aromatic cage” formed by Phe168, Trp119, Phe103, and the aromatic ring. The “gatekeeper” residue Ile199 and other residues like Cys172 and Ile198 support the side chain consisting of conjugated double bonds. The methylenedioxyphenyl group displayed hydrophobic contacts with Pro102, Phe103, Pro104, Trp119, Leu164, Phe168, and Ile316. The 4-formyl-2-methoxyphenyl part of 16 was noted to be sandwiched between Tyr188 and Phe343. In 2019, Murugan et al. performed molecular docking studies to identify binding sites of tau tracers to MAO-B (PDB ID: 2V5Z) [59]. The molecular docking analysis predicted that all tau positron emission tomography (PET) tracers bind to MAO-B in the same binding site with similar binding affinity as that of safinamide. Recently developed tau tracers JNJ-311, MK-6240, and PI-2620 displayed lowest relative affinity for MAO-B. In 2019, Kumar et al. reported new 4,6-diphenylpyrimidine derivatives possessing propargyl moiety and evaluated their inhibitory activity against MAO and AChE [60]. VB1 inhibited MAO-A, BuChE, and AChE activity with IC50 values of 18.34 ± 0.38 nM, 0.666 ± 0.03 μM, and 30.46 ± 0.23 nM, respectively. VB8 exhibited potent inhibitory activity against AChE (IC50 = 9.54 ± 0.07 nM) and MAO-A (IC50 = 1010 ± 70.42 nM). The molecular docking analysis depicted that VB1, VB3, and VB8 occupied the active site of MAO-A (PDB IDs: 2BXR and 2Z5X) and displayed hydrophobic, π-π stacking interactions with Tyr69, Phe208, Arg296, Ile335, Leu337, Phe352, Tyr407, Trp441, Tyr444, and FAD of MAO-A. The analysis of the docking poses indicated differences in the conformation of the propargyl group in VB8 as compared to VB1 and VB3, which is consistent with the observed lower activity of VB8 against MAO-A. In the case of VB1 and VB3, the propargyl group was directed toward the inner side of the cavity of MAO-A. In contrast, the propargyl group in VB8 was situated toward outer side of the cavity of MAO-A. The MD simulations depicted no major structural changes in MAO-A on the incorporation of VB1, and the orientation of pyrimidine moiety in the active site of MAO-A was identical to that obtained in the docking studies. VB1 displayed hydrogen bond interactions between NH2 group of pyrimidine fragment and Tyr407 and π-π stacking interactions

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

345

with Tyr444. The nitrogen atom of the morpholine ring was involved in the hydrogen bond interactions with Pro72 and Arg206 by water bridges. The MD simulations demonstrated the structural integrity of the MAO-A–VB1 complex. The study highlighted 4,6-diphenylpyrimidine derivatives as inhibitors of AChE and MAO-A. Later in 2022, the authors reported the design, synthesis, and evaluation of N-propargyl-substituted diphenylpyrimidine derivatives as MTDLs against AChE/BuChE and MAOs [61]. Among synthesized derivatives, VP1 exhibited good inhibitory activity against MAO-B (IC50 = 0.04 ± 0.002 μM), displayed selectivity over MAO-A, and also inhibited AChE activity with an IC50 value of 7.21 ± 0.15 μM. Compound VP1 possessing R as the morpholine and R1 as the -NH2 and propargylamine groups at the para position of the aromatic ring was noted to be the most potent MAO-B inhibitor (IC50 = 0.04 ± 0.002 μM) and displayed 248-fold selectivity over MAO-A. The morpholine and piperidine substituted compounds displayed high selectivity toward MAO-B, with IC50 values in the nanomolar to sub-micromolar range, whereas compounds possessing N,N-dimethyl and pyrrolidine substituents displayed similar activity against MAO-A and MAO-B. It was notable that most of the compounds exhibited higher selectivity against MAO-A as compared to MAO-B on the replacement of morpholine and piperidine with N,N-dimethyl and pyrrolidine moieties. The docked complexes of MAO-B–VP1 and MAO-B–VP2 depicted interactions of the pyrimidine moiety of VP1 and VP2 with gate residue, Ile199, of MAO-B (PDB ID: 2BYB), which resulted in the open-gate conformation of MAO-B. The 4-(2-phenoxyethyl)morpholine fragment of VP1 occupied the substrate cavity of MAO-B, whereas propargylamine-attached ring was directed to the entrance cavity. A cation-π interaction was observed between the protonated nitrogen atom of morpholine and Tyr326 of MAO-B. The NH2 group of VP1 was involved in the hydrogen bond interaction with Pro104. The ring attached to the propargylamine group and pyrimidine of VP1 was involved in the π-π stacking interactions with Phe103 and Trp119, respectively. In MAO-B–VP2 complex, the propargylamine-attached ring occupied the substrate cavity of MAO-B, and morpholine-containing ring was directed to the entrance cavity. The propargylamine group-containing ring of VP2 displayed π-π stacking interactions, and the other phenyl ring showed interactions with Trp119. The pyrimidine moiety in VP2 displayed interactions with the Ile199 of MAO-B. Molecular docking analysis highlighted that both VP1 and VP2 blocked the substrate as well as entrance cavity of MAO-B. Further, MD simulations predicted the structural stability of the MAO-B–VP1 and MAO-B–VP2 complexes. RMSD ranges

346

Gurmeet Kaur et al.

from 0.2 to 0.3 Å for MAO-B–VP1 and 0.2 to 0.4 Å for MAO-B– VP2. Notably, the pyrimidine moiety displayed a similar orientation in the active site of MAO-B as seen in the molecular docking. Overall, the study highlighted that the N-propargyl-substituted diphenylpyrimidines can be developed as potential MTDLs against AD. In 2018, Denya et al. designed and prepared a series of indole derivatives as inhibitors against MAO-A, MAO-B, eeAChE (ee, electric eel), and eqBuChE (eq, equine serum) to overcome the limitations of pharmacokinetic properties experienced in the case of a structurally similar compound, ladostigil [62]. The synthesized compounds possess propargylamine moiety (reported for its role as neuroprotector and inhibition of MAO activity) at the N1 position and a diethylcarbamate/urea moiety (known for its role in the inhibition of ChE) at the 5 or 6 position of the indole ring. Among synthesized indole derivatives, compound 18 displayed potent inhibitory activity against MAO-A (IC50 = 4.31 μM) and MAO-B (IC50 = 2.62 μM). As compared to ladostigil, compound 18 was found to be chemically as well as metabolically stable and displayed better enzyme inhibition in vitro and in vivo. Molecular docking was performed to elucidate the binding interactions of indole derivatives with MAO-A (PDB ID: 2BXS) and MAO-B (PDB ID: 2V5Z) using Molecular Operating Environment (MOE) software. The docking results displayed that compounds 21–24 lacked interactions with active site residues of MAO-A. The indole ring of 21–24 was positioned toward FAD in the substrate domain along with diethylcarbamoyl/urea moiety accommodated in the entrance cavity. The docked poses of compounds 17–20 with MAO-A depict interactions with active site residues along with the orientation of propargylamine moiety in close proximity to FAD, which is required to inhibit MAO activity [63]. Compounds 21–24 displayed interactions with the residues of the entrance cavity of MAO-B and lacked interactions with the residues of the substrate cavity, which indicate poor MAO-B activity of these compounds. In contrast, compounds 17–20 possessing propargylamine moiety displayed interactions with the substrate cavity of MAO-B. The 6-substituted indole derivatives 18 and 20 exhibited potent inhibition against MAO-B due to the π–H interactions between propargylamine moiety and FAD, whereas compounds 17 and 19 lacked π–H interactions due to the inverted orientation of the compounds. Although 18 and 20 were able to cross the BBB, 18 depicted higher BBB permeability as compared to 20 and ladostigil. Enhanced stability and good ADMET qualities of 18 along with a IC50 value of 4.31 μM and 2.62 μM for MAO-A and MAO-B, respectively, render compound 18 as a potential therapeutic candidate against MAOs.

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

4

347

Conclusions and Future Perspectives Neuropsychiatric disorders (NDs) such as AD, PD, and depression are responsible for the huge monetary and psychological burden on the affected persons. NDs are encouraged by various factors that are closely related to MAO. Numerous inhibitors have been designed against MAOs to inhibit its activity. Selegiline, an irreversible selective MAO-B inhibitor, depicted positive effects on cognitive functions in AD patients [64]; nevertheless, its therapeutic potential remains controversial due to laxity in clinical trials. Another, MAO-B inhibitor (lazabemide) showed a 20–40% reduction in cognitive decline as compared to the control in the phase 2 clinical trial [65]. Food and Drug Administration (FDA)-approved MAO inhibitors are available in the market; however, most of them are irreversible MAO inhibitors (selegiline, rasagiline, ladostigil, selective for MAO-B; tranylcypromine, non-selective MAO inhibitor) and have many side effects such as hepatotoxicity, cheese reaction, hypertension crisis, etc. [33, 66] Two FDA-approved reversible inhibitors, moclobemide and safinamide, possess selectivity for MAO-A and MAO-B, respectively [33]. A reversible MAO-B inhibitor, sembragiline, has failed in clinical trials due to less clinical efficacy and left AD patients discouraged [67]. In the last few years, researchers have shifted their attention from irreversible MAO inhibitors to reversible and less toxic MAO inhibitors. Computational studies on the MAO–inhibitor complexes provided detailed interaction of the inhibitors with the active site residues of MAOs, which, in turn, will be beneficial for the design and development of more potent MAOs inhibitors as antiAlzheimer drugs. Notably, studies on MAO–inhibitor complexes have greatly contributed to the understanding of aminergic neurotransmission and confirmed that MAO has a key role in the development and function of brain. In this chapter, an attempt has been made to review the key interactions of the synthetic and natural inhibitors with the substrate binding pocket and active site of MAOs. Among various synthesized and natural inhibitors of MAOs, few compounds have displayed potent inhibitory activity against MAOs. Computational studies have provided insights into the key interactions responsible for the observed potent in vitro activity of MAOs inhibitors. The halogen-bearing multi-conjugated dienones MK6 and MK12 exhibited significant inhibitory activity against MAO-B with an IC50 value of 2.82 nM and 3.22 nM, respectively [50]. These compounds possess huge potential as therapeutic leads against MAO activity. The residue-wise binding free energy analysis highlighted π–sulfur interaction, halogen interaction, and π–π stacked– T-shaped interactions between MK12 and MAO-B residues. MD simulations highlighted that MK6 and MK12 inhibit MAO-B

348

Gurmeet Kaur et al.

activity by distorting the conformational dynamics of MAO-B. Another compound coumarin derivative 3 possessing a -CF2H motif emerged as a water-soluble, orally bioavailable CNS-permeant potent dual inhibitor of AChE (IC50 = 550 nM) and MAO-B (IC50 = 8.2 nM, B/A selectivity >1200) [52]. Modelling studies depicted that the inhibitor is fully buried inside the enzymatic cavity of MAO-B. Compound 3 displayed π-π interaction with Tyr398 and is placed near to FAD. Additionally, compound 3 displayed interaction with Tyr188 via a bidentate hydrogen bonding with phenolic OH and carbonyl group of Cys172. Molecular docking and MD simulation studies of synthetic and natural inhibitors of MAOs significantly enriched our understanding of the key interactions between MAOs and inhibitors. The molecular modelling predicted that the addition of the propargylamine moiety in the 3,7-substituted coumarin derivatives resulted in compounds, which are unable to fit into MAO-A active site [54]. Molecular docking analysis predicted that compounds 9 and 10 lack the flexibility to fit and interact with the residues in the cavity’s active site, which is consistent with the low MAO-A inhibitory potential of compounds 9 and 10 (IC50 value of 3.86 μM for 9 and 20.80 μM for 10). The elongated active site in MAO-B permits better positioning of the 3-propargylamine derivatives [54]. The low nanomolar IC50 (29 nM for 9 and 101 nM for 10) and high SI values displayed in 9 and 10 highlight the critical role of propargylamine moiety for its ability to yield highly selective and potent inhibitors of MAO-B. The coumarin scaffold displayed promising activity against MAO due to its lipophilic character and structural flexibility to position itself in the active site of MAO. With the advances in the computational tools and techniques, a ligand-based virtual screening approach by employing coumarin derivative as reference compound can be utilized to screen various small-molecule databases to identify potential inhibitors of MAO. Structure–activity relationship (SAR) studies highlighted the presence of substituents at specific positions in the MAO inhibitors resulted in the higher selectivity toward MAO-B as compared to MAO-A. Various inhibitors have been designed to achieve high B/A selectivity to circumvent well-known side effects that are linked to the inhibition of peripheral MAO-A, termed “cheese effect” [68]. Compound VP1 possessing R as the morpholine and R1 as the -NH2 and propargylamine groups at the para position of the aromatic ring was noted to be the most potent MAO-B inhibitor (IC50 = 0.04 ± 0.002 μM) and displayed 248-fold selectivity over MAO-A. An (S)-2-(benzylamino)propanamide derivative possessing a chiral azacyclic amide moiety displayed highest in vitro activity (IC50 = 0.021 ± 0.002 μM) and promising selectivity (SI between MAO-A and MAO-B was 1227-fold), which was notably higher than safinamide (SI was 268-fold) [30]. The analysis

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents

349

of binding mode with molecular docking highlighted higher selectivity of (S)-2-(benzylamino)propanamide derivative toward MAO-B as the derivative displayed the steric clash with Phe208 of MAO-A. For future perspective, the introduction of hydrophobic moiety in the selected scaffolds will lead to better interaction with the entrance hydrophobic region enclosed with aromatic and aliphatic residues of the MAO cavity. The availability of high-resolution crystal structure of MAO isoforms and advanced computational approaches will greatly contribute in the rational design and development of new MAO inhibitors. In addition, insights from computational studies on MAO–inhibitor complexes will be beneficial for the design and development of highly efficacious and selective MAO inhibitors as anti-Alzheimer agents in the near future. References 1. (a) Buckner RL (2004) Memory and executive function in aging and AD: multiple factors that cause decline and reserve factors that compensate. Neuron 44(1):195–208; (b) Swart C, Haylett W, Kinnear C, Johnson G, Bardien S, Loos B (2014) Neurodegenerative disorders: dysregulation of a carefully maintained balance? Exp Gerontol 58:279–291 2. Gauthier S, Webster C, Servaes S, Morais JA, Rosa-Neto P (2022) World Alzheimer Report 2022: Life after diagnosis: navigating treatment, care and support. London, England, Alzheimer’s Disease International 3. Breijyeh Z, Karaman R (2020) Comprehensive review on Alzheimer’s disease: causes and treatment. Molecules 25(24):5789 4. (a) Barnham KJ, Masters CL, Bush AI (2004) Neurodegenerative diseases and oxidative stress. Nat Rev Drug Discov 3:205-214; (b) Nam E, Han J, Suh J-M, Yi Y, Lim MH (2018) Link of impaired metal ion homeostasis to mitochondrial dysfunction in neurons. Curr Opin Chem Biol 43:8-14; (c) Meraz-Rı´os MA, Toral-Rios D, Franco-Bocanegra D, Vil˜ a V (2013) leda-Herna´ndez J, Campos-Pen Inflammatory process in Alzheimer’s disease. Front Integr Neurosci 7:59 5. Cohen G, Kesler N (1999) Monoamine oxidase and mitochondrial respiration. J Neurochem 73(6):2310–2315 6. Youdim MB, Edmondson D, Tipton KF (2006) The therapeutic potential of monoamine oxidase inhibitors. Nat Rev Neurosci 7: 295–309 7. Cai Z (2014) Monoamine oxidase inhibitors: promising therapeutic agents for Alzheimer’s disease. Mol Med Rep 9:1533–1541

8. (a) Fowler CJ, KF Tipton (1984) On the substrate specificities of the two forms of monoamine oxidase. J Pharm Pharmacol 36(2):111–115; (b) Carradori S, Silvestri R (2015) New frontiers in selective human MAO-B inhibitors. J Med Chem 58(17):6717–6732 9. (a) Tripathi AC, Upadhyay S, Paliwal S, Saraf SK (2018) Privileged scaffolds as MAO inhibitors: retrospect and prospects. Eur J Med Chem 145:445–497; (b) Ramsay RR (2012) Monoamine oxidases: the biochemistry of the proteins as targets in medicinal chemistry and drug discovery. Curr Top Med Chem 12(20):2189–2209; (c) Yoshimoto M, Hirata M, Kagawa S, Magata Y, Ohmomo Y, Temma T (2019) Synthesis and characterization of novel radiofluorinated probes for positron emission tomography imaging of monoamine oxidase B. J Label Compd Radiopharm 62(9):580–587 10. Bortolato M, Chen K, Shih JC (2008) Monoamine oxidase inactivation: from pathophysiology to therapeutics. Adv Drug Deliv Rev 60(13–14):1527–1533 11. (a) Lan NC, Johnson DL, Abell CW, Bembenek ME, Kwan SW, Seeburg PH, Shih JC (1988) cDNA cloning of human liver monoamine oxidase A and B: molecular basis of differences in enzymatic properties. Proc Natl Acad Sci USA 85(13):4934–4938; (b) Chen K, Wu HF, Shih JC (2010) The deduced amino acid sequences of human platelet and frontal cortex monoamine oxidase B are identical. J Neurochem 61:187–190; (c) Kuwahara T, Takamoto S, Ito A (1990) Primary structure of rat monoamine oxidase

350

Gurmeet Kaur et al.

deduced from cDNA and its expression in rat tissues. Agric Biol Chem 54(1):253–257 12. Wouters J, Ramsay R, Goormaghtigh E, Ruysschaert JM, Brasseur R, Durant F (1995) Secondary structure of monoamine oxidase by FTIR spectroscopy. Biochem Biophys Res Commun 208(2):773–778 13. Wouters J (1998) Structural aspects of monoamine oxidase and its reversible inhibition. Curr Med Chem 5(2):137–162 14. (a) Binda C, Newton-Vinson P, Hubalek F, Edmondson DE, Mattevi A (2002) Structure of human monoamine oxidase B, a drug target for the treatment of neurological disorders. Nat Struct Biol 9:22–26; (b) De Colibus L, Li M, Binda C, Lustig A, Edmondson DE, Mattevi A (2005) Three-dimensional structure of human monoamine oxidase A (MAO A): relation to the structures of rat MAO A and human MAO B. Proc Natl Acad Sci USA 102(36):12684–12689 15. (a) Bradford MM, (1976) A rapid and sensitive method for the quantitation of microgram quantities of protein utilizing the principle of protein-dye binding. Anal Biochem 72(1–2):248–254; (b) Morris GM, Huey R, Lindstrom W, Sanner MF, Belew RK, Goodsell DS, Olson AJ (2010) AutoDock4 and AutoDockTools4: automated docking with selective receptor flexibility. J Comput Chem 30(16):2785–2791 16. Milczek EM, Binda C, Rovida S, Mattevi A, Edmondson DE (2011) The “gating” residues Ile199 and Tyr326 in human monoamine oxidase B function in substrate and inhibitor recognition. FEBS J 278(24):4860–4869 17. (a) Binda C, Li M, Huba´lek F, Restelli N, Edmondson DE, Mattevi A, (2003) Insights into the mode of inhibition of human mitochondrial monoamine oxidase B from highresolution crystal structures. Proc Natl Acad Sci USA 100(17):9750–9755; (b) Huba´lek F, Binda C, Khalil A, Li M, Mattevi A, Castagnoli N, Edmondson DE (2005) Demonstration of isoleucine 199 as a structural determinant for the selective inhibition of human monoamine oxidase B by specific reversible inhibitors. J Biol Chem 280(16):15761–15766 18. Kennedy B, Ziegler M, Alford M, Hansen L, Thal L, Masliah E (2003) Early and persistent alterations in prefrontal cortex MAO A and B in Alzheimer’s disease. J Neural Transm 110(7):789–801 19. Burke WJ, Li SW, Schmitt CA, Xia P, Chung HD, Gillespie KN (1999) Accumulation of

3, 4-dihydroxyphenylglycolaldehyde, the neurotoxic monoamine oxidase A metabolite of norepinephrine, in locus ceruleus cell bodies in Alzheimer’s disease: mechanism of neuron death. Brain Res 816(2):633–637 20. Schneier FR (2011) Pharmacotherapy of social anxiety disorder. Expert Opin Pharm 12(4):615–625 21. Carter SF, Scho¨ll M, Almkvist O, Wall A, Engler H, La˚ngstro¨m B, Nordberg A (2012) Evidence for astrocytosis in prodromal Alzheimer disease provided by 11C-deuterium-Ldeprenyl: a multitracer PET paradigm combining 11C-Pittsburgh compound B and 18F-FDG. J Nucl Med 53(1):37–46 22. Bar-Am O, Amit T, Weinreb O, Youdim MB, Mandel S (2010) Propargylamine containing compounds as modulators of proteolytic cleavage of amyloid protein precursor: involvement of MAPK and PKC activation. J Alzheimers Dis 21(2):361–371 23. Cochet M, Donneger R, Cassier E, Gaven F, Lichtenthaler SF, Marin P, Bockaert J, Dumuis A, Claeysen S (2013) 5-HT4 receptors constitutively promote the non-amyloidogenic pathway of APP cleavage and interact with ADAM10. ACS Chem Neurosci 4(1):130–140 24. Muck-Seler D, Presecki P, Mimica N, Mustapic M, Pivac N, Babic A, Nedic G, Folnegovic-Smalc V (2009) Platelet serotonin concentration and monoamine oxidase type B activity in female patients in early, middle and late phase of Alzheimer’s disease. Prog Neuro Psychopharmacol Biol Psychiatry 33(7):1226–1231 25. (a) Zheng H, Fridkin M, Youdim MB (2010) Site-activated chelators derived from antiParkinson drug rasagiline as a potential safer and more effective approach to the treatment of Alzheimer’s disease. Neurochem Res 35(12):2117–2123; (b) Weinreb O, Mandel S, Bar-Am O, Amit T (2011) Ironchelating backbone coupled with monoamine oxidase inhibitory moiety as novel pluripotential therapeutic agents for Alzheimer’s disease: a tribute to Moussa Youdim. J Neural Transm 118(3):479–492 26. Hong R, Li X (2019) Discovery of monoamine oxidase inhibitors by medicinal chemistry approaches. MedChemComm 10(1):10–25 27. Mathew B, Adeniyi AA, Dev S, Joy M, Ucar G, Mathew GE, Singh-Pillay A, Soliman ME (2017) Pharmacophore-based 3D-QSAR analysis of thienyl chalcones as a new class of human MAO-B inhibitors: investigation of combined

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents quantum chemical and molecular dynamics approach. J Phys Chem B 121(6):1186–1203 28. Ambure P, Bhat J, Puzyn T, Roy K (2018) Identifying natural compounds as multi-target directed ligands against Alzheimer’s disease: an in silico approach. J Biomol Struct Dyn 37(5):1282–1306 29. Is YS, Durdagi S, Aksoydan B, Yurtsever M (2018) Proposing novel MAO-B hit inhibitors using multidimensional molecular modeling approaches and application of binary QSAR models for prediction of their therapeutic activity, pharmacokinetic and toxicity properties. ACS Chem Neurosci 9(7):1768–1782 30. Jin C-F, Wang Z-Z, Chen K-Z, Xu T-F, Hao G-F (2020) Computational fragment-based design facilitates discovery of potent and selective monoamine oxidase-B (MAO-B) inhibitor. J Med Chem 63(23):15021–15036 31. Is YS, Aksoydan B, Senturk M, Yurtsever M, Durdagi S (2020) Integrated binary QSARdriven virtual screening and in vitro studies for finding novel hMAO-B-selective inhibitors. J Chem Inf Model 60(8):4047–4055 32. Mellado M, Gonza´lez C, Mella J, Aguilar LF, ˜a D, Uriarte E, Cuellar M, Matos MJ Vin (2021) Combined 3D-QSAR and docking analysis for the design and synthesis of chalcones as potent and selective monoamine oxidase B inhibitors. Bioorg Chem 108:104689 33. Koyiparambath VP, Prayaga Rajappan K, Rangarajan TM, Al-Sehemi AG, Pannipara M, Bhaskar V, Nair AS, Sudevan ST, Kumar S, Mathew B (2021) Deciphering the detailed structure-activity relationship of coumarins as monoamine oxidase enzyme inhibitors-an updated review. Chem Biol Drug Des 98(4):655–673 34. Fadaka AO, Taiwo OA, Dosumu OA, Owolabi OP, Ojo AB, Sibuyi NRS, Ullah S, Klein A, Madiehe AM, Meyer M, Ojo OA (2022) Computational prediction of potential druglike compounds from Cannabis sativa leaf extracts targeted towards Alzheimer therapy. J Mol Liq 360:119393 35. Jalal K, Khan K, Haleem DJ, Uddin R (2022) In silico study to identify new monoamine oxidase type a (MAO-A) selective inhibitors from natural source by virtual screening and molecular dynamics simulation. J Mol Struct 1254: 132244 36. Iqbal D, Rizvi SMD, Rehman MT, Khan MS, Bin Dukhyil A, AlAjmi MF, Alshehri BM, Banawas S, Zia Q, Alsaweed M, Madkhali Y, Alsagaby SA, Alturaiki W (2022) Soyasapogenol-B as a potential multitarget

351

therapeutic agent for neurodegenerative disorders: molecular docking and dynamics study. Entropy 24(5):593 37. Lee I-A, Park Y-J, Yeo H-K, Han MJ, Kim D-H (2010) Soyasaponin i attenuates TNBSinduced colitis in mice by inhibiting NF-KB pathway. J Agric Food Chem 58(20):10929–10934 38. Pan M, Li Z, Yeung V, Xu R-J (2010) Dietary supplementation of soy germ phytoestrogens or estradiol improves spatial memory performance and increases gene expression of BDNF, TrkB receptor and synaptic factors in ovariectomized rats. Nutr Metab 7:75 39. Ding J, Xi Y-D, Zhang D-D, Zhao X, Liu J-M, Li C-Q, Han J, Xiao R (2013) Soybean isoflavone ameliorates β-amyloid 1-42-induced learning and memory deficit in rats by protecting synaptic structure and sunction. Synapse 67(12):856–864 40. Hong S-W, Yoo D-H, Woo J-Y, Jeong J-J, Yang J-H, Kim D-H (2014) Soyasaponins Ab and Bb prevent scopolamine-induced memory impairment in mice without the inhibition of acetylcholinesterase. J Agric Food Chem 62(9):2062–2068 41. Rahman T, Rahmatullah M (2010) Proposed structural basis of interaction of piperine and related compounds with monoamine oxidases. Bioorg Med Chem Lett 20(2):537–540 42. Reniers J, Robert S, Frederick R, Masereel B, Vincent S, Wouters J (2011) Synthesis and evaluation of β-carboline derivatives as potential monoamine oxidase inhibitors. Bioorg Med Chem 19(1):134–144 43. Zhi K-K, Yang Z-D, Shi D-F, Yao X-J, Wang M-G (2014) Desmodeleganine, a new alkaloid from the leaves of Desmodium Elegans as a potential monoamine oxidase inhibitor. Fitoterapia 98:160–165 44. Dhiman P, Malik N, Khatkar A (2018) 3D-QSAR and in silico studies of natural products and related derivatives as monoamine oxidase inhibitors. Curr Neuropharmacol 16(6):881–900 45. Ramsay RR, Basile L, Maniquet A, Hagenow S, Pappalardo M, Saija MC, Bryant SD, Albreht A, Guccione S (2020) Parameters for irreversible inactivation of monoamine oxidase. Molecules 25(24):5908 46. Prajapati R, Park SE, Park HJ, Jung HA, Choi JS (2021) Identification of a potent and selective human monoamine oxidase-A inhibitor, glycitein, an isoflavone isolated from Pueraria

352

Gurmeet Kaur et al.

lobata flowers. ACS Food Sci Technol 1(4):538–550 47. Park J-S, Woo M-S, Kim D-H, Hyun J-W, Kim W-K, Lee J-C, Kim H-S (2007) Antiinflammatory mechanisms of isoflavone metabolites in lipopolysaccharide-stimulated microglial cells. J Pharmacol Exp Ther 320(3):1237–1245 48. Tsuchihashi R, Kodera M, Sakamoto S, Nakajima Y, Yamazaki T, Niiho Y, Nohara T, Kinjo J (2009) Microbial transformation and bioactivation of isoflavones from Pueraria flowers by human intestinal bacterial strains. J Nat Med 63(3):254–260 49. Prajapati R, Seong SH, Park SE, Paudel P, Jung HA, Choi JS (2021) Isoliquiritigenin, a potent human monoamine oxidase inhibitor, modulates dopamine D1, D3, and vasopressin V1A receptors. Sci Rep 11:23528 50. Mathew B, Oh JM, Abdelgawad MA, Khames A, Ghoneim MM, Kumar S, Nath LR, Sudevan ST, Parambi DGT, Agoni C, Soliman MES, Kim H (2022) Conjugated dienones from differently substituted cinnamaldehyde as highly potent monoamine oxidase-B inhibitors: synthesis, biochemistry, and computational chemistry. ACS Omega 7(9):8184–8197 51. Khan BA, Hamdani SS, Ahmed MN, Rashid U, Hameed S, Ibrahim MAA, Iqbal J, Granados CC, Macı´as MA (2022) Design, synthesis, crystal structures, computational studies, in vitro and in silico monoamine oxidase-A&B inhibitory activity of two novel S -benzyl dithiocarbamates. J Mol Struct 1265:133317 52. Rullo M, Cipolloni M, Catto M, Colliva C, Miniero DV, Latronico T, de Candia M, Benicchi T, Linusson A, Giacche` N, Altomare CD, Pisani L (2022) Probing fluorinated motifs onto dual AChE-MAO B inhibitors: rational design, synthesis, biological evaluation, and early-ADME studies. J Med Chem 65(5):3962–3977 53. Jia Z, Wen H, Huang S, Luo Y, Gao J, Wang R, Wan K, Xue W (2022) “Click” assembly of novel dual inhibitors of AChE and MAO-B from pyridoxine derivatives for the treatment of Alzheimer’s disease. Heterocycl Commun 28(2022):18–25 54. Mzezewa SC, Omoruyi SI, Zondagh LS, Malan SF, Ekpo OE, Joubert J (2021) Design, synthesis, and evaluation of 3,7-substituted coumarin derivatives as multifunctional

Alzheimer’s disease agents. J Enzyme Inhib Med Chem 36(1):1606–1620 55. Bru¨hlmann C, Ooms F, Carrupt PA, Testa B, Catto M, Leonetti F, Altomare C, Carotti A (2001) Coumarins derivatives as dual inhibitors of acetylcholinesterase and monoamine oxidase. J Med Chem 44(19):3195–3198 56. Joubert J, Foka G, Repsold B, Oliver DW, Kapp E, Malan SF (2017) Synthesis and evaluation of 7-substituted coumarin derivatives as multimodal monoamine oxidase-B and cholinesterase inhibitors for the treatment of Alzheimer’s disease. Euro J Med Chem 125:853–864 57. Pe´rez V, Marco JL, Ferna´ndez-Alvarez E, Unzeta M (1999) Relevance of benzyloxy group in 2-indolyl methylamines in the selective MAO-B inhibition. Br J Pharmacol 127(4):869–876 58. Dhiman P, Malik N, Khatkar A (2020) Natural based piperine derivatives as potent monoamine oxidase inhibitors: an in silico ADMET analysis and molecular docking studies. BMC Chem 14:12. https://doi.org/10.1186/ s13065-020-0661-0 59. Murugan NA, Chiotis K, Rodriguez-Vieitez E, Lemoine L, Ågren H, Nordberg A (2019) Cross-interaction of tau PET tracers with monoamine oxidase B: evidence from in silico modelling and in vivo imaging. Eur J Nucl Med Mol Imaging 46:1369–1382 60. Kumar B, Dwivedi AR, Sarkar B, Gupta SK, Krishnamurthy S, Mantha AK, Parkash J, Kumar V (2019) 4,6-Diphenylpyrimidine derivatives as dual inhibitors of monoamine oxidase and Acetylcholinesterase for the treatment of Alzheimer’s disease. ACS Chem Neurosci 10(1):252–265 61. Kumar B, Dwivedi AR, Arora T, Raj K, Prashar V, Kumar V, Singh S, Prakash J, Kumar V (2022) Design, synthesis, and pharmacological evaluation of N-Propargylated Diphenylpyrimidines as multitarget directed ligands for the treatment of Alzheimer’s disease. ACS Chem Neurosci 13(14):2122–2139 62. Denya I, Malan SF, Enogieru AB, Omoruyi SI, Ekpo OE, Kapp E, Zindo FT, Joubert J (2018) Design, synthesis and evaluation of indole derivatives as multifunctional agents against Alzheimer’s disease. Med Chem Commun 9: 357–370 63. Edmondson DE, Binda C, Mattevi A (2004) The FAD binding sites of human monoamine oxidases A and B. Neurotoxicology 25(1–2):63–72

Computational Modeling of MAO Inhibitors as Anti-Alzheimer Agents 64. Sano M, Ernesto C, Thomas RG, Klauber MR, Schafer K, Grundman M, Woodbury P, Growdon J, Cotman CW, Pfeiffer E, Schneider LS, Thal LJ (1997) A controlled trial of selegiline, alpha-tocopherol, or both as treatment for Alzheimer’s disease. N Engl J Med 336(17):1216–1222 65. Magni G, Meibach R (1999) Lazabemide for the long-term treatment of Alzheimer’s disease. Eur Neuropsychopharmacol 9:142 66. Cai Z (2014) Monoamine oxidase inhibitors: promising therapeutic agents for Alzheimer’s disease (review). Mol Med Rep 9(5):1533–1541

353

67. Park JH, Ju YH, Choi JW, Song HJ, Jang BK, Woo J, Chun H, Kim HJ, Shin SJ, Yarishkin O, Jo S, Park M, Yeon SK, Kim S, Kim J, Nam MH, Londhe AM, Kim J, Cho SJ, Cho S, Lee C, Hwang SY, Kim SW, Oh SJ, Cho J, Pae AN, Lee CJ, Park KD (2019) Newly developed reversible MAO-B inhibitor circumvents the shortcomings of irreversible inhibitors in Alzheimer’s disease. Sci Adv 5(3):eaav0316 68. Blackwell B, Mabbitt LA (1965) Tyramine in cheese related to hypertensive crises after monoamine-oxidase inhibition. Lancet 1(7392):938–940

Chapter 12 Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina Abstract Neurodegenerative diseases are pathological disorders inducing a gradual loss of neuronal functionality presenting a multifactorial character. Among them, Alzheimer’s disease (AD) causes the most well-known type of dementia and one of the major representatives. Due to the multifactorial etiology of AD, pleiotropic treatments are getting increasing importance. Phosphodiesterases (PDEs) are treated as molecular targets for many pathological situations. Neurodegenerative manifestations are among them. Regulation of the concentration of cAMP and/or cGMP is related to the inhibition of PDEs located in the human brain. In this chapter, we will discuss the results of computational modeling studies on PDE inhibitors as antiAlzheimer agents. Hydrogen bonds, π–π stacking, and volume are important for the interaction of the molecules with the catalytic site of PDEs. Key words Computational modeling studies, Phosphodiesterase inhibitors, Anti-Alzheimer agents

1

Introduction

1.1 Alzheimer’s Disease

Alzheimer’s disease (AD) is one of the major public health challenges worldwide and is considered to be the most common category of dementia, described by stepwise decrease and memory impairment, a deficit in language skills, and some neurodegenerative pathologies [1]. Data collected by the Alzheimer’s Association (USA) indicate that over 58 million people globally suffer from AD [2]. Neurodegeneration in AD and other neurodegenerative diseases is a pathological condition inducing an increasing loss of neuron functionality and appears to be multifactorial. However, several main pathological factors, such as accumulation of β- amyloids (Aβ), low concentrations of acetylcholine, nitric oxide (NO)/ soluble guanylyl cyclase (sGC), impaired homeostasis of biometals, inflammation, and oxidative stress, might be substantial in the development and progression of AD [3, 4].

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_12, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

355

356

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

Fig. 1 Drugs approved for AD therapy

Lipopolysaccharide (LPS) potently stimulates microglia producing pro-inflammatory cytokines in the brain [5, 6]. Accumulation of reactive oxygen and nitrogen species (ROS and RNS) detrimentally damages neurons inducing oxidative stress, which is caused by various underlying factors such as mitochondrial dysfunction or disturbance of metal ion homeostasis (e.g., redox-active Fe2+/Fe3+ and Cu+/Cu2+) [7]. Three acetylcholinesterase (AChE) inhibitors, donepezil [8], rivastigmine [9], galantamine [10], and one N-methyl-d-aspartate receptor antagonist, memantine [11], are the drugs (Fig. 1) currently used as AD’s therapeutics. However, the above drugs offer only moderate effects, and they do not delay the development of the neurodegenerative process. AD pathophysiology still is not well defined, and the molecular targets for the treatment of the disease are challenging for drug discovery. Cyclooxygenase-2 (COX-2), inducible nitric oxide synthase (iNOS), lipoxygenase (LOX), c-Jun N-terminal kinase 3 (JNK-3), and phosphodiesterases (PDEs) are referred to as promising AD targets. They have been related with the expression of anti-inflammatory mediators, neuroprotection, and ROS regulation [12–16]. Since AD is characterized by molecular complexity, pleiotropic therapeutic agents are of increasing importance, enhancing the therapeutic effect and decreasing the adverse effects associated with cocktail drugs [17]. 1.2 Phosphodiesterases and Their Role in AD

Phosphodiesterases (PDEs) are implicated in many diseases considering them therapeutic targets [16, 18]. They are the only known enzymes that hydrolyze cyclic nucleotides into adenosine monophosphate (AMP) or guanylate monophosphate (GMP). The second messengers, cyclic adenosine monophosphate (cAMP) and cyclic guanosine monophosphate (cGMP) are two such intracellular signaling targets with significant roles in intracellular signaling cascades, inflammation, smooth muscle contraction, steroid hormone function, and memory [18]. The PDE superfamily is coded by 21 identified genes and subdivided into 11 subtypes (PDE1–11), in relation to their structural and functional properties. Most of these families have more than one class of gene products (e.g., PDE4A, PDE4B, PDE4C,

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

357

PDE4D). In addition, each gene product may have multiple splice isoform variants (e.g., PDE4D1–PDE4D9). It has been estimated that more than 100 specific human PDEs exist. Due to the different expression of each subtype on the organs and tissues, PDE inhibitors have different specific therapeutic effects [16]. Among the 11 subtypes, PDEs 4, 7, and 8 specifically hydrolyze cAMP, whereas PDEs 5, 6, and 9 specifically hydrolyze cGMP. The other PDE families degrade both cGMP and cAMP. Since cAMP and cGMP are essential in cellular signaling and functionality, the increase of the cellular levels of cAMP and cGMP induced by PDE inhibitors regulates many biological processes [16, 18]. The decrease of cAMP may lower the concentration of cAMP-responsive element binding protein (CREB), which switches learning and memory, influencing the transcription of genes related to synaptic plasticity and survival, driving the loss of synaptic plasticity and memory. Nitric oxide activates cGMP via the NO/cGMP pathway, activating in continuation protein kinase G (PKG) and CREB phosphorylation, enhancing the level of the antiapoptotic protein Bcl-2. In preclinical and clinical studies, impaired CREB phosphorylation plays an important role in neurodegenerative diseases, especially in AD [13, 16]. Several PDE subfamilies are highly expressed in the human brain. As a result, their inhibition is involved in neurodegeneration by regulating the concentration of cAMP and/or cGMP. The evidence of phosphodiesterase (PDE) activity was first described in 1886, from the bronchodilator properties of caffeine. In the early 1960s, this effect was related to the cAMP and the inhibition of a specific PDE subtype. Thus, caffeine became the first known PDE inhibitor and turned out to be a nonselective PDE inhibitor since it also inhibits cGMP-specific PDEs including PDE type 5 (PDE5) [19]. Studies in rat and bovine tissue in the early 1970s demonstrated that PDEs hydrolyze the phosphodiesteric bond of cAMP and cGMP. A wide diversity of perturbations of the central nervous system (CNS) leads to structural damage, which in turn is followed by neurological dysfunction and abortive endogenous neurorepair [18]. Although PDE inhibitors are generally prescribed for peripheral indications, their influence on intracellular signaling makes them attractive tools for improving neuronal activity (Scheme 1). Up to 12 PDE inhibitors have been approved as therapeutics for the treatment of erectile dysfunction, pulmonary hypertension, chronic obstructive pulmonary disease, and heart failure. However, sildenafil, a PDE5 inhibitor, is the most successful example of this drug group, used for the treatment of male erectile dysfunction (Viagra) and pulmonary hypertension (Revatio) (Fig. 2) [16, 20]. During the last decade, encouraging effects on cognitive processes and neuroprotection in numerous animal studies lead to the design of selective PDEs inhibitors, which were subjected to clinical trials [17].

358

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

Scheme 1 The role of different phosphodiesterase subtypes in Alzheimer’s disease pathogenesis

2

Computational Drug Design Computational drug design tools have become important in the research of new drugs. Computational structure-based or ligandbased methods are applied in almost any new drug discovery strategy. These methods have diminished the cost as well as time of drug design and discovery. Molecular docking is a computational method for finding out binding modes of ligands to their receptors. Molecular docking and dynamics simulations were firstly applied for the rational design and the explication of structure–activity relationship of lead compounds, considering the improvement of the efficiency of the lead compound [21, 22].

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

359

3D-QSAR has been used to optimize the activity of ligands presenting specific structural characteristics. In some studies, unique applications of ligand-based methods, i.e., ligand-based homology modeling and exploration of selectivity toward specific receptor subtypes [21, 22], have been reported. Recently the discovery of novel and selective PDE inhibitors for the treatment of inflammatory disorders, CNS disorders, and many other diseases has been published. Structure-based methods, like docking, have been used for new lead identification for phosphodiesterase (PDE) inhibitors, while ligand-based methods such as quantitative structure-activity relationships (QSAR) have been mostly used for lead optimization. Ligand-based methods, such as pharmacophore development, as well as structure-based methods like molecular dynamics (MD) simulations played an important role in the analysis of the binding mode of PDE inhibitors with the enzyme [23]. In this chapter, we will review the reported research using computational modeling studies on phosphodiesterase inhibitors as anti-Alzheimer agents. 2.1 Computational Studies of PDE Inhibitors 2.1.1

PDE2 Inhibitors

PDE2 hydrolyzes both cAMP and cGMP, lowering the concentration of local cAMP. PDE2A is the only isoform of PDE2 that is highly expressed in most brain regions, including the caudate, nucleus accumbens, cortex, and hippocampus. In most peripheral tissues, except the spleen, the level of PDE2 is relatively low. This spatial distribution makes a PDE2 inhibitor an ideal therapeutic agent to co-front cognitive disorders but without cardiovascular and other side effects commonly existing in other PDE inhibitors. Several potent inhibitors for PDE2 were developed, but none succeeded to be in the market due to either lack of selectivity or poor pharmacokinetic properties [24]. BAY 60-7550 (Fig. 2) is a synthetic PDE2 inhibitor that was found to improve cognitive functions in rodents, rats, and mouse models of AD. EHNA gave BAY 60-7550 as an analog (Fig. 2), 100-fold more potent and highly selective for PDE2A. Other newly discovered selective PDE2 inhibitors are PDP (9-(6-phenyl-2-oxohex-3-yl)-2-(3,4-dimethoxybenzyl)-purin-6-one) and oxindole (Fig. 2) [24]. (i) Jiang et al. [25] designed a series of 1-aryl-4-methyl[1,2,4] triazolo[4,3-a]quinoxaline derivatives as novel multi-target ligands, antioxidants, and PDE2 inhibitors. The study was based on the design of a series of effective PDE2 inhibitors with pyrido[4,3-e][1,2,4] triazolo[4,3-a]pyrazines cores, reported by Rombouts et al. [26]. Their earlier study, according to the X-ray co-crystal structure of PDE2A with the lead compound 1 (Fig. 3), showed that the triazole core is sandwiched by Phe830 and Phe862 in the hydrophobic clamp, forming proper π–π stacking interactions. Hydrogen bond interactions can be observed between the [1,2,4]triazolo[4,3-a]quinoxaline core

360

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina O N N O

S

N

HN

O

NH2

O N

H

N

N

N

N

N

N

N

N

O

O

O

H N

N

N

O

OH

O Sildenafil

O

Vinpocetine

8-methoxymethyl-IBMX

O

O HN

O

O

N

HN

O

N N

EHNA

N

N N

O

O

N

O OH BAY 60-7550

Oxindole

PDP

O

Cl N

N

Cl

O

N N

N H Cilostazol

O

N

O

N H

N

O

O N

HN

O

N

N

O

F

NH

Cl H

O

Tadalafil

O

HN

N

F

O

N

O

N

O

N

O

Roflumilast

HN

O

Gebr-7b

N

O

N

O

N

Cl N H

GSK356278

F

O

N

O

Rolipram

F F

N

N

O

O

S

O O

HN

Anagrelide

N

N N

S

O

Cl

N

N

N

N

N

HN N

N

N

O

O PF-04447943

BAY 73-6691

N

N

BI-409306

Fig. 2 The chemical structure of PDE inhibitors

and Tyr655/Gln812 intermediated by two conservative water molecules. Moreover, the N5-position of the core also forms a hydrogen bond with Tyr827 through a third water molecule. Nbenzylamide moiety stretches outward from the catalytic pocket, and Leu774 provides space for the replacement of substituents with antioxidant activities. The oxygen atom in the amide group gives a hydrogen bond with a fourth water molecule that is unconservative. Jiang et al. [25] suggested the presence of a fragment of phenol at the [1,2,4] triazolo[4,3-a] quinoxaline core, considering the fact that a phenol structure is present in a variety of antioxidants (resveratrol, ferulic acid, and curcumin). These compounds might result in a series of potential anti-AD agents, PDE2 inhibition, and antioxidant activities simultaneously in a

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

361

Fig. 3 Structures of PDE2A inhibitors

multi-target profile. Thus, a number of different substituted phenyl moieties on [1,2,4]triazolo[4,3-a]quinoxaline core were designed and synthesized. However, the results undertaken by the molecular docking and dynamics simulations suggested that the phenol group did not impact the inhibitory activities of compounds against PDE2. Binding free energies were performed based on the molecular dynamics simulation results. The calculated binding free energies were all negative ranging from -21.88 to -23.80 kcal/ mol. These results were in accordance with the molecular docking studies [25], showing that all these compounds have strong interactions with PDE2A. Compound 2 (Fig. 3) gave the best PDE2A IC50 value among all the compounds. The [1,2,4]triazolo[4,3-a]quinoxaline moiety of 2 interacts indirectly through a hydrogen bond with residues Gln859/Tyr827/Gln812/Tyr655 via three water molecules. Direct π–π/π–σ interactions between Phe862/ Ile826/Phe830 and 2 are also observed. The 2-chlorophenyl group is nearly perpendicular to the triazoloquinoxaline core, which forms hydrophobic interactions with neighboring residues such as Leu770 and Tyr655. The benzylamine group at the 8-position of triazoloquinoxaline group is extended out of the catalytic pocket, possessing a similar pose to 1 in the co-crystal structure. It should be mentioned that the 6-F substituted phenolic group is suitable for a small pocket formed by four hydrophobic residues, Phe862/Ile866/Leu770/Lue774. In comparison to compound 2, compound 3 (Fig. 3) presents a relatively poor PDE2A IC50 value. The binding free energies between 3-PDE2A and 2- PDE2A complexes are -22.40 and 23.45 kcal/mol, respectively, with a difference of 1.05 kcal/mol

362

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

in accordance with the fourfold enhanced PDE2A inhibitory activity from 3 to 2 (IC50 values from 22 nM to 4.2 nM). The docking results underline the similar binding mode for compound 3 in relation to the triazoloquinoxaline moiety. However, the 3,5-Cl substituted phenolic group of 3 points in another direction due to steric hindrance. It fails to interact with the small hydrophobic pocket leading to a PDE2A inhibition decrease. The effect of steric hindrance also explains the overall decreased PDE2A inhibition with different m-substituted benzylamine groups, or with p-substituted ones. The polar pyridine ring diminishes the hydrophobic interactions with the small pocket and leads to decreased PDE2A inhibition in contrast to 2. Compounds 4 and 5 (Fig. 3) possess different binding modes in their planar imine groups compared to all the other compounds. They also did not succeed to interact with the small pocket. The better inhibition of 5 on PDE2A compared to 4 might be related to additional interactions with Ile826 and Phe830. The predicted results are in agreement with the experimental data, giving evidence for the further rational design. (ii) Zhang et al. [27] tried to synthesize selective potent benzo[cd] indol-2(1H)-ones as PDE2A inhibitors with improved pharmacokinetic properties. Novel PDE2A inhibitors were identified by structure-based virtual screening. They combined a pharmacophore model-based screening, molecular docking, MD simulations, and a bioassay in order to find new selective significant inhibitors. They organized a small-molecule database SPECS (comprising almost 200,000 small molecules) for virtual screening. Thirty molecules with proper binding modes and top binding energies made up the final dataset, followed by the corresponding enzymatic bioassay. Nine hits out of 30 molecules (a hit rate of 30%) were identified with less than 50 μM affinity for PDE2A. The predicted binding of 6 and 7 (Fig. 4) points to interactions with PDE2A via two hydrogen bonds and π–π stackng interactions. The molecules are embedded in the hydrophobic pocket: residues Phe862, Gln859, and Phe830 are on one end, while Leu770 and Ile866 form another side. The interactions

Fig. 4 Structures of PDE2A inhibitors

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

363

Fig. 5 Ferulic acid (FA) structure

Fig. 6 Structures of PDE4B inhibitors

between 7 and PDE2A include two hydrogen bonds with Gln859 and a π–π stack against Phe862. The molecule stretches into the abovementioned hydrophobic pocket. Based on these experiments and the binding analysis results, the researchers depicted the scaffold of 6 and 7 for further structural modifications. This study led to the discovery of the new compound 8 with IC50 = 570 nM (Fig. 5) presenting a novel benzo[cd]indol-2 (1H)-one scaffold among PDE2A inhibitors, which can be used as a new one for the design of potent inhibitors of PDE2A. 2.1.2

PDE4 Inhibitors

PDE4 subfamily hydrolyzes cAMP selectively. The four isoforms of PDE4 (PDE4A, PDE4B, PDE4C, and PDE4D) are widely expressed in the CNS and have been found to remain present in the aged and Alzheimer’s brain. PDE4C has very low expression in the brain. Rolipram and GEBR-7b (Fig. 2) target the PDE4D isoform, while GSK356278 (Fig. 2) inhibits PDE4B. Roflumilast (Fig. 2) is currently the only PDE4 inhibitor approved for the treatment of a subset of patients with severe chronic obstructive pulmonary disease (COPD) [28]. The PDE4 subtype PDE4B is believed to play an important role in inflammation. Inhibitors of PDE4B stimulate the cAMP/ CREB pathway, enhancing anti-inflammatory activity that induces the progression of AD. Some antioxidants such as ferulic acid (FA) (Fig. 6) and its ester derivatives decrease the levels of some inflammatory mediators, such as TNF-α and IL-1β. The development of selective PDE4B inhibitors has been a promising approach to treat neuro-inflammatory diseases such as AD. The PDE4B

364

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

isozyme family may be a useful therapeutic target for such cases [29]. (i) Huang et al. [29] examined how ferulic acid (Fig. 5) affects the production of PDE4B in response to LPS stimulation. FA is a cinnamic derivative of natural origin with antioxidant and antiinflammatory activities and can easily cross the blood–brain barrier being a useful agent for preventing neuroinflammatory diseases. In their study, molecular docking was used to identify the interaction between PDE4B2 and FA. A crystal structure of PDE4B2 (PDB ID: 1R06) was downloaded from the protein data bank. Molecular docking results revealed electrostatic and hydrophobic interactions between FA and PDE4B2. This analysis showed that FA strongly interacts with amino acid residues including Tyr233, His234, Met347, Asn395, Phe414, Gln443, and Phe446 at the FA-binding site of PDE4B2. FA can enter into the binding cavity of PDE4B2 interacting with the hydrophobic area and stabilized through π–π interactions with the amino acid residues Phe446 and Phe414. Hydrogen bonds may be found between FA and amino acid residues Gln443 and His234. These results indicate that there are hydrophobic and electrostatic interactions between FA and PDE4B2, which contribute to the free binding energy between FA and PDE4B2. The estimated free energy of binding (FEB) predicts the van der Waals energy and electrostatic and hydrogen bond energy. The most stable conformation exhibited the lowest FEB. The mean binding energy of FA with PDE4B2 is -6.36 kcal/mol, while the lowest binding energy among them is -6.53 kcal/mol. All the above findings support the use of FA as a lead structure for the design PDE4B inhibitors. (ii) The genus Uvaria Linn. (Family Annonaceae) consists a prolific source of pharmacologically active natural products exhibiting biological activities as anticancer, anti-inflammatory, antitubercular, antioxidant, and antimicrobial activities. Quimque et al. [30] performed a computational-assisted drug discovery experiment on the dichloromethane (DCM) sub-extract of U. alba to identify specific inhibitors for PDE4 B2B enzyme. Eighteen secondary metabolites were detected in silico considering strong binding affinities to PDE4 B2B. For the molecular docking studies, a crystal structure of PDE4B2 (PDB ID: 1R06) was downloaded from the protein data bank. The structure of PDEs comprises three domains, with the catalytic domain being the most conserved among the PDE isoenzymes. To assess the binding characteristics of the tested compounds, a molecular docking approach was performed directed to the binding domain of the enzyme. Dichamanetin (9) and grandifloracin (10) (Fig. 6) exhibited the highest binding affinity with a binding energy (BE) of 10.2 kcal/mol each. The attachment of compound 9 to the

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

365

binding site of the enzyme is related mostly to π-π interactions with four regions of the molecule: the chromanone moiety against Met347; a phenyl substituent against Phe414 and Phe446; 8-hydroxybenzyl against Tyr233 and Tyr410; and 6-hydroxyphenyl with Leu303. Hydrogen bonds were also observed between the Asp392 and the hydroxyl substituent of the chromanone core. Grandifloracin 10 is strongly bound to the active site of the enzyme through intramolecular conventional hydrogen bonds with a hydroxyl substituent to His234. The electronegative oxygens of the two ester groups are bound to Tyr233, His278, and Asn283. Bractelactones (11) in Fig. 6 conferred strong binding affinity BE of -10.1 kcal/mol against PDE4 B2B presenting hydrogen bonds between two hydroxyl substituents of the benzofuranone scaffold against His238, His274, Tyr233, and Asp392. Bractelactone (11) showed a better ADMETox profile and was further subjected to molecular dynamics (MD) simulations. The dynamics of the ligand bound system-bractelactone (11)-PDE4 B2B (Fig. 6) was found to be stable during the MD simulation at 50 ns, with a minute fluctuation between 20 and 30 ns. The average root mean square fluctuation (RMSF) of bractelactone (11) bound to PDE4 B2B was reported to be between 2.0 and 2.5 Å. The BE of bractelactone (11)-PDE4 B2B was reported to be -66.64 kcal/mol. (iii) A computational study was reported by Guariento et al. [31] using a comparative molecular field analysis (CoMFA) and molecular docking analysis. This study was performed in order to explore the key structural requirements of variant selective ligands of PDE4B. No crystallographic structures, including the catalytic site and the regulatory domains of PDE4D, have been used. It seems that the dynamic interaction between the regulatory domains (in particular UCR2 and CR3) and the active site prevents the development of a reliable binding with inhibitors protein model. In addition, although UCR2 arranges in proximity of the PDE4B and PDE4D catalytic sites in a similar way, the orientation of CR3 changes greatly with the chemical scaffold of a co-crystallized inhibitor. In this case, while 3G45 (PDE4B) and 3G4G (PDE4D) may be used for the study of the binding between UCR2 and inhibitors, only the first two studied scaffolds (group I and II) were co-crystallized with PDE4B including CR3 (PDB code, 4MYQ and 4NW7, respectively). On the other hand, no experimental data on the binding mode of those compounds included in group III (Fig. 7) were available. Only a preliminary manually analysis in PDE4B-UCR2 was performed for some of them.

366

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

Fig. 7 Structures of PDE4B inhibitors

Representative crystal structures from compounds of groups I and II (Fig. 7), within the PDE4B, were found to be available in PDB, 4MYQ and 4NW7, respectively. In the two X-ray complexes, the PDE4B inhibitor interacts with both the active site and CR3, explaining their selectivities between PDE4B and PDE4D. No experimental crystallographic data are available to explain the binding mode of inhibitors of group III with PDE4. Thus, automated docking calculations were performed, on inhibitors of group III, using two X-rays of PDE4B active site, including UCR2 and CR3, respectively. For the PDE4D enzyme, docking calculations were performed, using the PDE4D X-ray including the UCR2 domain. Compound 12 (Fig. 7) was used as a reference compound for group III. The docking results suggest that the interaction is stronger with UCR2 over CR3. In both cases, two conserved PDE4 inhibition interactions with the enzyme active site were retained (an H-bond with Q615 and a π-π stacking with F618). Nevertheless, compound 12 was predicted to interact differently with the regulatory domains surrounding the active site. Compound 12 forms a further π-π stacking interaction with Y274, which is normally displayed by UCR2-directed modulators, whereas in the presence of CR3, 12 appeared overturned placing the thiopyrano moiety toward the regulatory domain, displaying van der Waals contacts with L674. Furthermore, the crystallographic data indicate that the putative binding mode of sulphonyl

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

367

derivatives within the PDE4B-UCR2 seems to be comparable enough with the one putative binding mode observed by inhibitors belonging to groups I and II. Group III seemed to have the most effective scaffold to fit into the active site of PDE4B and, simultaneously, to interact with the surrounding regulatory domain. Indeed, despite the fact that all scaffolds displayed a strong H-bond between nitrogen of the central core and Q615, the thiopyrano moiety appeared to be better located inside the hydrophobic Q1 pocket compared to the substituents in the other two groups I and II. The presence of a sulphonyl group leads to additional possible H-bonds with Y405, related to the increase of PDE4B affinity. This interaction supports the two aromatic rings (R3 and R4). Regarding R4, its disposition in a quite large pocket in proximity of the regulatory domains allowed the insertion of multiple substituents on the ring. These findings are in agreement with CoMFA results. In all cases, it was important that the nitrogen involved in the H-bonding with key residue Q615 was not sterically hindered. Only group III (Fig. 7) was studied for selectivity focusing on the PDE4B and PDE4D UCR2 regulatory domains. The lack of the hydroxyl group of the TYR residue in PDE4D UCR2 seemed to influence the proper orientation of group III derivatives toward this regulatory region. Indeed, the R3-carboxylic group, in both benzoic and phenylacetic derivatives, interacted with the metalbinding pocket and placed the molecule in distance from Q535 (corresponding to Q615 in PDE4B). This fact prevents H-bonding with the aforementioned Q535 and Y325 (corresponding to Y405 of PDE4B), consequently destroying the stability of the compound inside the active site. On the other hand, the different dispositions of the analyzed compounds within PDE4D-active site could support the structural modifications predicted by CoMFA model C analysis. Actually, the higher distance between the central core of these compounds and Q535 may allow the insertion of a new fused ring in this position, keeping in this way the pivotal H-bond within the active site. Moreover, the R3 ring of the compounds within PDE4D was surrounded by a space larger compared to PDE4B, permitting the insertion of a bulkier moiety in R3. In addition, the contact with UCR2 seemed to be weaker, since only the π-π interaction between R4 and F196 was maintained. This binding hypothesis agrees with CoMFA data showing that the sulphonyl group and the phenylacetic moiety are pivotal for PDE4B selectivity. It seems that the latter fragment is responsible for the interaction between the inhibitors and PDE4B UCR2 helix, which is non-conserved regarding PDE4D, and therefore, this fragment consists the key point to address the selectivity of PDE4-inhibition. The researchers selected from the literature a number of 85 selective PDE4B inhibitors and analyzed them. The PDE4inhibitory potency of group I, II, and III was measured using

368

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

recombinant human PDE4B1, PDE4D3, PDE4B, PDE4D, PDE4B2, and PDE4D2, respectively. To gain a better insight of the chemical features underlying selective PDE4B inhibition, ligand-based CoMFA analyses were attempted. Three models (A, B, and C) were generated. Model A was calculated considering the PDE4B pIC50 inhibition values, Model B was performed using a sort of “weighted” PDE4B potency values, and Model C was determined using the inhibitory PDE4D pIC50 values as dependent variable. For QSAR analysis, IC50 values were transformed into pIC50 values, which were used as response variables. All the compounds were divided into a training set and a test set. In each model, molecules for the training and test set were chosen manually based in terms of biological activity and structural characteristics. For Model B, the inhibitory values were re-calculated considering the difference in pIC50 values between the PDE4B and PDE4D isoforms. A sort of weighted PDE4B pIC50 term was obtained according to Eq. (1): Weighted PDE4B pIC50 = PDE4B pIC50 þ PDE4B pIC50 - PDE4D pIC50

ð1Þ

CoMFA steric and electrostatic fields were used as independent variables, while pIC50 values were used as dependent variables. CoMFA model A showed an optimal number of components (ONC) of 5, a non-cross-validated r2 (r2ncv) = 0.93, a test set r2 (r2pred) = 0.62, Standard Error of Estimate (SEE) = 0.229, steric contribution = 0.404, and electrostatic contribution = 0.596. The statistical values for CoMFA model B displayed the following: ONC = 6, non-cross validated r2 (r2ncv) = 0.96, test set r 2 (r2pred) = 0.82, SEE = 0.275, steric contribution = 0.569, and electrostatic contribution = 0.431. CoMFA model C showed ONC = 5, non-cross-validated r2 (r2ncv) = 0.90, a test set r2 (r2pred) = 0.62, SEE = 0.218, steric contribution = 0.467, and electrostatic contribution = 0.533. The docking results and CoMFA analyses results are in agreement. The ligand-based approach was found to be better for PDE4 inhibitor drug design over the structure-based methodology. 3D-QSAR also unveiled important key structural requirements underlying for PDE4B and PDE4D inhibition. More potent PDE4B inhibitors could be developed based on a proper heteroaromatic core bearing one H-bond acceptor function (sulphonyl group), one aromatic ring, and also a benzoic acid moiety. The presence of: • A sulfur atom (rather preferred over the sulphonyl moiety), in the thiopyranopyrimidine moiety, and the introduction of a phenyl substituent lacking any acid function ( p-F-phenyl ring), drives to higher PDE4D affinity.

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

369

• A small aromatic ring bearing an acid group and a 2-benzo condensed substituent should increase affinity toward this isoform. • A bicyclic aromatic ring bearing a hindered ring and a 1-benzocondensed substituent at the corresponding positions of R3 and R4, respectively, increase PDE4D inhibition. 2.1.3

PDE5 Inhibitors

Phosphodiesterase-5 (PDE5) hydrolyzes cGMP and presents one isoform, PDE5A. Compared with the other PDE subfamilies, the expression of PDE5A in the brain is relatively low. However, several studies demonstrated that PDE5 inhibitors have a potential therapeutic effect on the treatment of AD stimulating nitric oxide (NO)/cGMP signaling, enhancing the concentration of cGMP. Since vasodilation in the brain increasingly activates other signaling pathways that impact neuroprotection [32, 33], it is generally accepted that selective PDE5 inhibitors can induce benefits on cognition and memory in pathophysiological conditions. (i) Mayara dos Santos Maia et al. [34] performed a QSAR and a molecular docking analysis of 34 ligands on PDE5. The 3D structure of PDE5 used in this study was obtained from PDB, with PDB code 3B2R. Six of the compounds were predicted to offer neuroprotection and antioxidant activity. Compounds 16 and 17 (Fig. 8) interact better with PDE5 and could be used for the development of new agents or lead molecules for AD. Compound 16 was able to form three hydrogen bonds with Met816, Tyr612, and Gln817 and four hydrophobic interactions with the amino acids Cys677, Val782, Phe786, and Phe820, whereas a steric interaction with Ile680 was exhibited. Compound 17 formed two hydrogen bonds with Tyr612 and Cys677 and five hydrophobic interactions with Ile680, Ala779, Val782, Phe786, and Phe820. The results of the ROC curve (receiver operating characteristic curve) and MCC (Matthews correlation coefficient) analyses revealed excellent results. The models achieved ROC curves greater than 0.78 during cross-validation, and the MCC values were also greater than 0.52. Lignans with a

Fig. 8 Structures of PDE5 inhibitors

370

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

probability of biological activity above 0.5 and that passed the applicability domain were considered active. However, the results showed that no lignans were considered active for PDE5. (ii) Ginger (Zingiber officinale [Z. officinale]) was known to contain several potentially bioactive phytochemicals, like gingerols and shogaols, with interesting biological activities, anti-inflammatory and antioxidant properties being among them. Ginger extract seems to inhibit the production of nitric oxide (NO) and pro-inflammatory cytokines in LPS-stimulated BV-2 microglial cells via the NF-kB (nuclear factor kappalight-chain-enhancer of activated B cells) pathway. Since in vitro data have shown that ginger’s active constituents protect nerve cells, these compounds might be potent to face AD. Azam et al. [35] studied via molecular docking the binding interactions of the bioactive ginger components, in various molecular targets implicated in AD. PDE5 with PDB code 1UDT was used. Twelve compounds 18–29 (Fig. 9) were tested, but the results did not show any suitable ligand for PDE5. (iii) Ribaudo et al. [36, 37] performed molecular docking studies on tadalafil (Fig. 10), a known drug targeting PDE5. PDB code 1UDU was used for PDE5 in complex with tadalafil. Tadalafil is bound to PDE5 with hydrophobic residues (Ala767, Ile778,

Fig. 9 Structures of PDE5 inhibitors

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

371

Fig. 10 Structures of PDE5 inhibitors

Fig. 11 Series of PDE5 inhibitors

Ala779, Ala783, Phe787, Phe786, Ile813, Phe820), as well as through some polar amino acids Gln775 and Gln817. The computational results showed that tadalafil can be used as a promising starting compound, for the development of innovative tools to counteract neurodegenerative disorders. The undertaken results agree with in vitro and in vivo findings. It should be noticed that a similar binding mode with PDE5 is given by icariside II (Fig. 3). The latter is bound to the identical region of PDE5, through the same hydrophobic residues (Leu725, Ala783, Phe786, Phe820). Thus, an aromatic and a hydrophobic moiety should be present in a molecule to target PDE5. (iv) Fiorito et al. [38] synthesized and performed a computational study for the binding site in a series of 1,2,3,4-tetrahydrobenzo [b][1,6]naphthyridines and 2,3-dihydro-1H-pyrrolo[3,4-b] quinolin-1-ones (Fig. 11). The most potent compounds, 30 and 31, were subjected to molecular docking studies on the PDE5A1 enzyme interacting with residues in the cGMP pocket of PDE5A1. Both compounds lack interaction with the structural Mg2+ and Zn2+ ions placed in the enzymatic pocket.

372

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

The significant impact of chemical changes on potency is explained from the best docking pose, in which the hydrogen bond between the nitrile group of compounds 30 or 31 and the amide group of residue Q775 is located on the binding pocket. Among the tested series, the best potency was related with the presence of a nitrile group (IC50  0.6 nM for both compound 30 and 31), while the weakest potency, within both series, was obtained when an acceptor group of any kind was absent. The presence of a weaker acceptor and/or of a shorter linker between the acceptor and the tricyclic ring moiety, placing the acceptor group away from the Q775 donor, leads to intermediate potencies. Metabolic optimization of these drugs will be investigated. The water bridge formed between the acetyl group and M816 backbone in the best docking pose of the naphthyridine series with groups lacking a good acceptor group, methyl group, and hydrogen can also explain the decreased potency. The hydrogen bonding of the acetyl group with the residue Q817 can also explain the reduced inhibition. However, this alternative binding mode is difficult to be used for the explanation of the importance of the role of the nitrile group. 2.1.4

PDE9 Inhibitors

PDE9 subfamily hydrolyzes selectively cGMP only. Among the 11 PDE subfamilies, only one isoform PDE9A exhibits the highest binding affinity for cGMP (Km = 170 nM). The expression of PDE9A occurs in most human tissues. A recent study reported that in the aged rat brain was found an increase in PDE9 expression and a decrease in cGMP concentration, which may cause Alzheimer’s disease (AD). Among all the PDE subfamilies, phosphodiesterase 9 (PDE9) presents unique advantages on AD therapy due to (a) its highest affinity with cGMP among all the PDE subfamilies and (b) its high expression level in the cortex, hippocampus, basal ganglia, and cerebellum of the brain [39]. Thus, PDE9 is nowadays characterized as a significant target for CNS pathological disorders. Today, several potent inhibitors of PDE9 are clinically tested for the treatment of AD. BAY 73-6691 (Fig. 2), developed by Bayer, is the first selective inhibitor of PDE9A for the treatment of Alzheimer’s disease. In addition, PDE9 inhibitor PF-04447943 of Pfizer (Fig. 2) completed Phase II clinical trials in patients with mild-to-moderate AD in 2013. Boehringer entered also Phase II trials for BI 409306 inhibitor (Fig. 2). (i) Zhang et al. [40] designed and synthesized a series of novel pyrazolopyrimidinone derivatives 32 with both PDE9 inhibition and antioxidant activity. The designed compounds were divided into three classes (Fig. 12). For the computational studies, PDE9A crystal structure complexed with the highly selective inhibitor 28 in Fig. 9 (PDB ID:4GH6) was used. All

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

373

Fig. 12 Structures of PDE9 inhibitors

the designed molecules were docked to the prepared structure of PDE9A. Twenty molecules with the highest docking scores and appropriate binding modes were subjected to MD simulations and binding free energy calculations. The crystal structures of PDE9A inhibitors showed that two conserved residues Gln453 and Phe456 play essential roles for their binding. Most PDE9A inhibitors contain a pyrazolopyrimidinone scaffold for hydrogen bonding with Gln453 and π–π interactions with Phe456. A narrow and long pocket next to pyrazolopyrimidinone is available in the PDE9 structure for the presence of an extra moiety with antioxidant activity. Tyr424, which is unique to PDE8 and PDE9 in this narrow pocket, has been shown to be the important feature for selectivity of PDE9 inhibitors over other PDEs. Structural characteristics from common antioxidants such as ferulic acid, caffeic acid, and lipoic acid were inserted to 6-position of pyrazolopyrimidinone with the help of various linkers to support antioxidant activities. Calculation results and dynamic stimulation (MD) point to the fact that compounds with an amide group as a linker showed stable MD simulations trajectories and high binding free energies, indicating that they might be more potent PDE9A inhibitors. All these compounds should have good inhibitory activities against PDE9A. Compound 33 formed an additional hydrogen bond with Tyr424 besides the interactions with Gln453 and Phe456. The interactions with Gln453 and Phe456 are responsible for the PDE9 affinity of inhibitors. Compound 33 with high PDE9 selectivity, potency, and antioxidant activities has been successfully developed with the support of molecular docking and dynamics simulation. (ii) Structure-based design and computational docking studies were performed on a series of PDE9 inhibitors by Luo et al. [41]. As revealed by the crystal structure, compound 34 (Fig. 13) is giving two hydrogen bonds with invariant Gln453 and π–π stacking with Phe456. Tyr424 of PDE9A2 does not form any hydrogen bond with 34. The fluoromethyl group, which is extended in another site of the small pocket, forms

374

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

Fig. 13 Structures of PDE9 inhibitors

another bond. Attempts were made to delineate the interaction with Tyr424. Thus, a linker chain with various lengths and different substitution groups was manually constituted in a graphic terminal. The linker contains an alanine, to secure that the methyl side chain or the Cβ atom will interact with Gly452 and Phe441 to prevent the substituent entrance to the neighboring small subpocket to steer the chain toward Tyr424. The amide group of the alanine linker is possible to form a hydrogen bond with Tyr424 to improve selectivity against other PDE families. The presence of a phenyl group at the end of the substitution occupies the partially open hydrophobic pocket composed of Met365, Phe441, and Val460 of PDE9A2 for further enhancement of the inhibitor affinity. The best compound 35 has an IC50 value of 21 nM and 3.3 μM, respectively, for PDE9 and PDE5 and about 3 orders of magnitude of selectivity against other PDE families. The crystal structure of the PDE9 catalytic domain complexed to compound 28 showed a hydrogen bond between 28 and Tyr424. This hydrogen bond might account for the 860-fold selectivity of 28 against PDE1B, compared to about 30-fold selectivity of BAY73-6691. Thus, it is suggested that Tyr424 is a unique residue of PDE8 and PDE9, for someone to improve the design of selective PDE9 inhibitors. This compound is a good example for someone to design PDE9 inhibitors. Compound 35 not only retains the affinity with PDE9 but also shows significant improved selectivity against other PDE families. As expected by the structure-based design and computational docking, compound 35 (Fig. 13) is bound to the active site of PDE9 in a similar way as other PDE9 inhibitors. The O4 and N5 atoms of pyrimidine for 35 are involved in two hydrogen bonds with the nitrogen and oxygen of side chain amide of Gln453 (2.8 and 2.9 Å). The amide oxygen of 35 is bound with tyrosyl OH of Tyr424 with a hydrogen bond as originally designed. The overall orientation of 34 and 35 is similar, but the pyrazolopyrimidinone ring shows about 0.7 Å positional translation. The pyrazolopyrimidine ring of 35 gives aromatic π–π stacking against Phe456 of PDE9A and van der Waals interactions with residues Phe251, His252, Met365, Ile403, Asn405, Leu420, and Phe441. It was unusual that a third

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

375

Fig. 14 Structures of PDE9 inhibitors

molecule of 35 was observed to be bound to molecule B in the dimer of the PDE9A catalytic domain. This molecule has weaker electron density than the two molecules that were bound to the active site of PDE9, indicating its partial occupancy. The third molecule of 35 interacts with van der Waals forces with Pro440, Phe441, Thr451, Ala452, Ile454, and Gly455 of molecule B in the PDE9 dimer and Ala499 and Glu502 of symmetry-related molecule A. Compound 34 does not fully occupy the binding pocket of PDE9. (iii) Tao Su et al. [37] reported the design, synthesis, and evaluation of a new series of multifunctional agents presented the pharmacophores of PDE9 inhibitors, acting as biometal chelators. The crystal structure of the catalytic domain of human PDE9 was complexed with 36 (PDB code: 4QGE). Compound 36 (Fig. 14) was used as a reference structure to define the active site of PDE9. Compounds 37 and 38 (Fig. 14) were selected for further studies as PDE9 inhibitors against PDE1B (500-fold and 236-fold, respectively) since they gave excellent in vitro results and presented high selectivity. The selectivity of 37 and 38 toward other PDEs, such as PDE2A3, PDE3A, PDE4D2, PDE5A1, PDE7A1, PDE8A1, and PDE10A2, was also evaluated. Compound 37 presented superior selectivity than that of 38 (toward all the tested PDEs, except PDE10A2). Thus, it was selected for further investigation.

376

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

The pyrazolopyrimidinone ring of 37 formed two hydrogen bonds, 2.9 and 3.3 Å, with the invariant Gln453 of PDE9 and aromatic π–π stacking interactions with Phe456. The newly introduced amine N2 atom of 37 supported a hydrogen bond, 3.0 Å, with the side chain of the unique Tyr424 in PDE9. This structural modification might explain its 500-fold better PDE9 selectivity over PDE1. Compound 36 showed a more negative docking score of -51.2 kcal/mol (CDOCKER-INTERACTION-ENERGY) than 37, which showed a score of 47.7 kcal/mol. This observation agrees with the found inhibitory activities of the two compounds (0.6 nM and 34 nM). Compound 37 might be used as a promising agent for the treatment of AD. The comparative molecular field analysis (CoMFA) method was performed to determine the quantitative relationship between the structures 39 and the IC50 values toward PDE9. Based on the IC50 values, the CoMFA results generated a reasonable/acceptable model (q2 = 0.554 and r2 = 0.996) at optimal component six. These findings showed that the steric and electrostatic fields in this CoMFA model were sufficient to explain the inhibitory effects of the target compounds 39 (Fig. 14). (iv) Since most of the inhibitors of PDE9A were based on the pyrazolopyrimidinone scaffold, Zhe Li et al. [42] looked for novel PDE9A inhibitors supported by new scaffolds. For this research, they used a combinatorial method including pharmacophores, molecular docking, molecular dynamics simulations, binding free energy calculations, and bioassays. The SPECS database containing about 200,000 compounds was screened using their combinatorial approach. The combination of ligandand structure-based methods was performed. Fifteen hits out of 29 molecules (a hit rate of 52%) with five novel scaffolds were identified to be PDE9A inhibitors different from the pyrazolopyrimidinones with inhibitory affinities no more than 50 mM to have enrichment and variety in the structure. The high hit ratio of 52% for this virtual screening method indicated that the combinatorial method is a good compromise between computational cost and accuracy. The binding analyses of those hits with non-pyrazolopyrimidinone moiety showed that they can bind to the same active site pocket of PDE9A as classical PDE9A inhibitors. Five novel scaffolds were discovered in this study and can be used for the rational design of PDE9A inhibitors with higher affinities. A crystal structure of PDE9A was complexed with a highly selective inhibitor (PDB ID: 4GH6) and used for dockingbased virtual screening. All molecules were docked to the prepared structure of PDE9A. In the binding site pocket of PDE9A, Gln453 and Phe456 are the two conservative amino

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

377

Fig. 15 Structures of PDE9 inhibitors

acid residues which play significant role in the receptor–ligand interactions. Only molecules interacting with these two residues were retained as potent PDE9 inhibitors. Two hundred molecules with the highest docking scores and appropriate binding modes were selected, which were subjected to the subsequent MD simulations and binding free energy calculations. Based on the 15 novel PDE9A inhibitors, 5 kinds of new scaffolds were summarized. The new scaffolds occupy the Q-pocket of PDE9A and can be subjected to further structural modification and optimization. Five new scaffolds (40–44) (Fig. 15), different from those of the classical PDE9A inhibitors, were gained from this analysis. (v) Tan et al. [43] performed a 3D-QSAR analysis of pyrazolopyrimidinones as PDE9A inhibitors. CoMFA, CoMSIA, docking-based structural alignment (DCBA), and local energy structure-based alignment (LESBA) analysis were performed on compounds 45, 46, 47, and 48 (Fig. 16). The crystal structure of the catalytic domain of two PDE9 complexes (PDB ID: 4GH6 and 4QGE) was used for docking studies. The pyrazolopyrimidinone moiety is bound through two hydrogen bonds with invariant Gln453 and π–π stacking interactions with Phe456. The isopropyl, isobutyl, cyclohexyl, or cyclopentyl ring at N1-position is orientating to the metal binding pocket and makes closely contacts with Met365 and Tyr424. Moreover, a contour map analysis suggested that the N1 position of the pyrazolopyrimidinone ring prefers bulkier hydrophobic groups, and in C6 position, moderately bulky electronegative groups are needed. Long linear linker between the phenyl and the pyrazolopyrimidinone rings is essential to localize the side chain into HC region.

378

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

Fig. 16 Structures of PDE9 inhibitors

2.1.5

PDE10 Inhibitors

PDE10A is a subfamily that hydrolyzes both cAMP and cGMP. The highest expression of PDE10A in the brain is taking place in the caudate nucleus together with PDE1B. Currently, PDE10 is considered a promising target for CNS neurological diseases. Pharmaceutical companies are currently re-evaluating the compounds for Huntington’s and Parkinson’s diseases [44], given the limited results for new drugs and PDE10 localization. (i) A series of PDE10 quinazolines inhibitors with antioxidant activities were designed and synthesized by Li et al. [42]. Molecular docking was performed for better understanding of their binding mode with PDE10. The binding site was defined by the co-crystallized PDE10A inhibitor (possessing the same quinazoline moiety) in PDB entry 3QPN. The structure of PDE10 complex with papaverine has been reported (PDB code: 2WEY). By the docking findings, it is shown that the quinazoline group occupies the same position as papaverine, and the oxygen of quinazoline presents hydrogen bonds with the residue of Gln716 and hydrophobic interactions with the residues of Phe719 and Phe686 through the quinazoline ring. These interactions play a crucial role in the binding capacity of PDE10 inhibitors. Thus, the introduction of substituents at the 6-position of the quinazoline ring could fill in the selective pocket of PDE10. No hydrogen bond was observed between compounds and Tyr683. Compounds 49 and 50 (Fig. 17) have the same 2-(1H-indol3-yl) ethyl group. The observation of the docked conformations of 49 and 50 complexed with PDE10A shows stretching of the side chains in different directions. Only compound 49 (Fig. 17), which presents the higher inhibitory activity, showed the suitable volume magnitude for the PDE10 selective pocket. The side chain of 49 resides in the selective pocket of PDE10A. On the contrary, the side chain of 50 is extended out of the catalytic site, possessing the same mode as papaverine. Compound 50, combining a significant inhibitory activity PDE10 and antioxidant activity, may structurally be modified and used as a lead molecule.

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents

379

Fig. 17 Structures of PDE10 inhibitors

3

Conclusion Phosphodiesterase inhibitors have been subjected to computational studies as anti-Alzheimer agents. Among them, not all isoforms are implicated. PDE2, PDE4, PDE5, PDE9, and PDE10 are the subfamilies for which inhibitors have been designed and modeling studies have been performed. In all cases, heteroaromatic scaffolds are crucial within the structure of inhibitors: Triazolo-quinoxalines were studied as PDE2 inhibitors. They showed that the triazole scafold is sandwiched by Phe830 and Phe862 in the hydrophobic pocket, giving proper π–π stacking interactions. Hydrogen bond interactions are given between the [1,2,4]triazolo[4,3-a]quinoxaline scaffold and Tyr655/Gln812. Molecular field analysis (CoMFA) and molecular docking were used to define the key structural characteristics of variant selective ligands of PDE4. Small aromatic rings bearing an acid group or a bicyclic group increase PDE4 inhibition. For PDE5 inhibition, three hydrogen bonds with Met816, Tyr612, and Gln817 and four hydrophobic interactions with the amino acids Cys677, Val782, Phe786, and Phe820 were found, whereas a steric interaction with Ile680 was observed in the modeling studies. Pyrazolopyrimidinone derivatives have been studied for PDE9 inhibition. The pyrazolopyrimidinone scaffold forms hydrogen bond with Gln453 and π–π interactions with Phe456. A narrow and long pocket next to pyrazolopyrimidinone group is available in the PDE9 structure. For PDE10 quinazolines inhibitors, the docking results indicate that the oxygen of the quinazoline core interacts through a hydrogen bond with the residue of Gln716. Hydrophobic interactions with the residues of Phe719 and Phe686 are also observed.

380

Ioanna-Chrysoula Tsopka and Dimitra Hadjipavlou-Litina

References 1. Jabir NR, Rehman MT, Alsolami K, Shakil S, Zughaibi TA, Alserihi RF et al (2021) Concatenation of molecular docking and molecular simulation of BACE-1, γ-secretase targeted ligands: in pursuit of Alzheimer’s treatment. Ann Med 53(1):2332–2344 2. Alzheimer’s association (USA) (2022) Alzheimer’s disease facts and figures 2022 3. Breijyeh Z, Karaman R (2020) Comprehensive review on Alzheimer’s disease: causes and treatment. Molecules (Basel, Switzerland). NLM (Medline) 25:5789 4. Fedele E, Ricciarelli R (2021) Memory enhancers for Alzheimer’s dementia: focus on cgmp. Pharmaceuticals. MDPI AG 14:1–14 5. Jaeger LB, Dohgu S, Sultana R, Lynch JL, Owen JB, Erickson MA et al (2009) Lipopolysaccharide alters the blood-brain barrier transport of amyloid β protein: a mechanism for inflammation in the progression of Alzheimer’s disease. Brain Behav Immun 23(4):507–517 6. Sayyah M, Javad-Pour M, Ghazi-Khansari M (2003) The bacterial endotoxin lipopolysaccharide enhances seizure susceptibility in mice: involvement of proinflammatory factors: nitric oxide and prostaglandins. Neuroscience 122(4):1073–1080 7. Huang WJ, Zhang X, Chen WW (2016) Role of oxidative stress in Alzheimer’s disease (review). Biomed Rep. Spandidos Publications 4:519–522 8. Birks JS, Harvey RJ (2018) Donepezil for dementia due to Alzheimer’s disease. Cochrane Database of Syst Rev. John Wiley and Sons Ltd 2018 9. Evans GJ (2009) Rivastigmine for Alzheimer’s disease (review) [Internet]. Available from: http://www.thecochranelibrary.com 10. Razay G, Wilcock GK (2008) Galantamine in Alzheimer’s disease. Expert Rev Neurotherap. 8:9–17 11. Robinson DM, Keating GM, Schmitt FA, van Dyck CH, Wenk GL, Wimo A, ADIS Drug Evaluation (2006) Memantine: a review of its use in Alzheimer’s disease. Drugs 66:1515–1534 12. Wang P, Guan PP, Wang T, Yu X, Guo JJ, Wang ZY (2014) Aggravation of Alzheimer’s disease due to the COX-2-mediated reciprocal regulation of IL-1β and Aβ between glial and neuron cells. Aging Cell 13(4):605–615 13. Luth HJ, Munch G, Arendt T (2002) Aberrant expression of NOS isoforms in Alzheimer’s disease is structurally related to nitrotyrosine formation. Brain Res 953:135–143

14. Kalra J, Kumar P, Majeed ABA, Prakash A (2016) Modulation of LOX and COX pathways via inhibition of amyloidogenesis contributes to mitoprotection against β-amyloid oligomer-induced toxicity in an animal model of Alzheimer’s disease in rats. Pharmacol Biochem Behav 146–147:1–12 15. Gourmaud S, Paquet C, Dumurgier J, Pace C, Bouras C, Gray F et al (2015) Increased levels of cerebrospinal fluid JNK3 associated with amyloid pathology: links to cognitive decline. J Psychiatry Neurosci 40(3):151–161 16. Nabavi SM, Talarek S, Listos J, Nabavi SF, Devi KP, Roberto de Oliveira M et al (2019) Phosphodiesterase inhibitors say NO to Alzheimer’s disease. Food Chem Toxicol. Elsevier Ltd 134: 110822 17. Ibrahim MM, Gabr MT (2019) Multitarget therapeutic strategies for Alzheimer’s disease. Neural Regen Res. Wolters Kluwer Medknow Publications 2018:437–440 18. Delaby C, Gabelle A, Blum D, SchraenMaschke S, Moulinier A, Boulanghien J et al (2015) Central nervous system and peripheral inflammatory processes in Alzheimer’s disease: Biomarker profiling approach. Front Neurol 6 (Aug):181 19. Encyclopedia of Psychopharmacology (2010) Encyclopedia of psychopharmacology. Springer, Berlin, Heidelberg 20. Nehra A, Colreavy F, Khandheria BK, Chandrasekaran K (2001) Sildenafil citrate, a selective phosphodiesterase type 5 inhibitor: urologic and cardiovascular implications. World J Urol 19:40–45 21. Prieto-Martı´nez FD, Lo´pez-Lo´pez E, Eurı´dice Jua´rez-Mercado K, Medina-Franco JL (2019) Computational drug design methods—current and future perspectives. In: In Silico drug design. Elsevier, pp 19–44 22. Schaduangrat N, Lampa S, Simeon S, Gleeson MP, Spjuth O, Nantasenamat C (2020) Towards reproducible computational drug discovery. J Cheminform. BioMed Central Ltd 12:1–30 23. Gaurav A, Xing M, Al-Nema M (2017) Computational approaches in the development of phosphodiesterase inhibitors. In: Quantitative structure-activity relationship. InTech 24. Sadek MS, Cachorro E, El-Armouche A, K€ammerer S (2020) Therapeutic implications for PDE2 and cGMP/CAMP mediated crosstalk in cardiovascular diseases. Int J Mol Sci. MDPI AG 21:1–30

Computational Modeling of Phosphodiesterase Inhibitors as Anti-Alzheimer Agents 25. Jiang MY, Han C, Zhang C, Zhou Q, Zhang B, Le ML et al (2021) Discovery of effective phosphodiesterase 2 inhibitors with antioxidant activities for the treatment of Alzheimer’s disease. Bioorg Med Chem Lett 41:128016 26. Rombouts FJR, Tresadern G, Buijnsters P, Langlois X, Tovar F, Steinbrecher TB et al (2015) Pyrido[4,3- e ][1,2,4]triazolo[4,3- a ] pyrazines as selective, brain penetrant phosphodiesterase 2 (PDE2) inhibitors. ACS Med Chem Lett 6(3):282–286 27. Zhang C, Feng LJ, Huang Y, Wu D, Li Z, Zhou Q et al (2017) Discovery of novel phosphodiesterase-2A inhibitors by structurebased virtual screening, structural optimization, and bioassay. J Chem Inf Model 57(2): 355–364 28. Li H, Zuo J, Tang W (2018) Phosphodiesterase-4 inhibitors for the treatment of inflammatory diseases. Front Pharmacol. Frontiers Media S.A. 9:1048 29. Huang H, Hong Q, Tan HL, Xiao CR, Gao Y (2016) Ferulic acid prevents LPS-induced up-regulation of PDE4B and stimulates the cAMP/CREB signaling pathway in PC12 cells. Acta Pharmacol Sin 37(12):1543–1554 30. Quimque MT, Notarte KI, Letada A, Fernandez RA, Pilapil DY, Pueblos KR et al (2021) Potential cancer- and Alzheimer’s diseasetargeting phosphodiesterase inhibitors from Uvaria alba: insights from in vitro and consensus virtual screening. ACS Omega 6(12): 8403–8417 31. Guariento S, Bruno O, Fossa P, Cichero E (2016) New insights into PDE4B inhibitor selectivity: CoMFA analyses and molecular docking studies. Mol Divers 20(1):77–92 32. Vignozzi L, Gacci M, Cellai I, Morelli A, Maneschi E, Comeglio P et al (2013) PDE5 inhibitors blunt inflammation in human BPH: a potential mechanism of action for PDE5 inhibitors in LUTS. Prostate 73(13):1391–1402 33. Ahmed WS, Geethakumari AM, Biswas KH (2021) Phosphodiesterase 5 (PDE5): structure-function regulation and therapeutic applications of inhibitors. Biomed Pharmacother. Elsevier Masson s.r.l. 134:111128 34. dos Santos MM, Rodrigues GCS, de Sousa NF, Scotti MT, Scotti L, Mendonc¸a-Junior FJB (2020) Identification of new targets and the virtual screening of lignans against Alzheimer’s disease. Oxidative Med Cell Longev 2020: 3098673

381

35. Azam F, Amer AM, Rabulifa A, Elzwawi MM (2014) Ginger components as new leads for the design and development of novel multitargeted anti-Alzheimer’s drugs: a computational investigation. Drug Des Devel Ther 8: 2045–2059 36. Ribaudo G, Ongaro A, Zagotto G, Memo M, Gianoncelli A (2020) Therapeutic potential of phosphodiesterase inhibitors against neurodegeneration: the perspective of the medicinal chemist. ACS Chem Neurosci. American Chemical Society 11:1726–1739 37. Su T, Zhang T, Xie S, Yan J, Wu Y, Li X et al (2016) Discovery of novel PDE9 inhibitors capable of inhibiting Aβ aggregation as potential candidates for the treatment of Alzheimer’s disease. Sci Rep 6:1–4 38. Fiorito J, Vendome J, Saeed F, Staniszewski A, Zhang H, Yan S et al (2017) Identification of a novel 1,2,3,4-tetrahydrobenzo[b][1,6] naphthyridine analogue as a potent phosphodiesterase 5 inhibitor with improved aqueous solubility for the treatment of Alzheimer’s disease. J Med Chem 60(21):8858–8875 39. Sivakumar D, Mudedla SK, Jang S, Kim H, Park H, Choi YW et al (2021) Computational study on selective pde9 inhibitors on pde9mg/mg, pde9-zn/mg, and pde9-zn/zn systems. Biomolecules 11(5):709 40. Zhang C, Zhou Q, Wu XN, Huang YD, Zhou J, Lai Z et al (2018) Discovery of novel PDE9A inhibitors with antioxidant activities for treatment of Alzheimer’s disease. J Enzyme Inhib Med Chem 33(1):260–270 41. Meng F, Hou J, Shao YX, Wu PY, Huang M, Zhu X et al (2012) Structure-based discovery of highly selective phosphodiesterase-9A inhibitors and implications for inhibitor design. J Med Chem 55(19):8549–8558 42. Li Z, Lu X, Feng LJ, Gu Y, Li X, Wu Y et al (2015) Molecular dynamics-based discovery of novel phosphodiesterase-9A inhibitors with non-pyrazolopyrimidinone scaffolds. Mol BioSyst 11(1):115–125 43. Tan C, Wu Y, Shao Y, Luo H, Zheng X, Wang L (2017) Docking-based 3D-QSAR studies of phosphodiesterase 9A inhibitors. Lett Drug Des Discov 14(9):986–998 44. Menniti FS, Chappie TA, Schmidt CJ (2021) PDE10A Inhibitors—clinical failure or window into antipsychotic drug action? Front Neurosci. Frontiers Media S.A. 14:600178

Chapter 13 Computational Methods for the Design and Development of Glutaminyl Cyclase Inhibitors in Alzheimer’s Disease Kiran Bagri, Ashwani Kumar, Parvin Kumar, Archana Kapoor, and Vikas Verma Abstract The amyloid β (Aβ) aggregates or senile plaques are the pathological hallmark of Alzheimer’s disease and are the main reason for neuronal death. The amyloidogenic cyclization of N-terminal glutaminyl of Aβ to pyroglutamate is catalyzed by a metalloenzyme, glutaminyl cyclase (QC). Clinical findings indicated that the pyroglutamate Aβ (pGlu-Aβ) is more neurotoxic, resistant to hydrolysis, and more soluble and have more tendency to form plaques. The activity of QC is upregulated during AD leading to the severity of the disease. Hence, QC inhibition has become a new target to treat AD, and substantial efforts have been put into the design and development of inhibitors of QC. PQ912, a QC inhibitor developed by Vivoryon Therapeutics, is currently in phase 2 of clinical trials. Preclinical results found that PQ912 reduced pGlu-Aβ levels and improved cognition and memory. Herein, we review the design and development of QC inhibitors by using modern computational approaches. Key words Amyloid β, Alzheimer’s disease, Glutaminyl cyclase, Docking, Molecular dynamics, QSAR

Abbreviations AD APP Aβ CADD CCL2 CNS MD MM-GBSA pGlu-Aβ QC QPLD QSAR

Alzheimer’s disease Amyloid precursor protein Amyloid beta Computer-aided drug design Chemokine (C–C motif) ligand 2 Central nervous system Molecular dynamics Molecular mechanics with generalized Born and surface area solvation Pyroglutamated amyloid beta Glutaminyl cyclase Glide quantum mechanics-polarized ligand docking Quantitative structure–activity relationship

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_13, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

383

384

1

Kiran Bagri et al.

Introduction The specific hallmark of Alzheimer’s affected brain is the extracellular senile plaques and intracellular neurofibrillary tangles. The accumulation of these two histopathological lesions in the brain triggers neurodegeneration, leading to the loss of cognition and memory as observed in Alzheimer’s disease. The characterization of senile plaques and genetic studies of mutant genes like presenilin 1 and presenilin 2 identified that amyloid β (Aβ) peptide is the main element of senile plaques [1, 2]. This leads to the postulation of the amyloid cascade hypothesis, according to which the imbalance of production and accumulation of amyloid β peptide resulted in the precipitation of AD [3]. Aβ is generated by proteolytic cleavage of amyloid precursor protein (APP) by enzymes β-secretase and γ-secretase. Reducing the production of Aβ by inhibiting β- secretase (BACE1) has been considered an effective therapeutic approach in AD treatment [4, 5]. A variety of Aβ of different lengths from 34 to 50 amino acids is generated from mutable APP. The N-terminal reduction of Aβ and the cyclization of terminal glutamate into pyroglutamate give rise to pGlu-Aβ. The resulting pGlu-Aβ was found to be more neurotoxic and hydrophobic and also has enhanced tendency to form aggregates [6]. Clinical studies revealed that the major proportion of Aβ deposits in the brain of AD patients is the pGlu-Aβ form [7]. In vitro and in vivo studies suggested that the cyclization of N-terminal glutamic acid resulting in the formation of pGlu-Aβ has been catalyzed by a metalloenzyme glutaminyl peptide cyclotransferase or glutaminyl cyclase (QC) [8, 9]. Mammalian QC is a metalloenzyme and is found to exist in two isoforms known as the secretory QC (sQC) and Golgi-resident QC (gQC). sQC consists of 361 amino acids and catalyzes the formation of pGlu-Aβ, whereas gQC consists of 382 amino acids and catalyze the pyroglutamation of inflammation mediator CCL2. Both the isoforms have catalytic domain of 331 residues and have approximately 45% sequence identity. The structure of both isoforms is globular with a mixed α/β framework in an open sandwich model. The active site consists of a zinc ion located at the bottom and is coordinated with three amino acid residues, namely, HIS 330, ASP 159, and GLU 202 [10–12]. Figure 1 shows the chain of A of the enzyme showing zinc and its corresponding residues. QC is widely distributed in the mammalian brain primarily in the hippocampus and cortex and is overexpressed in AD leading to increased formation of pGlu-Aβ, hence contributing to the severity of the disease [7, 13]. The involvement of QC and the therapeutic role of QC inhibitors in AD are shown in Fig. 2. PQ912 is the first QC inhibitor and is currently in clinical trials. Various clinical studies marked the efficacy of PQ912 in reducing the pGlu-Aβ

Computational Methods for the Design and Development of Glutaminyl Cyclase. . .

385

Fig. 1 3D structure of chain A of glutaminyl cyclase enzyme

Fig. 2 Role of glutaminyl cyclase and glutaminyl cyclase inhibitors in Alzheimer’s disease

levels in the brain and significantly improving in cognition and memory [14, 15]. Clinical studies clearly explain the pathogenic nature of QC; therefore, strategies aiming to reduce the generation and accumulation of pGlu-Aβ by inhibiting QC present a good alternative for the drug development of AD [16–19].

386

Kiran Bagri et al.

1.1 Application of Computational Methods in Drug Design of Glutaminyl Cyclase Inhibitors

2

Rational development of new drugs is a costly affair, but various approaches of CADD (computer-aided drug design) have assisted in the design and development of new drugs by significantly reducing the time and cost. Advanced computational approaches, along with chemical synthesis and biochemical evaluation, have considerably enhanced the drug discovery process [19, 20]. Various computational approaches have assisted in the drug discovery of QC inhibitors. Techniques like homology modeling, docking, molecular dynamics, virtual screening, and QSAR have been reported to explain the enzyme-ligand, active site conformation, ligand structure, reaction mechanism, etc. Hence, the structural advancement of QC inhibitors with the help of computational means and their outcomes have been reviewed here.

Homology Modelling Studies Understanding the protein-ligand interaction forms the basis of the structure-based drug design of inhibitors. But due to the lack of information on the 3D structure of protein, homology modeling was used to develop inhibitors of QC. QC was recognized as a metalloenzyme, whereas imidazole and imidazole derivatives were identified as its competitive inhibitors. Structure-based screening of imidazole derivatives demonstrated effective inhibition of the enzyme by N-1 derivatives of imidazole and reported two potent inhibitors, compound 1 and 2, with Ki values of 818 ± 1 nM and 295 ± 5 nM, respectively. Structural features of the active site identified using multiple sequence alignment projected HIS 140 and HIS 330 as important residues for catalysis. The studies also marked the importance of pH-dependent behavior of inhibitors on QC [21]. Structural homology with aminopeptidases characterized residues necessary for binding of substrate to catalytic zinc [22].

1

2

Further advancement of this study led to the development of imidazole-1-yl-alkyl thiourea as the first potent series of QC inhibitors [23]. The series of imidazole-propyl-thiourea and imidazolepropyl-thioamide derivatives were found to be effective inhibitors of QC. Compound 3 (PBD 150) and 4 were identified as the most potent inhibitors with Ki values of 60 nM and 90 nM. Due to the lack of 3D structure, molecular modeling predicted the binding

Computational Methods for the Design and Development of Glutaminyl Cyclase. . .

387

model by flexibly aligning the compound 3 and 4 with a tripeptide substrate H-GLN-PHE-ALA-NH2. The results of average strain energy and alignment score suggested effective binding between inhibitors and the active site of the enzyme. The study revealed an imidazole ring with an alkyl spacer (three methylene units) linking imidazole and thiourea/thioamide further linked to a hydrophobic residue (phenyl/benzyl ring) as a preferred skeleton for efficient inhibition of QC. Moreover, the homology modeling identified that QC inhibitors interacted with zinc through coordination bonding as imidazole exhibited only one coordination site. The thiourea moiety interacted via hydrogen bonding as it has two hydrogen donors at 1-N and 3-N positions. The hydrophobic interactions of the phenyl/benzyl moiety via π–π interactions were suggested to be significantly improved when substituted with electron-releasing groups in the para position. The structures of glutaminyl cyclase inhibitors identified three substrate-specific areas: (i) the zinc or metal binding moiety, (ii) the alkyl connector, and (iii) the aromatic or hydrophobic moiety.

3 (PBD 150)

4

The bio-isosteric substitutions of the thiourea moiety by cyanoguanidines, nitrovinyldiamine, and 2-thioxopyrimidin-4-(1H)ones led to the generation new class of inhibitors [24]. Compound 5 and 6 were identified as the most potent compounds of cyanoguanidines and nitrovinyldiamine derivatives but were found to be 10 times less potent than PBD 150. But the involvement of thiourea in the condensed ring as thioxopyrimidine led to potent inhibitors exhibiting activities in low micromolar range. The subsequent improvement in inhibitory potency derivatives was observed when methyl was substituted at position 5 of the imidazole ring. Compound 7 with Ki 2.6 nM was found to be the most potent inhibitor. The SAR of the developed inhibitors was supported by the findings of homology modeling with the active site of the enzyme by utilizing the X-ray structure of aminopeptidase from Aeromonas proteolytica. The structural ambiguities found in the loop region of the active site were modulated through MD simulations. The imidazole ring act as a zinc binding group, the length of spacer was restricted to three methylene units as a shorter linker resulted in steric clashes, and the 3,4-dimethoxyphenyl moiety was found interacting with TYR 299, PHE 325, PRO 326, and GLU 327. The inhibitors were biologically assessed by testing their ability to inhibit or block Aβ. The results were found to be in agreement with the inhibitory activity.

388

Kiran Bagri et al.

6

5

7

Based on the previously reported binding mode of a substrate of the enzyme (H-GLN-PHE-ALA-NH2), a new series of N- arylN-(5-methyl-1H-imidazol-1-yl) derivatives was synthesized by substituting the hydrophobic region with heterocycles and aromatic rings [25]. The various structural features were identified, and pharmacophore was divided into three main regions. Zinc binding imidazole with methyl substituent on fifth position represented region A, B region constituted of hydrogen bond donor that interact with C-terminal amide nitrogen of glutamine residue, and C region embraced the hydrophobic moiety interacting with the phenylalanine residue. Due to the high structural flexibility of the C region, it was considered the most favorable position for modification and conclusively investigating the SAR of synthesized inhibitors. The results concluded that inhibitors with hydrogen bond acceptor like oxygen and nitrogen atoms were found to be more potent. Compound 8 (IC50 58 nM) was found to be a promising inhibitor by significantly reducing the formation of amyloid plaques (pGlu-Aβ & Aβ).

8

3

Docking Studies with the Structure of Human Glutaminyl Cyclase Enzyme The design and development of QC inhibitors became more refined after the documentation of crystal structures of QC enzyme in protein data bank [26]. In line with structural developments, a series of phenol-modified apigenin derivatives were designed and

Computational Methods for the Design and Development of Glutaminyl Cyclase. . .

389

synthesized [27]. Compound 9 (pIC50 14.2 μM) was found to have the highest binding affinity. Molecular docking of compound 10 with an active site of 2AFX showed π–π stacking interactions between phenol and TRP 207, a hydrogen bond between C5-OH and GLN 304. The hydrogen bonding significantly favored the activity; hence, C5-OH derivatives had better inhibitory activity. The importance of OH at C-7 was concluded as it interacted with the zinc ion in the active pocket via hydrogen bonding, whereas C-7 methylated derivatives were found to be inactive. The results showed the apigenin derivatives with hydroxy (-OH) group at C-5 and C-7 as a new class of flavonoid-based glutaminyl cyclase inhibition.

10

9

In order to enhance the BBB penetrability and flexible binding, a series of diphenyl conjugated derivatives with varied substitutions and linkers was designed and synthesized [28]. Compound 11 with pIC50 of 1.23 μM was reported to be the most potent. SAR and docking studies highlighted that a propyl linker between imidazole and biphenyl seemed to be sufficient to guide the imidazole moiety to the zinc ion located in the active site. Reduction of linker length resulted in decreased activity. The appropriate linker length led to favorable π–π stacking interactions between aromatic moiety and PHE 325. 4-methyl substitution at the imidazole ring and fluorine or methyl substitution at the phenyl resulted in increased activity. Improved BBB permeability of analogues was proved by in vitro and in silico assays. Administration of the most potent compound resulted in decreased level of pyroglutamate Aβ in living cells. The studies concluded diphenyl conjugated derivatives may contribute to the development of novel anti-AD agents.

11

In order to mimic the arginine of N-terminal tripeptide (GLU-PHE-ARG) of Aβ3(pE)-42, a library of compounds with

390

Kiran Bagri et al.

Fig. 3 Pharmacophore of compound 12 and its binding characteristics with enzyme QC

various amine functional group were designed and synthesized [29]. The additional side chain occupies the region D of the newly designed scaffold (Fig. 3) and acts act as a positional anchor for better binding interactions. SAR evaluation of newly designed molecules identifies compounds with 5 to 40 times increased inhibitory potency as compared to PBD 150. The results were further supported by Aβ lowering effect in vivo and enhanced ability to penetrate BBB. The in vivo and in vitro results were further validated by performing long-term therapeutic studies of potent compound 12 in transgenic mice. The studies concluded the reduced brain concentration of Aβ3(pE)-42 and the brain levels of total Aβ were also reduced. Compound 12 docked well with the active site of glutaminyl cyclase, as the A region chelated with zinc, the C region interacted with TRP 299 via hydrophobic interactions, and the D region interacted with GLU 327. Hence, an overall improvement in activity was achieved by the additional D region. Therefore, 12 was found to be an effective inhibitor, and its studies provide better understanding of designing potent inhibitors. Based on the previous hypothesis that the D region provides additional binding interaction, a new series of analogs with phenyl and benzyl linker groups between the C and D regions were designed and synthesized [30]. The piperazine analog with benzyl linker compound 13 was found to be the most effective inhibitor having IC50 of 6.2 nM. Compound 13 scored well in in vivo studies as it significantly reduced the AβN3pE-40 in the brain extracts of mice. SAR studies concluded that the rigidification of the linker

Computational Methods for the Design and Development of Glutaminyl Cyclase. . .

391

between the C and D regions was not fruitful. Molecular docking findings suggested that the C region formed an additional hydrogen bond and the phenyl ring between C and D regions displayed an extra π–π interaction with PHE 325. SAR and molecular modeling studies on specific region D helped in further evolving the design and development of effective QC inhibitors.

13

Further optimization of the previous work led to the development of derivatives with 3-aminoalkyloxy-4-methoxyphenyl and 4-aminoalkyloxyphenyl groups at position 3 and 4 of the aromatic ring (region C), respectively [31]. The rationale for the design was to improvise the Arg-mimetic region and to conclude the importance of the 3-methoxy group in inhibition. The derivatives with primary amines and 4-piperidinyl groups showed maximum QC inhibition. The results of in vivo and in vitro studies revealed compound 14 and 15 with IC50 7.9 nM and 8.8 nM, respectively. The potent derivatives exhibited results surpassing PBD 150. The potency resulted due to the additional salt bridge interaction and hydrogen bonding with GLU 327 and PRO 326 as shown in molecular modeling studies. The substantial improvement in inhibitory activity and enhanced pharmacological effects marked compound 15 as a potential candidate in AD therapeutics. The docking studies of hQC inhibitors with PBD 150 revealed a Z-E conformation of ligands at the active site.

14

15

The knowledge of previously synthesized inhibitors led to the generation of N-substituted thiourea, urea, and α-substituted amide derivatives [32]. The modification led to the favorable bent

392

Kiran Bagri et al.

(Z-E) conformations of the ligand as shown by Glide quantum mechanics-polarized ligand docking (QPLD). Docking results revealed that imidazole chelated with zinc, indole NH interacted with TRP 302 via H-bonding, oxygen of urea showed H-bonding with GLN 304, and phenyl ring displayed π–π interactions with PHE 325. The N-substitutions showed additional hydrophobic interactions with TYR 299, ILE 303, and VAL 302. Urea compound 16 and thiourea compound 17 derivatives were found to possess the maximum inhibition with IC50 values of 1.3 and 1.39 nM, respectively. The in vivo, cytotoxicity, and metabolic studies concluded compound 18 (6.1 nM) as the most potent inhibitor. Compound 18 docked well in the active site in Z-E conformation as anticipated. The results helped in further elaborating the knowledge of binding pocket as well as conformational stability for favorable binding pose of QC inhibitors.

16

17

18

Based on the idea that urea- and thiourea-based derivatives had limited BBB penetration, Tran and coauthors computationally and experimentally designed two new series of inhibitors lacking the urea and thiourea moiety [33]. The inhibitory activity of new derivatives was assessed with the fluorogenic assay. A combination of molecular docking and molecular dynamics simulations was used to study the binding pose of inhibitors in a more refined form. Compound 19 was concluded to be most potent with IC50 of 0.11 μM and was found to possess a balance of electrostatic and van der Waals interaction energy as confirmed by the free energy perturbation method. The compound also formed two hydrogen bonds with a significant residue, GLU 202. Quantum chemical calculation helped in estimating the strong influence of the zinc ion on the binding affinity of 19 to the active site. It was emphasized

Computational Methods for the Design and Development of Glutaminyl Cyclase. . .

393

Fig. 4 Binding interactions of compound 20 with the active site of enzyme QC

that the total net charge of the designed ligand should be zero in order to overcome the repulsive forces of Zn+2. The BBB crossing ability and human intestinal absorption capacity foresee compound 19 as an ideal drug candidate.

19

A new series of N-(-4-aminoalkylphenyl) thiourea/urea derivatives by combining the structure of Arg-mimicking group and conformationally restricted group was designed and synthesized [34]. SAR studies revealed the combination of the two crucial pharmacophores into one scaffold had a synergistic effect on the activity and potency. The activity increased with the increase in size conformational blocker incorporated into the urea group like substitutions with cyclopentyl methyl and 1-methylpiperidinyl showed maximum activity. Among the 84 derivatives synthesized, the cyclopentyl methyl derivative compound 20 and benzimidazole derivative compound 21 were found to be the most potent inhibitors with IC50 of 0.1 nM and 9.9 nM, respectively. The crystal structure of hQC in the complex with 20 was determined and studied to highlight the ligand binding sites of hQC. The binding interactions revealed that compound 20 interacted with active site through metal coordination, hydrogen bonding, and π–π stacking interactions as depicted in Fig. 4. Compound 21 surpassed 21 in terms BBB permeability, metabolic stability, and hERG and CYP inhibition. Further in vivo evaluation of 21 showed its favorability in terms of efficacy, selectivity, and druggable profile. The novel scaffold may provide a promising therapeutic option in AD treatment.

394

4

Kiran Bagri et al.

High-Throughput Virtual Screening for Designing QC Inhibitors A pharmacophore with features from the previously identified inhibitors and substrate (H-GLN-PHE-ALA-NH2) was developed. The pharmacophore-based filtering of library of 6500 molecules identified benzimidazole attached to 1,3,4-oxadiazole as a hit scaffold, and based on this, derivatives of thiadiazole and triazole were reported [35]. SAR was thoroughly explored by systemic variation of substituents at the 5-membered ring, the spacer, and the terminal hydrophobic moiety. The structural changes resulted in potent inhibitors (22 and 23) with 1,3,4-thiadiazol/1,2,3-triazole ring, methylene spacer with sulfide/sulphone moiety linked to the terminal dimethoxyphenyl substituent. The IC50 value of 22 was found to be 0.07 μM and 23 was 0.63 μM. In silico docking analysis with in vitro mutagenesis studies helped in identifying the possible molecular alignment of the inhibitors as well as the contribution of amino acid residues in binding with the inhibitor. As the Zn atom interacted with benzimidazole ring, amino acids PHE 325, TRP329, and TRP207 were identified for interactions with the active site. Docking along with site-directed mutagenesis delivered convincing results and provided better insights of the active site.

22

23

Dileep and coauthors designed and identified a new class of QC inhibitors through pharmacophore-assisted highthroughput virtual screening [36]. The researchers generated a e-pharmacophore hypothesis based on the structure of receptorligand complex. Pharmacophore consists of a metal binding group, two hydrogen bond acceptors, and a stacking interaction. The generated hypothesis screened four million molecules from the database. The hits were further subjected to filtering based on the Lipinski rule and physiochemical parameters for a molecule to be CNS drug. The resulting hits were filtered by docking, and only 10% of the top scoring ligands were retrieved. MM-GBSA

Computational Methods for the Design and Development of Glutaminyl Cyclase. . .

395

calculations were performed on 183 ligands, resulting in the generation of 54 hits. Enzymatic assay and cytotoxic evaluation were performed to support the in silico screening results. Compound 24 (IC50 33.9 μM) with amide as metal binding group was identified as a novel and potent inhibitor. The docking interactions and MD simulation results confirmed that compound 24 interacted with the active site through a coordination bond with zinc, additional cation–π interactions with TRP 207 and 329, and crucial π-stacking interactions with TRP 329. This novel inhibitor with a piperidine4-carboxamide scaffold hold the future prospective for designing novel QC inhibitors.

24

5

QSAR-Based Drug Design for QC Inhibitors QSAR studies generate a mathematical correlation between structure and biological activity of molecules. The fundamental concept of QSAR is that variations in structural properties like physiochemical properties and biological properties result in variation of activity. Hence, QSAR studies involve the development of novel drug molecules through the application of mathematical and statistical methods by generating predictive QSAR models [37, 38]. In order to rationalize the QC inhibitory activity of 45 imidazole derivatives, a QSAR study was carried on them [39]. 2D AUTO and atom-centered fragment descriptors were selected to correlate the inhibitory activity of QC inhibitors with the help of the application of CP-MLR and PLS methods of model generation. 2D AUTO descriptors suggested that the polarizability of molecular fragments of one, six, and seven path lengths can modify the inhibitory activity. The presence of hydrogen on the unsaturated carbon was found unfavorable, whereas the presence of a greater number of hydrogens on unsubstituted carbon and methylsubstituted imidazole was found favorable for the activity. The descriptors generated verified the importance of molecular volume, refractivity, and molecular volume for the activity as well as the binding of inhibitors to the active site of receptor. The models developed were statistically validated and predictive for further development of novel QC inhibitors.

396

Kiran Bagri et al.

A QSAR modelling study on a dataset of 125 QC inhibitors was described [40]. Descriptors of correlation weight were used as optimal descriptors, and the QSAR modelling used SMILESbased Monte Carlo algorithm. The models developed were statistically validated according to OECD. The studies highlighted important structural features like methyl-substituted imidazole, propyl linker with urea, and methoxy-substituted phenyl ring, which contributed positively toward the activity, whereas halogen-substituted phenyl was found to have a negative effect on the activity. Novel molecules were designed with the help of structural features identified by QSAR models. Compound 25 was found to be the most potent with predicted pKi of 4.05. It was further docked into the active site, and the study showed all notable interactions for efficient binding like a coordination bond with zinc, π–π stacking interaction with PHE 325, and hydrogen bonding with TRP 207 and TRP 329. The results of the study could pave a new path for drug discovery in AD treatment.

25

The computational studies are summarized in Table 1.

6

Conclusions Over the past few years, there has been abundant clinical evidence regarding involvement of QC and pGlu-Aβ in the pathophysiology of AD. Thus, regulating the formation pGlu-Aβ by inhibiting QC has been seen as a promising disease-modifying approach in the treatment of AD. The process of designing and synthesizing inhibitors has been accelerated by the use of computational drug discovery approaches. Initially, synthetic studies, coupled with homology modelling, helped in recognizing imidazole and imidazole derivatives as effective QC inhibitors. The imidazole ring was identified crucial for activity as it coordinated with the zinc ion present at the active site. Subsequent modification in the structure, along with molecular modeling studies, highlighted specific key residues in the active site. Some of the important interactions were a coordination bond with zinc, hydrogen bonding with GLN 304, and stacking interactions with PHE 325 and TRP 299.

Notes

The combination of homology modelling and MD simulations predicted the possible binding pose as well as interactions

Compound 8 was found to be two Docking studies could predict the times more potent than PBD150 binding interactions as well as could correlate the SAR of synthesized molecules

Charmm22 forcefield (MOE), GOLD

–

Imidazole derivatives Homology modelling and molecular dynamics

N-aryl imidazolylpropyl thiourea derivatives

Scaffold design based on previously reported docking interactions

(continued)

[25]

[24] MD combined with homology modelling conserved the active site and possibly showed the situation of the actual QC active site. The structural differences could be further rectified with 3D protein structure

The hypothesis that zinc interacted [23] with imidazole nitrogen could only be proved with 3D structure of QC enzyme

(i) H-GLN-PHE-ALA-NH2 was identified as a substrate mimicking the active site (ii) Coordination bonding with zinc (iii) First potent QC inhibitor (compound 3/PBD 150) was identified

MOE

Homology modelling

References

The role of zinc in catalysis remains [21] (i) Potent N-1 imidazole unclear derivatives (ii) HIS 440 and HIS 330 vital for catalysis (iii) Enzyme inactivation by metal chelating groups and reactivation by divalent metal ions

Highlighted features

Imidazole-propylthiourea/ thioamide derivatives

Program/ method used ClustalW

Computational approach

Imidazole derivatives Homology modelling

Derivatives

Table 1 Summary of computational studies on glutaminyl cyclase enzyme inhibitors Computational Methods for the Design and Development of Glutaminyl Cyclase. . . 397

Glide SP, PyMOL

Glide SP, Maestro

Imidazole derivatives Docking and QPLD

Imidazole derivatives Docking and QPLD

MOE, GOLD

MOE and PyMOL

Docking study

Apigenin derivatives

Program/ method used

Diphenyl conjugated Docking study imidazole derivatives

Computational approach

Derivatives

Table 1 (continued)

QPLD helped in optimization and [30] Substantial change in the refinement of protein–ligand orientation of GLU 327 was complexes and supported the observed, and salt bridge hypothesis that D region showed interactions were reported with additional interactions with the QPLD-based optimization of QC active site docking studies. An additional hydrogen bond with TYR 299 and π–π interactions with PHE 325 was reported due to the benzyl linker between C and D regions

The interactions of D regions were [29] All notable interactions were visualized only by QPLD; observed, and the newly added D therefore, the docking results region showed interactions with were improvised and optimized GLU 327, hence justifying the by adopting QM-based docking rationale behind the design and synthesis

[28] Docking highlighted the importance of distance between imidazole and biphenyl as crucial parameters in order to achieve coordination bonding between zinc and imidazole and π–π stacking interactions between aromatic moiety and PHE 325

π–π stacking interactions between aromatic moiety and PHE 325 and chelation between imidazole and zinc ion were reported

References

Docking studies aided in analyzing [27] the binding mode as well as interactions with residues of active site

Notes

Binding interactions of C5-OH were π–π stacking interactions with TRP 207 and hydrogen bonding with GLN 304, while C7-OH derivatives also showed interactions with zinc

Highlighted features

398 Kiran Bagri et al.

Autodock4.2, GROMACS.

Imidazole derivatives Docking, molecular dynamics without urea/ thiourea scaffolds

Benzimidazole with High-throughput heteroaryl scaffold virtual screening and docking GOLD

Glide SP, Maestro, PyMOL

N-substituted urea/ Docking and QPLD thiourea imidazole derivatives

N-(-4Docking aminoalkylphenyl) thiourea/urea derivatives

Glide SP, Maestro, PyMOL

Imidazole derivatives Docking and QPLD

[33]

MD simulations presented an For efficient binding to an active advantage over docking by site, a balance of electrostatic and refining the three-dimensional van der Waals interactions is structure of protein–ligand required. Zinc ion was proposed complex to have a strong influence during binding

The interaction of zinc with benzimidazole recognized PHE 325, TRP 329, and TRP 207 as important residues of the active site

(continued)

Docking along with site-directed [35] mutagenesis study recognized the amino acids in the active site and identified the binding mode of ligand

[34]

[32]

The conformational restriction of The conformational analysis of inhibitor and additional N-substituted urea/thiourea interactions was observed by resulted in Z-E conformation using quantum mechanics that docked well into the active polarized ligand docking site. Additional hydrophobic interactions between imidazole and LEU 249 and TRP 207 and between N- substituted piperidinyl and TYR 299, ILE 303, and VAL 302 were observed

The synthesized inhibitor showed Docking studies aided in interactions with all key residues rationalizing the hypothesis of designed scaffolds

[31]

An enhanced inhibitory activity was The additional interactions were possible because of sequential due to the salt bridge interaction optimization through QPLD and hydrogen bonding of protonated amine group with GLU 327 and PRO 326

Computational Methods for the Design and Development of Glutaminyl Cyclase. . . 399

Pharmacophore-based virtual screening, docking, MD

QSAR

QSAR, docking

45 imidazole derivatives

125 Imidazole derivatives

Computational approach

Library of compounds

Derivatives

Table 1 (continued)

Highlighted features

Notes

References

Monte Carlo, AutoDock Vina

CP-MLR and PLS

Methyl-substituted imidazole, propyl linker, and methoxysubstituted phenyl ring contributed positively toward activity. Novel compounds were designed and showed interactions with all important residue of active site

[40] Statistically developed and validated models were used for in silico activity prediction of new compounds. Docking studies were performed to ascertain the binding of new compounds to active site

QSAR model identified that [39] The significance of molecular fragments of 1,6, and 7 path volume, molar refractivity, and lengths, sp3 carbon, and methylpolarizability toward QC activity substituted imidazole can modify was established. The statistical the inhibitory activity importance of developed MLR and PLS models could only be justified by designing and predicting the activity of new inhibitors

Schro¨dinger, Piperidine-4-carboxamide moiety The pharmacophore generated the [36] Namiki–-Shoji was identified as a potent scaffold advantage of both structuredatabase as it showed interactions with based and ligand-based zinc and π–π stacking approaches. Ligands were interactions with TRP 207 and precisely identified from standard TRP 329 and extra precision docking, and MD simulations were performed to evaluate the binding stability in the active site

Program/ method used

400 Kiran Bagri et al.

Computational Methods for the Design and Development of Glutaminyl Cyclase. . .

401

Quantum-based docking analysis (QPLD) revealed the favorable conformation of ligands, while binding to the active site and SAR studies identified the structural features for the preferred binding pose. Virtual screening with site-specific mutagenesis provided a new approach for designing potent QC inhibitors. Few QSAR models were reported and enhanced the knowledge regarding structural features for efficient QC inhibition. Although computational drug discovery has widened the knowledge, there is still room for more studies in order to have more potent GC inhibitors, which could be beneficial in modifying AD therapeutics. References 1. Kametani F, Hasegawa M (2018) Reconsideration of amyloid hypothesis and tau hpothesis in Alzheimer’s disease. Front Neurosci 12. https://doi.org/10.3389/fnins.2018.00025 2. Mendiola PJ, Berumen LC, Padilla K et al (2016) Therapies for prevention and treatment of Alzheimer’s disease. Biomed Res Int 2016: 1. https://doi.org/10.1155/2016/2589276 3. Hardy J, Selkoe DJ (2002) The amyloid hypothesis of Alzheimer’s disease: progress and problems on the road to therapeutics. Science 297(5580):353–356. https://doi.org/ 10.1126/science.1072994 4. Sathya M, Premkumar P, Karthick C et al (2012) BACE1 in Alzheimer’s disease. Clin Chim Acta 414:171–178. https://doi.org/ 10.1016/j.cca.2012.08.013 5. Evin G, Hince C (2013) BACE1 as a therapeutic target in Alzheimer’s disease: rationale and current status. Drugs Aging 30(10):755–764. https://doi.org/10.1007/s40266-0130099-3 6. Perez-Garmendia R, Gevorkian G (2013) Pyroglutamate-modified amyloid beta peptides: emerging targets for Alzheimer’s disease immunotherapy. Curr Neuropharmacol 11(5): 4 9 1 – 4 9 8 . h t t p s : // d o i . o r g / 1 0 . 2 1 7 4 / 1570159X11311050004 7. Bridel C, Hoffmann T, Meyer A et al (2017) Glutaminyl cyclase activity correlates with levels of Aβ peptides and mediators of angiogenesis in cerebrospinal fluid of Alzheimer’s disease patients. Alzheimer’s Res Ther 9(1): 38. https://doi.org/10.1186/s13195-0170266-6 8. Jawhar S, Wirths O, Schilling S et al (2011) Overexpression of glutaminyl cyclase, the enzyme responsible for pyroglutamate A{beta} formation, induces behavioral deficits, and glutaminyl cyclase knock-out rescues the behavioral phenotype in 5XFAD mice. J Biol Chem

286(6):4454–4460. https://doi.org/10. 1074/jbc.M110.185819 9. Schilling S, Appl T, Hoffmann T et al (2008) Inhibition of glutaminyl cyclase prevents pGluAbeta formation after intracortical/hippocampal microinjection in vivo/in situ. J Neurochem 106(3):1225–1236. https://doi.org/ 10.1111/j.1471-4159.2008.05471.x 10. Huang KF, Liu YL, Cheng WJ et al (2005) Crystal structures of human glutaminyl cyclase, an enzyme responsible for protein N-terminal pyroglutamate formation. Proc Natl Acad Sci U S A 102(37):13117–13122. https://doi. org/10.1073/pnas.0504184102 11. Vijayan DK, Zhang K (2019) Human glutaminyl cyclase: structure, function, inhibitors and involvement in Alzheimer’s disease. Pharmacol Res 147:104342. https://doi.org/10.1016/j. phrs.2019.104342 12. Xu C, Wang YN, Wu H (2021) Glutaminyl cyclase, diseases and development of glutaminyl cyclase inhibitors. J Med Chem 64(10): 6549–6565. https://doi.org/10.1021/acs. jmedchem.1c00325 13. Gunn AP, Wong BX, McLean C et al (2021) Increased glutaminyl cyclase activity in brains of Alzheimer’s disease individuals. J Neurochem 156(6):979–987. https://doi.org/10. 1111/jnc.15114 14. Hoffmann T, Meyer A, Heiser U et al (2017) Glutaminyl cyclase inhibitor PQ912 improves cognition in mouse models of Alzheimer’s disease-studies on relation to effective target occupancy. J Pharmacol Exp Ther 362(1): 119–130. https://doi.org/10.1124/jpet.117. 240614 15. Scheltens P, Hallikainen M, Grimmer T et al (2018) Safety, tolerability and efficacy of the glutaminyl cyclase inhibitor PQ912 in Alzheimer’s disease: results of a randomized, doubleblind, placebo-controlled phase 2a study.

402

Kiran Bagri et al.

Alzheimers Res Ther 10(1):107. https://doi. org/10.1186/s13195-018-0431-6 16. Cynis H, Scheel E, Saido TC, Schilling S et al (2008) Amyloidogenic processing of amyloid precursor protein: evidence of a pivotal role of glutaminyl cyclase in generation of pyroglutamate-modified amyloid-beta. Biochemistry 47(28):7405–7413. https://doi. org/10.1021/bi800250p 17. Morawski M, Schilling S, Kreuzberger M et al (2014) Glutaminyl cyclase in human cortex: correlation with (pGlu)-amyloid-β load and cognitive decline in Alzheimer’s disease. J Alzheimers Dis 39(2):385–400. https://doi.org/ 10.3233/JAD-131535 18. Coimbra JR, Sobral PJ, Santos AE et al (2019) An overview of glutaminyl cyclase inhibitors for Alzheimer’s disease. Future Med Chem 11(24):3179–3194. https://doi.org/10. 4155/fmc-2019-0163 19. Ramı´rez D (2016) Computational methods applied to rational drug design. Open Med Chem J 17–20:7. https://doi.org/10.2174/ 1874104501610010007 20. Yu W, MacKerell AD Jr (2017) Computeraided drug design methods. Methods Mol Biol 1520:85–106. https://doi.org/10. 1007/978-1-4939-6634-9_5 21. Schilling S, Niestroj AJ, Rahfeld JU et al (2003) Identification of human glutaminyl cyclase as a metalloenzyme. Potent inhibition by imidazole derivatives and heterocyclic chelators. J Biol Chem 278(50):49773–49779. https://doi.org/10.1074/jbc.M309077200 22. Booth RE, Lovell SC, Misquitta SA et al (2004) Human glutaminyl cyclase and bacterial zinc aminopeptidase share a common fold and active site. BMC Biol 2. https://doi.org/10. 1186/1741-7007-2-2 23. Buchholz M, Heiser U, Schilling S et al (2006) The first potent inhibitors for human glutaminyl cyclase: synthesis and structure-activity relationship. J Med Chem 49(2):664–677. https://doi.org/10.1021/jm050756e 24. Buchholz M, Hamann A, Aust S et al (2009) Inhibitors for human glutaminyl cyclase by structure based design and bioisosteric replacement. J Med Chem 52(22):7069–7080. https://doi.org/10.1021/jm900969p 25. Tran PT, Hoang VH, Thorat SA et al (2013) Structure-activity relationship of human glutaminyl cyclase inhibitors having an N-(5-methyl-1H-imidazol-1-yl)propyl thiourea template. Bioorg Med Chem 21(13): 3821–3830. https://doi.org/10.1016/j.bmc. 2013.04.005 26. Huang KF, Liaw SS, Huang WL et al (2011) Structures of human Golgi-resident glutaminyl

cyclase and its complexes with inhibitors reveal a large loop movement upon inhibitor binding. J Biol Chem 286(14):12439–12449. https:// doi.org/10.1074/jbc.M110.208595 27. Li M, Dong Y, Yu X et al (2016) Inhibitory effect of flavonoids on human glutaminyl cyclase. Bioorg Med Chem 24(10): 2280–2286. https://doi.org/10.1016/j.bmc. 2016.03.064 28. Li M, Dong Y, Yu X et al (2017) Synthesis and evaluation of diphenyl conjugated imidazole derivatives as potential glutaminyl cyclase inhibitors for treatment of Alzheimer’s disease. J Med Chem 60(15):6664–6677. https://doi. org/10.1021/acs.jmedchem.7b00648 29. Hoang VH, Tran PT, Cui M et al (2017) Discovery of potent human glutaminyl cyclase inhibitors as anti-Alzheimer’s agents based on rational design. J Med Chem 60(6): 2573–2590. https://doi.org/10.1021/acs. jmedchem.7b00098 30. Ngo VTH, Hoang VH, Tran PT et al (2018) Potent human glutaminyl cyclase inhibitors as potential anti-Alzheimer’s agents: structureactivity relationship study of Arg-mimetic region. Bioorg Med Chem 26(5):1035–1049. https://doi.org/10.1016/j.bmc.2018.01.015 31. Ngo VTH, Hoang VH, Tran PT et al (2018) Structure-activity relationship investigation of Phe-Arg mimetic region of human glutaminyl cyclase inhibitors. Bioorg Med Chem 26(12): 3133–3144. https://doi.org/10.1016/j.bmc. 2018.04.040 32. Hoang VH, Ngo VTH, Cui M et al (2019) Discovery of conformationally restricted human glutaminyl cyclase inhibitors as potent anti-Alzheimer’s agents by structure-based design. J Med Chem 62(17):8011–8027. https://doi.org/10.1021/acs.jmedchem. 9b00751 33. Tran PT, Hoang VH, Lee J, Hien TT et al (2019) In vitro and in silico determination of glutaminyl cyclase inhibitors. RSC Adv 9(51): 29619–29627. https://doi.org/10.1039/ c9ra05763c 34. Van MN, Hoang VH, Ngo VTH et al (2021) Discovery of highly potent human glutaminyl cyclase (QC) inhibitors as anti-Alzheimer’s agents by the combination of pharmacophorebased and structure-based design. Eur J Med Chem 226:113819. https://doi.org/10. 1016/j.ejmech.2021.113819 35. Ramsbeck D, Buchholz M, Koch B et al (2013) Structure-activity relationships of benzimidazole-based glutaminyl cyclase inhibitors featuring a heteroaryl scaffold. J Med Chem 56(17):6613–6625. https://doi.org/ 10.1021/jm4001709

Computational Methods for the Design and Development of Glutaminyl Cyclase. . . 36. Dileep KV, Sakai N, Ihara K et al (2021) Piperidine-4-carboxamide as a new scaffold for designing secretory glutaminyl cyclase inhibitors. Int J Biol Macromol 170:415–423. https://doi.org/10.1016/j.ijbiomac.2020. 12.118 37. Winkler DA (2002) The role of quantitative structure--activity relationships (QSAR) in biomolecular discovery. Brief Bioinform 3(1): 73–86. https://doi.org/10.1093/bib/3.1.73 38. Kwon S, Bae H, Jo J, Yoon S (2019) Comprehensive ensemble in QSAR prediction for drug discovery. BMC Bioinform 20(1):521. https://doi.org/10.1186/s12859-0193135-4

403

39. Kumar V, Gupta MK, Singh G et al (2013) CP-MLR/PLS directed QSAR study on the glutaminyl cyclase inhibitory activity of imidazole: rationales to advance the understanding of activity profile. J Enzyme Inhib Med Chem 28(3):515–522. https://doi.org/10.3109/ 14756366.2011.654111 40. Kumar A, Bagri K, Nimbhal M et al (2021) In silico exploration of the fingerprints triggering modulation of glutaminyl cyclase inhibition for the treatment of Alzheimer’s disease using SMILES based attributes in Monte Carlo optimization. J Biomol Struct Dyn 39(18): 7181–7193. https://doi.org/10.1080/ 07391102.2020.1806111

Part IV Special Topics

Chapter 14 Basic Information Science Methods for Insight into Neurodegenerative Pathogenesis Thomas H. W. Lushington, Mary I. Zgurzynski, and Gerald H. Lushington Abstract People have long sought to cure Alzheimer’s disease through the pursuit of a “know thine enemy” type of target discovery, but the better strategy might be to “know thyself.” Positing amyloid aggregation as the key pathogenic enemy has failed nearly every major Alzheimer’s clinical test to date. Better outcomes may emerge from intelligently prioritizing among the many coupled biochemical processes that dysfunction in Alzheimer’s patients—a “know thyself” philosophy that explores how local biochemical deviations cascade toward disease. Unfortunately, human brains are difficult to experiment on, while reliable animal or in vitro neurodegenerative models remain elusive. Fortunately, volumes of prior biomedical research contain clues on early premorbid risk factors that statistically presage neurodegeneration. This accumulated biochemical evidence may illustrate how non-neurological disorders may lay foundations of neurodegeneration long before clinical neuropathology manifests. It is challenging to intuit which of the many peripheral ailments may be relevant to downstream neuropathology. This chapter thus offers information-based suggestions for how to pursue novel target perception, outlining simple recipes and examples of how literature query tools and techniques may illuminate insight from emerging research trends. Such exploration may then spur sophisticated informatics (e.g., natural language processing, meta-analysis, etc.) and experimental verification. Key words Neurodegeneration, Alzheimer’s disease, Autoimmune disorders, Information science, Pathogenesis, Metabolic disorders, Endocrinology, Google Scholar, PubMed

1

Introduction The widely acknowledged tribulations of the biologic drug Aduhelm [1, 2] are only the most recent page in a long saga of disappointments in neurodegenerative drug development. Our chances for escaping these disappointments are contingent, more than anything, on two goals that seem remarkably self-evident: (1) We must prioritize new targeting paradigms over those that have led to perennial failure, and (2) these new interpretations must concur with markers that demonstrably distinguish neurological decline versus sustained health.

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_14, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

407

408

Thomas H. W. Lushington et al.

These statements may seem obvious but are a tacit admission of prolonged failure. Armed with a manifold of canonical paradigms, the neuromedical community has consistently failed to progress toward an effective treatment of terminal neuropathologies. Specifically, diseases such as Alzheimer’s disease (AD), amyotrophic lateral sclerosis (ALS) [3], frontotemporal dementia (FTD) [4], and dementia with Lewy bodies (DLB) [5] still lack practical treatment paradigms, and fairly few speculative scientific papers have emerged to foster optimism for novel insight capable of transforming the landscape. This stagnation was our primary reason for not reprising our earlier NeuroMethods chapter, which focused on simulating the pathology and potential pharmaceutical remediation of protein misfolding [6]. The previous chapter was an attempt aimed at addressing the amyloid aggregation observed in many degenerative neuropathologies and which formed the pharmacology motivation underlying the flawed Aduhelm paradigm. Instead, to contribute more meaningfully to the stimulation of renewed advancement in neurodegenerative research, it is the goal of this chapter to introduce information science strategies aimed at breaking away from unproductive research paradigms and finding novel facets of pathopharmacology space that avail tangible new opportunities. Finding promising new mechanistic frameworks for treating dementia often relies on adapting existing paradigms to produce new disease models that successfully harmonize prior analytical success with plausible work-arounds for past failures. In this respect, the amyloid/misfolding model does continue to provide useful insight in that amyloid plaques are definitely present in AD and proteins unquestionably do misfold during the many neurodegenerative pathologies, as well as other chronic autoimmune and metabolic disorders, but protein misfolding and aggregation are clearly not the whole story. From a pharmacological perspective, the recent clinical failures of the anti-plaque antibody Aduhelm have amplified a previously growing opinion that protein misfolding is more of a frequent symptom [7–10], rather than a practicable cause, of neurodegeneration. Relegating protein misfolding from cause to symptom opens up valuable rhetorical questions. Specifically, what sort of biochemical cascade(s) can account for the fundamental progression of gross chronic and degenerative neuropathology (i.e., excessive synaptic pruning, neuronal death, cognitive decline, motor control disturbances, etc.) while simultaneously also rationalizing elevated accumulations of degraded proteins? Can such a mechanism explain the former without inferring that it is caused by the latter? Do any of the other (i.e., non-misfolding) biochemical dysfunction lead to potential avenues for robust medical intervention? This volume of NeuroMethods seeks to instruct in the application of a robust range of different research methods capable of

Information Science Methods for Neurodegenerative Insight

409

addressing pointed research questions required for developing new treatments for AD and related neuropathologies. To further this discourse, the goal of our chapter is to address a different sort of question. Namely, how can we find the right question? As the “Information Age” unfolds, what new learning paradigms might illuminate Alzheimer’s targets where so many others have failed? If so many transformative technologies (e.g., molecular simulations, high-throughput screening, genomic profiling, brain imaging, machine learning, etc.) have failed to crack the code, what are we missing? In fact, it is quite reasonable to assert that prior published scientific studies contain numerous important clues that await our rediscovery. Finding the precise needles in the proverbial haystack poses a challenge, given the huge volume of prior publications that are either irrelevant to the task of treating neurodegeneration or an additional subset whose results or conjectures are ultimately irreproducible. Nonetheless, it is the assertion of this chapter that context-sensitive search algorithms such as those that drive resources like Google Scholar can, when flanked by sophisticated keyword ranking query tools like PubMed, shed a great deal of potential, insight when applied according to a human’s judicious and perceptive search strategies.

2

Chronic Disease Pathology as a Mineable Source of Neurodegenerative Insight In the years since the previous edition of our Alzheimer’s-oriented NeuroMethods chapter, challenging questions such as those articulated in the previous section have continued to propel a communal search for new ways to approach hypothesis discovery. Unfortunately, moving beyond old paradigms has proven slow and challenging, perhaps because the latter often sustain the momentum that can perpetuate a long string of disappointments, while totally new schemes are often suppressed by the “preliminary results” circularity, whereby funds are only allocated to prospects with tangible promising data, while producing tangible data requires preliminary funding. In terms of evolving away from unsuccessful paradigms, a key reason why the amyloid hypothesis has been supported for so long is that in addition to targeting plaques (the most obvious postmortem evidence of Alzheimer’s), the paradigm fell within the conventional “adversarial” class of drug development efforts. Adversarial strategies, in which a specific “pathogen” is posited and directly targeted for removal, have achieved long-standing success in areas like cancer and infectious disease but have displayed far less promise for matters of chronic autoimmune and neurological pathology. Cancers, microbes, and viruses are targeted adversarially by exploiting physiologically unique aspects of the pathogen

410

Thomas H. W. Lushington et al.

that can be differentially modulated via drugs designed to have a far less negative effect on healthy human physiology. Unfortunately, many forms of the chronic autoimmune and neurological disease respond poorly to adversarial targeting, since the pathology seems to manifest from marginal dysregulation of otherwise important physiological processes. The amyloid paradigm was formulated as a prospective adversarial strategy (i.e., remove the plaques or block the proteins that feed them), but unfortunately, the strategy has consistently failed to improve patient health. From a net assessment of clinical outcomes, Aduhelm has proven statistically successful in reducing plaques but broadly fails to reduce (and indeed often exacerbates) patient cognitive decline [1, 2]. Beyond amyloid, very few promising adversarial opportunities have emerged for mitigating degenerative dementias or comparable neuromuscular disorders. While many papers report evidence that a medical history of infections or cancer may influence susceptibility to neurodegenerative cascades, the case for causative links between known pathogens and the common forms of neurodegeneration is weak or indirect. Thus, any approach with demonstrable success in treating cancer and infectious disease remains of substantial value to medicine, but no antibiotics, antivirals, or cancer drugs have yet proven efficacious in treating Alzheimer’s, FTD, DLB, Parkinson’s, or ALS. Indeed the absence of evidence does not rigidly infer the evidence of absence, so the future may reveal successful adversarial approaches for treating diseases like Alzheimer’s, but the current impasse, and the surging global statistics for chronic and degenerative neuromedical caseloads, demands alternatives. One process worth considering is the systematic incremental adjustment of prior candidate treatments that exhibited preliminary, if inadequate, promise. Such candidates might have produced anecdotal success that did not weather fuller statistical analysis or might have achieved some therapeutic effect, but with untenable risks of side effects or toxicity. Indeed, biomedical research has achieved demonstrable practical success from incremental refinement efforts. For neurodegeneration, however, the basis for refinement is weak due to a paucity of studies demonstrating promising, replicated pharmacological effects. Thus, while strategies worthy of incremental refinement might eventually emerge, discovering them may require intuition, creativity, faith, or some combination of all three. Perhaps, for example, the path toward overcoming the lack of promising Alzheimer’s prospects will emerge from inferring possible benefits from candidate interventions that address pathologies that are distinct from, but biochemically similar to, the molecular dysfunction evident in Alzheimer’s. One practical requirement of using indirect incrementalism for new target rationalization entails establishing clear mechanistic evidence that the two endpoints are interrelated. Such mechanistic insight may be derived through rigorous molecular profiling (i.e., genomics, proteomics, and other omics analyses that resolve

Information Science Methods for Neurodegenerative Insight

411

comparable biochemistry), statistical and informational assessment (e.g., seeking significant levels of anecdotal consistency across inferences suggesting a plausible relationship), or, better yet, a combination of both via corroborative references that report rigorous experimental findings. In the specific case of AD, a substantial volume of evidence (ranging from biomedical publications to statistical trends in clinical data) has emerged to suggest that neurodegenerative vulnerability may be influenced by prior struggles with autoimmune, endocrine, and/or metabolic (AEM) disorders [11–14]. Superficially, these studies do not seem to converge toward a common biochemical mechanism, but objective examination of the underlying molecular basis for the various disorders may help draw analogies, especially if some common dysregulation can be identified that spans both the biochemical functions required for proper AEM function along with analogous brain physiology. In this sense, protein misfolding is noted in a diverse array of AEM disorders. The formation of pathological aggregates is most often regarded as a vehicle for the development of adjunct complications (e.g., renal pathology in diabetes [15] and lupus [16]). Aggregates do not, themselves, seem to directly disrupt crucial AEM physiology and are rarely considered to be an instigator of the primary etiological cascade; thus, it is not clear that the AEM manifestations of protein disordering should be considered crucial triggers of neurodegeneration. It may hopefully prove more productive to look for neurodegenerative targeting insight from specific biochemical dysfunctions that are considered to be fairly crucial differentiators between AEM physiology and pathology. One possible framework within which to begin such a search is a fundamental commonality that is crucial to the regulatory function of neurological, immune, endocrine, and metabolic processes, namely, the conserved reliance on cell polarization. Polarization is a vital evolutionary adaptation that permits cells to functionally adapt to the presence of other cells and to a diverse range of other environmental influences that impinge on a cell’s prospects to survive, thrive, and participate in a communal, organismal organization. Neurons, for example, function in an organismal nervous system to form and regulate responses (both with other neurons and with various other cells) to acquire and form patterned retention (memory) of environmental information and encode adaptive responses that serve as current or future instructions for cell function. Cell polarization is crucial to both the functional existence of multicellular eukaryotes and to the structure of the neural frameworks that regulate them. Polarization provides cells with the capacity to exploit, embrace, or evade entities in the extracellular environment inputs [17], thus comprising a fundamental capstone of sensory perception and information storage. Extensive

412

Thomas H. W. Lushington et al.

proteomic conservation between neurological, metabolic, developmental, and immune pathways [18–23] provides a strong basis for arguing that an organism’s neural machinery is not a unique segregated component of physiology, but rather a specialized adaptation of the same sort of polarization functions available for other adaptive requirements. Specifically, phylogenetic profiling suggests that rather than reinvent biochemical tools for each of the distinct environmentally sensitive physiological functions in an organism, evolution differentially adapted a core set of biochemical tools for similar yet distinct purposes manifest in neurological, autoimmune, endocrine, and metabolic organismal requirements [24–26]. There are both similarities and differences in how cell polarization processes support neural function, organismal metabolism, development, and immune action. The adaptive immune system sustains a secondary pathogen response that is bolstered by both an extensive and long-lasting memory [27, 28] (as is encoded via cell polarization [29]) and strategic cytokine-driven communication networks [30], which spur functional repolarization as a basis for customized immune response [31]. Interestingly, cytokines also play major roles in regulating metabolic and endocrine function [32, 33], but are not direct mediators of neurocognition and memory. Rather, neuronal communication occurs not by diffusive cytokine exchange but via privileged biophysical conduits known as axons. Nonetheless, it should be noted that axon structures are maintained by cytokines [34], which also foster synaptic potentiation through which axons reach their neuronal targets [35], and thus are important indirect actors in neurophysiology. The substantial overlap of both core physiological function and shared trends in co-pathological vulnerability raises the question of whether medical advances in endocrinology, metabolism, and immunology might be leveraged toward neurodegenerative research. If so, how might one prioritize targets? Answers to these questions are not obvious, since neurology, immunology, endocrinology, and metabolism are distinct subdisciplines, each with their own specialized journals, programs, and departments, with only modest levels of cross-disciplinary investigation. As in many scientific fields, research in these disciplines is often driven by focused specialists whose individual contributions rarely stray far from welldefined personal niches. This siloed approach to these key disciplines works against prospects for leveraging cross-disciplinary pathophysiologic analogies toward germinating new neurodegenerative hypotheses. In order to foster objective cross-disciplinarity, this chapter will outline how rudimentary information retrieval can aid in neuropharmacological hypothesis seeding by exploring clues present in established biochemical pathologies associated with chronic but treatable AEM diseases that have been posited as prospective neurodegenerative risk factors or co-morbidities.

Information Science Methods for Neurodegenerative Insight

413

3 Information Retrieval Strategies for Intuiting Neurodegenerative Mechanistic Hypotheses There is a growing consensus that Alzheimer’s disease, Parkinson’s, ALS, and the various terminal dementias are attributable at least in part to triggers dating back years (possibly decades) before gross neuropathology is clearly manifest [36–41], so careful mapping and assessment of clinical and biomolecular risk factors may foster new insight relevant to early diagnosis and preemptive therapeutics for delaying or mitigating eventual neurodegeneration. Fortunately, decades of medical record keeping have helped us perceive numerous health conditions that appear, statistically, to be tangible neurodegenerative risk factors. Without explicitly implying that a statistical risk factor for neurodegeneration is an outright cause, one can nonetheless assemble these factors into a framework suitable for postulating, codifying, and developing tests for prospective causal hypotheses via the expression below: AEM

BC ) ND

ð1Þ

where: • AEM are adverse risk factors, largely comprised of aspects of autoimmune, endocrine, and metabolic dysregulation, many of which are traceable to diet, gut dysbiosis, exposure to toxins, stress, or incomplete recovery from infectious pathology. •

indicates concurrence or correspondence.

• BC are biochemical dysregulations that may emerge from the AEM conditions. • ND indicates a neurodegenerative cascade that may hypothetically emerge from chronic exposure to the specific BC that originates from AEM. Expression (1) can form the basis for information retrieval through the use of common text-based research-oriented search tools, using intersection set queries that may deliberately and objectively sample both associations that we may already suspect (e.g., chronic inflammation as a precursor to dementia) as well as those that we may not have intuited (e.g., respiratory disease versus neuromuscular disorders). Such sampling might proceed via a systematic survey over specific [AEM + BC + ND] intersection set queries to see how many records are retrieved for a given query relative to controls, where a positive control might, for example, just be the term “Parkinson’s” without BC and AEM intersections (i.e., the set of all records that somehow reference Parkinson’s), whereas negative controls might correspond to the average number of hits obtained over a list of artificial queries that can be regarded as scientifically meaningless (e.g., “baseball” + “cabbage” + “Parkinson’s”). Such studies, although perhaps offering broad statistical

414

Thomas H. W. Lushington et al.

information, are not yet particularly helpful for novel discovery. The number of prospective BC terms (ranging from tens of thousands of genes to many millions of known proteoforms) is inconveniently large, which yields an impractical number of distinct searches to be run, of which many will necessitate some degree of supervision. Specifically, a massive search producing a large number of hits is potentially vulnerable to contextual inconsistency, whereby many hits may contain a desired search term (e.g., “depression”), but with a scattered mix of the intended connotation (i.e., the psychiatric manifestation) and many other spurious uses (i.e., a pit or hollow; the act of physically lowering something, etc.). There are further issues of whether even an instance of a search term with the desired connotation actually has any tangible association with other words or phrases in the query. For example, if the word “depression” appears only on page 2 of an article, while another query term “hypothyroidism” is only found on page 6, does their co-joint appearances in the same document imply any scientifically relevant association? Fortunately, more practicable strategies have emerged for balancing a desire for broad sampling with practical limitations of result supervision. As will be discussed in the next section, the context sensitivity of some search tools has evolved to the point where the search engine may be able to rank (and thus pre-filter) hits according to metrics that afford some confidence in true relevance, while some tools have produced mechanisms (filters or annotations) capable of assessing which BC terms are plausibly relevant to a given AEM + ND pairing, thus reducing the need for exhaustive systematic searches across broad ranges of BC candidates.

4

Resources In recent decades, information search engines have emerged to the extent that, given strategic care and attention to detail, it should be possible to use the associative schemes discussed in the last section to evolve new neuromedical hypotheses based on intriguing relationships spanning prior publications in the neurological and immunological sciences. At the heart of the pathogenetic discovery process outlined in this chapter is an assumption that informative search tools exist that are capable of identifying prior research that is potentially relevant to a current biomedical challenge, while not producing an overt bias toward direct assertions of relevance. This somewhat paradoxical requirement is explored in detail in a 2018 critical review that mapped out a concern that document retrieval tools could create a “bubble” effect, whereby community preconceptions could

Information Science Methods for Neurodegenerative Insight

415

Fig. 1 Conceptual query retrieval space (a) shows regions dominated by documents that match the query intent (green patches), versus those whose hits incorrectly interpret search focus (red), or contain homonymous terms that match those in the query but have different context (blue). For illustrative purposes, the theoretical coverages are shown for (b) a context NLP search and (c) a simpler keyword search. The intersection space between these two queries (shown as (d); also set aside to the lower right for clarity) retrieves only a subset of potentially useful material but tends to produce fewer mismatches

become embedded into search algorithms that retrieved hits that would be expected, rather than hits that were more objectively associative [42]. In a societal sense, this sort of self-reinforcement may have many practical applications (e.g., finding a popular restaurant), but in an intellectually stagnant biomedical discipline, the reinforcement of the status quo is less helpful. On the other hand, a balance is required to ensure that search hits share at least some plausibly actionable biomedical commonalities. A viable strategy for overcoming the prospective bubble effect is to employ a multifaceted scheme that weighs input from both very wide-net document scans (ideally with some intelligent bias to hopefully de-emphasize the more unreliable, anecdotal, and non-technical material) and narrower keyword matches (i.e., prioritize documents according to the frequency at which they match key search terms or close synonyms thereof) [43]. This input can be conceptually represented by a simple schematic (Fig. 1), which maps regions of consequential (green coloration; results that have true relevance to the intended search), incidental (blue color; results containing terms that have some conceptual overlap with the intended search but are not truly functionally relevant), and accidental (red color; results dominated by non-synonymous homonyms that are irrelevant to search intent) hits on the prospective query space. Superimposed on this space are the hypothetical results of two different query algorithms, of which one (yellow triangle) is drawn to suggest a keyword-oriented search that should

416

Thomas H. W. Lushington et al.

not be particularly vulnerable to incidental hits but is prone to retrieving homonymous accidentals whose meaning differs from the query intent (e.g., searching for “depression” with the intent of studying psychopathological disturbance but instead retrieving gene expression or metabolomic profiles reporting quantitative molecular underabundance), while a second query (purple) applies complex inferential logic to posit useful results, but with an inherent tendency to produce false-positive incidental hits according to imperfect artificial predictions of what the searcher might be interested in (a current example of this could be queries producing mostly material on COVID-19, even if the searcher had attempted to explicitly target research on the 2003 SARS epidemic). In order to provide protocols that are accessible to a majority of likely readers, the functionality reported in this chapter corresponds largely to resources that are available without licensing restrictions or geographic limits to researchers in academia, business, and the public sector. Mention will be made of several specialized services with some degree of restricted access, but the chapter’s focus is mainly on the freely available search engines, which are briefly outlined below. Google (www.google.com) The globally ubiquitous search engine services nearly every informational facet of modern life, with most searches occurring as rudimentary black-box (i.e., unfiltered, discipline-agnostic) queries. The greatest strength of Google—its unparalleled document coverage—is also a conceptual weakness, since, for any search, the search engine must discard a colossal set of documents that are irrelevant to the search before reporting only a small fraction that may be useful. Google achieves this via a very complex and computationally demanding collection of algorithms whose details are kept rigorously secret, but several aspects are understood, including the fact that sophisticated artificial intelligence is employed to place keywords within a natural language context (most Google algorithms are undocumented, but one transformative bidirectional processing tool called BERT was reported in detail in 2019 [44, 45]) and that search results depend on the searcher. In particular, based on the originating network address, Google may adapt the results based on whether the query was launched from a university, a hospital, a corporate setting, or a private home. Furthermore, if the searcher was logged into a Google account at the time, the query may evaluate which prior topics have been queried from that account. Given Google’s extensive generality and its application to an endless array of non-technical applications, it seems a stretch to regard it as a sophisticated biomedical research tool. Yet, many scientists consider it to be an essential part of their literature search activities and may use Google queries to inform or preliminarily vet

Information Science Methods for Neurodegenerative Insight

417

novel hypotheses. The process by which Google ranks hits is not intentionally customized for biomedical discovery, but researchers who use a dedicated Google account for their scientific search activities will likely find that the search engine adapts strategically to their usage patterns, translating browsing and query history into a customized profile that refines search results accordingly. Thus, although not rigorously developed for biomedical research, a Google profile that is regularly used for research-oriented queries will grow to emulate the contextual awareness commensurate with customized research tools. Notably, the black-box nature of Google has evolved away from earlier versions in which users could exert control over search foci via search syntax capable of Boolean-like logic. Old syntax (e.g., use “+” to require that hits contain specific terms, “-” to require term exclusion, and quotation marks to require instances of specific multi-word phrases) is blindly tolerated by Google, but the specialized search instructions are often ignored. Searchers may thus not get what they may have prejudicially been looking for, but there is usually a fair chance that they still find something interesting. Google Scholar (scholar.google.com/) As a compromise between the noisy “town square” environment of Google versus a more dignified academic lecture hall, Google Scholar is a more targeted and disciplined form of the search engine. Scholar supports advanced query filters and the functional equivalent of Boolean operators while also limiting queries to curated material. Searches thus span verifiably academic source material, including papers, theses, books, repositories, patents, legal documents, and web material from academically oriented sites such as universities, publishers, professional societies, etc. It has not been clearly documented whether Scholar is empowered with the same sort of search adaptability that refines regular Google searches. However, Scholar does enable users to create a profile with all personal publications, as well as a library spanning personal interests, the contents of which are used by the search engine to inform search contextuality. PubMed (pubmed.ncbi.nlm.nih.gov) The archetypal biomedical literature search utility that (for the purposes of this chapter) can be considered to include valuable adjunct services such as MedlinePlus (literature database and annotations) and PMC (open-access full-text paper repository). PubMed is generally acknowledged (ca. 09/2022) as the leading search engine for the biomedical sciences. It supports both a naive query interface (akin to Google) and sophisticated field and filter-based advanced searches. It also offers specialized iterative services such as

418

Thomas H. W. Lushington et al.

introducing a new modifier to refine the results of a prior search or finding the intersection set of multiple prior queries. To some fanfare in 2020 [46], PubMed took a major step away from simply prioritizing query hits based on how often a specific search term appeared in the document and instead began to intelligently assess the extent to which documents have keyword densities that exceed what is seen in across a wider document pool and also adjust document hit rankings adaptively, based on accumulated query statistics evaluating how often searchers actually clicked on a given document when it was presented in prior similar queries. While augmenting PubMed with PMC provides searchers with an invaluable source of free scientific literature, the integration with MedlinePlus arguably may add more immediate benefit to the chapter’s stated emphasis on hypothesis discovery, since MedlinePlus imbues PubMed with curated hit annotation, including a specific benefit of identifying important BC information found within search hits, even for queries that may not have specified any BC search instructions. The specific cross-indexing of biomedical concepts with corresponding biochemical terms is a lynchpin of our expedited hypothesis discovery scheme, whereby a given AEM + ND search may be subsequently mined in order to identify BC species (genes, proteins, metabolites, etc.) to intuit prospective mechanistic commonalities that might explain why a given autoimmune, endocrine, or metabolic condition might be featured in the same publication that discusses a neurodegenerative disease. Others Over the past two decades, several biomedical information search utilities have emerged as useful information mining tools, to varying degrees of longevity and success. Current examples of such resources include SpringerLink (link.springer.com/), Web of Science (access.clarivate.com), Embase (www.embase.com), ScienceDirect (www.sciencedirect.com), and Epistemic AI (epistemic.ai/). Some of these exist primarily to facilitate access to specific journals of biomedical relevance, while others are paid services with novel search capabilities and sources that augment PubMed and Google Scholar while also interoperating with the information provided by the aforementioned free services. An intriguing resource with substantial topical coverage that overlaps with, but also differs from, PubMed and Scholar is researchResearchGate (https://www.researchgate.net/)—a oriented networking site that was founded in 2008 for researchers to share and interact with scholarly p;ublications, via comments, questions, and answers. Although the publication repository (mostly abstracts, but a sizable number of full-text papers) is userdriven (i.e., all entries are either initiated or approved by a registered user), the net scholarly coverage is growing to rival

Information Science Methods for Neurodegenerative Insight

419

other top literature sites. Given the grassroot nature of the paper collection, it may not list all high-impact publications prioritized by PubMed and Scholar but is conversely an excellent place to find lower caliber studies that might not be indexed by traditional literature services. This latter attribute might somewhat devalue ResearchGate as a medium for finding definitive and authoritative studies to validate or guide an established research agenda, but it is the authors’ experience that ResearchGate has simple yet powerful search functionality that may yield substantial value in illuminating interesting speculative concepts that may not have entered the scientific mainstream. A drawback inherent in ResearchGate relative to PubMed is the absence of cross-indexing with BC/MeSH terms, but ResearchGate may nonetheless yield interesting opportunities for ideation, in a manner complementary to Scholar. As a general illustration of hypothesis-driven biomedical search strategies, this chapter will focus primarily on knowledge discovery derived from Google Scholar and PubMed queries. In referring back to Fig. 1b, the context-sensitive and user-sensitive nature of Google, when placed within the more scientifically rigorous constraints of the Scholar service, provides a dynamically adaptive resource for retrieving old and new publications of potential relevance, albeit with susceptibility to reporting documents of incidental similarity, while PubMed (Fig. 1c) applies keyword-matching strategies to retrieve a semi-overlapping set of documents, though with a greater tendency to find accidental homonymous matches. As illustrated in Fig. 1d, the difference in Scholar and PubMed retrieval strategies ultimately leads to a substantial narrowing of the set of consensus documents; however, this narrowing process affords a reasonable prospect for enhancing confidence in the true relevance of the intersection set. While the chapter as written relies on these two broadly available search utilities, researchers are encouraged to explore other tools that may expand query coverage (as per ResearchGate) or add novel contextual power to searches via natural language processing, as is employed to enhance search sensitivity in resources such as Epistemic AI.

5

Methods and Notes Essentially everyone reading this chapter will have some experience in conducting research-related reference searches; thus, it is not our goal to reprise rudimentary or intuitive skills but rather to illustrate how some novel query strategies can foster new prospective insight. This said, while Google searches and advanced Google Scholar queries are basic enough to proceed without further explanation, some readers may benefit from a discussion of some subtle aspects of advanced PubMed queries before proceeding to illustrative examples.

420

Thomas H. W. Lushington et al.

Figure 1 provides a sample screenshot of PubMed’s advanced query structure (upper right) and results readout (bottom left), including the following numbered features: 1. By default, search terms entered in the upper middle text box are biomedical terms that could be present in a diverse range of fields, but it is possible to steer the search by constraining that term to a specific field. More than 40 fields are available, covering biomedical terms, specific Medline annotations (biomarkers, diseases, drugs, genes, targets, etc.), author or investigator information, publication type, or publication date. 2. This button enables one specific search criterion to be entered into a composite query within a Boolean context. 3. Each click from #2 updates an explicit composite query. It is important to note that the text accruing in box 3 can be copied directly into the PubMed simple query line and run as a search, independent of the advanced query builder. 4. This box stores the structure and results of other queries conducted in the session. This is important because stored queries can be refined with additional search constraints. For example, the query “(#1) AND (metformin)” would mine the results from the prior search #1 to find those which also contained the term “metformin.” 5. Once the query has been submitted, the results can be further modified. For example, this slider bar can hone in on publications from specific date ranges. 6. Scrolling down on the left side (not shown in full) provides more advanced filtering options, representing a subset of restrictions available on the Advanced Query form (Fig. 2). In order to illustrate a practical scheme in which the more complex PubMed search refinement capabilities can be combined with a broader primitive knowledge discovery capacity of tools like Google and Google Scholar, the chapter authors were asked to pursue a multistep search process beginning with naive preliminary queries based on pairing a prospective neurodegenerative endpoint with some chronic AEM disorder, followed by refinement steps aimed at rough assessments of pairing significance, and final elucidation of potential BC molecular markers that might be characteristic of a potential AEM + ND co-pathology relationship. Four specific case study examples are elaborated below, wherein specific protocol steps are rendered in bold italics font to distinguish them from comments and analysis.

Information Science Methods for Neurodegenerative Insight

421

Fig. 2 PubMed advanced query options, showing the advanced query form (upper right) and interactive search results (bottom left)

Example 1: Diabetes + Alzheimer’s 1. Google Scholar query: diabetes mellitus Alzheimer’s disease Whereas a conventional Google search predominantly yielded recent website URLs for health advocacy groups and foundations, Scholar retrieved peer-review publications that, among the top 10 search hits, provided a historical sampling of studies analyzing prospective co-pathology or causal relationships between the two disorders. The same top hit (retrieved independently by searches using two different Google profiles) was entitled “Diabetes mellitus and Alzheimer’s disease: shared pathology and treatment?” [47]. It provided concerted biochemical evidence of pathological analogies between the two diseases, arguing that cognitive impairment from CNS insulin resistance might be the crucial neurodegenerative consequence of diabetic pathology [47]. This position differs from prior assertions that protein misfolding and aggregation are manifestations common to both diabetes and Alzheimer’s [48, 49]. To quantitatively assess these two distinct prospective co-pathology models, the rigorous PubMed search structure was employed via the following queries:

422

Thomas H. W. Lushington et al.

2A. PubMed Query: (Alzheimer’s Disease) AND (diabetes mellitus) AND (amyloidosis) AND 2020/01/01:2022/09/01[dp] 2B. PubMed Query: (Alzheimer’s Disease) AND (diabetes mellitus) AND (insulin resistance) AND 2020/01/01:2022/09/01 [dp] Tellingly, search 1.2b retrieved 184 publications, whereas 1.2a attained only 32. While restricting the search to very recent hits was intended to select for the current research trend, it should be noted that all-time search results yielded a similar ratio, with the [amyloidosis] constrained search finding 131 hits, while the corresponding [insulin resistance] found 684. Although not definitive, these results are suggestive of a trend, both current and historical, toward interest in insulin metabolic regulation as a potential avenue for understanding the metabolic underpinnings of Alzheimer’s pathology. While prospective opportunities for diabetes-inspired elucidation of Alzheimer’s targets might emerge from amyloidosis or other co-pathological foci, the elevated interest in insulin resistance makes it a reasonable choice for deeper analysis aimed at honing targeting prospects down to BC levels. This then leads to a refined search as follows: 3. PubMed Query: (Alzheimer’s Disease) AND (diabetes mellitus) AND (insulin resistance) AND 2020/01/01:2022/09/01 [dp] AND biomarkers[MeSH Terms] The biomarkers[MeSH Terms] query specification tends to focus the search more on causal analysis, with proposed or validated biomarkers serving as a prospect for identifying BC factors of interest. This search extracts 12 hits—a modest number that can be scrutinized manually for relevant molecular factors. While possible molecular terms may be identified from the title, abstract, and body of each paper, the most contextually objective and reliable BC information in PubMed searches likely emerges from MedlinePlus Subject Header (MeSH) annotations. From the lists of MeSH “substances” reported in these records, prospective co-pathological relevance may be inferred for a modest number of distinct terms, including Adiponectin [50], Apolipoprotein E4 [51], Cholesterol [52], Glutathione Peroxidase [53], Glycated Hemoglobin A [52, 54], Glycation End Products (Advanced) [55], Hemoglobin A1c protein [54], Insulin Glargine [52], Malondialdehyde [52], Microtubule-Associated Proteins [52, 56], Resistin [57], Secretases [52], Tau protein [52, 56], Triglycerides [52], Tumor Necrosis Factor-alpha [52], and Vitamin D [53]. Notes. It is worth relating that the list of prospective markers includes several that are commonly associated with Alzheimer’s research (in particular, apolipoprotein E4, secretases, and Tau) but also identifies more novel candidates. The veracity and potential

Information Science Methods for Neurodegenerative Insight

423

pharmacological relevance of the latter may be worth deeper literature scrutiny, as well as possible investigation through genomic profiling, pathway analytics, and various bioinformatics techniques. Example 2: Hypercholesterolemia + Alzheimer’s 1. Google Scholar Query: hypercholesterolemia Alzheimer’s disease The results from both Google and Google Scholar were stratified into two categories, including papers that commented on the extent to which high cholesterol tended to correlate to elevated amyloid risk [58], and on the tendency for anomalous low-density lipoprotein (LDL) receptor function to associate with both pathologies, especially in terms of familial hypercholesterolemia being a risk for both cognitive impairment and elevated Alzheimer’s risk [59]. The relative merits of these two categories were thus gauged in PubMed searches. 2A. PubMed Query: (Alzheimer’s Disease) AND (hypercholesterolemia) AND (amyloidosis) AND 2020/01/01:2022/09/01 [dp] 2B. PubMed Query: (Alzheimer’s Disease) AND (hypercholesterolemia) AND (LDL receptor) AND 2020/01/01:2022/09/01 [dp] Far fewer articles were retrieved for the above searches than in the comparable diabetes-related queries (1.2a and 1.2b). Query 2.2a yielded only two papers, which both modestly supported the prospect of amyloid aggregation being a joint marker for hypercholesterolemia and Alzheimer’s [60, 61], with prospective relevance to cognitive vitality [60, 61]. Query 2.2b, meanwhile, yielded five papers supporting a concerted framework under which the nexus of hypercholesterolemia and Alzheimer’s was contingent on the function of dysfunction of proteins responsible for vascular integrity [62–66], or in triggering aberrant apoptosis [65, 66]. Notes. The small number of hits in queries 2.2a and 2.2b obviates the need to filter further by biomarkers. Rather, the compiled from the MeSH terms may be manually extracted, which yielded the following prospective targeting candidates: Adaptor Proteins [65, 66], Apolipoproteins E [64, 66], Caspase 3 [65, 66], Hydroxymethylglutaryl-CoA Reductase [64, 66], LDL Receptor 1 [62, 65, 66], Receptor for Advanced Glycation End Products [62–66], and Sortilin [65, 66]. These markers are all reflective of vascular function and/or angiopathic effects, indicating that the peripheral hypertensive vasculopathy associated with hypercholesterolemia may translate into disruptive blood–brain barrier microvasculature and angiopathic consequences within the CNS [62, 67, 68]. As with the previous example, a precise rationalization of the plausible prospects for markers such as these to

424

Thomas H. W. Lushington et al.

potentially inspire new neurodegenerative target development may be assessed through deeper literature analysis and mining of molecular interrogation databases. Example 3: Rheumatoid Arthritis + Frontotemporal Dementia 1. Google Scholar Query: rheumatoid arthritis frontotemporal dementia Numerous recent publications have discussed a possible co-pathological relationship between rheumatoid arthritis (RA) and frontotemporal dementia (FTD). The Scholar search produced numerous semi-associated conceptual overlaps, including the link between RA and broader manifestations of dementia [69], which is conceptually plausible since RA may display clinical neuropsychiatric complications [70]. Google and Google Scholar queries also connected RA to amyotrophic lateral sclerosis (ALS), which, although pathologically distinct from FTD (ALS affects the primary motor cortex, the brainstem, and the spinal cord, but not the frontotemporal lobe), is often regarded as pathologically similar, largely due to shared molecular markers such as TDP-43 dysregulation [71, 72] and characteristic C9orf72 alleles [71, 72]. These relationships aside, the two most common rationalizations apparent from both Google and Google Scholar searches were evidence of anti-phospholipid pathology in both RA and FTD [69] and comparable dysfunction associated with the progranulin protein [73, 74]. These two concepts were subsequently examined in PubMed. 2A. PubMed Query: (frontotemporal dementia) AND (rheumatoid arthritis) AND (antiphospholipid) AND 2020/01/01: 2022/09/01[dp] 2B. PubMed Query: (frontotemporal dementia) AND (rheumatoid arthritis) AND (Progranulin) AND 2020/01/01:2022/ 09/01[dp] Rather than focus on the associations, more rigorous PubMed queries largely failed. Queries 3.2a and 3.2b each retrieved only a single reference, with PubMed retrieving the same antiphospholipid reference that featured prominently in the Scholar search [69], while a single paper arose for the progranulin-filtered query [75]. The very limited list of MeSH substances used to index these publications yields a comparably limited list of FTD mechanistic targeting opportunities: MFAP4 [75] and progranulin [75], where the latter of these two targets is already under investigation for dementia treatment [76]. This analysis does little to empower FTD drug discovery, since any scientific assertion on so few publications is distinctly tenuous, as reproducibility is a pillar of technical advancement. The discrepancy between the substantial and

Information Science Methods for Neurodegenerative Insight

425

diverse search results from Scholar and the bare singletons produced from PubMed spurs the question of whether Google’s associative algorithms might have overreached, or whether PubMed indexing, although robust for diabetes and passable for hypercholesterolemia, might be deficient for pairing RA with FTD. To examine this question, several further searches were undertaken as follows: 3A. PubMed Query: (rheumatoid arthritis) AND (Progranulin) AND 2020/01/01:2022/09/01[dp] 3B. PubMed Query: (frontotemporal dementia) AND (Progranulin) AND 2020/01/01:2022/09/01[dp] 3C. PubMed Query: (rheumatoid arthritis) AND (antiphospholipid) AND 2020/01/01:2022/09/01[dp] 3D. PubMed Query: (frontotemporal dementia) AND (antiphospholipid) AND 2020/01/01:2022/09/01[dp] These cross-validation queries are agnostic with respect to assertions of co-pathology. Instead, they assess the relative significance of specific BC terms that had been suggested from original Scholar co-pathology searches. Among the progranulin searches, it is interesting to note that query 3.3a retrieved only seven articles relevant to RA, whereas query 3.3b attained a far more robust 130 publications. The key difference is that the relevance of progranulin to FTD therapeutics has substantial current and emerging interest, but most publications relating progranulin to RA are older, with relatively little activity from 2020 onward. For the antiphospholipid criterion, trends are reversed: 97 recent publications are found for RA (query 3.3c), while the only paper retrieved for FTD is the previously cited article by Corzo et al. [69] Interestingly, whereas the term “antiphospholipid” refers to a distinct pathophysiological feature (a programmed autoimmune response to host phospholipids), greater success was attained through the simple act of deleting the “anti” in the latter query, to yield: 3E. PubMed Query: (frontotemporal dementia) AND (phospholipid) AND 2020/01/01:2022/09/01[dp] Notes. This query produces a more robust list of five references, within which we actually find a prospective bridge with the progranulin studies, via a paper reporting physiological stabilization of the phospholipid bis(monoacylglycero)phosphate as a promoter of the progranulin-mediated rescue of lysosomal storage function [77], while other papers in this list relate to the role of phospholipids in neuropathological lipid droplets [78], the specifically neurodegenerative effects of oxidized phospholipids [79], and, conversely, the neuroprotective attributes of N-acetylethanolamine phospholipids such as palmitoylethanolamide [80].

426

Thomas H. W. Lushington et al.

Collectively, these articles retrieved by searches 3.3a–e somewhat augment our understanding of the prospective pharmacological FTD- and RA-related benefits of upregulating granulin proteins, in terms of counteracting anemia-inducing erythropoiesis, and enhancing myeloid cell differentiation, with prospective value in treating neutropenia and myeloid leukemia [75]. Additional MeSH marker terms from search hits include Annexins [81], Cholesterol Esters [78], Lipofuscin [77], Palmidrol [80], Phosphatidylcholines [79], Transferrin [77], Trem2 [77], and Triglycerides [78]. The superset of these terms, plus MFAP4 and progranulin, provide a fairly robust list of candidates for target profiling, collectively slanting the pathology model to one of dysregulated phospholipid function. Example 4: Irritable Bowel Syndrome + Multiple Sclerosis 1. Google Scholar Query: irritable bowel syndrome multiple sclerosis The above Scholar search did not produce obvious candidates for BC anomalies shared between irritable bowel syndrome (IBS) and multiple sclerosis (MS) but rather retrieved numerous articles on conditions such as epilepsy and chronic lung diseases whose incidence may co-correlate with MS and IBS elevated risk, or which MS patients may experience co-pathologically [82– 84]. The query also featured articles discussing microbiotic risks prospectively shared between the two conditions [85, 86]. Lacking BC leads, a secondary series of queries was formulated, spanning three potentially associable co-pathologies, as follows: 2A. PubMed Query: (multiple sclerosis) AND (irritable bowel) AND (epilepsy) AND 2020/01/01:2022/09/01[dp] 2B. PubMed Query: (multiple sclerosis) AND (irritable bowel) AND (chronic lung) AND 2020/01/01:2022/09/01[dp] 2C. PubMed Query: (multiple sclerosis) AND (irritable bowel) AND (microbiome) AND 2020/01/01:2022/09/01[dp] Notes. The epilepsy query did not appear to converge toward commonalities, yielding only two hits, neither of which offered distinct opinions on shared pathology. Query 4.2b afforded even less mechanistic orientation, providing only a global public health study discussing patients with chronic health challenges [87]. The microbiome search modifier, however, did yield five references with interesting implications. None provided obvious biochemical target prospects, but several were suggestive of prospective pathological insight, including one exploring a prospective pancreatic nexus spanning from gastroenterological disorders to rheumatological, autoimmune, and neurological pathologies [88]. A second reference argued for the representation of the gut–brain axis as an

Information Science Methods for Neurodegenerative Insight

427

“enteric nervous system” to rationalize intestinal consequences toward neurodegenerative endpoints such as MS, ALS, Alzheimer’s, and Parkinson’s diseases [89] and the final paradigm aligning with the enteric nervous system model. Instead, the study specifically postulated lipopolysaccharide dysregulation as a key mechanism by which microbial pathology triggers neurodegeneration. These interpretations may not provide a wealth of insight into traditional drug targeting opportunities, but a better understanding of specific bacterial influences on neurological health could provide a basis for developing targeted antimicrobial schemes for prophylactic or adjuvant therapeutic strategies in countering neurodegenerative cascades.

6

Conclusions Modern biomedicine is a forest of information, replete with many diverse thickets and glades, yet most scientists focus intently on nearby boughs, missing answers that may be clearer in the next valley. In Alzheimer’s research, a great deal of attention is paid to the overtrodden stand of amyloid trees, with others focusing on nearby tau, neuroinflammatory markers, or excessive microglial synaptic pruning. These studies have revealed a great deal about amyloid, tau, neuroinflammation, and microglial pruning, but a fair bit less about Alzheimer’s pharmacology. Neurodegenerative diseases like AD appear to manifest as multifactorial conditions with pathological effects spread across multiple pathways. This implies informational complexity, but need not imply that the numerous disparate disease markers arose independently. Thus, while this complexity may seem an onerous barrier to effective treatment, it lends itself well to information science approaches that focus on data commonalities as seeds for mining etiological clues. The greater the diversity of credible symptoms and markers, the greater the informational basis. Just as information science can expose economic trends or terrorist threats, the assimilation of potentially relevant co-pathology and risk factor data could reveal key new aspects of neurodegenerative etiology and targetability. Productive application of broad-based information science techniques to Alzheimer’s research was proposed more than 7 years ago [90], but practical implementation has remained elusive. Highly practicable resources and tools are available to us, and the rudimentary examples employed in this chapter suggest that a concerted application of multiple intelligent query tools toward the same intellectual challenge can, under judicious human guidance, produce novel insight that might not be confidently posited when using a single query tool alone.

428

Thomas H. W. Lushington et al.

There are no guarantees that the interface between humans and a diversified array of artificial intelligence resources will cure neurodegenerative disease, but it is possible that the eureka spark leading to the first great Alzheimer’s target may emerge, not from genomics, high-throughput screening, antibody development, or drug repurposing but from just the right type of web search.

Acknowledgments The authors would like to thank Professor David Heeger (New York University), Dr. Heather Bowling, and Dr. Kenneth Ashworth (Epistemic AI) for opinions and discussions that helped steer the conceptualization of this chapter. References 1. Høilund-Carlsen PF, Alavi A (2021) Aducanumab (marketed as Aduhelm) approval is likely based on misinterpretation of PET imaging data. J Alzheimers Dis 84(4):1457–1460. https://doi.org/10.3233/JAD-215275 2. The approval of Aduhelm risks eroding public trust in Alzheimer research and the FDA. Nat Rev Neurol. Accessed 26 Aug 2022. https:// www.nature.com/articles/s41582-021-00 540-6 3. Caldwell AB, Liu Q, Zhang C et al (2022) Endotype reversal as a novel strategy for screening drugs targeting familial Alzheimer’s disease. Alzheimers Dement 18(11):2117. https://doi.org/10.1002/alz.12553 4. Panza F, Lozupone M, Seripa D et al (2020) Development of disease-modifying drugs for frontotemporal dementia spectrum disorders. Nat Rev Neurol 16(4):213–228. https://doi. org/10.1038/s41582-020-0330-x 5. Lee G, Cummings J, Decourt B, Leverenz JB, Sabbagh MN (2019) Clinical drug development for dementia with Lewy bodies: past and present. Expert Opin Investig Drugs 28(11): 9 5 1 – 9 6 5 . h t t p s : // d o i . o r g / 1 0 . 1 0 8 0 / 13543784.2019.1681398 6. Lushington G, Parker F, Lushington T, Wallace N (2018) Neuropharmacology in flux: molecular modeling tools for understanding protein conformational shifts in Alzheimer’s disease and related disorders. In: Roy K (ed) Computational modeling of drugs against Alzheimer’s disease, pp 573–611. https://doi. org/10.1007/978-1-4939-7404-7_20 7. Sturchio A, Dwivedi AK, Young CB et al (2021) High cerebrospinal amyloid-β 42 is associated with normal cognition in individuals

with brain amyloidosis. eClinicalMedicine:38. https://doi.org/10.1016/j.eclinm.2021. 100988 8. Alzheimer’s symptoms could be due to decline in brain protein not accumulation of amyloid plaques. News-Medical.net. Published June 28, 2021. Accessed 27 Aug 2022. https:// www.news-medical.net/news/20210628/ Alzheimers-symptoms-could-be-due-todecline-in-brain-protein-not-accumulation-ofamyloid-plaques.aspx 9. Makin S (2018) The amyloid hypothesis on trial. Nature 559(7715):S4–S7. https://doi. org/10.1038/d41586-018-05719-4 10. Why some people with brain markers of Alzheimer’s have no dementia. ScienceDaily. Accessed 27 Aug 2022. https://www. sciencedaily.com/releases/2018/08/18081 7093810.htm 11. Xiong J, Kang SS, Wang Z et al (2022) FSH blockade improves cognition in mice with Alzheimer’s disease. Nature 603(7901):470–476. https://doi.org/10.1038/s41586-02204463-0 12. Arshavsky YI (2020) Alzheimer’s disease: from amyloid to autoimmune hypothesis. Neuroscientist 26(5-6):455–470. https://doi.org/10. 1177/1073858420908189 13. Lim B, Prassas I, Diamandis EP (2021) Alzheimer disease pathogenesis: the role of autoimmunity. J Appl Lab Med 6(3):756–764. https://doi.org/10.1093/jalm/jfaa171 14. Gonza´lez A, Calfı´o C, Churruca M, Maccioni RB (2022) Glucose metabolism and AD: evidence for a potential diabetes type 3. Alzheimers Res Ther 14(1):56. https://doi.org/10. 1186/s13195-022-00996-8

Information Science Methods for Neurodegenerative Insight 15. Dı´ez R, Madero M, Gamba G, Soriano J, Soto V (2014) Renal AA Amyloidosis in patients with type 2 diabetes Mellitus. Nephron Extra 4(2):119–126. https://doi.org/10.1159/ 000363625 16. Huston DP, McAdam KP, Balow JE, Bass R, DeLellis RA (1981) Amyloidosis in systemic lupus erythematosus. Am J Med 70(2): 320–323. https://doi.org/10.1016/00029343(81)90768-3 17. Banavar SP, Trogdon M, Drawert B, Yi TM, Petzold LR, Campa`s O (2021) Coordinating cell polarization and morphogenesis through mechanical feedback. PLoS Comput Biol 17(1):e1007971. https://doi.org/10.1371/ journal.pcbi.1007971 18. Dantzer R (2018) Neuroimmune interactions: from the brain to the immune system and vice versa. Physiol Rev 98(1):477–504. https:// doi.org/10.1152/physrev.00039.2016 19. Gordan R, Gwathmey JK, Xie LH (2015) Autonomic and endocrine control of cardiovascular function. World J Cardiol 7(4):204–214. https://doi.org/10.4330/wjc.v7.i4.204 20. Han C, Rice MW, Cai D (2016) Neuroinflammatory and autonomic mechanisms in diabetes and hypertension. Am J Physiol - Endocrinol Metab 311(1):E32–E41. https://doi.org/10. 1152/ajpendo.00012.2016 21. Jha MK, Morrison BM (2018) Glia-neuron energy metabolism in health and diseases: new insights into the role of nervous system metabolic transporters. Exp Neurol 309:23–31. https://doi.org/10.1016/j.expneurol.2018. 07.009 22. Myers MG, Olson DP (2012) Central nervous system control of metabolism. Nature 491(7424):357–363. https://doi.org/10. 1038/nature11705 23. Gao Y, Li X, Zhao H, Liang et al (2021) Comprehensive analysis strategy of nervous– endocrine–immune-related metabolites to evaluate arachidonic acid as a novel diagnostic biomarker in depression. J Proteome Res 20(5):2477–2486. https://doi.org/10.1021/ acs.jproteome.0c00940 24. Porges SW (2001) The polyvagal theory: phylogenetic substrates of a social nervous system. Int J Psychophysiol 42(2):123–146. https:// doi.org/10.1016/S0167-8760(01)00162-3 25. Cohen N, Kinney K (2007) PROLOGUE: exploring the phylogenetic history of neuralimmune system interactions: an update. In: Psychoneuroimmunology, two-volume set, pp 1–38. https://doi.org/10.1016/B978012088576-3/50003-4 26. Hartenstein V, Giangrande A (2018) Connecting the nervous and the immune systems in

429

evolution. Commun Biol 1:64. https://doi. org/10.1038/s42003-018-0070-2 27. Marin I, Kipnis J (2013) Learning and memory . . . and the immune system. Learn Mem 20(10):601–606. https://doi.org/10.1101/ lm.028357.112 28. Ratajczak W, Niedz´wiedzka-Rystwej P, TokarzDeptuła B, Deptuła W (2018) Immunological memory cells. Cent-Eur J Immunol 43(2): 194–203. https://doi.org/10.5114/ceji. 2018.77390 29. Ma Q (2020) Polarization of immune cells in the pathologic response to inhaled particulates. Front Immunol:11. Accessed 4 Sept 2022. https://www.frontiersin.org/articles/10.33 89/fimmu.2020.01060 30. Altan-Bonnet G, Mukherjee R (2019) Cytokine-mediated communications: a quantitative appraisal of immune complexity. Nat Rev Immunol 19(4):205–217. https://doi.org/ 10.1038/s41577-019-0131-x 31. Tuzlak S, Dejean AS, Iannacone M et al (2021) Repositioning TH cell polarization from single cytokines to complex help. Nat Immunol 22(10):1210–1217. https://doi.org/10. 1038/s41590-021-01009-w 32. Kang YE, Kim HJ, Shong M (2019) Regulation of systemic glucose homeostasis by T helper type 2 Cytokines. Diabetes Metab J 43(5):549–559. https://doi.org/10.4093/ dmj.2019.0157 33. Spangelo BL (1997) Cytokines and endocrine function. In: Conn PM, Melmed S (eds) Endocrinology: basic and clinical principles. Humana Press, pp 115–128. https://doi.org/ 10.1007/978-1-59259-641-6_8 34. Vidal PM, Lemmens E, Dooley D, Hendrix S (2013) The role of “anti-inflammatory” cytokines in axon regeneration. Cytokine Growth Factor Rev 24(1):1–12. https://doi.org/10. 1016/j.cytogfr.2012.08.008 35. Prieto GA, Cotman CW (2017) Cytokines and cytokine networks target neurons to modulate long-term potentiation. Cytokine Growth Factor Rev 34:27–33. https://doi.org/10.1016/ j.cytogfr.2017.03.005 36. Benatar M, Wuu J, McHutchison C et al (2022) Preventing amyotrophic lateral sclerosis: insights from pre-symptomatic neurodegenerative diseases. Brain 145(1):27–44. https://doi.org/10.1093/brain/awab404 37. Mosconi L, Berti V, Dyke J et al (2021) Menopause impacts human brain structure, connectivity, energy metabolism, and amyloid-beta deposition. Sci Rep 11(1):10867. https://doi. org/10.1038/s41598-021-90084-y 38. Gime´nez-Llort L, Torres-Lista V, De la Fuente M (2014) Crosstalk between behavior and

430

Thomas H. W. Lushington et al.

immune system during the prodromal stages of Alzheimer’s disease. Curr Pharm Des 20(29): 4723–4732. https://doi.org/10.2174/ 1381612820666140130205500 39. Wang J, Knol MJ, Tiulpin A et al (2019) Gray Matter Age prediction as a biomarker for risk of dementia. Proc Natl Acad Sci U S A 116(42): 21213–21218. https://doi.org/10.1073/ pnas.1902376116 40. Jellinger KA (2015) Neuropathobiology of non-motor symptoms in Parkinson disease. J Neural Transm Vienna Austria 1996 122(10): 1429–1440. https://doi.org/10.1007/ s00702-015-1405-5 41. Ishida T, Tokuda K, Hisaka A et al (2019) A novel method to estimate long-term chronological changes from fragmented observations in disease progression. Clin Pharmacol Ther 105(2):436–447. https://doi.org/10.1002/ cpt.1166 ´ urkovic´ M, Kosˇec A (2018) Bubble effect: 42. C including internet search engines in systematic reviews introduces selection bias and impedes scientific reproducibility. BMC Med Res Methodol 18:130. https://doi.org/10.1186/ s12874-018-0599-2 43. Heath A, Levay P, Tuvey D (2022) Literature searching methods or guidance and their application to public health topics: a narrative review. Health Inf Libr J 39(1):6–21. https:// doi.org/10.1111/hir.12414 44. Devlin J, Chang MW, Lee K, Toutanova K (2019) BERT: Pre-training of deep bidirectional transformers for language understanding. Published online May 24, 2019. https:// doi.org/10.48550/arXiv.1810.04805 45. Understanding searches better than ever before. Google. Published October 25, 2019. Accessed 6 Nov 2022. https://blog.google/ products/search/search-language-understand ing-bert/ 46. Kiester L, Turp C (2022) Artificial intelligence behind the scenes: PubMed’s Best Match algorithm. J Med Libr Assoc 110(1):15–22. https://doi.org/10.5195/jmla.2022.1236 47. Akter K, Lanza EA, Martin SA, Myronyuk N, Rua M, Raffa RB (2011) Diabetes mellitus and Alzheimer’s disease: shared pathology and treatment? Br J Clin Pharmacol 71(3): 365–376. https://doi.org/10.1111/j. 1365-2125.2010.03830.x 48. Nicolls MR (2004) The clinical and biological relationship between type II diabetes mellitus and Alzheimer’s disease. Curr Alzheimer Res 1(1):47–54. https://doi.org/10.2174/ 1567205043480555 49. Go¨tz J, Ittner LM, Lim YA (2009) Common features between diabetes mellitus and

Alzheimer’s disease. Cell Mol Life Sci 66(8): 1321–1325. https://doi.org/10.1007/ s00018-009-9070-1 50. Khoramipour K, Chamari K, Hekmatikar AA et al (2021) Adiponectin: structure, physiological functions, role in diseases, and effects of nutrition. Nutrients 13(4):1180. https://doi. org/10.3390/nu13041180 51. Pekkala T, Hall A, Mangialasche F et al (2020) Association of peripheral insulin resistance and other markers of type 2 diabetes mellitus with brain amyloid deposition in healthy individuals at risk of dementia. J Alzheimers Dis JAD 76(4):1243–1248. https://doi.org/10.3233/ JAD-200145 52. Ali SK, Ali RH (2022) Effects of antidiabetic agents on Alzheimer’s disease biomarkers in experimentally induced hyperglycemic rat model by streptozocin. PLoS One 17(7): e0271138. https://doi.org/10.1371/journal. pone.0271138 53. Rajalakshmi R, Uthaiah CA, Ramya CM et al (2022) Comparative assessment of cognitive impairment and oxidative stress markers among vitamin D insufficient elderly patients with and without type 2 diabetes mellitus (T2DM). PLoS One 17(6):e0269394. https://doi.org/10.1371/journal.pone. 0269394 54. Pan Y, Chen W, Yan H, Wang M, Xiang X (2020) Glycemic traits and Alzheimer’s disease: a Mendelian randomization study. Aging 12(22):22688–22699. https://doi.org/10. 18632/aging.103887 55. Lynn J, Park M, Ogunwale C, AcquaahMensah GK (2022) A tale of two diseases: exploring mechanisms linking diabetes mellitus with Alzheimer’s disease. J Alzheimers Dis JAD 85(2):485–501. https://doi.org/10.3233/ JAD-210612 56. Ehtewish H, Arredouani A, El-Agnaf O (2022) Diagnostic, prognostic, and mechanistic biomarkers of diabetes mellitus-associated cognitive decline. Int J Mol Sci 23(11):6144. https://doi.org/10.3390/ijms23116144 57. Wang C, Huang X, Tian S et al (2020) High plasma resistin levels portend the insulin resistance-associated susceptibility to early cognitive decline in patients with type 2 diabetes mellitus. J Alzheimers Dis JAD 75(3): 8 0 7 – 8 1 5 . h t t p s : // d o i . o r g / 1 0 . 3 2 3 3 / JAD-200074 58. Umeda T, Tomiyama T, Kitajima E et al (2012) Hypercholesterolemia accelerates intraneuronal accumulation of Aβ oligomers resulting in memory impairment in Alzheimer’s disease model mice. Life Sci 91(23):1169–1176. https://doi.org/10.1016/j.lfs.2011.12.022

Information Science Methods for Neurodegenerative Insight 59. Zambo´n D, Quintana M, Mata P et al (2010) Higher incidence of mild cognitive impairment in familial hypercholesterolemia. Am J Med 123(3):267–274. https://doi.org/10.1016/j. amjmed.2009.08.015 60. Xu C, Apostolova LG, Oblak AL, Gao S (2020) Association of hypercholesterolemia with Alzheimer’s disease pathology and cerebral amyloid angiopathy. J Alzheimers Dis JAD 73(4): 1305–1311. https://doi.org/10.3233/ JAD-191023 61. Navas Guimaraes ME, Lopez-Blanco R, Correa J et al (2021) Liver X receptor activation with an intranasal polymer therapeutic prevents cognitive decline without altering lipid levels. ACS Nano 15(3):4678–4687. https://doi.org/10. 1021/acsnano.0c09159 62. Zhou R, Chen LL, Yang H et al (2021) Effect of high cholesterol regulation of LRP1 and RAGE on Aβ transport across the blood-brain barrier in Alzheimer’s disease. Curr Alzheimer Res 18(5):428–442. https://doi.org/10. 2174/1567205018666210906092940 63. O’Connell EM, Lohoff FW (2020) Proprotein convertase Subtilisin/Kexin Type 9 (PCSK9) in the brain and relevance for neuropsychiatric disorders. Front Neurosci 14:609. https://doi. org/10.3389/fnins.2020.00609 64. de Oliveira FF, Chen ES, Smith MC, Bertolucci PHF (2020) Selected LDLR and APOE polymorphisms affect cognitive and functional response to lipophilic statins in Alzheimer’s disease. J Mol Neurosci MN 70(10): 1574–1588. https://doi.org/10.1007/ s12031-020-01588-7 65. Mitok KA, Keller MP, Attie AD (2022) Sorting through the extensive and confusing roles of sortilin in metabolic disease. J Lipid Res 63(8): 100243. https://doi.org/10.1016/j.jlr.2022. 100243 66. de Oliveira J, Engel DF, de Paula GC et al (2020) LDL receptor deficiency does not Alter brain amyloid-β levels but causes an exacerbation of apoptosis. J Alzheimers Dis JAD 73(2):585–596. https://doi.org/10. 3233/JAD-190742 67. de Oliveira FF, Berretta JM, de Almeida Junior GV et al (2019) Pharmacogenetic analyses of variations of measures of cardiovascular risk in Alzheimer’s dementia. Indian J Med Res 150(3):261–271. https://doi.org/10.4103/ ijmr.IJMR_1209_17 68. de Oliveira FF, Chen ES, Smith MC, Bertolucci PH (2017) Associations of cerebrovascular metabolism genotypes with neuropsychiatric symptoms and age at onset of Alzheimer’s disease dementia. Rev Bras Psiquiatr Sao Paulo Braz 1999 39(2):95–103.

431

https://doi.org/10.1590/1516-44462016-1991 69. Corzo K, Farabi B, Lahoti L (2022) The link between frontotemporal dementia and autoimmunity: a case presentation and literature review. Cureus 14(4):e24617. https://doi. org/10.7759/cureus.24617 70. Joaquim AF, Appenzeller S (2015) Neuropsychiatric manifestations in rheumatoid arthritis. Autoimmun Rev 14(12):1116–1122. https:// doi.org/10.1016/j.autrev.2015.07.015 71. Lee SM, Asress S, Hales CM et al (2019) TDP-43 cytoplasmic inclusion formation is disrupted in C9orf72-associated amyotrophic lateral sclerosis/frontotemporal lobar degeneration. Brain Commun 1(1):fcz014. https://doi.org/10.1093/braincomms/ fcz014 72. Balendra R, Isaacs AM (2018) C9orf72mediated ALS and FTD: multiple pathways to disease. Nat Rev Neurol 14(9):544–558. https://doi.org/10.1038/s41582-0180047-2 73. Gass J, Prudencio M, Stetler C, Petrucelli L (2012) Progranulin: an emerging target for FTLD therapies. Brain Res 1462:118–128. https://doi.org/10.1016/j.brainres.2012. 01.047 74. Liu C, ju. (2011) Progranulin: a promising therapeutic target for rheumatoid arthritis. FEBS Lett 585(23):3675–3680. https://doi. org/10.1016/j.febslet.2011.04.065 75. Campbell CA, Fursova O, Cheng X et al (2021) A zebrafish model of granulin deficiency reveals essential roles in myeloid cell differentiation. Blood Adv 5(3):796–811. https://doi.org/10.1182/bloodadvances. 2020003096 76. Galimberti D, Fenoglio C, Scarpini E (2018) Progranulin as a therapeutic target for dementia. Expert Opin Ther Targets 22(7):579–585. https://doi.org/10.1080/14728222.2018. 1487951 77. Logan T, Simon MJ, Rana A et al (2021) Rescue of a lysosomal storage disorder caused by Grn loss of function with a brain penetrant progranulin biologic. Cell 184(18): 4651–4668.e25. https://doi.org/10.1016/j. cell.2021.08.002 78. Teixeira V, Maciel P, Costa V (2021) Leading the way in the nervous system: Lipid Droplets as new players in health and disease. Biochim Biophys Acta Mol Cell Biol Lipids 1866(1): 158820. https://doi.org/10.1016/j.bbalip. 2020.158820 79. Dong Y, Yong VW (2022) Oxidized phospholipids as novel mediators of neurodegeneration.

432

Thomas H. W. Lushington et al.

Trends Neurosci 45(6):419–429. https://doi. org/10.1016/j.tins.2022.03.002 80. Landolfo E, Cutuli D, Petrosini L, Caltagirone C (2022) Effects of Palmitoylethanolamide on neurodegenerative diseases: a review from Rodents to humans. Biomol Ther 12(5):667. https://doi.org/10.3390/biom12050667 81. Leoni TB, Gonza´lez-Salazar C, Rezende TJR et al (2021) A novel multisystem proteinopathy caused by a Missense ANXA11 variant. Ann Neurol 90(2):239–252. https://doi.org/10. 1002/ana.26136 82. Horton M, Rudick RA, Hara-Cleaver C, Marrie RA (2010) Validation of a self-report comorbidity questionnaire for multiple sclerosis. Neuroepidemiology 35(2):83–90. https:// doi.org/10.1159/000311013 83. Marrie RA, Hanwell H (2013) General health issues in multiple sclerosis: comorbidities, secondary conditions, and health behaviors. Contin Minneap Minn 19(4 Multiple Sclerosis):1046–1057. https://doi.org/10. 1212/01.CON.0000433284.07844.6b 84. Marrie RA, Yu BN, Leung S et al (2013) The utility of administrative data for surveillance of comorbidity in multiple sclerosis: a validation study. Neuroepidemiology 40(2):85–92. https://doi.org/10.1159/000343188 85. Jangi S, Gandhi R, Cox LM et al (2016) Alterations of the human gut microbiome in multiple

sclerosis. Nat Commun 7(1):12015. https:// doi.org/10.1038/ncomms12015 86. Tremlett H, Waubant E (2017) The multiple sclerosis microbiome? Ann Transl Med 5(3): 53. https://doi.org/10.21037/atm.2017. 01.63 87. Green BM, Van Horn KT, Gupte K, Evans M, Hayes S, Bhowmick A (2020) Assessment of adaptive engagement and support model for people with chronic health conditions in online health communities: combined content analysis. J Med Internet Res 22(7):e17338. https:// doi.org/10.2196/17338 88. Gesualdo M, Rizzi F, Bonetto S et al (2020) Pancreatic diseases and microbiota: a literature review and future perspectives. J Clin Med 9(11):E3535. https://doi.org/10.3390/ jcm9113535 89. Niesler B, Kuerten S, Demir IE, Sch€afer KH (2021) Disorders of the enteric nervous system - a holistic view. Nat Rev Gastroenterol Hepatol 18(6):393–410. https://doi.org/10. 1038/s41575-020-00385-2 90. Erdelez S, Howarth LC, Gibson T (2015) How can information science contribute to Alzheimer’s disease research? Proc Assoc Inf Sci Technol 52(1):1–4. https://doi.org/10. 1002/pra2.2015.14505201006

Chapter 15 Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug Development Raju Dash, Yeasmin Akter Munni, Sarmistha Mitra, Nayan Dash, and Il Soo Moon Abstract Alzheimer’s disease (AD) has become a public health emergency due to its complexity and heterogeneity; therefore, therapeutic regimens must focus on cure rather than symptom management. Alternative strategies, such as repositioning existing drugs to treat AD, have been increasingly applied recently due to the sluggish pace and rising failure rate of traditional drug discovery. Reevaluating existing drugs for a new indication is known as “drug repositioning,” which may save money, time, and effort throughout the drug development process. Computational strategies have been providing excellent facilities for the effective prediction of drug repositioning, especially the integration of the network pharmacology method, which offers a novel approach to drug discovery by creating models that account for the broad physiological or pathophysiological context of protein targets and the effects of changing them without compromising the essential molecular details. Network pharmacology guides and assists drug repositioning by identifying new drug targets, disease mechanisms, multi-target drugs, drug combinations, and adverse drug reactions through the analysis and molecular visualization of multilayer omics data on the drug-target-disease association. This chapter discusses the importance and success of drug repositioning in AD development and the prospects and methodologies of network pharmacology in understanding various aspects of drug repositioning. Key words Drug repositioning, Alzheimer’s disease, Network pharmacology, Network analysis, Omics

1

Background Over the last century, neurodegenerative diseases (NDDs), such as Alzheimer’s disease (AD), Huntington’s disease (HD), Parkinson’s disease (PD), and amyotrophic lateral sclerosis (ALS), have presented several novel obstacles to efficient drug discovery and development. It was estimated in 2016 that more than 43.8 million are affected by AD, and other NDDs associated dementias, which have become a significant health concern worldwide, and the prevalence is higher in the older than the young generation [1]. This

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_15, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

433

434

Raju Dash et al.

prevalence is expected to increase twofold every 20 years until 2050 [2]. Among these NDDs, a significant cause of mortality in the elderly population is AD, which was first described by Alois Alzheimer as a neurological condition characterized by progressive cognitive failure and memory loss [3]. The loss of neurons usually occurs due to intracellular neurofibrillary tangles (NFTs) [4] and amyloid plaques in the extracellular space, which interferes with the diverse functions of neurons that are essential for cell viability, including synaptic communication and intracellular trafficking [5, 6]. While understanding mechanistic insights on disease development and risk factors in AD prevalence has rapidly improved in recent years because of multi-national AD genome projects [7, 8] and high-throughput technologies, the drugs that eliminate the disease still have none. Nearly 200 clinical studies have been conducted to identify disease-modifying therapies for AD, and most have failed due to a lack of effectiveness or excessive side effects [9]. Furthermore, many other hurdles are included in the drug discovery of AD, including failure of blood–brain barrier (BBB) crossing, animal models for AD failing to replicate human pathophysiology but specific symptoms, designing safe doses with practical outcomes, and many more [10]. Still, only six drugs are currently approved by the FDA for managing AD symptoms, including N-methyl-D-aspartate (NMDA) receptor antagonist (memantine), cholinesterase inhibitors (galantamine, donepezil, tacrine, and rivastigmine), and a fixed combination of memantine and donepezil [11]. Thus, there is an anticipated demand for the prompt discovery of effective drugs for AD therapy with a gradual increase in disease burden. When a new novel molecular entity failed in clinical studies, a considerable amount of money and time was also invested. Therefore, considering repositioning the existing FDA-approved drug rather than designing and developing a new entity is now an attractive option for pharmaceutical companies to avoid severe toxicities, reduce time and cost, and have a greater success rate, known as drug repositioning [12]. While drug repositioning has emerged as a superior technology to get new and effective drugs, this process is also accelerated with more accuracy through the integration of in silico methods. The integration of novel technologies in AD research, such as “omics,” not only produces vast data on disease understanding [13] but also improves the ability of decision-making on toxicity and efficacy when incorporated in drug repositioning methodologies [14]. The purpose of this chapter is to provide some instances of the use of the drug repositioning technique in the rational design of innovative drug candidate prototypes for the treatment of AD and to highlight some recent developments in this area, especially the integration of network pharmacology approaches for recognizing and confirming repositioning targets and ideas.

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

2

435

Drug Repositioning in the AD Drug Discovery Drug repositioning or drug repurposing, also known as reprofiling or re-tasking, involves identifying novel indications for existing drugs, which can be active pharmaceutical ingredients or compounds already present in the market or in any clinical phases (including clinically failed drugs). Following this approach, the existing drugs or bioactive molecules can provide potential therapeutic benefits to the patient more rapidly and at a lower cost than producing new drugs since their safety has previously been established in various clinical studies for other purposes. Drug repositioning is an all-encompassing strategy that hastens the translation of information toward treating human diseases by avoiding typical roadblocks in the drug development process. If successful, drug repositioning might enable compounds to bypass phase II of the drug development process, cutting development time and costs. Furthermore, the risks associated with continued development are considerably decreased due to the availability of clinical data at the outset of a research project [15]. Another positive aspect of drug repositioning is the high rate at which its safety has been shown via preliminary studies. The progression of drug research is further facilitated by preclinical data and clinical characteristics for a treatment that has already been authorized [16]. In the drug discovery approach for AD, drug repositioning offers significant advantages over conventional methodologies. For instance, designing a drug targeting the brain is more complicated than the other part of human disease due to issues of translation, e.g., behavioral pattern differences between animal models and humans [17], overcoming the BBB [18], target selectivity, drug metabolism and pharmacokinetics (DMPK), side effects, and toxicities. Addressing these issues requires substantial animal and human studies [19]. Still, it is not problematic in the case of repositioned drugs, and even selectivity is no longer necessary if the drug is efficacious and safe [20]. Success in repositioning drugs has previously resulted from coincidental occurrences in the lab and the clinic, which is included for AD also (Fig. 1). For example, retinoic acid, the physiologically active metabolite of vitamin A, is the ligand of a family of receptors known as retinoic acid (RARα, β, and γ) and retinoid x receptors (RXRα, β, and γ) [21, 22]. These receptors control the formation of new neurons, survival of neurons, and synaptic plasticity in immature and fully developed animal brains [23]. Rats deficient in vitamin A were found to have an aggregation of Aβ in their cerebral arteries and a downregulation of RAR expression [24, 25]. An RXR agonist, bexarotene (BEXA), was granted approval by the FDA for use in the treatment of cutaneous T-cell lymphoma (CTCL), and it was discovered to be particularly effective for AD. AD is connected

436

Raju Dash et al.

Fig. 1 Representative examples of targeted drug candidates for repurposing in Alzheimer’s disease (AD). Several types of retinoic acid and retinoic X receptor agonists (bexarotene, acitretin, and tamibarotene); glucagon-like peptide-1 (GLP-1) analogue (liraglutide); alkylating agent (carmustine); tetracycline antibiotic (minocycline); calcium-channel antagonist (nitrendipine, nilvadipine); tyrosine protein kinase ALB (imatinib); angiotensin II receptor blocker (valsartan, candesartan, telmisartan); acetylcholinesterase (AChE) inhibitors (phenserine); and Rho-kinase inhibitor (fasudil) showing beneficial effect in AD

to brain failure to eliminate amyloid formation, associated with apolipoprotein E (apoE) reduction. BEXA has been proven in research to be effective in treating AD because of its ability to boost ApoE-related lipoprotein particles and promote the clearance of Aβ. In the advanced stages of AD in mice, BEXA significantly reduced the Aβ amount [26]. Acitretin, a synthetic retinoid, is often

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

437

used by people living with psoriasis as oral medication [27]. The JAK-STAT signaling system regulates major psoriasis-related cytokines, including INF-γ, IL-6, IL-17, and IL-22. Acitretin dramatically reduced the expression of STAT1 and STAT3, inhibiting the JAK-signaling pathway [28]. Treatment with acitretin has also been shown to have an anti-amyloidogenic impact in patients with AD [29]. Endres et al. revealed that acitretin (30 mg/day) has an antiamyloidogenic effect in the AD model (animal) and human patients without any side effects [29]. CSF APPs levels were significantly higher in the acitretin group than in the placebo group [30]. Therefore, these preliminary findings suggest that acitretin is a promising candidate for AD treatment. Drug candidates like tamibarotene (AM80), which are said to have fewer adverse effects than other retinoids, are considered favorable for treating AD [31]. Presently, it is only available in Japan, but preliminary trials have shown that it is generally more well tolerated than all-trans retinoic acid (ATRA) [32, 33]. AM80 decreased Aβ plaques by inhibiting microglial and astrocyte activation in APP/PS1 transgenic mice. AM80 ameliorated the expression of BACE1 regulated by the NF-kB pathway due to having anti-inflammatory properties. In addition, AM80 also regulated a large number of target genes that are implicated in the pathophysiology of AD [34]. Drugs like liraglutide is a fatty acid acylated analog of glucagonlike peptide-1 (GLP-1) marketed under the trade name Victoza used as an antidiabetic to treat type 2 diabetes, obesity, and metabolic syndrome [35]. AD has been called “type 3 diabetes,” and drugs that were prescribed for diabetes have been explored as possible treatments for AD [36]. The neuroprotective effects of central insulin signaling are diminished in AD brains due to insulin receptor degradation. The neuroprotective effects of liraglutide are especially noteworthy in light of the growing evidence linking AD to dysregulation of central insulin signaling. The treatment of liraglutide restored memory impairment caused by AD-linked amyloid oligomers in mice, and the administration of liraglutide prevented the loss of brain insulin receptors and synapses [37]. Liraglutide also increased the number of immature neurons in the dentate gyrus. Furthermore, Holubova et al. reported that the neuroprotective effects of liraglutide could reduce the Aβ plaque and caspase-3 in the hippocampus in AD mice (aged 7–8 months, 0.2 mg/kg) [38]. Therefore, liraglutide may be a promising drug for treating AD and diabetes. Chloroethylnitrosourea (or carmustine) is the most commonly used drug in chemotherapy, which is marketed under the brand name BiCNU [39]. Carmustine is successfully used in treating AD and has been licensed for use by the FDA in treating brain cancer [40]. Based on MD simulation, carmustine destabilizes the β-strand of Aβ protofibrils via binding to the hydrophobic grooves [41]. Researchers found that treating amyloid precursor protein

438

Raju Dash et al.

751WT expressing cells with carmustine at 5–20 μM reduced the accumulation of Aβ plaques and increased the expression of transforming growth factor β (TGFβ) [42]. Also, carmustine interacts with secretase in a manner that is separate from other enzymes, allowing it to escape off-target consequences. Because of this, there is less risk associated with using carmustine as an anti-Aβ medication. Minocycline, also known as minocin, is a tetracycline antibiotic used to treat a wide variety of bacterial infections as pneumonia. It is also employed in the therapy of psoriasis and rheumatoid arthritis [43, 44]. Due to its anti-neuroinflammatory actions and crossing BBB, minocycline is currently being thought to be a possible therapeutic for the cure of AD. The production of several proinflammatory cytokines, such as TNF-α, and IL-6, were inhibited by minocycline induced by the Aβ42 peptide [45]. People with moderate AD who took minocycline (200 mg/day or 400 mg/day) for 2 years did not experience a delay in the progression of cognition or functional impairment due to the drug [46]. However, research trials have shown that different dosages of minocycline have varying efficacy compared to placebo, where higher doses of minocycline did not improve its effectiveness. Hence, more studies on drug repositioning are required for minocycline. A dihydropyridine calcium-channel antagonist called nitrendipine is prescribed to patients suffering from hypertension. Nitrendipine has been shown to lessen the occurrence of dementia in experimental studies [42]. As most dementia (60–80%) cases are caused by AD [35], nitrendipine use in treating AD could be due to the ability of this drug to prevent dementia and the unregulated influx of calcium into neurons. Nilvadipine is a prescribed drug for high blood pressure that has also been demonstrated in animal models of AD to reduce amyloid and improve memory performance. In 18 months of randomized clinical trials, when tested on a population with mild-to-severe AD, nilvadipine (8 mg/day) had no significant effect in lowering the rate of cognitive loss compared with matched placebo control [47]. Hence the findings do not provide evidence to support the use of nilvadipine as a therapy for mild-to-moderate AD in the studied population. Oral chemotherapy drug imatinib, also known by the brand names Gleevec and Glivec (both of which are marketed internationally by Novartis), is used to treat cancer and is offered under a variety of other brand names [48]. The strategy of repositioning imatinib is connected to polypharmacology. Imatinib decreased tau phosphorylation and phospho-Cdk5 and thus interrupted the c-Abl/p73 signaling pathway [49]. Since Cdk5 is activated by Aβ neurotoxicity, imatinib could be a potential treatment option for AD due to its ability to disrupt the c-Abl/p73 signaling pathway. Although imatinib has been shown to effectively lower the production of Aβ in both in vitro and in vivo models, this has not been

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

439

demonstrated to be the case in humans; therefore, additional research is required. Valsartan is classified as an angiotensin II receptor blocker (ARB), a group of drugs used to treat diabetic kidney disease, high blood pressure, and heart failure, offered under the brand name Diovan, among other names. Valsartan inhibited the oligomerization of Aβ peptides into high-molecular-weight (HMW) oligomeric peptides, which are linked to cognitive deterioration in an in vitro AD model. Valsartan (10 and 40 mg/kg/day) reduced AD-type neuropathology and brain-soluble HMW extracellular oligomeric Aβ in vivo. Interestingly, a dose of valsartan around twofold lower than that used for antihypertensive therapy in humans significantly prevented the development of Aβ-mediated cognitive impairment [50]. Among the ARBs tested for potential repositioning, candesartan was found to readily cross the BBB and alter glial activities linked to AD. Candesartan inhibits the activity of AT1 and produces an anti-inflammatory response in the brain of AD mice and decreases the amyloid load by promoting the phagocytosis of Aβ42 in the microglia [51]. As a whole, candesartan reduces glial cell reactivity and AD risk variables, suggesting its utility as a repurposed treatment for AD. Furthermore, brain blood flow was improved by the administration of telmisartan (0.35 mg/kg), while the Aβ-induced upregulation of cytokines, including TNF-α and inducible NO, was suppressed [52]. AChE inhibitors such as phenserine have been made available for treating AD, showing neurotrophic and neuroprotective effects. For example, phenserine has been demonstrated in several preclinical investigations to significantly decrease APP levels in vitro and in vivo [53–55]. In two phase II placebo-controlled trials, phenserine was trialed in people with mild-to-moderate AD. The group of participants who received the drug (10–15 mg twice daily) showed markedly improved cognitive function than the placebo member [56, 57]. In summary, the pharmacological effects of phenserine on the treatment of AD are strongly supported by preclinical studies. Similarly, it has a proven track record of clinical success with a low risk of adverse effects. A vasodilator with powerful Rho-kinase inhibitor properties, fasudil, has been used for decades. Since its discovery, it has been implemented to treat cerebral vasospasm in Japan and China and to reverse the cognitive deterioration that occurs in stroke patients [58]. Fasudil was reported to suppress cognitive impairment and synaptic loss mediated by Aβ mediated by Dkk1-driven Wnt–PCP pathway [59–61]. Meanwhile, daily doses of fasudil (30 mg per day) combined with nimodipine were found to be acceptable, and compared to a placebo group, those who were given fasudil had considerably higher MMSE (Mini-Mental Status Examination) scores [62]. Therefore, preliminary research has concluded that fasudil inhibits the development of AD-related neuropathology.

440

Raju Dash et al.

Using late-onset AD-associated genes, including ABCA7, APOE, CLU, BIN1, and PICALM, and AD drug targets such as TREM2, GABRA1, CHRNA2, CD33, PRSS8, TKT, ACE, and APP from GWAS, researchers planned to find out prospective anti-AD drugs by conducting enrichment analysis of transcriptional effects of drugs [63]. Eight drugs have been studied as potential treatments for AD: tomelukast, ouabain, ellipticine, lorglumide, ginkgolide A, alsterpaullone, chrysin, and sulindac sulfide. Four drugs (ouabain, ginkgolide A, alsterpaullone, and chrysin) have shown drug repositioning potential for AD, as indicated by preclinical evidence.

3

Approaches to Drug Repositioning In selecting possible drug candidates for a new indication, drug repositioning methods are categorized into three steps. The first step involves hypothesis generation, selecting prospective molecules for a specific indication. This step is considered the most crucial step for choosing the right drugs for the appropriate target for the selected new indication, requiring more advanced and robust techniques, including experimental or computational approaches [12]. Among these methods, computational approaches mainly deal with significant data analysis. The advent of the so-called “big data” age has been marked by the exponential growth in the volume of data in fields like biology and chemistry, opening up new possibilities for researchers to establish a connection between diseases to drugs, although one that is indirect and dependent on intricate mechanisms of action [64]. Integration of computational methods in the big data analysis provides a clear picture of the drug-driven physiological or pathophysiological mechanism, which includes systemic analysis of any biological data such as electronic health records (EHRs), chemical structure, genotype or proteomic data, and gene expression, providing a platform for hypothesis generation [65]. Based on the data type, computational approaches are broadly categorized into molecular and real-world data approaches [66]. In real-world data approaches, the analysis is based on individual health information (such as electronic health records) that has not been influenced by outside factors or skewed by the methods used to acquire the information. On the other hand, the molecular approach involved analyzing various omics data to link the relationship between drugs to disease. In the second step, newly identified indications and drugderived effects are assessed mechanistically in various preclinical models. If the tested drug has enough safety profiles (phase I) for its original indication, the efficacy of the new indication has been further evaluated by phase II clinical trials in the final steps.

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

4

441

Network Pharmacology in the AD Drug Discovery Recently, it has been identified that the underlying challenge in drug development may not be technical or scientific concerns but rather guiding tactics [67]. Biological systems, however, are resistant to single-point disruptions due to compensation mechanisms and redundant processes. From this point of view, the disease is the failure of ordinarily strong physiological systems, which may be triggered by a combination of hereditary and environmental causes [68]. For this reason, the simultaneous regulation of many targets improves the efficacy of treating or alleviating many complicated diseases. It has been more apparent in recent years that many potent drugs across a wide range of therapeutic domains, including cancer, psychiatry, and anti-infectives, work on many different targets rather than just one [69, 70]. Considering the complexity of living organisms, and more specifically, the patho/physiological environment where protein targets are (mal)functioning and where drugs have to exert their restorative action, the multi-target approach has recently burst into the drug discovery scenario from a network biology perspective on human diseases and treatments. Therefore, it is acknowledged that drug discovery often calls for a systems-level polypharmacology strategy to address issues such as insufficient therapeutic effectiveness and the development of resistance to single-targeted drugs. Questions also have been raised concerning the old paradigm of “one drug, one target, one disease” due to the persistent failure of drug development in complex NDDs like AD. It is now well recognized that AD is one of the multifactorial disorders which shares different pathophysiological features of neuronal degeneration and death as a result of oxidative stress [71, 72], neuroinflammation [73], accumulation and aggregation of the amyloid oligomers [74, 75], neurofibrillary tangles formation due to tau hyperphosphorylation [76–78], dysfunction of the cholinergic system [79, 80], proteostasis failure [81, 82] and genetic variants (which covered minimum 80% of AD cases [83]). Thus inhibitors blocking a single AD target are ineffective (such as NMDAR or AchE), addressing just symptoms rather than halting disease development, highlighting the importance of polypharmacological techniques to drug discovery [84]. The current polypharmacology paradigm, which is replacing the traditional drug discovery view (to a “many-to-many” relationship from “one drug, one target, one disease” model), can be represented by a network in which most drugs have many common targets and most targets have many common drugs, interlinked by each other [68]. Such network-based representation and analysis using topological parameters are known as the “network pharmacology” method, in which two or more drugs, functioning on different targets but in the same causative signaling, can modify

442

Raju Dash et al.

disease pathology synergetically [67, 85, 86]. In the network pharmacology approach, individual biomolecular components, such as genes, proteins, compounds, signalings, or target pathways, are represented by a node, and their associate interaction, either direct or indirect, is represented by an edge. By positioning compounds within the setting of underlying biochemical processes and mechanisms influencing interactions between chemical compounds and their cellular targets, network pharmacology permits a more comprehensive, systems-level understanding of the pharmacological mode of action [87, 88].

5

Drug Repositioning Using Network Pharmacology Nacher and Schwartz [89] provide an example of a typical network pharmacology-based drug repositioning, in which a drug-therapy network is first constructed based on known drug interaction with the known therapeutic indication. In this network, an edge could be established between a therapeutic indication and two drugs if both drugs share the same indication. Similarly, a disease-based network can be developed if two disease therapies are linked with the same drug. Using this framework, whether a drug is a candidate for repositioning is determined by using some statistical measurement, such as betweenness centrality and closeness, which indicate the relative importance of a node in a network. Thus, a drug with a high betweenness centrality score in a drug network is likely to be engaged in more than one therapeutic indication or multiple molecular targets of the same therapeutic indication, increasing the probability of a successful repositioning effort (Fig. 2). However, if targets for drugs were located in peripheral nodes, their perturbation would have little impact on disease phenotype and would be redundant. Based on the fact that many drug targets are found to be highly interconnected but not crucial, it is possible that statistical network analysis might be a helpful tool for prioritizing drug targets [90]. Network pharmacology approaches several advantages in drug repositionings, such as identifying drug targets and multi-diseasespecific drugs and understanding the modified mechanism of action (MoA), disease mechanism, and side effects or adverse effects reactions (Fig. 3).

5.1 In Drug Target Identification

A wider variety of potential pharmacologically valuable targets is made available through the network pharmacology approach for a disease. For example, if the desired target protein is not druggable to treatment, a nearby target protein may fill the void. According to the PISCES [91] dataset, it has been estimated that there are around 39 proteins on average available for a single registered small molecule drug because of binding pocket similarity (residue conserveness) and binding affinity [91, 92]. Using the network

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

443

Fig. 2 Illustration showing a network model with various topological features. Node, target (any biological entity); edge, association, interaction, any relationship; hub, a node with a high degree; degree, the number of edges connected to a node; betweenness, the number of shortest paths that go through a node; shortest path, a short path between any to nodes in a network; module, a group of nodes that act in concert to perform a specific function

pharmacology approach, new suitable drug targets can be extracted by incorporating disease target knowledge (genes/proteins involved in a specific disease, as a seed node) into known drug– target interaction [86]. Another aspect of network pharmacology is identifying target combinations, or protein complexes, that will be most effective and safe when targeted together. Because network analysis highly depends on the interconnection between proteins, a pair or group of proteins, even in a small network, can be a potential target by a single drug, which can increase drug efficiency in a particular disease. Studies by Ruths et al. [93] and Dasika et al. [94] showed that a network-based approach effectively identifies a group of targets for blocking undesirable processes while maintaining the anticipated ones. Targeting this group of targets will allow the designing of pharmacological combinations, wherein the individual drug would be ineffective at blocking the process on its own but would be effective when combined [95–97]. It has been identified that a large portion of pharmacological targets has multiple disease associations. Thus, analyzing network pharmacology can predict multi-disease-specific drug targets by superimposing the network onto the human disease-gene network, as a single drug often acts on many targets, many of which are implicated in several diseases [98].

444

Raju Dash et al.

Fig. 3 Graphical outline of network pharmacology approach in drug repositioning

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

445

5.2 Understanding the MoA of a Drug

The molecular mechanisms by which a drug produces its pharmacological effect are called MoA [88]. Drug effects in diseasemodifying mechanisms can be either “on-target” or “off-target” interaction. In “on-target” interaction, the modified disease mechanism of action is desired due to the direct interaction of the drug to a single target (gene/protein). An off-target protein interacts with the intended target protein yet has the potential to modulate a different disease, which could be a prospective new target. Using the traditional approach, it has been a major undertaking to separate on- and off-targets and to identify and characterize the offtarget effects. However, implementing the network pharmacology approach provides a better understanding of the MoA by integrating a multilayer of OMICS data in the network model, such as genomics, transcriptomics, proteomics, epigenomics, and metabolomics, especially in separating off-target effects [86]. Furthermore, the development of computational techniques, such as molecular docking simulation and drug similarity approach, and their integration into network pharmacology, also accelerated off-target identification. The molecular docking approach predicts the binding pose and affinity of the ligand to a protein target and is routinely used for computational hit discovery and virtual screening. However, this approach is followed by reversed way off-target identification, where a ligand is subjected to screening against a library of targets using the information of their ligand binding sites for an off-target screening [99]. On the other hand, drug similarity is used for target identification because drugs bearing the same structural similarities tend to interact with similar targets [100].

5.3 Understanding Disease Mechanisms

The mechanism of drug effects is not only limited to target binding but also changes transporter proteins, metabolic enzymes, and the other components of subsequent downstream signaling pathways. Studies also showed that drug effects could be changed by genetic variants, known as pharmacogenomics. By integrating publicly available expression profiles (Table 1), for example, Online Mendelian Inheritance in Man (OMIM), which is a comprehensive collection of Mendelian disorders, and Gene Expression Omnibus (GEO), a large repository for transcriptomic data, network-based analysis can differentiate the changes of downstream signaling from the actual intended effect [86, 101]. Furthermore, drug response modifier genes, or pharmacogenetics, can also be identified with network analysis. Network pharmacology with the integration of drug-disease profiles is also considered a “reverse disease profile,” which is based on the idea that drug-induced changes in expressions that are highly inversely linked to a disease profile might reverse some of the disease effects [102–105], providing a better way to understand disease induced mechanistic changes.

446

Raju Dash et al.

Table 1 List of publicly available resources that are frequently used for data integration in network pharmacology-based drug repurposing Name

Features

Links

References

Drug–target interaction DrugBank

Drug Profiles/SMILES Sequences of the target proteins Target proteins Drug interactions

http://www.drugbank.ca

[118]

ChemProt

2 million target proteins Bioactivity for 15,290 proteins and interactions

http://www.cbs.dtu.dk/ services/ChemProt

[119]

KEGG DRUG DB

4900 drug entries are available with 1467 FDA-approved drugs and their known targets Prediction of Drug–Target network Integration of GenBank, KEGG, PubChem, and PubMed

http://genome.jp/kegg/ drug

[120]

STITCH

Known and predicted interactions of chemicals and proteins Off-target identification Linking the off-target to related disease outcomes other than the original indication of the drug

http://stitch.embl.de/

[121]

NCGC Pharmaceutical Collection (NPC)

Large collection of registered drugs, human pathways, and diseases Bioassay base measurement of drug– target relationship Prediction of off-target and related diseases

http://ncgc.nih.gov/

[122]

PubChem

112 M Compounds and 298MSubstances 301 M Bioactivities of compounds

http://pubchem.ncbi.nlm. nih.gov/

[123]

chEMBLdb

Chemical properties ADME/T Data Literature-based bioactivity data

https://www.ebi.ac.uk/ chembldb/

[124]

BindingDB

44,801 compounds 3132 target proteins 108,510 binding affinity conformation 4826 experimental assays

https://www.bindingdb. org/bind/index.jsp

[125]

TTD

Targets including protein/DNA/ RNA Association with biomarkers (1755) for various disorders (365) and scaffolds (210 of 714 drugs)

http://bidd.nus.edu.sg/ group/ttd/ttd.asp

[126]

(continued)

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

447

Table 1 (continued) Name

Features

Links

References

MATADOR

Drug interaction: multiple direct and http://matador.embl.de indirect modes Manually annotated database of protein and ligand

SuperTarget

32,828 drug–target interactions Cytochrome P450s binding information

http://bioinf-apache.charite. [128] de/supertarget_v2/

TDR Targets

Chemo genomics databank Drugs for neglected disease agents

http://tdrtargets.org/

[129]

Integrity

Large number of clinical drugs Targets and diseases information

http://integrity.thomsonpharma.com

[130]

PDSP_Ki

Ki, or affinity values of drug binding Molecular targets

http://pdsp.med.unc.edu/ kidb.php

DTome

Drug–target networks Integrating the drug–drug interactions, drug–target interactions Integrating drug–gene associations and target/gene–protein interactions

http://bioinfo.mc. vanderbilt.edu/DTome

Drug2Gene

Drug–target interactions of compounds with genes from different organisms

http://www.drug2gene.com

[127]

Drug target prediction TarFisDock

Reverse docking Large database for potential drug targets User-friendly Web sever

http://www.dddc.ac.cn/ tarfisdock/

[131]

PharmMapper

Pharmacophore mapping Auto-generation of 3D structure for small molecules Fast (fit for batch processing) User-friendly web sever

http://59.78.96.61/ pharmmapper

[132]

miRTarCLIP

Sequence analysis https://tools4mirs.org/ miRNA target prediction using CLIP software/target_ sequencing data prediction/mirtarclip/ User-friendly Web sever

[133]

Drug repositioning by Li et al.

Cross-docking Large-scale docking to predict novel drug–target interactions

[134]

(continued)

448

Raju Dash et al.

Table 1 (continued) Name

Features

Links

References

SEA

Ligand-based similarity Compare targets by the similarity of the ligands that bind to them User-friendly Web sever

https://sea.bkslab.org/

[135]

NBI, EWNBI, and NWNBI

Network-based inference Interaction of novel proteins or chemicals can be predicted

[136]

Swiss Target Prediction

Target prediction combining 2D and http://www. swisstargetprediction.ch 3D similarity measures User-friendly Web sever

[137]

DDI-CPI

User-friendly Web sever http://cpi.bio-x.cn/ddi/ Real-time docking study Chemical–protein interactome (CPI)

[138]

TargetHunter

Target prediction using chemogenomic database BioassayGeoMap

http://cbligand.org/ TargetHunter

[139]

Omics databases for integration in drug–target interaction MANTRA 2.0

Deduce similarities amid druginduced transcriptional profiles

http://mantra.tigem.it

[140]

Connectivity Map

Microarray-based genome-wide expression profiles Expression response of four cancer cell lines to over 1300 bioactive, drug-like molecules

http://www.broadinstitute. org/cmap/

[105]

GEO

Functional genomics data repository Array- and sequence-based data

http://www.ncbi.nlm.nih. gov/geo/

[141]

OMIM

Timely compendium of human genes http://omim.org/ and genetic phenotypes The disease–gene relationship

[142]

atBioNet

Contains seven public PPI databases, http://www.fda.gov/ ScienceResearch/ which include BioGRID, The BioinformaticsTools/ DIPTM database, HPRD, IntAct, ucm285284.htm MINT, REACTOME, and SPIKE Conversion of drug–target relationship into drug–pathway relationship

[143]

KEGG

Providing drugs, diseases http://www.genome.jp/ Pathway with systematic connections kegg/ and annotation Drug–diseases–pathway relationship

[144]

(continued)

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

449

Table 1 (continued) Name

Features

Links

HMDB

Human metabolic, drug, and disease https://hmdb.ca/ pathways Chemical data, clinical data, molecular biology/biochemistry data SNP and mutation

DISEASES

Disease–gene associations from automatic text mining

References [145]

http://diseases.jensenlab. org

Genetic Association Polymorphism data Database Gene association with disease

http://geneticassociationdb. [146] nih.gov/

DisGeNET

http://www.disgenet.org/

[147]

Genes and variants associated with human diseases

Drug toxicity and side effects SIDER

Information on side effects of 888 approved drugs Adverse effects (1450) curated from public documents and package inserts

http://sideeffects.embl.de/

[144]

ACToR

Toxicity data on 500,000 chemicals Information on health and environmental risks Integration of ToxRefDB and ToxCastDB

http://actor.epa.gov/

[148]

PROMISCUOUS

Association of drug–protein, protein– http://bioinformatics. charite.de/promiscuous protein, and drug–side-effect relations

[149]

PharmGKB

Genetic variation on drug response Drug–gene relations

https://www.pharmgkb. org/

[150]

Comparative Toxicogenomics Database

45 million toxicogenomic relationships 16,300 chemicals 51,300 genes 5500 phenotypes 7200 diseases

http://ctdbase.org/

[151]

ToxCast and Tox21 Publicly available high-throughput toxicity data on thousands of chemicals

http://www2.epa.gov/ chemical-research/ toxicity-forecastertoxcasttm-data

[152]

Carcinogenic Toxicological data from National Potency Database Cancer Institute/National Toxicology Program Data like dosage, strain, histopathological study, level, and route of drug administration.

https://www.nlm.nih.gov/ toxnet/index.html

[153]

(continued)

450

Raju Dash et al.

Table 1 (continued) Name

Features

Links

References

FDALabel

A database of about 30,000 drug labels Text-mining and assessment of the drug similarity from a side effect perspective

https://nctr-crs.fda.gov/ fdalabel/ui/search

[154]

FAERS

US Food and Drug Administration (FDA) Adverse Event Reporting System 4 million reports

[155]

Network visualization tool Cytoscape

Network visualization and analysis

VisANT

Network visualization and analysis of http://visant.bu.edu functional genomics study

Pathview

Error-free, analytical, and visualizing tool Drug pathway analysis

https://pathview.uncc.edu/

[158]

DAVID

Gene ontology Gene enrichment analysis

https://david.ncifcrf.gov/

[159]

5.4 Understanding Drug Toxicity and Adverse Reactions

https://cytoscape.org/

[156] [157]

Network pharmacology helps understand drug-induced adverse reactions by incorporating metabolic, signaling, and gene regulatory networks. It has been recognized that both intended and unintended drug-induced effects emerged from the drug-mediated disruption of the intricate network landscape. In addition, “offtarget” effects might occur even when a drug is intended for a specific target, which is often undesirable and cannot be explained if considered only the primary drug binding target. Integration of network pharmacology provides an opportunity to identify secondary targets of drugs [106]. Moreover, environmental variables and genetic diversity will likely impact the therapeutic response to drugs, needed dosage, and susceptibility to adverse events and side effects based on individual differences in cellular networks [88, 107]. Integrative network pharmacology techniques will therefore be able to define fundamental biological mechanisms leading to drug-induced organ failures by utilizing genome-wide expression data from drug-perturbed states and clinical data on human phenotypic response to drug treatment. In theory, these methods might help anticipate the potential side effects of newly indicated drugs [88].

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

6

451

Recent Efforts in Network Pharmacology-Based AD Drug Repositioning Lee et al. recently made an effort to identify potential drugs for AD therapy using computational drug repositioning methods integrating pharmacogenomics knowledge and the inverse relationship between drug-induced or disease-induced genes and protein perturbation patterns. Using the data from L1000 CMap and CMap, the authors constructed a Drug-induced Gene Perturbation Signature Database (DGPSD) consisting of 1520 drugs and their perturbed gene signatures (61,019 signatures). Also, they developed a scoring system called Drug Repositioning Perturbation Score/ Class (DRPS/C). DRPS/C was used to classify the drugs into three DRPCs (Drug Repositioning Perturbation Class, such as low, intermediate, and high) based on the DRPS (Drug Repositioning Perturbation Score), where each score was calculated using the signature (disease-and drug-induced gene perturbation) from The Cancer Genome Atlas (TCGA) and Drug-induced Gene Perturbation Signature Database (DGPSD). To identify possible AD-targeted drug, signatures of AD-mediated perturbation were obtained from RNA-seq, microarray, and proteomic datasets in the GEO, CMAP, and Synapse database. In contrast, DGPSD was used for predicting drug response scores. Resultantly, 31 had promising possibilities with high scores; however, only four of these drugs were considered: monoamine oxidase inhibitors (iproniazid, selegiline) and voltage-gated sodium channel blockers (topiramate, bupivacaine) were classified as CNS active compounds [108]. Zhang et al. enlisted 524 anti-AD targets by integrating publicly available “omics” data, such as epigenomics, genomics, proteomics, and metabolomics, and the disease pathogenesis data from PubMed and OMIM databases, which could be used for drug development and as biomarkers. Using drug-target information, drug-target, name, indication, stage, and MoA from Therapeutic Target Database and DrugBank, they identified 75 drugs associated with 18 protein targets that can be subjected to new AD treatment. A modified weighted sum model algorithm was used to score targets based on fold change (AD-induced changed expression), reporting paper (AD pathogenesis), citation (Google Scholar), and the number of literature that evidence the target-AD relationship. Using this score, two AD targets, CD33 and MIF, were identified as the strongest candidates for seven known drugs. The authors also found SOD3, SPON1, INPP5D, PICALM, CELF1, BIN1, APOE, and ABCA7 promising new drug discovery targets. Furthermore, all seven known drugs could be used to alleviate cognitive symptoms as they inhibit acetylcholinesterase [109]. To measure the correlation between drug-induced alteration of neuronal cell gene expression and molecular alterations in the brains of AD patients at various stages, Rodriguez et al. developed

452

Raju Dash et al.

a machine learning-based framework called Drug Repurposing In AD (DRIAD). They applied for 80 drugs with clinical testing and FDA approval to generate a prioritized list of potential repurposing candidates. Using the data from mRNA expression profiles of postmortem brain specimens and AMP-AD datasets, the authors identified 33 FDA-approved drugs, which could directly be applied for therapy. To train and assess a predictor of AD pathology stage from AMP-AD gene expression data, DRIAD employs a custombuilt drug-associated gene list (DGL) for feature selection, which allows for an evaluation of drug MoA relating to the degenerative processes underlying AD. DLGs were generated by analyzing the responses of neurons to the small molecule (drug) using RNAseq and then identifying genes that expressed differentially. Predictive accuracy was highest for drugs that target proteins (including JAK, NEK, and ULK) involved in signaling networks controlling innate immunity, autophagy, and microtubule dynamics, all of which are novel avenues for AD therapy [110]. Xie et al. used combined gene expression from the CMap dataset and inverse molecular docking approach to identify possible hits from FDA-approved drugs. For selecting hits from FDA-approved drugs, seven major targets of AD, such as betasecretase 1 (BACE1), acetylcholinesterase (AChE), glycogen synthase kinase-3 beta (GSK-3β), N-methyl-d-aspartate (NMDA), and butyrylcholinesterase (BChE) were chosen, and subjected to molecular docking with hit selection criteria of -10 kcal mol-1, which resulted in 211 FDA-approved drugs that bound tightly to these seven targets. Among these top hits, the authors could only analyze gene expression perturbation signatures of 74 drugs. KEGG analysis showed that several perturbated genes by drugs were involved in various signaling AD, such as oxidative stress, dyshomeostasis of biometals, low acetylcholine levels, tau-protein aggregation, and amyloid deposits. Considering the criteria of multi-target modulation, if any drug is included at two signaling pathways in AD, it was selected for further studies. Using in vitro validation by Aβ25–35-induced cytotoxicity model, three potential multi-targeted drugs were further identified, including risperidone, glimepiride, and droperidol [111]. To discover potential candidates for repositioning drugs for AD, Hsieh et al. built a multi-task machine learning pipeline that integrates a complete knowledge network on pharmacological/ biological interactions and multi-level evidence on drug effectiveness. Multiple lines of evidence, including transcriptome profiles following drug perturbation, clinical trial and preclinical trial data, and population-based treatment impact assessment, were used to identify potential repurposable drug candidates and identified several effective drug combinations in maintaining cell morphology, decreasing oxidative stress, and increasing viability, for example, mifepristone in combination with galantamine [112].

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

453

Based on the idea that disease-causing genes have unique molecular characteristics and process a small set of pathobiological pathways in the human proteome network, including potential targets and disease-causing modulators, Xu et al. introduced NETTAG, a network topology-based deep learning framework, to identify potential drug targets for AD using OMICS and genetic information. Genetic features include the impacts of non-coding GWAS loci on open chromatin, promoter regions, enhancers and CpG islands, promoter flanking regions, and quantitative trait loci using the protein-protein interactome. Using NETTAG, the authors highlighted MEF2D and CPLX2 as potential AD-associated genes and 156 AD-risk genes as druggable targets. APOE, BIN1, and PICALM were also identified as AD-risk genes. Using Drugbank database and closest-based network proximity measure, the authors identified 118 candidate drugs for AD treatment, where four drugs, including ibuprofen, gemfibrozil, cholecalciferol, and ceftriaxone, appeared promising in reducing AD consequences, especially gemfibrozil, which reduced risk of AD to 43% [113]. Savva et al. proposed a network-based drug-repurposing methodology for filtering and re-ranking drugs based on a networkbased score derived from the weighted total of connections in a network mimicking the structural similarity currently underway, failed, or authorized drugs in different stages of AD severity. This methodology also filters candidate drugs using structural, functional, and theoretical information, followed by ranking on the blood–brain barrier (BBB) permeability. Applying this methodology, the authors identified 30 candidates for AD drug repositioning, where six drugs, such as, saracatinib, pioglitazone, nicotine, vorinostat, candesartan, and curcumin, were identified to appear in recent clinical trials [114]. Multiple platforms have been developed recently, particularly organizing the multimodal and comprehensive interactions derived from public data sets and experimental data created by many consortiums [115–117]. For example, AlzPlatform (www.cbligand. org/AD/) is a cloud-based platform that integrates chemogenomics databases, such as TargetHunter, HTDocking, and BBB Predictor, in identifying novel AD targets and polypharmacology. The platform complied with the 405,188 compounds, 194 drugs that have been licensed or used in clinical trials for AD, AD-related genes (928) and protein (320), and the reported bioactivities (1023137) and relevant bioassays (38284). AlzPlatform has also included case studies as examples of findings of multi-targets and polypharmacology for FDA-approved drugs and identifying new drug targets and small molecules using integrated platforms (TargetHunter and HTDocking) [115].

454

Raju Dash et al.

AlzGPS (https://alzgps.lerner.ccf.org) is another example, a genome-wide positioning systems platform for AD drug discovery, which facilitates drug repositioning by providing network-based analysis of multi-omics data. This platform aids in the search, visualization, and analysis of multi-omics, various heterogeneous biological networks, and clinical databases to pinpoint potential therapeutic and preventative avenues for AD. AlzGPS contains 100 AD multi-omics data of different AD stages, treatment information for 3000 FDA-approved drugs, and knowledge of over 1000 AD clinical trials [116]. Su¨gis et al. created HENA, a heterogeneous network-based dataset for AD, combining information from public knowledge databases with experimental datasets to better understand AD pathological mechanisms. HENA combines 64 experimental and computational datasets. This information includes GWAS results, the networks of gene co-expression, and protein-protein interaction networks. Furthermore, HENA allows for finding diseasecausing genes by constructing a gene-centric network, analyzing through graph convolutional networks [117].

7

Conclusions Given the rising prevalence of AD worldwide, the current treatments are inadequate to address this epidemic. In this context, drug repositioning has emerged as a highly effective method in drug research because of its lower costs and more expedited approval processes. The cost of repositioning a drug is significantly less than the cost of developing a drug from the beginning presents a new opportunity for research institutions. The overview of this chapter is to use drug repositioning to find out the effective treatments against AD as this technique is a valuable method for identifying candidate drugs for future exploration. Many existing drugs can target risk genes in AD, but due to a lack of sufficient information and in-depth investigation, their novelty remains hidden. Therefore, a compelling scientific strategy is necessary to screen existing drugs and conduct biological, epidemiological, and clinical research to find new prospective candidates for repurposing in AD. “Single drug, single target, single disease” is the pharmacologic paradigm associated with the lack of drug efficiency. Recent and extensive contributions of system pharmacology to drug design emphasize the importance of realizing that no single drug can treat all symptoms, but understanding their interconnected nature can shed light on everything from drug development to clinical symptom clustering. The current insight of network pharmacology of drug design is to relocate one or more targets through a complete biological network, speed up the finding of pharmacological targets, and aid in creating an entirely new drug. Using the network analysis approach, researchers also can perform technical analysis on

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . .

455

an already-established network, acquire more relevant information, and boost productivity. This is especially helpful when dealing with a large biological network whose elaborate data connections are simplified into a visually appealing map. Moreover, while it is true that drug repositioning can cut down on development costs and timelines, it is essential to remember that it has its limits and needs in in vivo and in vitro experimental settings. As a result, combining the computational approach with network pharmacology in drug repositioning can considerably increase drug screening efficiency by predicting novel uses for existing drugs and simultaneously acquiring their intrinsic properties. To sum up, this chapter proposes integrating experimental and network pharmacology-based screening to enhance the efficacy of drug repositioning in aid of safe, cost-effective, and efficient patient care by accelerating more effective, safer, and improved drug discovery in such complex diseases like AD.

Acknowledgments This work was supported by the National Research Foundation of Korea (NRF) grant (No. NRF-2021R1A2C1008564) to I.S.M. and funded by the Korean Ministry of Science and ICT. References 1. GBD 2016 Dementia Collaborators (2019) Global, regional, and national burden of Alzheimer’s disease and other dementias, 19902016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol 18(1):88–106. https://doi.org/10.1016/ s1474-4422(18)30403-4 2. Erkkinen MG, Kim MO, Geschwind MD (2018) Clinical neurology and epidemiology of the major neurodegenerative diseases. Cold Spring Harb Perspect Biol 10(4). https://doi. org/10.1101/cshperspect.a033118 3. Alzheimer A (1907) Uber eine eigenartige Erkrankung der Hirnrinde. Zentralbl Nervenh Psych 18:177–179 4. Karch Celeste M, Cruchaga C, Goate AM (2014) Alzheimer’s disease genetics: from the bench to the clinic. Neuron 83(1): 11–26. https://doi.org/10.1016/j.neuron. 2014.05.041 5. Association As (2019) 2019 Alzheimer’s disease facts and figures. Alzheimers Dement 15(3):321–387. https://doi.org/10.1016/j. jalz.2019.01.010 6. Cummings JL, Tong G, Ballard C (2019) Treatment combinations for Alzheimer’s disease: current and future pharmacotherapy

options. J Alzheimers Dis 67(3):779–794. https://doi.org/10.3233/jad-180766 7. GBD 2016 Neurology Collaborators (2019) Global, regional, and national burden of neurological disorders, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol 18(5):459–480. https://doi.org/10.1016/s1474-4422(18) 30499-x 8. Beecham GW, Bis JC, Martin ER, Choi SH, DeStefano AL, van Duijn CM et al (2017) The Alzheimer’s Disease Sequencing Project: study design and sample selection. Neurol Genet 3(5):e194. https://doi.org/10.1212/ nxg.0000000000000194 9. Mehta D, Jackson R, Paul G, Shi J, Sabbagh M (2017) Why do trials for Alzheimer’s disease drugs keep failing? A discontinued drug perspective for 2010-2015. Expert Opin Investig Drugs 26(6):735–739. https://doi. org/10.1080/13543784.2017.1323868 10. Cummings J (2018) Lessons learned from Alzheimer disease: clinical trials with negative outcomes. Clin Transl Sci 11(2):147–152. https://doi.org/10.1111/cts.12491 11. Fang J, Pieper AA, Nussinov R, Lee G, Bekris L, Leverenz JB et al (2020) Harnessing

456

Raju Dash et al.

endophenotypes and network medicine for Alzheimer’s drug repurposing. Med Res Rev 40(6):2386–2426. https://doi.org/10. 1002/med.21709 12. Pushpakom S, Iorio F, Eyers PA, Escott KJ, Hopper S, Wells A et al (2019) Drug repurposing: progress, challenges and recommendations. Nat Rev Drug Discov 18(1):41–58. https://doi.org/10.1038/nrd.2018.168 13. Li Z, Jiang X, Wang Y, Kim Y (2021) Applied machine learning in Alzheimer’s disease research: omics, imaging, and clinical data. Emerg Top Life Sci 5(6):765–777. https:// doi.org/10.1042/etls20210249 14. Pulley JM, Rhoads JP, Jerome RN, Challa AP, Erreger KB, Joly MM et al (2020) Using what we already have: uncovering new drug repurposing strategies in existing omics data. Annu Rev Pharmacol Toxicol 60:333–352. https:// doi.org/10.1146/annurev-phar mtox010919-023537 15. Sleigh SH, Barton CL (2010) Repurposing strategies for therapeutics. Pharm Med 24(3):151–159. https://doi.org/10.1007/ BF03256811 16. Novac N (2013) Challenges and opportunities of drug repositioning. Trends Pharmacol Sci 34(5):267–272. https://doi.org/10. 1016/j.tips.2013.03.004 17. Millan MJ, Goodwin GM, Meyer¨ S (2015) Learning Lindenberg A, Ove O from the past and looking to the future: emerging perspectives for improving the treatment of psychiatric disorders. Eur Neuropsychopharmacol 25(5):599–656. https:// doi.org/10.1016/j.euroneuro.2015.01.016 18. Alavijeh MS, Palmer AM (2010) Measurement of the pharmacokinetics and pharmacodynamics of neuroactive compounds. Neurobiol Dis 37(1):38–47. https://doi. org/10.1016/j.nbd.2009.09.025 19. Waring MJ, Arrowsmith J, Leach AR, Leeson PD, Mandrell S, Owen RM et al (2015) An analysis of the attrition of drug candidates from four major pharmaceutical companies. Nat Rev Drug Discov 14(7):475–486. https://doi.org/10.1038/nrd4609 20. Bastos LFS, Coelho MM (2014) Drug repositioning: playing dirty to kill pain. CNS Drugs 28(1):45–61. https://doi.org/10. 1007/s40263-013-0128-0 21. Lerner AJ, Gustaw-Rothenberg K, Smyth S, Casadesus G (2012) Retinoids for treatment of Alzheimer’s disease. Biofactors 38(2): 84–89. https://doi.org/10.1002/biof.196 22. Krezel W, Kastner P, Chambon P (1999) Differential expression of retinoid receptors in the adult mouse central nervous system. Neuroscience 89(4):1291–1300. https://doi. org/10.1016/s0306-4522(98)00342-x

23. Olson CR, Mello CV (2010) Significance of vitamin A to brain function, behavior and learning. Mol Nutr Food Res 54(4): 489–495. https://doi.org/10.1002/mnfr. 200900246 24. Corcoran JP, So PL, Maden M (2004) Disruption of the retinoid signalling pathway causes a deposition of amyloid beta in the adult rat brain. Eur J Neurosci 20(4): 896–902. https://doi.org/10.1111/j. 1460-9568.2004.03563.x 25. Shudo K, Fukasawa H, Nakagomi M, Yamagata N (2009) Towards retinoid therapy for Alzheimer’s disease. Curr Alzheimer Res 6(3): 302–311. https://doi.org/10.2174/ 156720509788486581 26. Tai LM, Koster KP, Luo J, Lee SH, Wang YT, Collins NC et al (2014) Amyloid-β pathology and APOE genotype modulate retinoid X receptor agonist activity in vivo. J Biol Chem 289(44):30538–30555. https://doi.org/10. 1074/jbc.M114.600833 27. Zito PM, Mazzoni T (2022) Acitretin. StatPearls. StatPearls Publishing Copyright © 2022, StatPearls Publishing LLC, Treasure Island (FL) 28. Qin X, Chen C, Zhang Y, Zhang L, Mei Y, Long X et al (2017) Acitretin modulates HaCaT cells proliferation through STAT1and STAT3-dependent signaling. Saudi Pharm J 25(4):620–624. https://doi.org/ 10.1016/j.jsps.2017.04.034 29. Endres K, Fahrenholz F, Lotz J, Hiemke C, Teipel S, Lieb K et al (2014) Increased CSF APPs-α levels in patients with Alzheimer disease treated with acitretin. Neurology 83(21): 1930–1935. https://doi.org/10.1212/wnl. 0000000000001017 30. Dos Santos GM, Stoye NM, Rose-John S, Garbers C, Fellgiebel A, Endres K (2019) The synthetic retinoid Acitretin increases IL-6 in the central nervous system of Alzheimer disease Model Mice and human patients. Front Aging Neurosci 11:182. https://doi. org/10.3389/fnagi.2019.00182 31. Fukasawa H, Nakagomi M, Yamagata N, Katsuki H, Kawahara K, Kitaoka K et al (2012) Tamibarotene: a candidate retinoid drug for Alzheimer’s disease. Biol Pharm Bull 35(8):1206–1212. https://doi.org/10. 1248/bpb.b12-00314 32. Adis International Limited (2004) Tamibarotene: AM 80, retinobenzoic acid, Tamibaro. Drugs R D 5(6):359–362. https://doi.org/ 10.2165/00126839-200405060-00010 33. Miwako I, Kagechika H (2007) Tamibarotene. Drugs Today (Barc) 43(8):563–568. https://doi.org/10.1358/dot.2007.43.8. 1072615

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . . 34. Qiao A, Li J, Hu Y, Wang J, Zhao Z (2021) Reduction BACE1 expression via suppressing NF-κB mediated signaling by Tamibarotene in a mouse model of Alzheimer’s disease. IBRO Neurosci Rep 10:153–160. https:// doi.org/10.1016/j.ibneur.2021.02.004 35. 2015 Alzheimer’s disease facts and figures (2015) Alzheimers Dement 11(3):332–384. https://doi.org/10.1016/j.jalz.2015. 02.003 36. Li X, Song D, Leng SX (2015) Link between type 2 diabetes and Alzheimer’s disease: from epidemiology to mechanism and treatment. Clin Interv Aging 10:549–560. https://doi. org/10.2147/cia.S74042 37. Batista AFB, Forny-Germano L, Clarke JR, Lyrae Silva NM, Brito-Moreira J, Boehnke SE et al (2018) The diabetes drug liraglutide reverses cognitive impairment in mice and attenuates insulin receptor and synaptic pathology in a non-human primate model of Alzheimer’s disease. J Pathol 245:85–100 38. Holubova´ M, Hruba´ L, Popelova´ A, Bencze M, Prazˇienkova´ V, Gengler S et al (2019) Liraglutide and a lipidized analog of prolactin-releasing peptide show neuroprotective effects in a mouse model of β-amyloid pathology. Neuropharmacology 144:377– 3 8 7 . h t t p s : // d o i . o r g / 1 0 . 1 0 1 6 / j . neuropharm.2018.11.002 39. Silverman RB, Holladay MW (2014) Chapter 6 - DNA-interactive agents. In: Silverman RB, Holladay MW (eds) The organic chemistry of drug design and drug action, 3rd edn. Academic Press, Boston, pp 275–331 40. Brent TP, Remack JS (1988) Formation of covalent complexes between human O6-alkylguanine-DNA alkyltransferase and BCNUtreated defined length synthetic oligodeoxynucleotides. Nucleic Acids Res 16(14b):6779–6788. https://doi.org/10. 1093/nar/16.14.6779 41. Jahan I, Nayeem SM (2021) Destabilization of Alzheimer’s Aβ42 protofibrils with acyclovir, carmustine, curcumin, and tetracycline: insights from molecular dynamics simulations. New J Chem 45(45):21031–21048. https://doi.org/10.1039/D1NJ04453B 42. Hayes CD, Dey D, Palavicini JP, Wang H, Patkar KA, Minond D et al (2013) Striking reduction of amyloid plaque burden in an Alzheimer’s mouse model after chronic administration of carmustine. BMC Med 11: 81. https://doi.org/10.1186/1741-701511-81 43. Gold LS, Dhawan S, Weiss J, Draelos ZD, Ellman H, Stuart IA (2019) A novel topical minocycline foam for the treatment of moderate-to-severe acne vulgaris: results of

457

2 randomized, double-blind, phase 3 studies. J Am Acad Dermatol 80(1):168–177. https://doi.org/10.1016/j.jaad.2018. 08.020 44. Langevitz P, Livneh A, Bank I, Pras M (2000) Benefits and risks of minocycline in rheumatoid arthritis. Drug Saf 22(5):405–414. https://doi.org/10.2165/00002018200022050-00007 45. Seabrook TJ, Jiang L, Maier M, Lemere CA (2006) Minocycline affects microglia activation, Abeta deposition, and behavior in APP-tg mice. Glia 53(7):776–782. https:// doi.org/10.1002/glia.20338 46. Howard R, Zubko O, Bradley R, Harper E, Pank L, O’Brien J et al (2020) Minocycline at 2 different dosages vs placebo for patients with mild Alzheimer disease: a randomized clinical trial. JAMA Neurol 77(2):164–174. https://doi.org/10.1001/jamaneurol.2019. 3762 47. Lawlor B, Segurado R, Kennelly S, Olde Rikkert MGM, Howard R, Pasquier F et al (2018) Nilvadipine in mild to moderate Alzheimer disease: a randomised controlled trial. PLoS Med 15(9):e1002660. https://doi. org/10.1371/journal.pmed.1002660 48. Iqbal N, Iqbal N (2014) Imatinib: a breakthrough of targeted therapy in cancer. Chemother Res Pract 2014:357027. https://doi. org/10.1155/2014/357027 49. Cancino GI, Perez de Arce K, Castro PU, Toledo EM, von Bernhardi R, Alvarez AR (2011) c-Abl tyrosine kinase modulates tau pathology and Cdk5 phosphorylation in AD transgenic mice. Neurobiol Aging 32(7): 1249–1261. https://doi.org/10.1016/j. neurobiolaging.2009.07.007 50. Wang J, Ho L, Chen L, Zhao Z, Zhao W, Qian X et al (2007) Valsartan lowers brain beta-amyloid protein levels and improves spatial learning in a mouse model of Alzheimer disease. J Clin Invest 117(11):3393–3402. https://doi.org/10.1172/jci31547 51. Torika N, Asraf K, Apte RN, FleisherBerkovich S (2018) Candesartan ameliorates brain inflammation associated with Alzheimer’s disease. CNS Neurosci Ther 24(3): 231–242. https://doi.org/10.1111/cns. 12802 52. Tsukuda K, Mogi M, Iwanami J, Min LJ, Sakata A, Jing F et al (2009) Cognitive deficit in amyloid-beta-injected mice was improved by pretreatment with a low dose of telmisartan partly because of peroxisome proliferatoractivated receptor-gamma activation. Hypertension 54(4):782–787. https://doi.org/10. 1161/hypertensionaha.109.136879

458

Raju Dash et al.

53. Lahiri DK, Chen D, Maloney B, Holloway HW, Yu QS, Utsuki T et al (2007) The experimental Alzheimer’s disease drug posiphen [(+)-phenserine] lowers amyloid-beta peptide levels in cell culture and mice. J Pharmacol Exp Ther 320(1):386–396. https://doi.org/ 10.1124/jpet.106.112102 54. Marutle A, Ohmitsu M, Nilbratt M, Greig NH, Nordberg A, Sugaya K (2007) Modulation of human neural stem cell differentiation in Alzheimer (APP23) transgenic mice by phenserine. Proc Natl Acad Sci U S A 104(30):12506–12511. https://doi.org/10. 1073/pnas.0705346104 55. Lahiri DK, Alley GM, Tweedie D, Chen D, Greig NH (2007) Differential effects of two hexahydropyrroloindole carbamate-based anticholinesterase drugs on the amyloid beta protein pathway involved in Alzheimer’s disease. NeuroMolecular Med 9(2):157–168. https://doi.org/10.1007/bf02685889 56. Winblad B, Giacobini E, Fro¨lich L, Friedhoff LT, Bruinsma G, Becker RE et al (2010) Phenserine efficacy in Alzheimer’s disease. J Alzheimers Dis 22(4):1201–1208. https:// doi.org/10.3233/jad-2010-101311 57. Kadir A, Andreasen N, Almkvist O, Wall A, Forsberg A, Engler H et al (2008) Effect of phenserine treatment on brain functional activity and amyloid in Alzheimer’s disease. Ann Neurol 63(5):621–631. https://doi. org/10.1002/ana.21345 58. Shibuya M, Suzuki Y (1993) Treatment of cerebral vasospasm by a protein kinase inhibitor AT 877. No To Shinkei 45(9):819–824 59. Sellers KJ, Elliott C, Jackson J, Ghosh A, Ribe E, Rojo AI et al (2018) Amyloid β synaptotoxicity is Wnt-PCP dependent and blocked by fasudil. Alzheimers Dement 14(3): 306–317. https://doi.org/10.1016/j.jalz. 2017.09.008 60. Killick R, Ribe EM, Al-Shawi R, Malik B, Hooper C, Fernandes C et al (2014) Clusterin regulates β-amyloid toxicity via Dickkopf-1driven induction of the wnt-PCP-JNK pathway. Mol Psychiatry 19(1):88–98. https:// doi.org/10.1038/mp.2012.163 61. Elliott C, Rojo AI, Ribe E, Broadstock M, Xia W, Morin P et al (2018) A role for APP in Wnt signalling links synapse loss with β-amyloid production. Transl Psychiatry 8(1):179. https://doi.org/10.1038/ s41398-018-0231-6 62. Yan B, Sun F, Duan L-h, Pen Q-l, Zhao W-x, Zhou G-q (2011) Curative effect of Fasudil injection combined with Nimodipine on Alzheimer disease of elderly patients. J Clin Med Pract 14:7–9

63. Xu Y, Kong J, Hu P (2021) Computational drug repurposing for Alzheimer’s disease using risk genes from GWAS and single-cell RNA sequencing studies. Front Pharmacol 12:617537. https://doi.org/10.3389/fphar. 2021.617537 64. Costa FF (2014) Big data in biomedicine. Drug Discov Today 19(4):433–440. https:// doi.org/10.1016/j.drudis.2013.10.012 65. Pilarczyk M, Fazel-Najafabadi M, Kouril M, Shamsaei B, Vasiliauskas J, Niu W et al (2022) Connecting omics signatures and revealing biological mechanisms with iLINCS. Nat Commun 13(1):4678. https://doi.org/10. 1038/s41467-022-32205-3 66. Cha Y, Erez T, Reynolds IJ, Kumar D, Ross J, Koytiger G et al (2018) Drug repurposing from the perspective of pharmaceutical companies. Br J Pharmacol 175(2):168–180. https://doi.org/10.1111/bph.13798 67. Hopkins AL (2008) Network pharmacology: the next paradigm in drug discovery. Nat Chem Biol 4(11):682–690. https://doi.org/ 10.1038/nchembio.118 68. Yıldırım MA, Goh K-I, Cusick ME, Baraba´si A-L, Vidal M (2007) Drug—target network. Nat Biotechnol 25(10):1119–1126. https:// doi.org/10.1038/nbt1338 69. Roth BL, Sheffler DJ, Kroeze WK (2004) Magic shotguns versus magic bullets: selectively non-selective drugs for mood disorders and schizophrenia. Nat Rev Drug Discov 3(4):353–359. https://doi.org/10.1038/ nrd1346 70. Paolini GV, Shapland RHB, van Hoorn WP, Mason JS, Hopkins AL (2006) Global mapping of pharmacological space. Nat Biotechnol 24(7):805–815. https://doi.org/10. 1038/nbt1228 71. Markesbery WR (1997) Oxidative stress hypothesis in Alzheimer’s disease. Free Radic Biol Med 23(1):134–147. https://doi.org/ 10.1016/s0891-5849(96)00629-6 72. Pratico` D (2008) Oxidative stress hypothesis in Alzheimer’s disease: a reappraisal. Trends Pharmacol Sci 29(12):609–615. https://doi. org/10.1016/j.tips.2008.09.001 73. Heneka MT, Carson MJ, El Khoury J, Landreth GE, Brosseron F, Feinstein DL et al (2015) Neuroinflammation in Alzheimer’s disease. Lancet Neurol 14(4):388–405. https://doi.org/10.1016/s1474-4422(15) 70016-5 74. Hardy J, Allsop D (1991) Amyloid deposition as the central event in the aetiology of Alzheimer’s disease. Trends Pharmacol Sci 12(10): 383–388. https://doi.org/10.1016/01656147(91)90609-v

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . . 75. Games D, Adams D, Alessandrini R, Barbour R, Berthelette P, Blackwell C et al (1995) Alzheimer-type neuropathology in transgenic mice overexpressing V717F betaamyloid precursor protein. Nature 373(6514):523–527. https://doi.org/10. 1038/373523a0 76. Goedert M, Spillantini MG, Crowther RA (1991) Tau proteins and neurofibrillary degeneration. Brain Pathol 1(4):279–286. https://doi.org/10.1111/j.1750-3639. 1991.tb00671.x 77. Iqbal K, Liu F, Gong CX, Grundke-Iqbal I (2010) Tau in Alzheimer disease and related tauopathies. Curr Alzheimer Res 7(8): 656–664. https://doi.org/10.2174/ 156720510793611592 78. Brunden KR, Trojanowski JQ, Lee VM (2009) Advances in tau-focused drug discovery for Alzheimer’s disease and related tauopathies. Nat Rev Drug Discov 8(10): 783–793. https://doi.org/10.1038/ nrd2959 79. Bartus RT, Dean RL 3rd, Beer B, Lippa AS (1982) The cholinergic hypothesis of geriatric memory dysfunction. Science 217(4558): 408–414. https://doi.org/10.1126/science. 7046051 80. Hampel H, Mesulam MM, Cuello AC, Khachaturian AS, Vergallo A, Farlow MR et al (2019) Revisiting the cholinergic hypothesis in Alzheimer’s disease: emerging evidence from translational and clinical research. J Prev Alzheimers Dis 6(1):2–15. https://doi. org/10.14283/jpad.2018.43 81. Dash R, Jahan I, Ali MC, Mitra S, Munni YA, Timalsina B et al (2021) Potential roles of natural products in the targeting of proteinopathic neurodegenerative diseases. Neurochem Int 145:105011. https://doi.org/10. 1016/j.neuint.2021.105011 82. Dash R, Ali MC, Jahan I, Munni YA, Mitra S, Hannan MA et al (2021) Emerging potential of cannabidiol in reversing proteinopathies. Ageing Res Rev 65:101209. https://doi. org/10.1016/j.arr.2020.101209 83. Van Cauwenberghe C, Van Broeckhoven C, Sleegers K (2016) The genetic landscape of Alzheimer disease: clinical implications and perspectives. Genet Med 18(5):421–430. https://doi.org/10.1038/gim.2015.117 84. Albertini C, Salerno A, de Sena Murteira Pinheiro P, Bolognesi ML (2021) From combinations to multitarget-directed ligands: a continuum in Alzheimer’s disease polypharmacology. Med Res Rev 41(5):2606–2633. https://doi.org/10.1002/med.21699

459

85. Gao J, Barzel B, Baraba´si A-L (2016) Universal resilience patterns in complex networks. Nature 530(7590):307–312. https://doi. org/10.1038/nature16948 86. Nogales C, Mamdouh ZM, List M, Kiel C, Casas AI, Schmidt HHHW (2022) Network pharmacology: curing causal mechanisms instead of treating symptoms. Trends Pharmacol Sci 43(2):136–150. https://doi.org/ 10.1016/j.tips.2021.11.004 87. Iorio F, Bosotti R, Scacheri E, Belcastro V, Mithbaokar P, Ferriero R et al (2010) Discovery of drug mode of action and drug repositioning from transcriptional responses. Proc Natl Acad Sci U S A 107(33):14621–14626. h t t p s : // d o i . o r g / 1 0 . 1 0 7 3 / p n a s . 1000138107 88. Berger SI, Iyengar R (2009) Network analyses in systems pharmacology. Bioinformatics 25(19):2466–2472. https://doi.org/10. 1093/bioinformatics/btp465 89. Nacher JC, Schwartz J-M (2008) A global view of drug-therapy interactions. BMC Pharmacol 8(1):5. https://doi.org/10.1186/ 1471-2210-8-5 90. Liu Z, Fang H, Reagan K, Xu X, Mendrick DL, Slikker W et al (2013) In silico drug repositioning – what we need to know. Drug Discov Today 18(3):110–115. https://doi. org/10.1016/j.drudis.2012.08.005 91. Wang G, Dunbrack RL Jr (2003) PISCES: a protein sequence culling server. Bioinformatics 19(12):1589–1591. https://doi.org/10. 1093/bioinformatics/btg224 92. Chartier M, Morency LP, Zylber MI, Najmanovich RJ (2017) Large-scale detection of drug off-targets: hypotheses for drug repurposing and understanding side-effects. BMC. Pharmacol Toxicol 18(1). https://doi.org/ 10.1186/s40360-017-0128-7 93. Ruths DA, Nakhleh L, Iyengar MS, Reddy SA, Ram PT (2006) Hypothesis generation in signaling networks. J Comput Biol 13(9): 1546–1557. https://doi.org/10.1089/cmb. 2006.13.1546 94. Dasika MS, Burgard A, Maranas CD (2006) A computational framework for the topological analysis and targeted disruption of signal transduction networks. Biophys J 91(1): 382–398. https://doi.org/10.1529/ biophysj.105.069724 95. Boran AD, Iyengar R (2010) Systems pharmacology. Mt Sinai J Med 77(4):333–344. https://doi.org/10.1002/msj.20191 96. Sobie EA, Lee YS, Jenkins SL, Iyengar R (2011) Systems biology--biomedical

460

Raju Dash et al.

modeling. Sci Signal 4(190):tr2. https://doi. org/10.1126/scisignal.2001989 97. van Hasselt JGC, Iyengar R (2019) Systems pharmacology: defining the interactions of drug combinations. Annu Rev Pharmacol Toxicol 59:21–40. https://doi.org/10. 1146/annurev-pharmtox-010818-021511 98. Goh K-I, Cusick ME, Valle D, Childs B, Vidal M, Baraba´si A-L (2007) The human disease network. Proc Natl Acad Sci 104(21):8685–8690. https://doi.org/10. 1073/pnas.0701361104 99. Ye H, Wei J, Tang K, Feuers R, Hong H (2016) Drug repositioning through network pharmacology. Curr Top Med Chem 16(30): 3646–3656. https://doi.org/10.2174/ 1568026616666160530181328 100. Keiser MJ, Setola V, Irwin JJ, Laggner C, Abbas AI, Hufeisen SJ et al (2009) Predicting new molecular targets for known drugs. Nature 462(7270):175–181. https://doi. org/10.1038/nature08506 101. Kwon OS, Kim W, Cha HJ, Lee H (2019) In silico drug repositioning: from large-scale transcriptome data to therapeutics. Arch Pharm Res 42(10):879–889. https://doi. org/10.1007/s12272-019-01176-3 102. Dudley JT, Sirota M, Shenoy M, Pai RK, Roedder S, Chiang AP et al (2011) Computational repositioning of the anticonvulsant topiramate for inflammatory bowel disease. Sci Transl Med 3(96). https://doi.org/10. 1126/scitranslmed.3002648 103. Sirota M, Dudley JT, Kim J, Chiang AP, Morgan AA, Sweet-Cordero A et al (2011) Discovery and preclinical validation of drug indications using compendia of public gene expression data. Sci Transl Med 3(96): 96ra77–96ra77. https://doi.org/10.1126/ scitranslmed.3001318 104. Hu G, Agarwal P (2009) Human diseasedrug network based on genomic expression profiles. PLoS One 4(8):e6536. https://doi. org/10.1371/journal.pone.0006536 105. Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ et al (2006) The connectivity map: using gene-expression signatures to connect small molecules, genes, and disease. Science 313(5795):1929–1935. https://doi.org/10.1126/science.1132939 106. Campillos M, Kuhn M, Gavin AC, Jensen LJ, Bork P (2008) Drug target identification using side-effect similarity. Science 321(5886):263–266. https://doi.org/10. 1126/science.1158140 107. Hoffmann P, Warner B (2006) Are hERG channel inhibition and QT interval

prolongation all there is in drug-induced torsadogenesis? A review of emerging trends. J Pharmacol Toxicol Methods 53(2):87–105. https://doi.org/10.1016/j.vascn.2005. 07.003 108. Lee SY, Song M-Y, Kim D, Park C, Park DK, Kim DG et al (2020) A Proteotranscriptomicbased computational drug-repositioning method for Alzheimer’s disease. Front Pharmacol:10. https://doi.org/10.3389/fphar. 2019.01653 109. Zhang M, Schmitt-Ulms G, Sato C, Xi Z, Zhang Y, Zhou Y et al (2016) Drug repositioning for Alzheimer’s disease based on systematic ‘omics’ data mining. PLoS One 11(12):e0168812. https://doi.org/10. 1371/journal.pone.0168812 110. Rodriguez S, Hug C, Todorov P, Moret N, Boswell SA, Evans K et al (2021) Machine learning identifies candidates for drug repurposing in Alzheimer’s disease. Nat Commun 12(1):1033. https://doi.org/10.1038/ s41467-021-21330-0 111. Xie H, Wen H, Qin M, Xia J, Zhang D, Liu L et al (2016) In silico drug repositioning for the treatment of Alzheimer’s disease using molecular docking and gene expression data. RSC Adv 6(100):98080–98090 112. Hsieh K-L, Plascencia-Villa G, Lin K-H, Perry G, Jiang X, Kim Y (2021) Deep learning for Alzheimer’s disease drug repurposing using knowledge graph and multi-level evidence. medRxiv:2021120321267235. https://doi.org/10.1101/2021.12.03. 21267235 113. Xu J, Mao C, Hou Y, Luo Y, Binder JL, Zhou Y et al (2022) Interpretable deep learning translation of GWAS and multi-omics findings to identify pathobiology and drug repurposing in Alzheimer’s disease. Cell Rep 41(9): 111717. https://doi.org/10.1016/j.celrep. 2022.111717 114. Savva K, Zachariou M, Bourdakou MM, Dietis N, Spyrou GM (2022) Networkbased stage-specific drug repurposing for Alzheimer’s disease. Comput Struct Biotechnol J 20:1427–1438. https://doi.org/10.1016/j. csbj.2022.03.013 115. Liu H, Wang L, Lv M, Pei R, Li P, Pei Z et al (2014) AlzPlatform: an Alzheimer’s disease domain-specific chemogenomics knowledgebase for polypharmacology and target identification research. J Chem Inf Model 54(4): 1050–1060. https://doi.org/10.1021/ ci500004h 116. Zhou Y, Fang J, Bekris LM, Kim YH, Pieper AA, Leverenz JB et al (2021) AlzGPS: a genome-wide positioning systems platform

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . . to catalyze multi-omics for Alzheimer’s drug discovery. Alzheimers Res Ther 13(1):24. https://doi.org/10.1186/s13195-02000760-w 117. Su¨gis E, Dauvillier J, Leontjeva A, Adler P, Hindie V, Moncion T et al (2019) HENA, heterogeneous network-based data set for Alzheimer’s disease. Sci Data 6(1):151. https://doi.org/10.1038/s41597-0190152-0 118. Wishart DS, Feunang YD, Guo AC, Lo EJ, Marcu A, Grant JR et al (2018) DrugBank 5.0: a major update to the DrugBank database for 2018. Nucleic Acids Res 46(D1): D1074–D1d82. https://doi.org/10.1093/ nar/gkx1037 119. Kim Kjærulff S, Wich L, Kringelum J, Jacobsen UP, Kouskoumvekaki I, Audouze K et al (2013) ChemProt-2.0: visual navigation in a disease chemical biology database. Nucleic Acids Res 41(Database issue):D464–9. https://doi.org/10.1093/nar/gks1166 120. Goto S, Okuno Y, Hattori M, Nishioka T, Kanehisa M (2002) LIGAND: database of chemical compounds and reactions in biological pathways. Nucleic Acids Res 30(1):402–404. https://doi.org/10.1093/ nar/30.1.402 121. Kuhn M, Szklarczyk D, Franceschini A, Campillos M, von Mering C, Jensen LJ et al (2010) STITCH 2: an interaction network database for small molecules and proteins. Nucleic Acids Res 38(Database issue): D552–6. https://doi.org/10.1093/nar/ gkp937 122. Huang R, Southall N, Wang Y, Yasgar A, Shinn P, Jadhav A et al (2011) The NCGC pharmaceutical collection: a comprehensive resource of clinically approved drugs enabling repurposing and chemical genomics. Sci Transl Med 3(80):80ps16. https://doi.org/ 10.1126/scitranslmed.3001862 123. Kim S, Chen J, Cheng T, Gindulyte A, He J, He S et al (2021) PubChem in 2021: new data content and improved web interfaces. Nucleic Acids Res 49(D1):D1388–D1D95. https://doi.org/10.1093/nar/gkaa971 124. Gaulton A, Bellis LJ, Bento AP, Chambers J, Davies M, Hersey A et al (2012) ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Res 40(Database issue): D1100–7. https://doi.org/10.1093/nar/ gkr777 125. Gilson MK, Liu T, Baitaluk M, Nicola G, Hwang L, Chong J (2016) BindingDB in 2015: a public database for medicinal chemistry, computational chemistry and systems pharmacology. Nucleic Acids Res 44(D1):

461

D1045–D1053. https://doi.org/10.1093/ nar/gkv1072 126. Wang Y, Zhang S, Li F, Zhou Y, Zhang Y, Wang Z et al (2020) Therapeutic target database 2020: enriched resource for facilitating research and early development of targeted therapeutics. Nucleic Acids Res 48(D1): D1031–D1d41. https://doi.org/10.1093/ nar/gkz981 127. Gu¨nther S, Kuhn M, Dunkel M, Campillos M, Senger C, Petsalaki E et al (2008) SuperTarget and Matador: resources for exploring drug-target relationships. Nucleic Acids Res 36(Database issue): D919–22. https://doi.org/10.1093/nar/ gkm862 128. Hecker N, Ahmed J, von Eichborn J, Dunkel M, Macha K, Eckert A et al (2012) SuperTarget goes quantitative: update on drug-target interactions. Nucleic Acids Res 40(Database issue):D1113–7. https://doi. org/10.1093/nar/gkr912 ˜ os MP, Carmona SJ, Crowther GJ, 129. Magarin Ralph SA, Roos DS, Shanmugam D et al (2012) TDR targets: a chemogenomics resource for neglected diseases. Nucleic Acids Res 40(Database issue):D1118–27. https://doi.org/10.1093/nar/gkr1053 130. Emig D, Ivliev A, Pustovalova O, Lancashire L, Bureeva S, Nikolsky Y et al (2013) Drug target prediction and repositioning using an integrated network-based approach. PLoS One 8(4):e60618. https:// doi.org/10.1371/journal.pone.0060618 131. Li H, Gao Z, Kang L, Zhang H, Yang K, Yu K et al (2006) TarFisDock: a web server for identifying drug targets with docking approach. Nucleic Acids Res 34(Web Server issue):W219–24. https://doi.org/10.1093/ nar/gkl114 132. Liu X, Ouyang S, Yu B, Liu Y, Huang K, Gong J et al (2010) PharmMapper server: a web server for potential drug target identification using pharmacophore mapping approach. Nucleic Acids Res 38(Web Server issue):W609–14. https://doi.org/10.1093/ nar/gkq300 133. Chou CH, Lin FM, Chou MT, Hsu SD, Chang TH, Weng SL et al (2013) A computational approach for identifying microRNAtarget interactions using high-throughput CLIP and PAR-CLIP sequencing. BMC Genomics 14 Suppl 1(Suppl 1):S2. https:// doi.org/10.1186/1471-2164-14-s1-s2 134. Li YY, An J, Jones SJ (2011) A computational approach to finding novel targets for existing drugs. PLoS Comput Biol 7(9):e1002139. https://doi.org/10.1371/journal.pcbi. 1002139

462

Raju Dash et al.

135. Keiser MJ, Roth BL, Armbruster BN, Ernsberger P, Irwin JJ, Shoichet BK (2007) Relating protein pharmacology by ligand chemistry. Nat Biotechnol 25(2):197–206. https://doi.org/10.1038/nbt1284 136. Cheng F, Zhou Y, Li W, Liu G, Tang Y (2012) Prediction of chemical-protein interactions network with weighted networkbased inference method. PLoS One 7(7): e41064. https://doi.org/10.1371/journal. pone.0041064 137. Daina A, Michielin O, Zoete V (2019) SwissTargetPrediction: updated data and new features for efficient prediction of protein targets of small molecules. Nucleic Acids Res 47 (W1):W357–WW64. https://doi.org/10. 1093/nar/gkz382 138. Luo H, Zhang P, Huang H, Huang J, Kao E, Shi L et al (2014) DDI-CPI, a server that predicts drug-drug interactions through implementing the chemical-protein interactome. Nucleic Acids Res 42(Web Server issue):W46–52. https://doi.org/10.1093/ nar/gku433 139. Wang L, Ma C, Wipf P, Liu H, Su W, Xie XQ (2013) TargetHunter: an in silico target identification tool for predicting therapeutic potential of small organic molecules based on chemogenomic database. AAPS J 15(2): 395–406. https://doi.org/10.1208/ s12248-012-9449-z 140. Carrella D, Napolitano F, Rispoli R, Miglietta M, Carissimo A, Cutillo L et al (2014) Mantra 2.0: an online collaborative resource for drug mode of action and repurposing by network analysis. Bioinformatics 30(12):1787–1788. https://doi.org/10. 1093/bioinformatics/btu058 141. Edgar R, Domrachev M, Lash AE (2002) Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res 30(1):207–210. https://doi.org/10.1093/nar/30.1.207 142. Amberger JS, Bocchini CA, Schiettecatte F, Scott AF, Hamosh A (2015) OMIM.org: Online Mendelian Inheritance in Man (OMIM®), an online catalog of human genes and genetic disorders. Nucleic Acids Res 43(D1):D789–DD98. https://doi.org/ 10.1093/nar/gku1205 143. Ding Y, Chen M, Liu Z, Ding D, Ye Y, Zhang M et al (2012) atBioNet– an integrated network analysis tool for genomics and biomarker discovery. BMC Genomics 13(1): 325. https://doi.org/10.1186/1471-216413-325 144. Kanehisa M, Araki M, Goto S, Hattori M, Hirakawa M, Itoh M et al (2007) KEGG for

linking genomes to life and the environment. Nucleic Acids Res 36(suppl_1):D480–D4. https://doi.org/10.1093/nar/gkm882 145. Wishart DS, Tzur D, Knox C, Eisner R, Guo AC, Young N et al (2007) HMDB: the human metabolome database. Nucleic Acids Res 35(Database issue):D521–6. https://doi. org/10.1093/nar/gkl923 146. Pletscher-Frankild S, Palleja` A, Tsafou K, Binder JX, Jensen LJ (2015) DISEASES: text mining and data integration of disease– gene associations. Methods 74:83–89. https://doi.org/10.1016/j.ymeth.2014. 11.020 ˜ ero J, Sau¨ch J, Sanz F, Furlong LI (2021) 147. Pin The DisGeNET cytoscape app: exploring and visualizing disease genomics data. Comput Struct Biotechnol J 19:2960–2967. https:// doi.org/10.1016/j.csbj.2021.05.015 148. Judson R, Richard A, Dix D, Houck K, Elloumi F, Martin M et al (2008) ACToR — aggregated computational toxicology resource. Toxicol Appl Pharmacol 233(1): 7–13. https://doi.org/10.1016/j.taap. 2007.12.037 149. von Eichborn J, Murgueitio MS, Dunkel M, Koerner S, Bourne PE, Preissner R (2010) PROMISCUOUS: a database for networkbased drug-repositioning. Nucleic Acids Res 39(suppl_1):D1060–D6. https://doi.org/ 10.1093/nar/gkq1037 150. Thorn CF, Klein TE, Altman RB (2013) PharmGKB: the pharmacogenomics knowledge base. Methods Mol Biol 1015:311– 320. https://doi.org/10.1007/978-162703-435-7_20 151. Davis AP, Grondin CJ, Johnson RJ, Sciaky D, Wiegers J, Wiegers TC et al (2020) Comparative Toxicogenomics Database (CTD): update 2021. Nucleic Acids Res 49(D1): D1138–D1D43. https://doi.org/10.1093/ nar/gkaa891 152. Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ (2006) The ToxCast program for prioritizing toxicity testing of environmental chemicals. Toxicol Sci 95(1): 5–12. https://doi.org/10.1093/toxsci/ kfl103 153. Young J, Tong W, Fang H, Xie Q, Pearce B, Hashemi R et al (2004) Building an organspecific carcinogenic database for SAR analyses. J Toxicol Environ Health A 67(17): 1363–1389. https://doi.org/10.1080/ 15287390490471479 154. Fang H, Harris SC, Liu Z, Zhou G, Zhang G, Xu J et al (2016) FDA drug labeling: rich resources to facilitate precision medicine,

Network Pharmacology for Drug Repositioning in Anti-Alzheimer’s Drug. . . drug safety, and regulatory science. Drug Discov Today 21(10):1566–1570 155. Food U, Administration D (2020) Questions and answers on FDA’s adverse event reporting system 156. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D et al (2003) Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 13(11):2498–2504. https:// doi.org/10.1101/gr.1239303 157. Hu Z, Mellor J, Wu J, Yamada T, Holloway D, Delisi C (2005) VisANT: dataintegrating visual framework for biological networks and modules. Nucleic Acids Res 33

463

(Web Server issue):W352–7. https://doi. org/10.1093/nar/gki431 158. Luo W, Pant G, Bhavnasi YK, Blanchard SG Jr, Brouwer C (2017) Pathview web: user friendly pathway visualization and data integration. Nucleic Acids Res 45(W1):W501– W5W8. https://doi.org/10.1093/nar/ gkx372 159. Dennis G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC et al (2003) DAVID: database for annotation, visualization, and integrated discovery. Genome Biol 4(9): R60. https://doi.org/10.1186/gb-2003-49-r60

Chapter 16 Web Services for the Prediction of ADMET Parameters Relevant to the Design of Neuroprotective Drugs Valentin O. Perkin, Grigory V. Antonyan, Eugene V. Radchenko, and Vladimir A. Palyulin Abstract The ADMET properties (absorption, distribution, metabolism, excretion, and toxicity) of a compound are key factors in its success as a drug. While many experimental approaches to their measurement are available, the computational approaches are in high demand due to their cost- and time-efficiency. Some ADMET properties are especially relevant for Alzheimer’s disease treatments and other neuroprotective compounds. This review provides an overview of 13 online tools that can be used to predict the ADMET profile of a compound, with a special focus on the properties such as lipophilicity, aqueous solubility, hERG inhibition, blood–brain barrier permeability, Caco-2 permeability, human intestinal absorption, and Ames mutagenicity. The functionality of different tools is compared, and the correlations of their predictions with one another and with the available experimental data are analyzed. Key words ADMET, Drug design, In silico prediction, Alzheimer’s disease, Neuroprotective drugs

1

Introduction Designing new drugs requires large investments and is characterized by a rapid decrease in the number of drug candidates as the development pipeline progresses, and, at the same time, by rapid increase of the costs of each subsequent development phase. Therefore, the principle “fail fast, fail cheap” is highly relevant for pharmaceutical development. As a result, various approaches to the prediction and optimization of properties of biologically active compounds have been proposed and implemented that are commonly known as rational drug design. In particular, ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties of a molecule are a crucial component of the success of a substance as a drug candidate [1]. Substances that have favorable ADMET properties are more likely to be approved as medicines. Before 2000, the problems with ADMET accounted for the failure

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3_16, © The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

465

466

Valentin O. Perkin et al.

of up to 40% of drug candidates. However, after 2000, a number of test procedures that weed out substances that do not have favorable ADMET profile during the early stages of testing have been introduced by the pharmaceutical companies into their drug development pipelines, which significantly reduced the attrition related to the pharmacokinetics and toxicity issues [2, 3]. However, if their accuracy is similar, the computational approaches to the prediction of ADMET properties are preferable to the experimental ones due to lower costs and greater time and resource efficiency, allowing one to predict the properties and filter the compounds in silico even before they are synthesized and tested and switch to the in vitro studies only at the later stages of the pipeline and for fewer structures [4–9]. Therefore, the in silico prediction of ADMET properties is a field that has for a long time attracted the attention of many teams of researchers employing very diverse methods—from classical QSAR studies [10] to the models using modern machine learning techniques [11]. Traditionally, such models and prediction algorithms were distributed in the source code form or as components of commercial or free closed-source chemical software. A new step in simplifying access to the predictive models for ADMET properties was made with the appearance of the open access services. They make it possible to quickly test the hypotheses and evaluate the compounds on dozens of in silico models from one’s Web browser, without special hardware and software, and before their synthesis and in vitro testing. This greatly reduces the expenses and delays, as well as minimizes the need for animal testing. A large number of citations (e.g., 3971 for the SwissADME [12] service) confirm the high demand for such services among the academic researchers. Quite a few services have been created that predict various sets of properties based on different methodologies. However, as was previously shown, the predictions from different in silico models are not always consistent, and it is often difficult to rank their expected reliability in the absence of diverse and representative validation data obtained from the experiments [13]. While many ADMET properties are important for the druglikeness, bioavailability, and safety of any drug compound, some properties are particularly relevant for the agents intended for the Alzheimer’s disease treatment, as well as other neuroprotective compounds. These include the lipophilicity that greatly affects the distribution of a compound in tissues, especially the blood–brain barrier (BBB) permeability that determines the possibility of acting in the brain. In this chapter, we give an overview of the available online tools that can be used to predict the ADMET properties of a compound, compare the functionality of the most popular ADMET prediction services, and evaluate the consistency of their predictions with each

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

467

other and (if possible) with experimental data for a number of properties relevant to the design of anti-Alzheimer’s and other neuroprotective drugs, using a test set of drugs and drug-like compounds focused on the chemical space of potential drug candidates for the treatment of the Alzheimer’s disease.

2 2.1 2.1.1

Materials and Methods Data Collection Prediction Services

2.1.2 Prediction Test Set and Evaluation Procedure

For this study, it was decided to collect the widest available set of links to the prediction services, which could then be filtered according to their operational status, performance, and convenience for the researchers. As a starting point, the lists compiled in the Web page [14] and the review [15] were used; some sites were also identified using Web search results. After filtering out the obsolete and irrelevant links, as well as the links to defunct sites and to online code repositories for the stand-alone property prediction programs, the list of links to 22 online services for predicting ADMET properties from the structural formula of a substance was obtained. Next, the services (Web sites) that did not respond to requests or had limited availability depending on the region, the services that imposed obstacles for a large-scale use (in particular, required solving a CAPTCHA to proceed), and the services that accepted the prediction request and did not send a result within a week were excluded. Some services were also unavailable during the data collection period or seemed to have a very long response time. The remaining 13 services (Table 1) were employed to predict the properties of the test set compounds. In addition, since two stand-alone chemical data management software tools, ChemAxon (MarvinSketch 22.9, ChemAxon Kft., Budapest, Hungary, https://chemaxon.com/) and DataWarrior 5.5.0 [16], as well as the OCHEM online platform [17], are commonly used to predict some physicochemical and ADMET properties in workflows similar to the ones employing Web-based prediction services, their prediction results were also included in the comparison to make it more representative. For testing of the prediction services, a representative and diverse set of structures of 184 biologically active compounds with various action profiles was compiled. They were selected among three groups of compounds: – Drug candidates reported in the ClinicalTrials.gov database [18] for 2020, in particular, for the treatment of the Alzheimer’s disease (79 molecules).

468

Valentin O. Perkin et al.

Table 1 Web services and software tools included in the testing for prediction of ADMET properties Namea

URL

Input

Properties predicted

ADMETlab

https://admet. scbdd.com/ calcpre/index/

SMILES

logP, Ames test, BBB, Caco2, logS, hERG

ADMETlab 2.0 (ADMETMesh) [23]

https://admetmesh. scbdd.com/ service/ evaluation/index

SMILES

(53 models) Ames test, BBB, Caco2, CYP, hERG, logP, logS, Pgp, HIA, toxicity, NR/SR

admetSAR2 [24]

http://lmmd.ecust. edu.cn/ admetsar2/

SMILES

(48 models) Ames toxicity, BBB Caco2, CYP inhibition/ substrate, hERG, HIA, Pgp, solubility

AMED [25]

https://drugdesign. riken.jp/ hERGdb/

SDF from sketch (JSME), paste SMILES, SDF

hERG, Nav1.5, Kv1.5, Cav

BBB Predictor (BBBpred) [26]

https://www. cbligand.org/ BBB/predictor. php

SDF from sketch (JSME), paste SMILES, SDF

BBB

ADME@NCATS (ADMENCATS) [27]

https://opendata. ncats.nih.gov/ adme/

SMILES CSV or sketch

PAMPA permeability, HLC, RLM stability, logS, CYP inhibition

pkCSM [28]

https://biosig.lab. uq.edu.au/ pkcsm/prediction

Up to 100 SMILES (31 models) logP, solubility, Caco2, HIA, BBB, skin permeability, in file Pgp, HERG, CYP inhibition, Ames test, toxicity

PreADMET [29]

Sketch SDF https://preadmet. (ChemDoodle), qsarhub.com/ paste MOL https://preadmet. webservice.bmdrc. org/

logP, BBB, solubility, Caco2, CYP inhibition

ADMET Prediction Service (QSARMSU) [30]

http://qsar.chem. msu.ru/admet/

SDF from sketch (JSME), paste SMILES, SDF

HIA, hERG, BBB

SwissADME [12]

http://www. swissadme.ch/ index.php

List of SMILES

(27 models) logP, logS, HIA, BBB, Pgp, CYP inhibition, skin permeation

vNN-ADMET [31]

https://vnnadmet. bhsai.org/ vnnadmet/

SMILES

Ames test, BBB, CYP inhibition, cytotoxicity, hERG, HLM, MRTD, Pgp (continued)

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

469

Table 1 (continued) Namea

URL

Input

Properties predicted

SSL-ToxGCN

https://app.cbbio. online/ssl-gcn/ home/

SMILES FASTAlike

7 nuclear and 5 stress receptors response

Way2Drug [32]

http://www.way2 drug.com/

SMILES

hERG, CYP inhibition / substrate

ChemAxon

https://chemaxon. com/

Sketch (ChemAxon), SMILES, SDF

logP, logD, pKa, solubility

DataWarrior [16]

https:// openmolecules. org/datawarrior/

Sketch (Idorsia), SMILES, SDF

logP, solubility, mutagenicity

OCHEM [17]

https://ochem.eu/

Sketch (JSME), SMILES, SDF

logP, solubility, Ames test, CYP inhibition, and other models

a

Short convenient names that are often used to refer to a service in this paper (if necessary) are listed in parentheses

– Compounds from the PubChem BioAssays database [19] involved in the studies of the ADMET properties from 2019 to 2022 (74 molecules). – Compounds listed as representative examples of the drugs of various pharmaceutical groups (both neuroactive and non-neuroactive) in a medicinal chemistry textbook (31 molecules) [20]. Presumably, the latter subset of compounds should (or at least could) have been represented in the training sets of the models implemented in the reviewed Web services; the first two subsets were specifically limited to the most recent data to minimize their chances of being taken into account by some (but not all) of the models, thus providing sufficient data for a fair comparison. For a small subset of compounds (54 structures), the experimental data on lipophilicity (logP) were available, making it possible to validate the predictions by comparison to the experiment. The structure of each compound was represented as a SMILES string (obtained from the PubChem database) and an SDF file obtained from SMILES using the Open Babel 3.0.0 [21] software. The molecules in the resulting dataset were characterized by the descriptors involved in the Lipinski’s rule of five using the tools available in the RDKit 2021.09.2 [22] software. The distribution plots of the calculated logP, molecular weight, and numbers of the H-bond donor and acceptor groups (Fig. 1) confirm that the test set compounds cover rather broad regions of the drug-like

470

Valentin O. Perkin et al.

Fig. 1 Properties of molecules in the test dataset. (a) Predicted logP. (b) Molecular weight. (c) Number of H-bond donor groups. (d) Number of H-bond acceptor groups

chemical space and should thus belong to the applicability domains of the prediction models. In the course of the testing, the structures were submitted to the Web services, and the results were collected and processed automatically by means of the specially developed Python3 scripts using the Pandas, Requests, and Selenium libraries. Next, the results were summarized in the tables of predictions of similar properties, and the Pearson correlation and Spearman rank correlation coefficients between predictions of various services were analyzed.

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

2.2 Featured Services

471

The services that were included in the testing and comparison are listed in Table 1. ADMETlab and ADMETlab2 [23] services developed by Xiangya School of Pharmaceutical Sciences at Central South University (China) allow predicting a wide range of physicochemical and ADMET properties of compounds. The input is done by uploading SDF and writing SMILES string or drawing a structure; the output is downloadable as a comma-separated values (CSV) file. admetSAR2 [24] is an ADMET prediction service by School of Pharmacy, East China University of Science and Technology that provides interface for various machine learning-derived models. Available input methods are SMILES string or drawing a molecule, and results can be downloaded as CSV files. AMED [25] is a server developed by researchers at the RIKEN Center for Life Science Technologies (Japan) for prediction of several properties of compounds important for the heart function: inhibition of potassium, sodium, and calcium channels. The input is done by drawing the structure and converting it to SMILES or by uploading an SDF file, and the output is downloaded as a CSV (comma-separated values) file. Online BBB Predictor (BBBpred) [26] is a service that predicts the permeability of a substance through the blood–brain barrier from its structure. Input methods available are uploading file in SDF, MOL, SMILES format, or drawing a molecule on the Web page. ADME@NCATS (ADMENCATS) [27] is a service developed by NCATS researchers for predicting a number of ADME properties: stability, solubility, parallel artificial membrane permeability assay (PAMPA), and interactions with cytochrome P450 family of enzymes. The input is done by drawing a structure or by uploading a CSV file, and the output is downloadable as a CSV file. pkCSM [28] is an ADMET property prediction service developed by the University of Melbourne that is using for prediction the physicochemical and graph-based molecular descriptors. It allows up to 100 SMILES strings for molecules to be entered in a batch file, and output is available in CSV format. PreADMET [29] is a service by the Bioinformatics and Molecular Design Research Center at Yonsei University. In the two versions of the site, more than 30 various ADMET properties can be predicted such as solubility, Ames test, or CYP inhibition. The only input method is drawing a molecule with the ChemDoodle applet on the Web site, and output can be downloaded as Excel, CSV, or PDF files. ADMET Prediction Service (QSARMSU) [30] from the Laboratory of Medicinal Chemistry, Lomonosov Moscow State University is a service that provides models predicting three properties of compounds: BBB permeability [33], human intestinal absorption [34], and inhibitor activity against hERG [35]. The input is

472

Valentin O. Perkin et al.

carried out by drawing the structure in the JS editor on the Web site. SwissADME [12] is a huge set of services for predicting various ADMET properties that is developed by the Swiss Institute of Bioinformatics. It provides batch upload up to 200 SMILES, and the output is in CSV format. vNN-ADMET [31] is a server developed by the Biotechnology HPC Software Applications Institute (USA) researchers that predicts the metabolic and toxic properties of substances based on the similarity of the studied structure with structures from the training set. The predicted value is the weighted average of the corresponding properties of the nearest neighbors of the compound selected by structural similarity. The input is carried out by uploading CSV with structures encoded as SMILES or by drawing the structure with an applet on the Web page; the output is available for download in CSV format. SSLToxGCN is a model based on the methods of semisupervised learning developed by a team from the University of Macau that predicts the activity of substances to nuclear receptors by converting a structure-representing SMILES string to a graph and applying the prediction technique involving Graph Convolution Neural Networks. The input is via SMILES upload in FASTAlike format, and the output is downloadable in CSV format. Way2Drug [32] is a set of services developed at the Orekhovich Institute of Biomedical Chemistry (Moscow) predicting various physicochemical, ADMET, and bioactivity properties of compounds. Among them are models that predict the toxicity of a substance and its penetration through the BBB, hERG inhibition, and CYP inhibition/substrate. The input methods available are drawing the structure and uploading SMILES strings or MOL file, and the output is in text format. ChemAxon stand-alone software for structure editing and structure database management, search, and analysis (in particular, Marvin Sketch and Instant JChem) includes a number of prediction modules for properties such as logP, pKa, logD, and solubility. The structures can be drawn in the editor or pasted/ imported in a variety of formats including SMILES and SDF. In Instant JChem, batch and dynamic calculations are available. The output can be exported as SDF, text file, spreadsheet, etc. or copied directly. DataWarrior [16] is a powerful stand-alone open-source program for data visualization and analysis with chemical intelligence that offers efficient tools for structure database management, search, and analysis, including prediction of a number of properties such as logP, solubility, and mutagenicity. The structures can be drawn in the editor or pasted/imported in a variety of formats including SMILES and SDF; batch calculations are available. The output can be exported as SDF or text file or copied directly.

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

473

OCHEM (Online Chemical Modeling Environment) [17] is a Web-based platform that aims to automate and simplify the typical steps required for management of the structure-property data and for the QSAR modeling based on a variety of structural descriptors and modern machine learning methods. Using any available public or private predictive model, one can predict various physicochemical, ADMET and bioactivity properties of compounds including logP, solubility, Ames test, CYP inhibition, and others. The structures can be drawn in the editor or pasted/imported in a variety of formats including SMILES and SDF; batch calculations are available. The output can be exported as SDF, text, or spreadsheet file or copied directly. Among the considered services, SwissADME, pkCSM, and ADMETMesh have the most extensive sets of predictable properties. In the vNN-ADMET service, the predicted values are mainly represented in the output of the service in a binary “yes/no” format. There is also an output mode that shows whether the uploaded compounds are within the limits of the model applicability domain, determined by the conformance of the calculated values of logP, molecular weight, and the numbers of hydrogen bond donors and acceptors in the molecule to the value ranges of these parameters in the training set. In this mode, it was revealed that a significant part of the compounds from the test set used in this review is apparently beyond the model applicability limits to a large extent in terms of at least some of the individual properties. The admetSAR2 service also primarily represents the result data as binary but also adds the probability that the compound belongs to a given class. In addition to notifications that a compound is within the applicability limits, it also highlights the results that are important for determination of the toxicity or bioavailability of the compound.

3

Results and Discussion

3.1 Data Peculiarities and the Properties Analyzed

One of the issues that complicated the analysis of the results and their trends was caused by a large number and wide variety of sometimes disparate properties predicted by various services and models. Taking this into account, only a few key groups of “recurring” properties were identified that are available in several different models and can typically be important for evaluation and optimization of new drugs, especially neuroprotective agents. The following groups of similar or related properties were selected: logP/logD, solubility, Caco-2 permeability, human intestinal absorption, Ames toxicity test, blood–brain barrier permeability, hERG inhibition, and CYP inhibition. The results for each group were compiled into tables, and the graphs and/or heat maps of Pearson correlations and Spearman rank correlations were obtained.

474

Valentin O. Perkin et al.

Another complication that made it difficult to directly compare various predictions arose from the differences in the scales, measurement units, and/or methodologies for determination of the predicted categories of compounds (e.g., different cutoff values could be applied to the predicted probability or property values in order to determine if a compound will have “good” solubility). If possible, the obtained values were assigned or transformed to a consistent scale (e.g., by replacing the predicted probabilities p with 1 – p if a model outputs a probability that the compound belongs to a “negative” class, while other models predict the probability of belonging to a “positive” class). The actual value scales used in the analysis of correlations between predictions are reported in specific tables. In many cases, this standardization allowed the possibility of at least an indirect comparison. Nevertheless, we had to exclude from analysis the results from several models that provide only binary (categorical) predictions, without reporting the underlying continuous probability estimates. When evaluating and benchmarking the predictions obtained from the in silico models, one obviously would be interested primarily in the comparison—and correspondence—of the predicted and experimental values. Unfortunately, this approach was difficult to implement due to a severe lack of test compounds that had experimental measurements suitable for such a comparison and obtained only recently (not older than 2019), so that one could be reasonably sure that these values were not available when various models were trained and can thus be confidently used in a validation dataset. In fact, we were able to collect such experimental data even for a small set of compounds (54 structures) only for the lipophilicity (logP) values. For other predicted properties, there are no recently released reasonably sized datasets (and it would be meaningless to test a model on the same data that was used to train it), and only the correlation coefficients between the quantitative results obtained by different models for the entire test set were calculated. However, as was shown previously [13], the data on the mutual consistency and cross-correlations between different predictions for a (supposedly) the same property are also very important and could provide indirect indications of the general accuracy and reliability of the models—especially if the predicted values are to be used to select or filter the compounds in any way. 3.2 Analysis of Predictions for Specific Properties: Implications and Patterns 3.2.1

Lipophilicity

Apparently, the most popular predicted property among various services was the (calculated) logP value, i.e., the logarithm of the partition (distribution) coefficient of a compound in the octanol– water system, traditionally used as a key indicator of its lipophilicity. The logD value, which is close in meaning, characterizes the (apparent) distribution in such a system for the ionizable substances. Lipophilicity has a significant impact on the penetrating ability of the drugs and their distribution in tissues, explaining its role as one

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

475

of the classic predictors of drug-likeness. The logP and their associated logD values turned out to be the most comparable among various prediction services and models, which could also rely on systematic highly accurate experimental datasets for a large number of compounds. Taking advantage of the available recent experimental measurements of logP for a subset of compounds (54 structures), we were able to calculate the correlation coefficients of predictions from different services not only with one another but also with the experimental data. These coefficients varied widely between different models (from -0.09 to 0.74; see Table 2) and demonstrated strong differences when the measurements conducted by different teams were considered. In addition, the experimental dataset was pretty small and had very uneven coverage of chemical space. All of these issues prevented us from confidently judging the comparability and quality of the prediction results.

Table 2 (a) Pearson correlations of the logP predictions. (b) Spearman rank correlations of the logP predictions

476

Valentin O. Perkin et al.

Fig. 2 Ranked experimental data points vs. predictions by ADMETMesh

In particular, the experimental dataset comprised the data from two PubChem bioassays as well as a number of data points extracted from the public DrugBank database. For the analyzed models predicting logP and logD, the values of correlation coefficients of the in silico predictions with the experimental data and among themselves differed significantly between these three subsets. For example, for the subset based on the PubChem AID 1698306 (Octanol/water partition coefficient, logP measured by potentiometric titration analysis), the maximum correlation coefficient (for the pair “experimental logP – ADMETMesh logP”) was 0.75, while for the subset from AID 1278580 (Octanol/water partition coefficient, logP measured by shake flask method), it was only 0.15, and for the subset of classical bioactive compounds, it was 0.89. The maximum rank correlation for the entire dataset was obtained with ADMETMesh and amounted to 0.59, maximum Pearson correlation amounted to 0.74 (Figs. 2 and 3). Regarding the comparison between different predictions (marked with a double frame in Table 2), one can note that, with the exception of PreADMET AlogP98 that is based on a rather crude Ghose approach, their cross-correlations for the full test set were quite decent and varied from moderate to very high. This result is also in agreement with the conclusion [13] that the models of the state-of-the-art quality tend to correlate well. In addition, these models face a very well-defined logP property and can build upon very big and diverse structure–lipophilicity datasets with accurately measured values. All this probably serves to level the competition and ensure high or at least reasonable quality of all the models. Interestingly, for the pairs with medium-strength correlations, the rank correlation coefficients tend to be lower than the continuous ones, probably reflecting ranking disruptions due to the random spread of similar values.

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

477

Fig. 3 Correlation of experimental data and predictions by ADMETMesh 3.2.2 hERG Inhibition

Predictions of the hERG inhibitory activity also were quite common in the services reviewed herein. hERG, or human ether-a-gogo-related gene, encodes a protein forming a potassium ion channel that participates in the electrical activity of the heart. This channel is responsible for regulating the repolarization of the heart cells, which plays an important role in maintaining regular heart rhythm. Activity against hERG (Table 3) is a predictor of potentially fatal side effects of a substance, such as the appearance of a long QT interval syndrome leading to a chaotic heartbeat, and this makes it an important anti-target to avoid. In the considered services, such as vNN-ADMET and AMED, the IC50 value for hERG = 10 μM was used to cut off active substances from the inactive ones. Some models predict the activity in terms of actual IC50 values, while others report the estimated probability that a compound would act as an hERG blocker. As can be seen from Table 3, the cross correlations between the predicted values in both the IC50 and probability scales varied from moderate to low, with rather similar values of the rank and continuous correlation coefficients.

3.2.3 Blood–Brain Barrier (BBB) Permeability

Predictions of the BBB permeability were available in several of the reviewed services. The function of the blood–brain barrier is to protect the brain from the penetration of (potentially dangerous or disruptive) xenobiotics; thus, it is important to know the

478

Valentin O. Perkin et al.

Table 3 (a) Pearson correlations of the hERG inhibition predictions. (b) Spearman rank correlations of the hERG inhibition predictions

compound’s ability to penetrate this barrier (BBB permeability, blood–brain barrier permeability) when one aims to minimize the side effects of a non-CNS drug or, conversely, to obtain agents acting in the brain. This penetration ability is usually characterized by the logarithm of the ratio of the compound concentration in the brain to its concentration in the blood (logBB). As the experiments to determine the logBB values in vivo are quite laborious and expensive, they are often substituted by various in vitro permeability models [36]. The analysis of correlations between the predicted values (Table 4) reveals rather uneven picture, with correlation coefficients varying from moderate to rather low. In part, this could be related to differences in the size and representativeness of the training sets. Here, the rank correlations generally tend to be higher than the continuous ones, indicating better ranking consistency despite differences in the predicted values. 3.2.4

Caco-2 Permeability

3.2.5 Human Intestinal Absorption (HIA)

The Caco-2 permeability characterizes the penetration of a compound through a monolayer of the cultured human colorectal adenocarcinoma cells. It is widely used as an in vitro model of the barrier function of the human intestinal mucosa. In the in silico models considered here, the logarithm of the apparent permeability coefficient (measured in centimeters per second) is predicted. The correlation coefficients between predictions for this property varied from moderate to rather low, with roughly similar levels of the rank and continuous correlations (Table 5). The human intestinal absorption (HIA) value represents the percentage (fraction) of a substance that is absorbed while passing through the small intestine, the main route of absorption of the orally administered drugs. This is the parameter that largely determines the oral bioavailability of a drug. In the models considered

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

479

Table 4 (a) Pearson correlations of the LogBB predictions. (b) Spearman rank correlations of the LogBB predictions

Table 5 (a) Pearson correlations of the Caco-2 permeability predictions. (b) Spearman rank correlations of the Caco-2 permeability predictions

here, either the probability that the compound would belong to the high absorption category or the actual percentage of the absorbed dose were predicted. ADMETMesh places a compound into HIA+ category if its predicted HIA is less than 30% (poor absorption, usually classified as HIA–), which may surprise a new user. The continuous and rank correlation coefficients for different HIA prediction varied widely, from 0.48 to 0.03 (Table 6).

480

Valentin O. Perkin et al.

Table 6 (a) Pearson correlations of the HIA predictions. (b) Spearman rank correlations of the HIA predictions

3.2.6 Water Solubility

Water solubility, commonly predicted as logS (logarithm of molar concentration of a substance in the saturated aqueous solution), is also often considered during drug design as it impacts the pharmacokinetic profile and the pharmaceutical formulation requirements of a compound, influencing its absorption, distribution, and bioavailability. The correlation coefficients between the solubility predictions from different models represented rather uneven picture, with values varying from -0.03 to 0.94 (Table 7). Interestingly, the DataWarrior and OCHEM predictions, while strongly correlated to each other, were very dissimilar to the results from all other models. On the other hand, the correlations between different Web services varied from decent to very high. Similarly to lipophilicity, this could to some extent be explained by the fact that the solubility is a very well-defined property, and rather big and diverse structure–solubility datasets with accurately measured values are available.

3.2.7 Ames Mutagenicity

In the Ames mutagenicity (toxicity) test, the mutagenicity of a substance is evaluated using as an experimental object the Salmonella typhimurium bacteria with disabled histidine synthesis pathway. Appearance of mutations in bacteria under the influence of the test substance leads to the colony growth on a medium poor in histidine [37]. This test is used as a predictor allowing an approximate estimate of the carcinogenicity in humans. Although it has been pointed out that there might be significant differences in mutagenicity between humans and Salmonella because the latter cells are prokaryotic, in fact, in early studies, the Ames test correctly detected up to 90% of the known carcinogens [38]. The models considered herein usually produced either a binary prediction of a positive or negative test result or the probability that the test would turn out to be positive. The correlation coefficients between the

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

481

Table 7 (a) Pearson correlations of the solubility predictions. (b) Spearman rank correlations of the solubility predictions

Table 8 (a) Pearson correlations of the Ames toxicity test predictions. (b) Spearman rank correlations of the Ames toxicity test predictions

three available probability-type estimates (Table 8) turned out to be rather low or, at best, moderate. 3.2.8 CYP Interactions

The “CYP interactions” category comprises the effects of a substance that are related to the ability of the heme-dependent monooxygenases of the cytochrome P450 superfamily to carry out the oxidation of xenobiotics. The services considered here offer the predictions for two types of the interactions of substances with

482

Valentin O. Perkin et al.

CYPs, as their inhibitors or as substrates. For each of these interactions, the probability for a substance to be in an “active” group or category of activity is predicted, usually for specific CYP isoforms. If a compound interacts with CYP as a substrate, the resulting oxidation may provide a mechanism for the metabolic transformation of a drug to an active or bioavailable form (e.g., for codeine) or, conversely, cause its inactivation by transforming it to inactive or even toxic metabolites. CYP inhibition may be an undesirable activity for a substance in some cases since it could prolong the action of other drugs and effectively increase their doses in an unpredictable way. On the other hand, the same effect could be used beneficially in pharmacology in order to protect and “boost” the doses of some drugs that would otherwise be quickly cleared from the body. Although CYP activity predictions are supported by many services, among the common groups of predicted properties, they turned out to be poorly comparable between services because of the wide variety of modeled protein inhibition processes, which are often unique to the specific models.

4

Conclusions The computer-aided drug design is a rapidly developing and expanding field, and the implementation of open-access Web servers that host predictive structure–activity and structure–property models is an important approach to making its fruits more easily accessible. We have reviewed 13 open-access services for ADMET prediction that vary in their scope and accuracy of the models. The ADMETlab2, admetSAR2, and pkCSM services provide the broadest variety of the ADMET prediction models (53, 48, and 31, respectively), with a modest level of usability of the input and output functions. The most convenient input method for the practical purposes (multiple SMILES upload) is available in the SwissADME and pkCSM services. For a practical testing of the services and the evaluation of consistency of their predictions, we focused on a number of properties relevant to the design of anti-Alzheimer’s and other neuroprotective drugs, such as lipophilicity, aqueous solubility, hERG inhibition, blood–brain barrier permeability, Caco-2 permeability, human intestinal absorption, and Ames mutagenicity, although many of these properties would also be important for the druglikeness, bioavailability, and safety of any drug compound. The test set of 184 drugs and drug-like compounds, while diverse, also was focused on the chemical space of potential drug candidates for the treatment of the Alzheimer’s disease. For the lipophilicity (logP) prediction, taking advantage of the available recent experimental measurements for a small subset of

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . .

483

54 compounds, we were able to evaluate their correlations with the predicted values. The logP predictions provided by ADMETMesh and SwissADME, as well as by the popular DataWarrior and ChemAxon software, were quite accurate. An extensive analysis of cross-correlations between the predictions from different services revealed a rather uneven picture. For some well-defined properties with big available training datasets, such as lipophilicity and aqueous solubility, most of the services provide very consistent predictions. For many properties (e.g., hERG inhibition, blood–brain barrier permeability, Caco-2 permeability, human intestinal absorption), the correlations are generally moderate, although both low and high values can sometimes be observed for specific pairs of models. Finally, for the Ames mutagenicity, the correlation coefficients turned out to be rather low or, at best, moderate. The differences between predictions from various models are probably caused by different underlying training sets, modeling methodologies, and applicability domains. Overall, the open-access predictive ADMET models and Web services can provide useful data for the drug discovery and development processes, but their performance may sometimes be severely limited, especially for relatively complicated properties and undersized or insufficiently representative training sets. Moreover, a researcher should not be deceived by the idea that all in silico predictions are basically equal. Before relying on them for decisionmaking, one should carefully evaluate the available solutions and choose an option providing acceptable results in the relevant problem space.

Acknowledgments This work was supported by the State Assignment of the Department of Chemistry, Lomonosov Moscow State University (project no. 121021000105-7). References 1. Smith DA, Allerton C, Kalgutkar AS, van de Waterbeemd H, Walker DK (2012) Pharmacokinetics and metabolism in drug design, 3rd edn. Wiley-VCH, Weinheim 2. Khanna I (2012) Drug discovery in pharmaceutical industry: productivity challenges and trends. Drug Discov Today 17:1088–1102. https://doi.org/10.1016/j.drudis.2012. 05.007 3. Waring MJ, Arrowsmith J, Leach AR, Leeson PD, Mandrell S, Owen RM, Pairaudeau G, Pennie WD, Pickett SD, Wang J, Wallace O, Weir A (2015) An analysis of the attrition of

drug candidates from four major pharmaceutical companies. Nat Rev Drug Discov 14:475– 486. https://doi.org/10.1038/nrd4609 4. Rognan D (2017) The impact of in silico screening in the discovery of novel and safer drug candidates. Pharmacol Ther 175:47–66. https://doi.org/10.1016/j.pharmthera.2017. 02.034 5. Ferreira LLG, Andricopulo AD (2019) ADMET modeling approaches in drug discovery. Drug Discov Today 24:1157–1165. https://doi.org/10.1016/j.drudis.2019. 03.015

484

Valentin O. Perkin et al.

6. Hemmerich J, Ecker GF (2020) In silico toxicology: from structure-activity relationships towards deep learning and adverse outcome pathways. WIREs Comput Mol Sci 10:e1475. https://doi.org/10.1002/wcms.1475 7. Kumar A, Kini SG, Rathi E (2021) A recent appraisal of artificial intelligence and in silico ADMET prediction in the early stages of drug discovery. Mini Rev Med Chem 21:2788– 2 8 0 0 . h t t p s : // d o i . o r g / 1 0 . 2 1 7 4 / 1389557521666210401091147 8. Ota R, Yamashita F (2022) Application of machine learning techniques to the analysis and prediction of drug pharmacokinetics. J Control Release 352:961–969. https://doi. org/10.1016/j.jconrel.2022.11.014 9. Pantalea˜o SQ, Fernandes PO, Gonc¸alves JE, Maltarollo VG, Honorio KM (2022) Recent advances in the prediction of pharmacokinetics properties in drug design dtudies: a review. ChemMedChem 17:e202100542. https:// doi.org/10.1002/cmdc.202100542 10. Toshio F, Junkichi I, Corwin H (1964) A new substituent constant, π, derived from partition coefficients. J Am Chem Soc 86:5175–5180. https://doi.org/10.1021/ja01077a028 11. Wenzel J, Matter H, Schmidt F (2019) Predictive multitask deep neural network models for ADME-Tox properties: learning from large data sets. J Chem Inf Model 59:1253–1268. https://doi.org/10.1021/acs.jcim.8b00785 12. Daina A, Michielin O, Zoete V (2017) SwissADME: a free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules. Sci Rep 7: 42717. https://doi.org/10.1038/srep42717 13. Sosnina EA, Osolodkin DI, Radchenko EV, Sosnin S, Palyulin VA (2018) Influence of descriptor implementation on compound ranking based on multiparameter assessment. J Chem Inf Model 58:1083–1093. https://doi. org/10.1021/acs.jcim.7b00734 14. Villoutreix B ADMET and physchem predictions and related tools. http://www.vls3d. com/index.php/links/chemoinformatics/ admet/admet-and-physchem-predictionsand-related-tools. Accessed 22 Sept 2022 15. Kar S, Leszczynski J (2020) Open access in silico tools to predict the ADMET profiling of drug candidates. Expert Opin Drug Discov 15: 1473–1487. https://doi.org/10.1080/ 17460441.2020.1798926 16. Sander T, Freyss J, von Korff M, Rufener C (2015) DataWarrior: an open-source program for chemistry aware data visualization and analysis. J Chem Inf Model 55:460–473. https:// doi.org/10.1021/ci500588j

17. Sushko I, Novotarskyi S, Ko¨rner R, Pandey AK, Rupp M, Teetz W, Brandmaier S, Abdelaziz A, Prokopenko VV, Tanchuk VY, Todeschini R, Varnek A, Marcou G, Ertl P, Potemkin V, Grishina M, Gasteiger J, Schwab C, Baskin II, Palyulin VA, Radchenko EV, Welsh WJ, Kholodovych V, Chekmarev D, Cherkasov A, Aires-de-Sousa J, Zhang Q-Y, Bender A, Nigsch F, Patiny L, Williams A, Tkachenko V, Tetko IV (2011) Online chemical modeling environment (OCHEM): web platform for data storage, model development and publishing of chemical information. J Comput Aided Mol Des 25:533–554. https://doi.org/10.1007/s10822-0119440-2 18. ClinicalTrials.gov. https://clinicaltrials.gov/. Accessed 22 Sept 2022 19. Kim S, Chen J, Cheng T, Gindulyte A, He J, He S, Li Q, Shoemaker BA, Thiessen PA, Yu B, Zaslavsky L, Zhang J, Bolton EE (2021) PubChem in 2021: new data content and improved web interfaces. Nucleic Acids Res 49:D1388– D1395. https://doi.org/10.1093/nar/ gkaa971 20. Patrick G (2017) An introduction to medicinal chemistry, 6th edn. Oxford University Press, New York 21. O’Boyle NM, Banck M, James CA, Morley C, Vandermeersch T, Hutchison GR (2011) Open babel: an open chemical toolbox. J Cheminform 3:33. https://doi.org/10.1186/ 1758-2946-3-33 22. RDKit: open-source cheminformatics software. https://www.rdkit.org/. Accessed 10 Oct 2022 23. Xiong G, Wu Z, Yi J, Fu L, Yang Z, Hsieh C, Yin M, Zeng X, Wu C, Lu A, Chen X, Hou T, Cao D (2021) ADMETlab 2.0: an integrated online platform for accurate and comprehensive predictions of ADMET properties. Nucleic Acids Res 49:W5–W14. https://doi.org/10. 1093/nar/gkab255 24. Yang H, Lou C, Sun L, Li J, Cai Y, Wang Z, Li W, Liu G, Tang Y (2019) admetSAR 2.0: web-service for prediction and optimization of chemical ADMET properties. Bioinformatics 35:1067–1069. https://doi.org/10.1093/bio informatics/bty707 25. Sato T, Yuki H, Ogura K, Honma T (2018) Construction of an integrated database for hERG blocking small molecules. PLoS One 13:e0199348. https://doi.org/10.1371/jour nal.pone.0199348 26. Liu H, Wang L, Lv M, Pei R, Li P, Pei Z, Wang Y, Su W, Xie X-Q (2014) AlzPlatform: an Alzheimer’s disease domain-specific chemogenomics knowledgebase for

Web Services for the Prediction of ADMET Parameters Relevant to the Design. . . polypharmacology and target identification research. J Chem Inf Model 54:1050–1060. https://doi.org/10.1021/ci500004h 27. Williams J, Siramshetty V, Nguye˜ˆ n Ð-T, Padilha EC, Kabir M, Yu K-R, Wang AQ, Zhao T, Itkin M, Shinn P, Mathe´ EA, Xu X, Shah P (2022) Using in vitro ADME data for lead compound selection: an emphasis on PAMPA pH 5 permeability and oral bioavailability. Bioorg Med Chem 56:116588. https:// doi.org/10.1016/j.bmc.2021.116588 28. Pires DEV, Blundell TL, Ascher DB (2015) pkCSM: predicting small-molecule pharmacokinetic and toxicity properties using graphbased signatures. J Med Chem 58:4066– 4 0 7 2 . h t t p s : // d o i . o r g / 1 0 . 1 0 2 1 / a c s . jmedchem.5b00104 29. Lee SK, Lee IH, Kim HJ, Chang GS, Chung JE, No KT (2003) The PreADME Approach: Web-based program for rapid prediction of physico-chemical, drug absorption and druglike properties. In: EuroQSAR 2002 designing drugs and crop protectants: processes, problems and solutions. Blackwell Publishing, Massachusetts, USA, pp 418–420 30. Radchenko EV, Palyulin VA, Zefirov NS (2017) Advanced approaches to prediction of ADMET properties of drug compounds. In: 3rd Russian conference on medicinal chemistry, Kazan, Russia, September 28 – October 03, 2017, p 98 31. Schyman P, Liu R, Desai V, Wallqvist A (2017) vNN web server for ADMET predictions. Front Pharmacol 8:889. https://doi.org/10. 3389/fphar.2017.00889 32. Filimonov DA, Lagunin AA, Gloriozova TA, Rudik AV, Druzhilovskii DS, Pogodin PV,

485

Poroikov VV (2014) Prediction of the biological activity spectra of organic compounds using the PASS Online Web resource. Chem Heterocycl Comp 50:444–457. https:// doi.org/10.1007/s10593-014-1496-1 33. Dyabina AS, Radchenko EV, Palyulin VA, Zefirov NS (2016) Prediction of blood-brain barrier permeability of organic compounds. Dokl Biochem Biophys 470:371–374. h t t p s : // d o i . o r g / 1 0 . 1 1 3 4 / S1607672916050173 34. Radchenko EV, Dyabina AS, Palyulin VA, Zefirov NS (2016) Prediction of human intestinal absorption of drug compounds. Russ Chem Bull 65:576–580. https://doi.org/10. 1007/s11172-016-1340-0 35. Radchenko EV, Rulev YA, Safanyaev AY, Palyulin VA, Zefirov NS (2017) Computer-aided estimation of the hERG-mediated cardiotoxicity risk of potential drug components. Dokl Biochem Biophys 473:128–131. https://doi. org/10.1134/S1607672917020107 36. Katt ME, Shusta EV (2020) In vitro models of the blood-brain barrier: building in physiological complexity. Curr Opin Chem Eng 30:42– 52. https://doi.org/10.1016/j.coche.2020. 07.002 37. Mortelmans K, Zeiger E (2000) The Ames Salmonella/microsome mutagenicity assay. Mutat Res 455:29–60. https://doi.org/10. 1016/s0027-5107(00)00064-6 38. McCann J, Choi E, Yamasaki E, Ames BN (1975) Detection of carcinogens as mutagens in the Salmonella/microsome test: assay of 300 chemicals. Proc Natl Acad Sci U S A 72: 5135–5139. https://doi.org/10.1073/pnas. 72.12.5135

INDEX A Aβ fibril model ......................................53, 54, 57, 65, 66 Aβ oligomer model ......................................55–60, 62, 66 Aβ plaques ......................................... 14, 19, 29, 36, 128, 234, 285, 286, 291, 437 Aβ42 peptide ................................................................. 438 Aβ protofibrils ............................................................... 437 Ab-initio modeling............................................... 196, 199 Absorption, distribution, metabolism, excretion, and toxicity (ADMET) .......................82, 86, 142, 143, 219, 315, 319, 333, 338, 346, 465–469, 471–473, 482, 483 Acetylcholinesterase (AChE) ..........................6, 7, 20, 23, 35–37, 39, 79, 82–84, 88, 138, 188, 211–214, 218, 219, 232, 233, 235, 249–253, 259–261, 333, 335–340, 342, 344, 345, 348, 356, 436, 439, 441, 451, 452 Acetylcholinesterase inhibitor (AchEI) .................. 22, 23, 129, 232, 235, 249 Acitretin ................................................................ 436, 437 Activation loop ............................................ 133, 135, 168 Active site.....................................59, 76, 77, 79, 82, 105, 108, 110, 112, 114, 115, 139, 140, 145, 147, 153, 168–170, 173, 174, 179–182, 200, 202, 206, 208, 210, 217, 234, 235, 239, 246, 254–257, 327, 328, 330, 331, 336–340, 343, 344, 346–348, 365–367, 374–376, 384, 386, 387, 389–401 Adenosine triphosphate (ATP)........................... 129, 135, 138, 153, 166–168, 172, 178, 181, 182, 190, 312, 316, 321 ADME .................................. 79, 84, 139, 147, 150, 289, 319, 446, 468, 471 Adrenergic receptors ....................................................... 19 Aduhelm ...............................................20, 23, 40, 52, 75, 188, 408, 410 Adversarial drug development...................................... 409 Aggregation ...............................7, 15, 16, 18, 22, 28, 31, 33, 36, 40, 52, 53, 58, 61, 66, 68, 128–130, 165, 188, 191, 209, 233, 236, 249–251, 253, 284, 326, 327, 331, 408, 421, 423, 435, 441, 452 Allosteric sites ....................................................... 335, 340 AlzGPS........................................................................... 454

Alzheimer’s disease (AD)................................ 3–8, 12–33, 35–41, 52, 53, 68, 73–76, 79, 86, 88, 99, 100, 113, 119, 121, 127–130, 132, 133, 135, 136, 140, 141, 143, 146, 147, 153, 155, 158, 159, 165, 181, 188, 190–192, 195–197, 201, 208–222, 225, 231–236, 238, 249–253, 259, 267, 268, 279, 280, 282, 284–287, 289, 291, 295, 296, 319, 322, 325–327, 330, 331, 335, 338, 339, 341, 346, 347, 355–357, 359, 360, 364, 370, 376, 384, 385, 391, 393, 396, 401, 408, 409, 411, 427, 433–441, 451–455 AlzPlatform ................................................................... 453 Ames test ........................... 315, 468, 469, 471, 473, 480 Amyloid ................................. 4, 5, 12–14, 18–20, 30–32, 36, 37, 40, 52, 74, 88, 89, 99, 111, 128, 165, 233, 234, 250, 251, 253, 260, 284, 285, 327, 330, 355, 384, 388, 408–410, 423, 427, 434, 436–439, 441, 452 Amyloid beta (Aβ)..................................15, 20, 128, 130, 165, 232, 285–287, 289, 384 Amyloid cascade hypothesis ..........................52, 233, 384 Amyloid precursor protein (APP) ........................ 4, 8, 12, 13, 15, 16, 29, 33, 36, 37, 52, 74, 75, 99, 110, 111, 113, 128, 130, 192–194, 233–235, 327, 330, 331, 339, 384, 437, 439, 440 Amyloidosis .......................................................... 422, 423 Amyotrophic lateral sclerosis (ALS)................... 410, 413, 424, 427, 433 Angiopathic ................................................................... 423 Angiotensin II receptor blocker (ARB) ................. 38, 39, 233, 436, 439 Anti-Alzheimer agent................................... 76, 188, 191, 196, 220, 225, 349, 359, 379 Anti-amyloidogenic................................................ 79, 437 Anti-inflammatory............................. 129, 192, 193, 338, 356, 360, 364, 370, 437, 439 Anti-inflammatory.............................................. 21, 22, 29 Antiphospholipid.................................................. 424, 425 Apoptosis ..................................22, 34, 36, 134, 189, 212 Autoimmune ......................................216, 408–413, 418, 425, 426 Autoimmune, endocrine and metabolic (AEM) disorders.................................................... 411, 420 Axons ................................................................5, 129, 412

Kunal Roy (ed.), Computational Modeling of Drugs Against Alzheimer’s Disease, Neuromethods, vol. 203, https://doi.org/10.1007/978-1-0716-3311-3, © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Science+Business Media, LLC, part of Springer Nature 2023

487

COMPUTATIONAL MODELING OF DRUGS AGAINST ALZHEIMER’S DISEASE

488 Index B

BACE1 inhibitor peptide ............................................... 81 Basic Local Alignment Search Tool (BLAST) ............197, 238, 241 BERT algorithm............................................................ 416 Best subset selection (BSS)........................................... 287 β-amyloid aggregation .................................................... 35 β-secretase....................................4, 8, 74, 197, 234, 319, 327, 331, 339, 384 β-secretase 1 (BACE1)................................. 4, 28, 29, 35, 73–89, 99–121, 232, 234, 249–252, 259, 339, 340, 384, 437, 452 β-strand .......................................................................... 437 Betweenness ......................................................... 442, 443 Bexarotene (BEXA) ............................................. 435, 436 Big data.......................................................................... 440 Binding energy .....................................86, 100, 103, 106, 108–110, 136–139, 144, 145, 150, 153, 154, 202, 203, 207, 218, 242, 254, 257, 259, 268, 288–290, 316, 333, 335, 336, 338, 340–342, 362, 364 Biochemical dysregulations .......................................... 413 BIOVIA Discovery Studio.................................. 198, 201, 202, 207, 223 Blood–brain barrier (BBB) .................. 19, 21, 28, 30–32, 75, 81–83, 88, 100, 141, 209, 214, 215, 218, 284, 319, 333, 346, 389, 390, 392, 393, 434, 435, 438, 439, 453, 466, 468, 471, 472, 477, 478 Boolean ................................................................. 417, 420 Brain-derived neurotrophic factor (BDNF) .......... 15, 18, 189, 192, 193, 338 Bubble effect ................................................................. 415 Butyrylcholinesterase (BuChE).......................... 211, 252, 253, 335, 337–340, 342, 344, 345, 452

C Caco-2 permeability................................... 472, 478, 479, 482, 483 Calcium calmodulin-activated protein kinase II (CaM kinase II) ................................................. 129 Calcium-channel antagonist ................................ 436, 438 CAMP and cGMP...................................... 189, 190, 208, 357, 359, 378 CAMP response element binding protein (CREB) ..... 15, 18, 189, 190, 193, 214, 357, 360 Capped VQIVYK peptides .......................................53–68 Casein kinase (CK)..................................... 129, 134, 135, 154, 233 Catalytic domain ..................................76, 167, 168, 217, 220, 296, 328, 364, 374, 375, 377, 384 Cavity shaping loop ............................................. 327, 328 Cdc2-like kinase 1 (CLK-1) ...................... 168–172, 178, 181, 182

CDOCKER ................................................ 54, 58, 59, 68, 140, 210 Cell metabolism ............................................................ 166 Cell polarization ................................................... 411, 412 Central nervous system (CNS).................. 17, 18, 29, 39, 74, 75, 129, 130, 133, 165, 207, 221, 235, 284, 342, 357, 359, 360, 370, 373, 394, 421, 451 Cerebrovascular disease .................................................... 4 Chemical read-across .............................. 8, 196, 205–206 Chemotherapy...................................................... 437, 438 Cholinergic neurons ....................................................... 17 Cholinesterase inhibitors ...............................40, 129, 434 Chronic disease .................................................... 409–412 Chronic lung diseases ................................................... 426 C-Jun N-terminal kinase (JNK) ......................... 129, 133, 134, 146, 149, 356 Clinical trials .................................. 5, 7, 8, 21, 23–28, 35, 38–40, 52, 75, 88, 100, 101, 110, 175, 188, 191, 195–197, 219, 232, 233, 250, 251, 296, 347, 357, 370, 384, 438, 440, 452–454 ClustalW ...................................................... 197, 223, 397 ClustalX ................................................................ 197, 223 CMAP ................................................................... 451, 452 CMGC family ....................................................... 134, 168 CODES program .......................................................... 216 Cognitive abilities........................................ 4, 20, 22, 214 Cognitive decline ............................................13, 21, 325, 347, 408 Coincidence time window ............................................ 282 Comparative molecular field analysis (CoMFA) ........106, 120, 140, 143, 144, 148, 246, 263, 264, 365, 367, 368, 376, 377, 379 Comparative molecular similarity indices analysis (CoMSIA)....................................... 120, 143, 144, 148, 377 Computer-aided drug design (CADD) ..................76–79, 88, 89, 103, 172, 196, 207, 220, 234, 240, 386, 482 Conformational structure-based design ........................ 76 Convolutional networks ............................................... 454 Co-pathology ......................................420, 421, 425–427 Coumarin..........................8, 79, 81, 197, 249, 251, 336, 342, 343, 348 COVID-19 ............................................................. 12, 416 CpG islands ................................................................... 453 C-terminal .............................. 64, 66, 76, 130, 166, 167, 234–236, 288, 327, 328, 388 Cyclic adenosine monophosphate (cAMP) ......... 18, 136, 189, 190, 192, 193, 208, 209, 216, 356, 357, 359, 360, 373 Cyclic guanosine monophosphate (cGMP)................189, 190, 208, 214, 356, 357, 359, 364, 370, 371, 373 Cyclin-dependent protein kinase (CDK)....................129, 130, 134, 140

COMPUTATIONAL MODELING

OF

DRUGS AGAINST ALZHEIMER’S DISEASE Index 489

Cyclin-dependent protein kinase-5 (CDK-5) ............131, 132, 141–144, 168, 174–179, 181, 182, 233, 236, 259, 438 Cytochrome P450................................................ 471, 481 Cytokine communication networks ............................. 412

364–366, 371, 384–388, 390, 393, 397, 438, 445, 471 Epilepsy.......................................................................... 426 Epistemic AI ......................................................... 418, 419 ERRAT ................................................................. 198, 223

D

F

Dementia ...............................3–5, 12, 13, 15, 19, 21, 23, 40, 73, 74, 99, 127, 188, 195, 220, 232, 235, 267, 279, 280, 284, 285, 325, 355, 408, 410, 413, 424, 433, 438 Dementia with Lewy bodies (DLB).................... 408, 410 DFG sequence............................................................... 167 Diabetes ................................................29, 232, 421, 422, 425, 437 Disease-modifying agents (DMA) ............................... 121 Docking ....................................................... 181, 201, 289 Dopamine (DA) ....................................... 5, 19, 120, 233, 284, 326, 327 Double cross validation (DCV) ................................... 287 Drug-associated gene list (DGL) ................................. 452 DrugBank ............................................446, 451, 453, 476 Drug development ........................ 8, 100, 146, 197, 234, 259, 265, 268, 319, 385, 407, 409, 435, 441, 451, 454, 466 Drug-induced Gene Perturbation Signature Database (DGPSD)........................................... 451 Drug metabolism and pharmacokinetics (DMPK) ..... 435 Drug perturbation ........................................................ 452 Drug repositioning ........................... 300, 434, 435, 438, 440, 442, 447, 451–455 Drug Repositioning Perturbation Score/Class (DRPS/C) ......................................................... 451 Drug repurposing ...................................... 155, 253, 428, 435, 446, 453 Drug repurposing in AD (DRIAD)............................. 452 Dual-specificity Yak-related kinase (DYRK) ...............129, 134, 135, 296 Dynamic cross-correlation (DCC) maps ..................... 141 DYRK1A........................... 134, 135, 150, 152, 153, 158, 168–172, 174, 178, 181, 182, 296, 297, 300–302, 311, 312, 315, 316, 319, 321, 322 DYRK1A inhibitor ..................... 152, 155, 157, 295–322 Dyshomeostasis .................................................... 326, 452

FDA-approved drug...................................................... 434 FlexX ...........................................202, 223, 241, 242, 256 Fold recognition..................................196, 198, 240, 241 Food and Drug Administration (FDA) ..................20–23, 27, 40, 52, 76, 188, 191, 232, 284, 347, 434, 435, 437, 450, 452 Fragment-based de novo design (FBDND)...............196, 201, 202 Frontotemporal dementia (FTD) .............. 410, 424, 425

E

Half-life...........................................................28, 282–284 Harmine............................................. 153, 154, 172, 174, 335, 339, 340 HENA............................................................................ 454 Heterogeneous .............................................................. 454 Hidden Markov Models (HMM) ................................ 197 Hinge region .............................155, 156, 166, 168, 169, 172, 175, 177 Histamine (H3) receptor....................................... 35, 189 Histone deacetylase 6 (HDAC6) ..................16, 214, 215

Edges ........................................................... 246, 442, 443 Electronic health records (EHRs)................................ 440 Endocrinology............................................................... 412 Enzymes.................................. 4, 6, 7, 18, 20, 22, 31, 35, 39, 52, 59, 74, 78, 99, 100, 110, 112, 129, 138, 142, 145, 153–156, 166, 182, 189, 190, 208, 209, 214, 219, 220, 233–238, 250, 252, 253, 261, 326, 327, 335, 339, 340, 342, 346, 359,

G γ-secretase ...........................................4, 21, 52, 232–235, 319, 327, 331, 384 Gatekeeper residues ............................................. 172, 344 Gene-centric network ................................................... 454 Gene Expression Omnibus (GEO) ............ 445, 448, 451 Genotypes...................................................................... 440 Glide quantum mechanics-polarized ligand docking (QPLD) ...................................392, 398, 399, 401 Glucagon-like peptide-1 (GLP-1)....................... 436, 437 Glutamate .......................................................5–7, 18, 384 Glutaminyl cyclase...................... 384–387, 389, 390, 397 Glycogen synthase kinase (GSK)....................33, 39, 129, 136–140 Glycogen synthase kinase 3 beta (GSK-3β)..............7, 35, 37, 39, 40, 88, 131, 136–140, 157, 168, 174–178, 180–182, 189, 233, 236, 250, 259, 335, 338, 339, 452 Google ................................................416, 417, 419–421, 423–425 Google profiles ..................................................... 417, 421 Google Scholar...................................409, 417–421, 423, 424, 426, 451 GROMACS 5.1.2 ................................................ 204, 223 GWAS .......................................................... 440, 453, 454

H

COMPUTATIONAL MODELING OF DRUGS AGAINST ALZHEIMER’S DISEASE

490 Index

HMG-CoA reductase inhibitors .................................... 22 Homology modeling .........................136, 196–198, 238, 241, 359, 386, 387 Hotspots ...........................................................54, 56, 256 HTDocking ................................................................... 453 Hub................................................................................ 443 Human ether-a-go-go-related gene (hERG) .... 393, 468, 469, 471, 472, 477, 478, 482, 483 Human intestinal absorption (HIA) .................. 319, 393, 468, 471, 472, 478–480, 482, 483 Huntington’s disease (HD).......................................... 433 Hydrogen bond acceptor (HBA)....................... 104, 108, 109, 142, 147, 156, 206, 289, 388, 394 Hydrogen bond donor (HBD) ................. 104, 108, 109, 119, 142, 144, 145, 147, 156, 206, 388, 473 Hydrophobic (HYP) ....................... 31, 57, 59, 108, 109, 115, 119, 139, 142, 144–146, 148–151, 155, 157, 158, 175, 182, 206, 208, 211, 213, 218, 219, 234, 235, 246, 258, 288, 289, 313, 314, 316, 321, 340, 344, 349, 359, 361–364, 367, 370, 371, 373, 374, 377, 379, 384, 387, 388, 390, 392, 394, 399, 437 Hydrophobicity ...................................176, 181, 182, 246 Hydrophobic contact.............................57, 64, 210, 335, 339, 340, 342, 344 Hydrophobic region ........................................... 109, 328, 349, 388 Hydrophobic zone ........................................................ 166 5-Hydroxytryptamine (5-HT) receptors ............... 17, 35, 38, 39, 188, 326, 327 Hypercholesterolemia .......................................... 423, 425 Hyperphosphorylation............................. 22, 36, 74, 128, 130, 134, 136, 141, 166, 168, 169, 181, 182, 251, 330

I Induced fit docking (IFD).......................... 136, 137, 139 Information science ............................................. 408, 427 In silico ..................................68, 79, 81–84, 86, 88, 136, 145, 146, 153, 154, 156, 158, 190, 199, 205, 207, 208, 220, 237, 238, 252–254, 268, 280, 290, 291, 296, 300, 315, 319, 364, 389, 394, 395, 400, 434, 466, 474, 476, 478, 483 In silico modeling ................................... 8, 197, 208–220 Insulin resistance .................................................. 421, 422 Interleukin-1 ................................................................... 14 Irritable bowel syndrome (IBS) ................................... 426 I-TASSER ............................................198, 199, 223, 241

K KEGG .......................................................... 446, 448, 452 Kinase...................................38, 129–136, 139–141, 145, 146, 153, 155, 156, 158, 159, 166, 168, 169, 172, 174, 175, 178, 181–183, 233, 235, 251, 296, 357

L Ligand-based drug design (LBDD)................... 196, 203, 244–248 LigandScout ........................................201, 207, 223, 246 LigBuilder V3................................................................ 202 Line of response (LOR)................................................ 282 Lipophilicity ..............................262, 466, 469, 474, 476, 480, 482, 483 Lipopolysaccharide dysregulation ................................ 427 Literature query ............................................................ 416 Lupus ............................................................................. 411 Low-density lipoprotein (LDL) receptor .................... 423

M Machine learning (ML) .........................84, 86, 136, 137, 147, 205, 253, 258, 262–265, 281, 298, 310, 409, 452, 466, 473 Maestro Version 10.7.014 ............................................ 201 Many-to-many............................................................... 441 MAO-B selectivity......................................................... 343 MAO inhibitors.......................................... 250, 330, 338, 347–349 MAO isoforms...................................................... 327, 349 MD simulation ......................... xi, 55, 57, 61–67, 76–79, 81–83, 88, 106, 107, 110, 112, 113, 138–140, 142, 143, 150, 153, 196, 202–204, 210, 221, 222, 243, 250, 253, 260, 268, 287, 289, 290, 333, 335–339, 341, 344, 345, 348, 359, 365, 373, 377, 395, 397, 399, 400, 437 MedlinePlus.......................................................... 417, 418 MedlinePlus Subject Header (MeSH) ................... 63, 65, 113, 263, 419, 422–424, 426 Medusa .......................................................................... 200 Mendelian disorders...................................................... 445 Metabolic disorders....................................................... 408 Metabolomics .............................................. 416, 445, 451 Metadynamics simulation .................................... 287, 288 7-Methoxytacrine derivative.....................................22–23 Microtubule-associated protein (MAP)................. 5, 129, 134, 145, 146, 165 Microtubules ...............................5, 7, 13, 129, 130, 135, 166, 232, 236, 295, 452 Mitogen-activated protein kinase (MAPK) ................129, 132–134, 144–147 MMTK software................................................... 204, 223 Modified mechanism of action (MoA) ........................ 442 Module ............................... 55–58, 60, 62, 65, 133, 137, 138, 140, 180, 181, 204, 224, 312, 315, 319, 443, 472 Molecular docking ................................. 8, 53–55, 57, 58, 66, 68, 76, 78, 79, 81–84, 86, 88, 100, 103, 107–109, 113, 120, 136–145, 147–151, 153–156, 158, 178, 196, 201, 202, 207, 208, 210–215, 217–221, 240–242, 250, 252–255, 258–261, 268, 281, 287–291, 300, 301, 312,

COMPUTATIONAL MODELING

OF

DRUGS AGAINST ALZHEIMER’S DISEASE Index 491

313, 315, 316, 321, 322, 333, 335–346, 348, 349, 357, 361, 362, 364, 365, 370, 371, 373, 376, 379, 389, 391, 392, 445, 452 Molecular dynamics (MD) ...........................8, 25, 53–55, 57, 62–68, 76–79, 81–83, 86, 88, 106, 107, 109, 110, 112–114, 136, 138–143, 150, 153, 156, 158, 196, 203, 204, 209, 210, 217, 219–221, 238, 243, 244, 250, 254, 256, 260, 268, 281, 287, 289–291, 322, 333, 335–339, 341, 344, 345, 347, 348, 359, 361, 362, 365, 373, 376, 377, 386, 387, 392, 395, 397, 399, 400, 437 Molecular Mechanics-Generalized Born surface area (MM-GBSA)................................... 78, 79, 81, 83, 136, 203, 204, 290, 333, 394 Molecular Mechanics Poisson-Boltzmann Surface Area (MM-PBSA)............................. 79, 103, 150, 154, 203, 204, 210, 341 Molecular Operating Environment (MOE) ...............144, 201, 207, 241, 246, 263, 311, 346, 397, 398 Monoamine oxidase (MAO) .................35, 39, 129, 250, 251, 326–328, 330–332, 334–336, 338–340, 342–349, 451 Multiple sclerosis (MS) .........................14, 216, 426, 427 Multi-target-directed ligands (MTDLs) .............. 35, 336, 342, 345, 346 Muscarinic receptors .................................................17, 40

NMDA receptors ...................................7, 18, 20, 35, 38, 129, 233, 434 N-methyl-D-aspartate (NMDA) ........................ 6, 18, 39, 99, 129, 135, 188, 232, 335, 338, 339, 356, 452 Non Proline Directed Protein Kinase (Non-PDPK) ............................................ 129, 131 N-terminal ..................................... 15, 57, 60, 64–66, 76, 130, 135, 166, 234, 235, 384, 389 NuBBE database ........................297, 315, 319, 321, 322

N

P

NAMD 2.13 ......................................................... 204, 224 Nanomaterials .................................................... 28–29, 40 Natural compounds ................................... 29, 35, 36, 79, 86, 142, 143, 153, 315, 339 Natural language processing ........................................ 419 NAViGaTOR................................................................. 200 NETTAG ....................................................................... 453 Network pharmacology ............................. 434, 441–445, 450, 454, 455 Neurodegeneration ................................6, 15, 35, 37, 74, 128, 130, 134–136, 189, 327, 330, 355, 357, 384, 408–411, 413, 427 Neurodegenerative cascades ....................... 410, 413, 427 Neurodegenerative diseases ............................22, 89, 129, 147, 154, 165, 166, 279, 284, 340, 355, 357, 418, 427, 428, 433 Neurofibrillary tangle (NFT) .......................4, 12–14, 18, 19, 22, 74, 127–130, 132, 133, 135, 138, 165, 188, 232, 233, 251, 279, 295, 326, 384, 434, 441 Neuromedical .............................................. 408, 410, 414 Neuron ...................................... 4, 13–15, 17, 19, 20, 34, 36, 37, 74, 129, 130, 150, 166, 167, 190, 265, 327, 355, 356, 411, 434, 435, 437, 438, 452 Neuropathology .................... 4, 158, 408, 409, 413, 439 Neurophysiology ........................................................... 412 Nicotinic receptors.......................................................... 17

Paired helical filaments (PHFs) ................... 53, 128, 130, 166, 235 PanDDA ............................................................... 201, 202 Parkinson’s disease (PD) ......................... 4, 19, 159, 191, 195, 196, 284, 285, 325, 341, 347, 433 Pathogenesis ................................5, 6, 15, 17, 41, 74, 88, 99, 130, 132, 133, 191, 232, 250, 358, 451 Pathological aggregates ................................................ 411 Patterned retention ....................................................... 411 PDEs inhibitors ............................................................. 357 Pharmacophore mapping ............................ 76, 136, 207, 250, 268, 281, 447 Pharmacophore modeling ....................... 8, 79, 142, 145, 158, 196, 200, 201, 206–207, 246, 247, 256, 259, 261, 291 PharmaGIST......................................................... 207, 224 Phosphatases................................................ 130, 166, 233 Phosphate ............................................129, 135, 168, 425 Phosphodiesterase (PDE).................... 35, 188–192, 195, 196, 208–210, 214, 216, 219–221, 225, 356–360, 364, 370, 373–375, 379 Phospholipids .................................................31, 425, 426 Phosphorylation ............................ 7, 17, 33, 36, 37, 129, 130, 132–136, 150, 155, 158, 166, 168, 182, 190, 192–194, 214, 232, 235, 236, 296, 357, 438

O Off-target...................................... 86, 438, 445, 446, 450 Omics........................ 410, 434, 440, 445, 448, 451, 453 One disease.................................................................... 441 One drug ...................................................................6, 441 One target ............................................................ 249, 441 Online Mendelian Inheritance in Man (OMIM) .......445, 448, 451 Online services .............................................................. 467 On-target ................................................................ 39, 445 Organization for Economic Co-operation and Development (OECD) .......................84, 89, 115, 205, 287, 299, 396 Oxidative stress.............................. 14–16, 20, 28, 31, 36, 37, 74, 128, 133, 249–251, 326, 327, 330, 355, 356, 441, 452

COMPUTATIONAL MODELING OF DRUGS AGAINST ALZHEIMER’S DISEASE

492 Index

PMC...................................................................... 417, 418 Polypharmacology....................................... 438, 441, 453 Positron emission tomography (PET) ............... 279–291, 337, 344 Presenilin 1-2 (PS 1-2) ...................................15, 16, 128, 129, 235, 384 Principal component analysis (PCA).................... 81, 111, 114, 141, 147 PROCHECK............................................... 198, 224, 241 Progranulin........................................................... 424–426 Proline-directed protein kinase (PDPK)............. 129–134 Propargylamine moiety ....................................... 338, 342, 343, 346, 348 Protein database (PDB) ........................... 68, 76, 77, 137, 138, 142, 145, 148, 150, 153, 155, 172, 178, 181, 197, 199, 201, 210, 219, 220, 255, 300, 302, 312–315, 327, 328, 333, 338–346, 364–366, 370, 373, 375–377 Protein kinase A (PKA).............................. 129, 134, 136, 153, 190, 193, 233, 236 Protein misfolding ...................................... 408, 411, 421 Protein–protein interactions network ........ 196, 200, 239 Proteomics.................................................. 410, 411, 440, 445, 451 PSIPRED server................................................... 198, 241 PubChem.....................................84, 120, 207, 224, 446, 469, 476 PubMed ...................................... 409, 417–426, 446, 451 Pyroglutamate Aβ (pGlu-Aβ) .............384, 385, 388, 396

Q Quantitative structure-activity relationship (QSAR) ..... 8, 76, 79, 89, 103, 106, 115, 118, 120, 121, 136, 144, 145, 147–149, 153, 159, 196, 203, 205, 220, 224, 247, 248, 260–267, 281, 286, 287, 289, 291, 297, 298, 301, 302, 311, 312, 315, 321, 322, 331, 359, 364, 368, 386, 395, 396, 400, 401, 466, 473 Quantum chemical interaction..................................... 392 Query ................................ 113, 141, 197, 199, 207, 240, 246, 409, 413–427 Query builder ................................................................ 420

R Radionuclides .............................................. 280, 282, 283 Radiotracers .......................................................... 283, 284 Radius of gyration (Rg) ................................................ 204 Ramachandran plots...................................................... 198 Reprofiling..................................................................... 435 ResearchGate ........................................................ 418, 419 Re-tasking...................................................................... 435 Retinoic acid......................................................... 435–437 Reverse disease profile................................................... 445

Rheumatoid arthritis................................... 424, 425, 438 RNA-seq ........................................................................ 451 Robetta meta-server...................................................... 198 Root mean square deviation (RMSD) ................... 63, 65, 66, 108, 111, 139, 140, 142, 143, 146, 150, 153, 154, 181, 204, 219, 244, 245, 255, 258, 313, 345 Root mean square fluctuation (RMSF) ..................63–66, 110, 139–141, 143, 146, 153, 156, 204, 244, 245, 333, 335, 339, 365

S Safinamide ......................... 333, 335, 337, 344, 347, 348 Schro¨dinger-Glide......................................................... 202 Seed node ...................................................................... 443 Serotonin ............................. 5, 17, 18, 21, 221, 326, 333 Shortest paths................................................................ 443 Similarity search..........................196, 207, 246, 247, 256 Single photon emission computed tomography (SPECT) ................................................... 279–291 Solubility...............................................29, 214, 468, 469, 471–474, 480–483 Solvent accessible surface area (SASA)................... 54, 60, 61, 139, 150, 153 Structure-activity relationship (SAR) ................... 79, 115, 136, 207, 211–213, 217, 251, 297–299, 301, 302, 310, 315, 321, 322, 348, 387–391, 393, 394, 397, 401, 416 Structure-based drug design (SBDD) ............... 196, 197, 202, 204, 211, 216, 237–244, 386 Synapse database ........................................................... 451 Synaptic potentiation .................................................... 412 Synaptic pruning .................................................. 408, 427 Systems-level polypharmacology .................................. 441

T Target........................................................... 169, 188, 280 TargetHunter ....................................................... 448, 453 Target-specific medicines (TSMs) .................................. 35 Tau .......................................... 4, 5, 7, 13, 16, 31, 33, 36, 37, 74, 128, 130, 133–136, 155, 158, 165–168, 181, 188, 190, 192–194, 232, 235, 236, 251, 285, 287–290, 295, 296, 330, 337, 344, 422, 427, 438 Tau hyperphosphorylation .......................... 5, 8, 30, 130, 135, 138, 193, 197, 441 Tau kinases..................................................................... 130 Tau protein ............................5, 13, 22, 30, 53, 128–130, 135, 141, 165, 188, 232, 233, 284, 287, 296, 452 Tetracycline antibiotic.......................................... 436, 438 Text query ..................................................................... 413 The Cancer Genome Atlas (TCGA) ............................ 451 Threading methods....................................................... 196 TOUCHSTONE-II...................................................... 199

COMPUTATIONAL MODELING Transcriptomics ............................................................. 445 Tumor necrosis factor (TNF)................................ 14, 189 Type 3 diabetes ............................................................. 437 Tyrosine protein kinase (TPK) ............................ 131, 436

V Van der waals energy..................................................... 364 Vascular .............................. 5, 12, 28, 195, 280, 285, 423

OF

DRUGS AGAINST ALZHEIMER’S DISEASE Index 493

Verify3D ............................................................... 198, 224 Virtual screening ....................... 78, 79, 84, 88, 100–103, 107, 109, 113, 114, 119, 139, 141, 147, 148, 150, 152–154, 156, 202, 206, 207, 217, 219, 220, 237, 250, 258, 259, 268, 289, 313, 333, 348, 362, 376, 386, 394, 399–401, 445 Vitamin A ...................................................................... 435 VMD....................................................................... 77, 204