Complete accounts of integrated drug discovery and development. Volume 2: recent examples from the pharmaceutical industry 9780841234314, 0841234310

Citation preview

1332

STORIES OF DRUGS FROM THE BENCH These engaging accounts walk readers through the drug discovery and development processes, from identification of a target compound to the development of large-scale processes. The authors hail from leading pharmaceutical companies, grounding the text in real-world applications. These accounts also touch on reaction safety and development costs, providing insight into often closed-door procedures. These relevant examples from industry are informative to chemists in both industry and academia, especially those interested in discovering and developing drug candidates.

PUBLISHED BY THE

American Chemical Society SPONSORED BY THE

ACS Division of Organic Chemistry

B I O L O G I C A L

VOLUME 1332

COMPLETE ACCOUNTS OF INTEGRATED DRUG DISCOVERY AND DEVELOPMENT RECENT EXAMPLES FROM THE PHARMACEUTICAL INDUSTRY VOLUME 2

ACS SYMPOSIUM SERIES

ACS SYMPOSIUM SERIES

COMPLETE ACCOUNTS OF INTEGRATED DRUG DISCOVERY AND DEVELOPMENT RECENT EXAMPLES FROM THE PHARMACEUTICAL INDUSTRY VOLUME 2

PESTI et al.

PESTI, ABDEL-MAGID & VAIDYANATHAN

Complete Accounts of Integrated Drug Discovery and Development: Recent Examples from the Pharmaceutical Industry Volume 2

ACS SYMPOSIUM SERIES 1332

Complete Accounts of Integrated Drug Discovery and Development: Recent Examples from the Pharmaceutical Industry Volume 2 Jaan A. Pesti, Editor EnginZyme AB, Stockholm, Sweden

Ahmed F. Abdel-Magid, Editor Therachem Research Medilab, LLC, Chelsea, Alabama, United States

Rajappa Vaidyanathan, Editor Bristol-Myers Squibb, Bangalore, India

Sponsored by the ACS Division of Organic Chemistry

American Chemical Society, Washington, DC

Library of Congress Cataloging-in-Publication Data Library of Congress Cataloging in Publication Control Number: 2019033049

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences—Permanence of Paper for Printed Library Materials, ANSI Z39.48n1984. Copyright © 2019 American Chemical Society All Rights Reserved. Reprographic copying beyond that permitted by Sections 107 or 108 of the U.S. Copyright Act is allowed for internal use only, provided that a per-chapter fee of $40.25 plus $0.75 per page is paid to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. Republication or reproduction for sale of pages in this book is permitted only under license from ACS. Direct these and other permission requests to ACS Copyright Office, Publications Division, 1155 16th Street, N.W., Washington, DC 20036. The citation of trade names and/or names of manufacturers in this publication is not to be construed as an endorsement or as approval by ACS of the commercial products or services referenced herein; nor should the mere reference herein to any drawing, specification, chemical process, or other data be regarded as a license or as a conveyance of any right or permission to the holder, reader, or any other person or corporation, to manufacture, reproduce, use, or sell any patented invention or copyrighted work that may in any way be related thereto. Registered names, trademarks, etc., used in this publication, even without specific indication thereof, are not to be considered unprotected by law. PRINTED IN THE UNITED STATES OF AMERICA

Foreword The purpose of the series is to publish timely, comprehensive books developed from the ACS sponsored symposia based on current scientific research. Occasionally, books are developed from symposia sponsored by other organizations when the topic is of keen interest to the chemistry audience. Before a book proposal is accepted, the proposed table of contents is reviewed for appropriate and comprehensive coverage and for interest to the audience. Some papers may be excluded to better focus the book; others may be added to provide comprehensiveness. When appropriate, overview or introductory chapters are added. Drafts of chapters are peer-reviewed prior to final acceptance or rejection. As a rule, only original research papers and original review papers are included in the volumes. Verbatim reproductions of previous published papers are not accepted. ACS Books Department

Contents Foreword .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

ix

Preface .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

xi

1. Synthetic Routes for Venetoclax at Different Stages of Development .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  Yi-Yin Ku and Michael D. Wendt

1

2. Discovery and Development of Lorlatinib: A Macrocyclic Inhibitor of EML4-ALK for the Treatment of NSCLC .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  27 Robert Dugger, Bryan Li, and Paul Richardson 3. From Discovery to Market Readiness: The Research and Development of the βSparing Phosphatidylinositol 3-Kinase Inhibitor Taselisib .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  61 Rémy Angelaud, Steve Staben, Timothy Heffron, Andreas Schuster, and Frédéric St-Jean 4. Discovery and Development of the First Antibody–Antibiotic Conjugate LinkerDrug.  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  85 Stefan G. Koenig and Thomas H. Pillow 5. The Discovery of the Nav1.7 Inhibitor GDC-0276 and Development of an Efficient Large-Scale Synthesis .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  107 Andreas Stumpf, Daniel Sutherlin, Christoph M. Dehnhardt, and Rémy Angelaud 6. Discovery and Development of AMG 333: A TRPM8 Antagonist for Migraine.  . . . . . . . . . . . .  125 Neil F. Langille and Daniel B. Horne 7. The Discovery and Chemical Development of PF-06273340: A Potent, Selective, and Peripherally Restricted Pan-Trk Inhibitor for Pain .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  155 David C. Blakemore, Thomas Brandt, Craig Knight, and Sarah E. Skerratt 8. Optimization of an Azaindazole Series of CCR1 Antagonists and Development of a Semicontinuous-Flow Synthesis .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  185 Christian Harcken, Joshuaine Grant, Hossein Razavi, Maurice A. Marsini, Frederic G. Buono, Jon C. Lorenz, and Jonathan T. Reeves 9. Discovery and Development of Non-Covalent, Reversible Bruton’s Tyrosine Kinase Inhibitor Fenebrutinib (GDC-0853) .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  239 James J. Crawford and Haiming Zhang 10. Discovery and Early Development of Small Molecule Proprotein Convertase Subtilisin/Kexin Type 9 (PCSK9) Inhibitors.  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  267 David W. Piotrowski and Emma L. McInturff vii

11. Rational Design to Large-Scale Synthesis: Development of GSK8175 for the Treatment of Hepatitis C Virus Infection .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  297 Andrew J. Peat and Shiping Xie Editors’ Biographies .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  323 Indexes Author Index .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  327 Subject Index .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  329

viii

Foreword For every drug that is sold on the market, there will be dozens of other drugs that failed somewhere in development, most often as the benefit/risk was simply not good enough to warrant commercialization. Every drug, but also all those which do not crown their development with commercialization, have a story. It starts with a biological hypothesis on the cause for a disease, the development of biological assays, and after the identification of the first hits, the development of these to leads. Then comes the further evolution of the leads to candidates and then finally their progression through the many phases of the development. And when all obstacles are overcome, ultimately we attain commercialization to serve the patient. Every step of the way is complex and involves the interplay of many disciplines, making the discovery and development of a new drug into the biggest and most challenging “science experiment” one can imagine. Each phase of the genesis of a new drug requires the combination of significant scientific insight with the ability to find the creative solutions to the problems and challenges that threaten to derail the process at every step of the way. Many scientific disciplines are involved, but it is the synthetic chemist that is the essential factor. A highly remarkable and insightful description of the uniqueness of chemistry amount the sciences was already suggested in the 19th century by the French scientist Marcellin Berthelot: “La chimie crée son objet. Cette faculté créatrice, semblable à celle de l’art lui-même, la distingue essentiellement des sciences naturelles et historiques.” (“Chemistry creates its object of study. Such a creative power is analogous to the power of art; it essentially distinguishes chemistry from natural and historical sciences.”) Marcelin Berthelot, La synthèse chimique Alcan, Paris, 1887 It clearly describes the central role that we play as chemists. We are the scientists that create. We are not just descriptive and find ourselves with an insurmountable problem, but we have the means to design the solution. This is the same creative process that gives us art, but we chemists are providing a beautifully conceived and prepared organic molecule. The beauty of the creative solution is a molecule with the right properties which ultimately provide the indispensable tool that cures a disease. And the complexity of the creative task, from initially designing the molecule to actually being able to make it safely and economically in large amounts, is covered and described for several examples in this book. The topics encompass drugs that are covering the whole gamut of diseases that inflict mankind: they reach from infectious diseases, systemic disorders, over pain and CNS diseases to various metabolic malfunctions. The long list of diseases clearly demonstrates the strong need for better medicines, created and made by organic chemists.

ix

Anybody reading the book will realize the central role that organic chemistry plays in the discovery and development of new drugs. As organic chemists we should be proud of the central role of our science and hope that that the message of Berthelot from 1887 is being recognized again. Kai Rossen, Editor Organic Process Research & Development

x

Preface

As for yin and yang, the relationship between Drug Discovery and Process Chemistry in the pharmaceutical industry is both interdependent and complementary. Neither can produce viable pharmaceuticals in reasonable times and quantities without the other. Discovery Chemists produce small quantities of identified lead drug candidates. While these quantities are enough for early testing and selection, the development process would remain impractical without enough substance to first test the concept in human clinical trials and eventually treat the patients who need them worldwide. The skills of Process Chemists would not be of use with no guidance as to what to make. The fusion of these two major functions transfers the drug discovery from theory to application and reality; it helps produce nearly all the pharmaceutical entities known today. The eventual production or screening of candidates for a pharmaceutical activity begins with the Discovery Chemists. Organic synthesis has progressed dramatically since the pioneering work of Friedrich Wöhler and the synthesis of the first organic molecule: urea. Thanks to innovation both in academia and industry, today’s organic chemists have ever-growing arrays of reactions, reagents, catalysts and techniques at their disposal to handle the synthesis of increasingly complex drug candidates. The skills of organic chemists in drug discovery and development have contributed significantly to the advancement of the pharmaceutical industry and have helped introduce new effective life-preserving medicines. Indeed, most drugs found in the modern pharmacy arise from synthetic origins, entirely or partially. Nearly all drugs that populate the contemporary pharmacopeia required the skills of Discovery Chemists to prepare the first few grams. When a drug candidate displays efficacy and shows a promising improvement over the existing therapies, the expertise of process chemists is brought to bear to render the synthesis safe, scalable, and economical. A good example is Gilead’s Harvoni, a drug that comprises two discrete active molecules (ledipasvir and sofosbuvir) with different mechanisms of action used to treat hepatitis C. Gilead’s scientists expended much effort to discover the efficacy of this paired drug, as well as to devise reasonable means to develop a scalable synthesis. As a result of these intense efforts, a serious previously uncontrolled disease is now curable. The advent of advanced, ubiquitous computational tools to identify active sites and binding principles has led to the evolution of structure-based rational drug design concepts in Discovery Chemistry. Very exciting is the emergence of CRISPR technology – a number of companies have been established that may exploit this new knowledge to produce the next generation of drugs. Discovery Chemistry has as its raison d'être: the identification and synthesis of new drugs that improve upon existing therapies, and treat unmet medical needs. While the initial synthesis developed by the Discovery Chemists may be capable of producing hundreds of grams of a drug xi

candidate with minimal tweaks, it is generally not designed to be the most synthetically efficient or economically viable approach. Similarly, Process Chemists have been at the forefront of innovating new technologies to efficiently assemble complex molecules, and have utilized advanced tools to gain a better appreciation of the ‘reaction space’ around a particular chemical transformation. The use of advanced computing and automated reaction arrays permits the examination of a significantly wider scope of reaction conditions. An interesting manifestation of this ability is the understanding of ‘cliffs’ – areas in the reaction space where a small change in a parameter (e.g., temperature) will cause a large change in the yield or purity. Needless to say, a robust process would not exist near such a cliff where a small inadvertent change might result in an unfortunate loss of yield or purity! Process Chemistry’s central role is to produce high quality drug substances at commercially acceptable costs on multi-ton scales in an environmentally responsible manner. The chapters in these books (Volume 1 came out in 2018) are written to accentuate the interdependency and synergy between drug discovery and process development disciplines to advance a new chemical entity into clinical trials and eventually to the market. Due to the success of our previous books (Comprehensive Accounts of Pharmaceutical Research and Development: From Discovery to Late-Stage Process Development, Volume 1 and Volume 2), we sought to further arm experienced modern synthetic organic chemists, and budding researchers perhaps still in universities with real world examples from the pharmaceutical industry. In addition, the chapters contain citations of a large number of valuable selected references to the primary literature. The book highlights the tireless efforts of Discovery and Process Chemists, and their roles in the advancement of drug discovery and development. We were motivated to create this book by our appreciation of the value of chemical research by both Discovery and Process Chemists in producing new pharmaceutical entities. Their combined efforts make it possible to introduce novel and effective drugs into the market to treat hundreds of millions of patients and alleviate their suffering, improve their quality of life and possibly save their lives from diseases and disorders. The chapters presented in this book are written by a selected group of outstanding, highly accomplished Medicinal and Process Chemists with noted experiences and diverse backgrounds representing some of the top pharmaceutical companies. The chapters highlight examples of emerging concepts, new developments and challenges arising in the discovery of new drug candidates and the development of new practical synthetic chemistry processes to produce these drug candidates on large scale. The story leading up to the discovery of each drug or drug candidate is presented by the Discovery Chemist(s), and then the Process Chemist(s) describe the development of the same drug to give the reader a complete story of drug discovery and development. The reader will experience a rare and unique opportunity to obtain the complete perspectives of Discovery and Process Chemistry in a single book. In this volume, we are incorporating chapters on different therapeutic areas for the treatment of many diseases and disorders including pain, migraine, cancer, inflammation, autoimmune arthritis, LDL regulation, bacterial infections and hepatitis C. While most of these topics have appeared in the primary literature where space is critical and brevity valued, the book setting permits us to tell the complete tales as stories, from start to finish or to the current state of the drug development. We aimed to increase the value of this book by imposing fewer limits on relevant details. Our special thanks to all the authors who are acknowledged in the chapters listed below. All have volunteered their efforts and time to sculpt this book. Their willingness to contribute, the demanding work expended in the writing, and the result at the finish bear witness to their outstanding

xii

contributions to this book. Some made this book their second experience with us editors. We cannot say enough about them. Yi-Yin Ku and Michael D. Wendt of AbbVie Process Chemistry and Oncology Medicinal Chemistry, respectively, both of Global Pharmaceutical Research and Development, discuss the discovery and development of venetoclax, a selective Bcl-2 inhibitor approved for the treatment of chronic lymphocytic leukemia as well being examined for other forms of leukemia (Chapter 1). The authors narrate their work to identify selective Bcl-2 inhibitors that culminated in the discovery of venetoclax. They then described the multiple improvements in the process that led to the first largescale synthesis of the drug. Robert Dugger and Bryan Li of Pfizer’s Process Development, Medicinal Sciences, Chemical Research and Development and Paul Richardson of Oncology Medicinal Chemistry, Medicine Design relate the discovery and large-scale synthesis of lorlatinib, a selective EML4-ALK inhibitor for the treatment of non-small cell lung cancers (Chapter 2). This was approved for general use in late 2018 and has demonstrated a degree of efficacy on brain tumors as well. Rémy Angelaud and Frederic St-Jean of Genentech’s Department of Small Molecule Process Chemistry, Steve Staben and Timothy Heffron of Genentech’s Department of Discovery Chemistry, and Andreas Schuster of Hoffmann-La Roche Ltd. Small Molecules Technical Development relate the discovery and development of taselisib, an inhibitor of PI3Kα, δ, and γ (Chapter 3). This drug inhibits the PI3K signaling pathway as an approach to cancer treatments. Stefan G. Koenig of Genentech’s Small Molecule Process Chemistry and Thomas H. Pillow of Discovery Chemistry report a novel and robust process to a complex linker–antibiotic (Chapter 4). Antibody conjugate therapeutics represent a sophisticated treatment method to deliver potent small molecules to the desired site of action. The chapter on the antibody-antibiotic conjugate linker-drug will describe the medicinal chemistry to a complex quaternary salt to facilitate the potent conjugate against intracellular S. aureus. Daniel Sutherlin of Genentech’s Department of Discovery Chemistry, Christoph M. Dehnhardt of Xenon Pharmaceuticals Medicinal Chemistry, and Andreas Stumpf and Rémy Angelaud of Genentech’s Department of Small Molecule Process Chemistry reveal the compound GDC-0276 (Chapter 5). This is a potent inhibitor of the channel Nav1.7 that is an important mediator of pain signals and has emerged as a promising therapeutic target for the treatment of chronic pain. Neil F. Langille and Daniel B. Horne of Amgen’s Pivotal Drug Substance Synthetic Technologies and Discovery Research, respectively, present research conducted on AMG 333, a potent inhibitor of the TRPMS ion channel (Chapter 6). This is a potential therapeutic for migraine pain. Multikilogram quantities of the drug were eventually made by adapting the process to exclude chromatography. David C. Blakemore of Pfizer’s Medicine Design, Thomas Brandt of Pfizer’s Chemical Research and Development, Craig Knight of Pfizer UK Research and Development and Sarah E. Skerratt of Vertex’s Department of Medicinal Chemistry report the discovery and chemical development of PF-06273340, a potent, selective and peripherally restricted pan-Trk inhibitor for the treatment of pain (Chapter 7). This drug has demonstrated anti-hyperalgesic effects pre-clinically and in humanevoked pain models in the clinic. Christian Harcken of Boehringer Ingelheim’s R&D Project Management, Joshuaine Grant of Translational Medicine and Clinical Pharmacology, Hossein Razavi of Medicinal Chemistry, and Maurice A. Marsini, Frederic G. Buono, Jon C. Lorenz and Jonathan T. Reeves of Chemical Development discuss a novel series of azaindazole CCR1 antagonists for the treatment of xiii

inflammatory and autoimmune disease (Chapter 8). CCR1 is involved in the trafficking of immune cells to sites of inflammation. James J. Crawford of Genentech’s Department of Discovery Chemistry and Haiming Zhang of the Department of Small Molecule Process Chemistry present the discovery and multihundred kilogram preparation of fenebrutinib, a selective, non-covalent, reversible Bruton’s Tyrosine Kinase (Btk) inhibitor. This therapy is for the treatment of autoimmune diseases such as rheumatoid arthritis and systemic lupus erythematosus (Chapter 9). David W. Piotrowski of Pfizer’s Medicinal Chemistry and Emma L. McInturff of Chemical Research and Development describe the discovery and early development of small molecule proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors (Chapter 10). These inhibitors can potentially provide a newer option for regulating the plasma concentration of LDL cholesterol. Andrew J. Peat of GlaxoSmithKline’s Medicine Design and Shiping Xie of API Chemistry write about a second-generation non-nucleoside RNA-dependent RNA polymerase (NS5B) inhibitor (Chapter 11). This drug is for the treatment of hepatitis C virus infection and produces a robust drop in plasma viral RNA levels in HCV infected people. We would be remiss if we did not acknowledge the many referees who made creative suggestions for improvement and correction. We must thank our colleagues at ACS Books who encouraged and facilitated the compilation of this book: Elizabeth Hernandez, Sara Tenney, Amanda Koenig, and Arlene Furman. Special thanks go to Kai Rossen whose foreword elegantly states the interlocking importance of pharmaceuticals and organic chemistry, and whose role as the Editor-in-Chief of Org. Process Res. Dev. has contributed significantly to the advancement of process chemistry. Finally, we thank the Division of Organic Chemistry of the American Chemical Society for their sponsorship of our biennial symposium from which this book sprung. Jaan A. Pesti EnginZyme AB Stockholm 114 28, Sweden Ahmed F. Abdel-Magid Therachem Research Medilab, LLC Chelsea, Alabama 35043, United States Rajappa Vaidyanathan Bristol-Myers Squibb Bangalore 560099, India

xiv

Chapter 1

Synthetic Routes for Venetoclax at Different Stages of Development Yi-Yin Ku*,1 and Michael D. Wendt2 1Process Chemistry R-450, PRD, Global Pharmaceutical Research and Development,

AbbVie, 1 North Waukegan Road, North Chicago, Illinois 60064, United States 2Oncology Medicinal Chemistry R4AB, Global Pharmaceutical Research and Development,

AbbVie, 1 North Waukegan Road, North Chicago, Illinois 60064, United States *E-mail: [email protected].

We describe the discovery of venetoclax, a selective Bcl-2 inhibitor that has been approved for the treatment of chronic lymphociytic leukemia, and is in clinical studies to treat other forms of leukemia. We begin by briefly discussing the biological rationale for Bcl-2 inhibition and the background of the Bcl-2 family inhibitors program at AbbVie (previously known as Abbott), including earlier projects aimed at discovery of dual Bcl-xL/Bcl-2 inhibitors. Subsequently, we present a synopsis of the later effort to find selective Bcl-2 inhibitors, which culminated in the discovery of venetoclax, and the synthetic route used at that time, with emphasis on the steps responsible for changes to earlier acylsulfonamide series compounds. We then describe how the process team discovered multiple improvements to the initial route and led to the first-generation large-scale synthesis. A new and more efficient synthetic route was then strategically designed to overcome the chemistry challenges associated with the first-generation synthesisIt features a Buchwald–Hartwig amination followed by a uniquely effective saponification reaction using anhydrous hydroxide generated in situ. The new synthesis improved the process convergence, overall yield, and manufacturing robustness. In addition, the new synthesis enabled consistent and reproducible large-scale manufacture of venetoclax to produce high quality active pharmaceutical ingredient (API) with >99% area purity.

© 2019 American Chemical Society

Introduction Venetoclax (Venclexta®) is a first-in-class selective Bcl-2 inhibitor approved for the treatment of chronic lymphociytic leukemia (1). Venetoclax 1 (Figure 1) disrupts blockage of the intrinsic apoptosis pathway mediated by Bcl-2 family proteins. This is the first protein–protein interaction (PPI) inhibitor approved for cancer treatment. Here we describe the discovery and initial medicinal chemistry synthesis of venetoclax, along with modifications to this route by the process group, and the final large-scale synthesis.

Figure 1. Structure of venetoclax.

Biology of Bcl-2 Family Proteins and Inhibitors Programmed cell death, or apoptosis, is a key mechanism for the disposal of old, damaged, or no longer needed cells (2). The Bcl-2 family of proteins has long been known to play a vital role in the maintenance of the intrinsic apoptosis pathway. The antiapoptotic members of this family are strongly associated with tumor maintenance, disease progression, and resistance to chemotherapy (3, 4). Bcl-2 in particular has been shown to be of great importance to many hematologic malignancies (5, 6). Bcl-2 family proteins modulate apoptosis largely through their interactions with each other. Antagonism of antiapoptotic members of this family (e.g., Bcl-2, Bcl-xL, and Mcl-1) involves mimicking pro-apoptotic family members with small molecules, and disrupting their protein–protein interactions. The nature of these interactions, involving two large, hydrophobic surfaces, despite the long-standing biological rationale for inhibition of Bcl-2 family proteins prevented a large investment by the pharma industry into discovering effective inhibitors of these proteins. In the late 1990s scientists at AbbVie (then Abbott) began work on these targets by obtaining structures of Bcl-xL (7). Soon after, a medicinal chemistry group began to seek small molecule dual Bcl-xL/Bcl-2 inhibitors. This eventually resulted in the discovery of ABT-737 2, a proof-of-concept compound (8), and later, ABT-263 (navitoclax, 3), an oral inhibitor that is still in clinical studies for cancer (Figure 2) (9–12). These studies revealed positive signs of activity against lymphoid malignancies thought to be Bcl2-dependent (11, 12). Additionally, both preclinical and clinical data indicated that Bcl-xL inhibition was responsible for a rapid decrease of circulating platelets. This thrombocytopenia proved to be the dose-limiting toxicity of single-agent navitoclax. As a result of these studies, the idea emerged that a Bcl-2-only inhibitor should have a larger therapeutic window with regard to blood-borne cancers, 2

thus allowing higher drug concentrations and better efficacy. In light of this, a new effort to find a selective agent began.

Figure 2. Structures of ABT-737 (2) and ABT-263 (navitoclax, 3), and X-ray structure of navitoclax bound to Bcl-2, with P2 and P4 regions noted.

Discovery of Venetoclax The binding modes of AbbVie inhibitors are primarily characterized by an electrostatic interaction between the charged acylsulfonamide and an arginine residue on the target protein, and two hydrophobic interactions with well-defined pockets (P2 and P4 in Figure 2). The P2 pocket is filled by AbbVie inhibitors with a chlorophenyl group, while we had a wealth of data from earlier work, and a large number of X-ray structures of inhibitors, we were able to remove moieties more responsible for Bcl-xL activity than Bcl-2 activity from navitoclax and related structures. In particular, removing the amine existing outside of the P4 binding pocket, and destroying the pistacking interaction that fills the P4 pocket by removing the phenylthio group, gave analogs that exhibited much lower affinity to Bcl-2 but much better selectivity versus Bcl-xL. Incorporation of these changes, along with replacement of the dimethylcyclohexenyl group with the synthetically simpler phenyl, and replacement of the triflyl group with the more versatile nitro, gave us canonical lead compound 4.

Figure 3. Compounds 4, 5, and 6. 3

The reinstallation of the pi-stacking arrangement for the P4 ligand through a different connection was a key discovery that enabled more fully elaborated compounds that had improved Bcl-2 affinity and also maintained or improved selectivity. We began with a phenyl group linked by a single oxygen atom to give 5 (Figure 3). This compound demonstrated that the new arrangement had intrinsically better selectivity for Bcl-2. We were also informed by a serendipitous Bcl-2-bound X-ray structure that a hydrogen bond with an aspartic acid within the P4 pocket might be a source of additional affinity and selectivity. Bcl-xL differs slightly from Bcl-2 by having a glutamic acid at that position, resulting in exploitable differences in distance and geometry for a hydrogen bond. This was best accessed via indoles, as in 6. Later, an adjacent-in-space arginine was used to add an interaction with a pyridyl nitrogen and remained to switch the P2 phenyl back to a dimethylcyclohexenyl group; combining these modifications gave 1, named ABT-199, and ultimately became venetoclax (13).

Medicinal Chemistry of Venetoclax Much of the syntheses of venetoclax and other project compounds were able to closely follow similar syntheses of compounds from earlier projects; however, introduction of the –OAr groups required new chemistry. A scan of then-available methods of forming diaryl ethers led to the use of a copper-mediated O-arylation reaction, which at first provided products in 60–80% yields (Figure 4, eq. 1) (14). As the final step of compound syntheses was in almost all cases formation of the acylsulfonamide from acids and sulfonamides, the completed acid portion of the target compounds was constructed with the reactive bromide in place to facilitate a more efficient synthesis of analogs (Figure 4, eq. 2). The bromide of this acid was somewhat less reactive toward O-arylation than that of the 4-fluorophenyl ester, but nevertheless, this substrate performed well initially. Unfortunately, when more elaborated aryl rings were present, the yields dropped to an average of 40%. The reaction proved even more disappointing with indoles, as N-arylation was a common side-product. While this could have been prevented through installation of a protecting group on the indole nitrogen, we were keen to avoid two more steps (protection–deprotection) to the overall scheme. Even if the substrate was switched back to the more reactive 4-fluorobenzoic acid, some phenols failed to provide any product at all. Moreover, in many cases, this Cu-mediated protocol required the use of higher catalyst loading, and large amounts of additives in the form of 1-naphthoic acid and sieves to help solubilize the phenols. This became increasingly troublesome at scales much larger than 1 mmol (14). At that point we were primarily interested in indoles and indole analogues, and as the syntheses of indole analogue were working very poorly via a late installation using the fully elaborated ester (Figure 4, eq. 2), we investigated the introduction of the aryloxy groups at an earlier stage. Eventually a switch was made to an SNAr approach. The use of 4-bromo-2-fluoroesters led to reactions that were sluggish enough to result in competing amount of N-arylation. However, 2,4-difluorophenylesters provided much faster and cleaner reaction profiles (Figure 4, eq. 3). Polar solvents gave the desired product primarily, but also delivered up to 25% of the regioisomer resulting from attack at the alternate fluoride, as well as small amounts of the bis-addition product. We temporarily employed diglyme as solvent (necessary to achieve high enough temperatures) with tripotassium phosphate that gave products cleanly and in excellent yields with only trace amounts of the regioisomer. Ridding the product of solvent could be cumbersome, but products were often collected by trituration and, in other cases, column chromatography usually served to remove residual diglyme. 4

Figure 4. Iterations of the key aryl ether formation step. Eventually we settled on a 7-azaindole as the ideal substituent. The aza-nitrogen interacts with an arginine side chain on the Bcl-2 protein, leading to both better affinity and improved selectivity, presumably by strengthening the indole N–H hydrogen bond. The synthesis of this derivative required we produce multigram quantities of the 5-hydroxy-7-azaindole (Scheme 1). Fortunately, halogenation of the parent azaindole occurs at the 5-position and the relatively inexpensive and widely commercially available 5-bromo derivative 7 served as the most appropriate starting material. Conversion of the bromo to a hydroxyl group via a boronate seemed straightforward; it only needed to be determined whether the intermediate boronate would be best installed via palladium-mediated coupling or transmetalation. Both methods worked; the latter method was employed successfully and quickly on a moderate scale, proceeding in one-pot to give 9 from the TIPS-protected starting bromide 8. The process group found that, for the addition step, switching the base to the more soluble potassium tert-butoxide allowed a lower temperature to be used and reactions could be run in 2methyltetrahydrofuran rather than diglyme. This provided 11 more readily in up to 99:1 regioselectivity.

Scheme 1. Synthesis of diaryl ether 11. 5

During this time, replacement of the biaryl P2 fragment (Figure 3a) with an ABT-263-era P2 fragment began. This featured a dimethylcyclohexene as a replacement for the central phenyl ring and used chemistry developed during the navitoclax program (Scheme 2).

Scheme 2. Synthesis of the aldehyde 18. With the fluoroester 11 and the aldehyde 18 available, it remained to complete the synthesis of the acid portion of the molecule. In principle, the central piperazine could be added first to either the “top” ester piece 11, followed by reductive amination with 18; alternately, the “bottom” aldehyde piece 18 could be appended first, followed by an SNAr on 11. In practice, an SNAr worked much better with multiple equivalents of amine and resulted in a very clean reaction. The synthesis was therefore followed as shown in Scheme 3.

Scheme 3. First synthesis of venetoclax. This synthesis was very reliable and could be performed on up to hundred-gram quantities with little difficulty. Many of the steps required no chromatography and this route was frequently used in the medicinal chemistry lab to bring up smaller fragments for SAR work, as well as complete 6

acid fragments. For synthesis of the sulfonamide portion, the chlorosulfonamide 19b worked well and only required higher temperatures and dimethylformamide (DMF) as solvent; but the fluorosulfonamide 19a was commercially available in large enough quantities for the medicinal chemistry group. A multitude of amines and alcohols were used in this reaction with no difficulties. The final acylsulfonamide formation worked reliably with EDCI and poorly (or not at all) with other coupling reagents.

Process Chemistry for First-Generation Large-Scale Synthesis In general, developing an optimized chemical process for a drug candidate requires substantial amounts of time and resources, which may not be appropriate for the early stages of development as the physical supply of API is often on the criticalpath. Expeditious delivery of sufficient quantities of API to enable advancement of a program oftentimes takes priority, especially considering the high attrition rate in early development prior to demonstrating proof-of-concept and intended efficacy (15–17). With this in mind, an initial assessment of the synthetic route indicated that the quickest path to the development of a scalable process for the complex drug candidate 1 was to improve upon the existing medicinal chemistry approach (Scheme 3). In the medicinal chemistry synthesis, the final coupling of the core acid 25 and the sulfonamide 21 worked reasonably well. However, the SNAr reaction for assembling the key building block 11 had regio- and chemoselectivity concerns, both of the aryl fluorides at 4- and 6-postion in the starting material could undergo the SNAr reaction, and the N-arylation of azaindole could also occur. In addition, the reaction suffered from reproducibility issues caused by strong scale-dependence of agitation upon scale up of the heterogenous reaction mixture. Additionally, the synthesis of the key aldehyde intermediate 18 was lengthy and included a tedious purification. Finally, 18 was a thick syrup that made the isolation of acceptable quality material difficult for large-scale production. Therefore, early process chemistry efforts were focused on improving these problematic and inefficient reactions and isolations, as well as to overcome scale-up challenges associated with the medicinal chemistry synthesis. A two-step, one-pot synthesis of aldehyde intermediate 18 (Scheme 4) (18) was developed as a streamlined alternative. Starting with the same starting material 3,3-dimethylcyclohexanone 12, the ketone was converted via Vilsmeier reagent to the chloroaldehyde intermediate 26, which was directly subjected to the Suzuki coupling with boronic acid 15 without need for prior isolation. Aldehyde 18 was isolated in 37% yield over the two steps after column chromatography purification.

Scheme 4. New synthesis for aldehyde 18. Although the synthesis of the aldehyde 18 was significantly improved from the five-step sequence, its isolation and purification remained challenging due to its syrup-like physical state, which made removal of process impurities difficult. The Suzuki cross-coupling (19–21) reactions for 7

both the initial and improved syntheses introduced several highly toxic mutagenic (15 in Figure 5) and carcinogenic impurities (27 and 28 in Figure 5) that required control to part per million (ppm) and part per billion (ppb) levels, respectively (22, 23). Consequently, purification to reduce these impurities to desirable levels required tedious column chromatography.

Figure 5. Mutagenic and carcinogenic impurities. Attempts were made to form a crystalline bisulfite adduct of aldehyde 18 in an effort to eliminate the chromatographic purification. Although crystallization of the bisulfite adduct was possible, multiple crystallizations were required to achieve the desired purity profile, resulting in substantial product losses. The significantly higher isolated yield of 18 from chromatography led us to proceed with this method of purification in spite of the resource and time-intensity required. Investigation of the initial SNAr reaction using N-triisopropylsilyl-protected hydroxyl azaindole 9 (Scheme 5) (24) revealed that substantial levels of desilylation occurred during the reaction and produced a mixture of products with and without the silyl protecting group. The mixture of products could be converted to the desired intermediate 11 by treatment with tetrabutylammonium fluoride. The major byproducts were then 29 and 30, generated from nonregioselective displacement of the fluoro group at both 4- and 6-positions. Interestingly, only a minimal amount of byproduct 31 from N-arylation of the deprotected azaindole, was observed.

Scheme 5. Initial SNAr reaction using the TIPS-protected azaindole 9. Based on this observation, we reasoned that the protection/deprotection steps could be completely avoided and the SNAr reaction could be performed using the unprotected azaindole 32 (Scheme 6). Conditions for the SNAr reaction with azaindole 32 were briefly screened. After evaluating solvent, base, and reaction temperature it was discovered that the reaction conditions using a combined solvent system of 2-Me-THF/DMF (10:1) with KOt–Bu afforded comparable results to the synthesis using TIPS-protected 9. During the scale-up of this reaction, the product purity profile was discovered to be extremely sensitive to agitation and was heavily scale-dependent, because the reaction mixture initiated as a 8

heterogeneous mixture. On a 2 kg scale, the reaction proceeded similarly to lab experiments with respect to yield and product purity profile; however, when the reaction scale was increased 10-fold, a substantial increase of the undesired byproducts was observed, resulting in a significantly diminished yield of 33%. Chemical engineering modeling to understand the agitation power in relation to the reactor geometry was performed, but reproducibility challenges with respect to product purity profile and yield persisted.

Scheme 6. The new SNAr using unprotected azaindole. The second SNAr reaction of 11 with piperazine 22 (Scheme 7) for the preparation of intermediate 23 was initially carried out at 100 °C in dimethyl sulfoxide (DMSO) (25), but the safety evaluation of this reaction revealed that exothermic decomposition of the reaction mixture occured at 200 °C with an onset temperature at 130 °C. To address this safety concern for manufacture, the reaction temperature was reduced to 50 °C while still achieving full reaction conversion in ≤10 h.

Scheme 7. The second SNAr reaction to synthesize intermediate 23. As expected, the bis-adduct impurity 33 (Figure 6) was also generated during the reaction and needed to be controlled to 99%) and high yield (>93%). It was discovered that >3 equiv of NaOt–Bu were required (Scheme 13). To develop a cost-effective manufacturing process, attempts were made to lower the Pd catalyst loading for the C–N coupling. When the amount of [Pd(P(t-Bu)3)Br]2 used was reduced to 1 mol% (for 2 mol% of total Pd), the reaction conversion stalled out at ~40% when 3.2 equiv of NaOt–Bu were used (Figure 7). However, full conversion (>99%) could be achieved with larger 13

excesses of base (~7 equiv). We noted that less NaOt–Bu was required with higher Pd loadings. At present, we hypothesize this phenomenon is a result of unidentified, but subtle organometallic interactions; the π-basic N-1 and N-7 of the azaindole may complex to Pd to form catalytically inactive, off-cycle Pd–N species and the excess NaOt–Bu may be necessary to recover the catalyst via a Pd alkoxide (41). Previously reported methods for aminations with LiHMDS in the presence of similar unprotected heterocycles gave poor conversion in this case (43, 44). Use of a large excess of NaOt–Bu in this coupling reaction has little impact to the overall cost of goods as NaOt–Bu is relatively inexpensive and could be readily sourced.

Scheme 13. Initial Buchwald–Hartwig amination for 42.

Figure 7. Impact of NaOt–Bu equiv on C–N coupling conversion. The final saponification of the tert-butyl ester 42 turned out to be more challenging than expected. The previous saponification conditions (LiOH in 1:1:1 ratio of THF/EtOH/H2O at 50 °C) were ineffective for 42. The reaction was extremely slow even with extended reaction times. When the reaction temperature was raised, significant amounts of degradation were observed. Other commonly used basic saponification conditions were also examined without success. Switching to acid-mediated conditions with trifluoroacetic acid in dichloromethane or phosphoric acid in acetonitrile gave low yields of product and was accompanied by substantial amounts of decarboxylation.

14

Alternatively, it was found that the tert-butyl ester 42 could be cleanly converted to the methyl ester 24 under acidic conditions, which subsequently underwent clean saponification to the desired acid 25 as described before. Although both reactions proceeded cleanly, this was non-ideal due to the additional step required and we sought a direct hydrolysis of the tert-butyl ester . Our analysis of the literature revealed that “anhydrous hydroxide,” generated from the reaction of KOt–Bu with a carefully controlled stoichiometry of water, could be used to carry out the saponification of a variety of bulky esters in organic solvent (45). We thought this under-utilized procedure could be promising. In fact, we observed clean, but incomplete saponification when 42 was treated under the reported conditions using 2 equiv of KOt–Bu and 1 equiv of water in THF at room-temperature. After some optimization, the reaction proceeded to completion to give acid 25 in >90% yield when the saponification was performed with 5 equiv of KOt–Bu and 3 equiv of water at 55 °C (Scheme 14).

Scheme 14. Saponification of tert-Butyl ester with “anhydrous hydroxide.” Having demonstrated that the Buchwald–Hartwig coupling could be used to successfully prepare the advanced intermediate 42, we required a cost-effective synthesis of key tert-butyl ester 43. As previously mentioned, methyl benzoate 37 was expensive and unavailable in large quantities, which was also true for the tert-butyl analogue 43. So, a new synthesis for 43 that utilized the inexpensive and readily available raw material 44 was developed (Scheme 15). Selective metalation of 4-bromo-2-fluoro-1-iodobenzene 44 with iPrMgCl afforded the Grignard intermediate derived from the iodide, which was then treated with Boc2O at -5 °C to produce the desired tert-butyl ester 43 in 88% yield. A simple aqueous workup afforded product in >98% area purity for use in the following SNAr reaction without further purification. The timely development of this straightforward and economical synthesis of 43 provided a reliable source of the starting material and was an important factor in the decision to pursue this new synthetic route.

Scheme 15. New synthesis for the tert-butyl ester 43.

15

These proof-of-concept studies demonstrated the improved efficiency and convergence of the new synthetic route. A Buchwald–Hartwig amination reaction was used to connect the two key building blocks bromobenzoate 43 and chloropiperazine 38. Moreover, highly scalable processes for both 43 and 38 were implemented. This new synthesis addressed all of the aforementioned challenges associated with the first-generation process, while also importantly reducing the cost of goods. Having achieved the goals set forth, the decision was made to further optimize this route to synthesize 1.

New Route Development for Large-Scale Synthesis Robust control strategies that define the process parameters and operating ranges must be established to ensure consistent process performance and product quality. They are based on thorough examination and fundamental understanding of each reaction and the corresponding risk assessment. Control of material attributes and purity profiles for each intermediate and the RSMs (29) are also important parts of the overall strategy and are established based on process understanding of impurity formation and their downstream fate. The development of a robust manufacturing process for venetoclax 1 was focused on further optimization and understanding of the reactions, starting from the designated RSMs (19b, 38, and 41). One of the more important aspects of this endeavor was to optimize and study the Buchwald–Hartwig amination between 41 and 38. The first step was to identify a more efficient catalyst. Based on the promising initial result with the [Pd(P(t–Bu)3)Br]2 precatalyst, a focused ligand screen was performed by evaluating tert-butyl-ester and other bulky phosphines (Figure 8). From this ligand screen, where 2.0 mol% of total Pd was used, not surprisingly, PtBu3 (A in Figure 8) was still effective, but both AmPhos (F in Figure 8) and DiTBPF (I in Figure 8) were also observed to be effective for the coupling of 41 and 38, each giving >80% conversion to 42. In order to better differentiate among these three ligands, the amination was performed under more challenging conditions using only 1.0 mol% of total Pd (Figure 9). Under these stressed conditions, AmPhos (43) was discovered to be the superior ligand (F and G in Figure 9). Although [Pd(πcinnamyl)Cl]2 gave a slightly more active catalyst than Pd2dba3, the decision was made to move forward with the process using Pd2dba3, due to cost and commercial availability. The starting material charge ratio of benzoate 41 to chloropiperazine 38 was found to be important with respect to formation of impurities. When the chloropiperazine 38 was used in excess, increased levels of Cl-amination impurity 45 was generated from further coupling of 38 with the desired product 42 (Scheme 16). Although the Pd catalyst reacted preferentially with the aryl bromide at low conversion, while approaching high conversion of bromobenzoate 41, the low concentration of aryl bromide relative to the higher concentration of chloride 42, led to the undesired coupling. Therefore, 41 was used in slight excess (1.04 equiv) to ensure that the impurity 45 was formed in an acceptable level of impurity 45 and its daughter impurities were not readily purged downstream. As previously observed, efficient cross-coupling was dependent on the addition of excess equivalents of NaOt–Bu, which was also found to be the case with the Pd/AmPhos catalyst. However, as few as 4 equivalents of NaOt–Bu were sufficient to achieve full conversion with 1.5 mol% of the Pd/AmPhos catalyst.

16

Figure 8. Focused ligand screen for C–N coupling.

Figure 9. Evaluation of Pd precursors for the key amination. 17

Scheme 16. The key impurity generated from the amination. The quality of NaOt–Bu also had a significant impact on the reaction performance due to the large amount used. When old lots of NaOt–Bu were used, the reaction proceeded to generate high levels of benzoic acid impurities from the saponification of both 41 and 42. We assumed that the old lots of base contained higher levels of NaOH generated from quenching of NaOt–Bu with atmospheric moisture, analogous to the “anhydrous hydroxide” generated in the saponification of the tert-butyl ester 42 to the core acid 25. Saponification of 41 generated a sodium carboxylate that, in theory, could couple with chloropiperazine 38 to give core acid 25 directly. However, due to its poor solubility in tetrahydrofuran (THF), the carboxylate did not undergo this amination and the saponification of 41 was an unproductive pathway resulting in lower yield of 42 and higher levels of impurities. Analytical control of the quality of NaOt–Bu to ensure low levels of NaOH would be challenging and impractical, especially given that the base would be re-exposed to atmospheric conditions following the quality testing. Therefore, it was necessary to develop a robust operational control strategy to ensure minimal amounts of NaOH would be introduced to the reaction. Given the very low-solubility of NaOH in THF, filtration of the NaOt–Bu/THF solutions to remove the sodium hydroxide in laboratory experiments demonstrated that the amination reaction proceeded consistently with lots of varying quality. Incorporation of a filtration unit operation during manufacturing for removal of NaOH was straightforward and offered a practical and robust strategy to enable this sensitive reaction to be carried out reproducibly. The coupling reaction was found to be sensitive to the level of oxygen in the system, the high level of oxygen adversely affected the reaction performance and produced variable results. This concern was addressed by controlling and monitoring the oxygen level of the headspace using an oxygen monitor. It was determined that an oxygen level of ≤40 ppm in the headspace was sufficient to ensure consistent reaction performance and could be achieved by subsurface sparging of solvents with dry nitrogen and active exchange of reactor atmospheres with vacuum/N2-purge cycles. With control of the oxygen level, under the optimized conditions (6.5 equiv NaOt–Bu, 0.8 mol% Pd2dba3, 1.6 mol% AmPhos in THF) the amination proceeded in >99% conversion with about 90% isolated yield following aqueous workup and crystallization. The palladium content was effectively reduced to a desirable level (~120 ppm) with three aqueous washes using a solution of 5% L-cysteine, 5% NaHCO3, and a crystallization. The two more chemical steps in the downstream process further reduce the palladium to an acceptable level for the drug substance. 18

The saponification using the “anhydrous hydroxide” was also further studied and optimized. Ultimately, we chose 2-Me-THF as the reaction solvent because it is less hygroscopic than THF, which provides better control of water levels in the reaction. In addition, the aqueous extractive workup and crystallization purification could be performed more effectively in 2-Me-THF. When a series of saponification experiments were performed with 5 equiv of KOt–Bu and varying amounts of water (Figure 10), the reaction proceeded to completion only when the ratio of water to KOt–Bu was ≤1 (i.e., an excess of KOt–Bu relative to water). Incomplete reactions were observed when water was used in excess relative to KOt–Bu. The optimal molar ratio was determined to be 0.6 using 5 equiv of KOt–Bu.

Figure 10. Impact of water amount using 5 equiv of KOt–Bu. Higher temperatures drastically reduced the reaction time; the saponification proceeded to completion cleanly within 2 h at 50 °C, while only 43% conversion was achieved during that time period at 21 °C. Since there was no detrimental impact on the impurity profile, the reaction was performed at 50 °C to keep the reaction cycle time low. Under the optimized conditions (6 equiv) KOt–Bu and 3.6 equiv H2O at 50 °C), the reaction proceeded with >98% conversion and ultimately afforded about 90% isolated yield. Developing robust strategies for the final sulfonamide coupling reaction and isolation was critical to ensure the API quality. The low-solubility of the API 1 in most organic solvents limited the solvent options for the amide coupling and isolation. Moreover, the poor nucleophilicity of the sulfonamide resulted in a slow reaction and competing byproduct formation. Screening of coupling reagents (e.g., CDI, HATU, T3P and EDAC) and different amine bases (e.g., Hunig’s base, Et3N, 4-(Dimethylamino)pyridine (DMAP), 1,8-Diazabicyclo[5.4.0]undec-7-ene (DBU), 1,4diazabicyclo[2.2.2]octane (DABCO), pyridine, collidine, lutidine, and N-methylmorpholine) confirmed that the initial conditions using 1-Ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDAC) and DMAP/Et3N produced the best results with respect to conversion and purity. Solvent screening revealed that dichloromethane was the optimal solvent and it was also discovered that the coupling was best run at room-temperature. Although increased temperatures (35 °C to reflux) decreased the reaction times, they also increased the levels of impurities. Efforts were also focused on understanding the formation of impurities and their rejection. Several process impurities were identified, and their structures were determined using 2D NMR and MS analyses (Figure 11); the low-level impurities 46, 47, and 48 could be readily removed during 19

aqueous work-up and crystallization. The major bisamide 49 impurity was difficult to reject and its level varied between batches. It slowly increased during the reaction, meanwhile the reaction rate for forming 1 decreased.

Figure 11. Impurities generated in the sulfonamide coupling reaction. The order of addition of the reagents also impacted this bisamide impurity level. The bisamide 49 was formed in 8% when EDAC was added last to a suspension of the sulfonamide 21, core acid 25, and DMAP/Et3N. In contrast, slowly adding core acid 25 to the reaction mixture last resulted in only 5% of the bisamide. The addition time of the core acid 25 solution also played a key role because longer addition times (8 h) led to incomplete reactions with up to 9.0% of core acid 25 remaining. After a series of experiments (Figure 12), it was determined that the best reaction profile was obtained with slow addition of a mixture of core acid 25 and Et3N in dichloromethane to a suspension of sulfonamide 21, DMAP and EDAC in dichloromethane with an optimal addition time of ~2–6 h. It was also discovered that treatment of bis-amide 49 with N,N-dimethylethylediamine (DMEDA) led to the re-formation of desired product 1 along with the corresponding DMEDAamide impurity 50, which was readily rejected (Scheme 17). This discovery further ensured the robustness of the process in respect to impurity rejection. The API 1 was found to readily form solvates that had low-solubilities, making the work-up of the amide coupling product challenging. After examination of different solvents and solvent mixtures, we opted to use dichloromethane for the work-up and isolation. A dichloromethane solvate was found to spontaneously crystallize from solution during aqueous extractions; therefore, the work-up was performed at high dilution (20 L/kg) and at 35 °C to keep the API soluble. In addition, a small amount of methanol (0.5 L/kg) was added to the organic layer after each extraction to increase product solubility and prevent product crystallization. The extractive work-up used aqueous acetic acid to remove excess DMAP followed by a sodium bicarbonate wash (to remove residual acetic acid) and an aqueous wash with 5% NaCl (to prevent emulsions). 20

Figure 12. Impact of the addition time of the core acid (25) solution.

Scheme 17. Derivatization of the major dimer API impurity.

The final product purification by crystallization effectively rejected product-related impurities and consistently produced 1 in acceptable yield and high purity as the desired solid polymorphic form. It was achieved by slowly adding ethyl acetate as antisolvent to attain a 10/1/9 mixture of dichloromethane/methanol/ethyl acetate, with a targeted total volume of 16 L/kg. As previously mentioned, the initial addition of methanol (1:10 ratio to dichloromethane) increased the product solubility. Additionally, the presence of methanol also increased the solubility of the reaction impurities and resulted in a robust purification with rejection of up to 5% DMEDA-amide. This optimized process was used for producing venetoclax 1 with consistent high quality.

21

Scheme 18. Optimized new synthesis of venetoclax 1.

Summary To support the start of clinical development of 1, early process chemistry efforts were directed towards rapidly developing a scalable synthesis. This first-generation synthesis was based on the medicinal chemistry route and incorporated several key improvements to address scalability challenges. A more efficient and robust synthesis that supported the increased demands for API was developed (Scheme 18) and offered significant advantages as well as addressed known scale-up challenges associated with the first-generation synthesis. The convergent second-generation synthesis included many notable process chemistry innovations: (1) Development of a highly efficient and cost-effective synthesis to ensure a reliable supply of key benzoate starting material 43;

22

(2) Identification of a regioselective and chemoselective SNAr reaction for key building block 41, which also overcame the scale-dependent agitation and reproducibility issues of the initial SNAr reaction; (3) Implementation of a new reaction sequence for key building block 38, which circumvented the challenges of handling and isolating a syrup-like aldehyde intermediate 18; (4) Development of a Buchwald–Hartwig amination protocol and control strategy to ensure the execution of this reaction; (5) Development of a uniquely effective saponification using anhydrous hydroxide for the highly hindered tert-butyl ester 42; (6) Establishing robust protocols and control strategies for the final sulfonamide coupling and the isolation/purification to ensure consistent manufacture of 1. The newly developed process led to a significant reduction in the cost of goods by more than doubling the overall yield.

References Venclexta (venetoclax) label and information for patients and US healthcare professionals: https://www.venclexta.com. 2. Hanahan, D.; Weinberg, R. A. Cell 2000, 100, 57–70. 3. Adams, J. M.; Cory, S. Oncogene 2007, 26, 1324–1337. 4. Youle, R. J.; Strasser, A. Nat. Rev. Mol. Cell. Biol. 2008, 9, 47–59. 5. Vaux, D. L.; Cory, S.; Adams, J. M. Nature 1988, 335, 440–442. 6. Huang, J. Z.; Sanger, W. G.; Greiner, T. C.; Staudt, L. M.; Weisenburger, D. D.; Pickering, D. L.; Lynch, J. C.; Armitage, J. O.; Warnke, R. A.; Alizadeh, A. A.; Lossos, I. S.; Levy, R.; Chan, W. C. Blood 2002, 99, 2285–2290. 7. Sattler, M.; Liang, H.; Nettesheim, D.; Meadows, H. P.; Harlan, J. E.; Eberstadt, M.; Yoon, H. .; Shuker, S. B.; Chang, B. S.; Minn, A. J.; Thompson, C. B.; Fesik, S. W. Science 1997, 275, 983–986. 8. Oltersdorf, T.; Elmore, S. W.; Shoemaker, A. R.; Armstrong, R. C.; Augeri, D. J.; Belli, B. A.; Bruncko, M.; Deckwerth, T. L.; Dinges, J.; Hajduk, P. J.; Joseph, M. K.; Kitada, S.; Korsmeyer, S. J.; Kunzer, A. R.; Letai, A.; Li, C.; Mitten, M. J.; Nettesheim, D. G.; Ng, S.-C.; Nimmer, P. M.; O’Connor, J. M.; Oleksijew, A.’; Petros, A. M.; Reed, J. C.; Shen, W.; Tahir, S. K.; Thompson, C. B.; Tomaselli, K. J.; Wang, B.; Wendt, M. D.; Zhang, H.; Fesik, S. W.; Rosenberg, S. H. Nature 2005, 435, 677–681. 9. Park, C.-M.; Bruncko, M.; Adickews, J.; Bauch, J.; Ding, H.; Kunzer, A.; Marsh, K. C.; Nimmer, P.; Shoemaker, A. R.; Song, X.; Tahir, S. K.; Tse, C.; Wang, X.; Wendt, M. D.; Yang, X.; Zhang, H.; Fesik, S. W.; Rosenberg, S. H.; Elmore, S. W. J. Med. Chem. 2008, 51, 6902–6915. 10. Tse, C.; Shoemaker, A. .; Adickes, J.; Anderson, M. G.; Chen, J.; Jin, S.; Johnson, E. F.; Marsh, K. C.; Mitten, M. J.; Nimmer, P.; Roberts, L.; Tahir, S. K.; Xiao, Y.; Yang, X.; Zhang, H.; Fesik, S.; Rosenberg, S. H.; Elmore, S. W. Cancer Res. 2008, 68, 3421–3428. 11. Wilson, W. H.; O’Connor, O. A.; Czuczman, M. S.; LaCasce, A. S.; Gerecitano, J. F.; Leonard, J. P.; Tulpule, A.; Dunleavy, K.; Xiong, H.; Chiu, Y. L.; Cui, Y.; Busman, T.; Elmore, 1.

23

12.

13.

14. 15. 16. 17. 18.

19. 20. 21. 22. 23. 24.

25.

26. 27. 28.

29.

S. W.; Rosenberg, S. H.; Krivoshik, A.; Enschede, S. H.; Humerickhouse, R. A. Lancet Oncol. 2010, 11, 1149–1159. Roberts, A. W.; Seymour, J. F.; Brown, J. R.; Wierda, W.; Kipps, T. J.; Khaw, S. L.; Carney, D. A.; He, S. Z.; Huang, D. C.; Xiong, H.; Cui, Y.; Busman, T. A.; McKeegan, E. M.; Krivoshik, A. P.; Enschede, S. H.; Humerickhouse, R. J. Clin. Oncol. 2012, 30, 488–396. Souers, A. J.; Leverson, J. D.; Boghaert, E. R.; Ackler, S. L.; Catron, N. D.; Chen, J.; Dayton, B. D.; Ding, H.; Enschede, S. H.; Fairbrother, W. J.; Huang, D. C. S.; Hymowitz, S. G.; Jin, S.; Khaw, S. L.; Kovar, P. J.; Lam, L. T.; Lee, J.; Maecker, H. L.; Marsh, K. C.; Mason, K. D.; Mitten, M. J.; Nimmer, P. M.; Oleksijew, A.; Park, C. H.; Park, C.-M.; Phillips, D. C.; Roberts, A. W.; Sampath, D.; Seymour, J. F.; Smith, M. L.; Sullivan, G. M.; Tahir, S. K.; Tse, C.; Wendt, M. D.; Xiao, Y.; Xue, J. C.; Zhang, H.; Humerickhouse, R. A.; Rosenberg, S. H.; Elmore, S. W. Nature Med. 2013, 19, 202–208. Marcoux, J.-F.; Doye, S.; Buchwald, S. L. J. Am. Chem. Soc. 1997, 119, 10539–10540. Smietana, K.; Siatkowski, M.; Møller, M. Nat. Rev. Drug Discovery 2016, 15, 379–380. Mullard, A. Nat. Rev. Drug Discovery 2016, 15, 447. Carter, P. H.; Berndt, E. R.; DiMasi, J. A.; Trusheim, M. Nat. Rev. Drug Discov. 2016, 15, 673–674. Chan, V. S.; Christesen, A. C.; Grieme, T. A.; Ku, Y.; Mulhern, M. M.; Pu, Y. M. Processes for the Preparation of an Apoptosis-Inducing Agent. U.S. Patent Appl. Publ. US20140275540A1, September 18, 2014. Miyaura, N.; Suzuki, A. Chem. Rev. 1995, 95, 2457–2483. Suzuki, A. Pure Appl. Chem. 1991, 63, 419–422. Suzuki, A. J. Organomet. Chem. 1999, 576, 147–168. Teasdale, A.; Elder, D.; Chang, S.-J.; Wang, S.; Thompson, R.; Benz, N.; Flores, I. H. S. Org. Process Res. Dev. 2013, 17, 221–230. Teasdale, A.; Fenner, S.; Ray, A.; Ford, A.; Phillips, A. Org. Process Res. Dev. 2010, 14, 943–945. Doherty, G. A.; Elmore, S. W.; Hansen, T. M.; Hasvold, L. A.; Mantei, R.; Souers, A. J.; Tao, Z.; Wang, G. T.; Wang, L. Apoptosis-Inducing Agents for the Treatment of Cancer and Immune and Autoimmune Diseases, U.S. Patent Appl. Publ. US20110124628A1, May 26, 2011. Doherty, G. A.; Hansen, T. M.; Hexamer, L. A.; Mantei, R. A.; Sullivan, G. M.; Tao, Z.; Wang, G. T.; Wang, L.; Wang, X. Apoptosis-Inducing Agents for the Treatment of Cancer and Immune and Autoimmune Diseases, U.S. Patent Appl. Publ. US20100298323A1, November 25, 2010. Catron, N. D.; Lindley, D. J.; Miller, J.; Schmitt, E. A.; Tong, P. Solid Dispersions Containing an Apoptosis-Inducing Agent, PCT Int. Appl. WO2012058392A1, May 03, 2012. Catron, N. D.; Chen, S.; Gong, Y.; Zhang, G. G. Salts and Crystalline Forms of an ApoptosisInducing Agent, PCT Int. Appl. WO2012071336A1, May 31, 2012. ICH Q1A (R2) Stability Testing of New Drug Substances and Products. http://www.ich.org/ fileadmin/Public_Web_Site/ICH_Products/Guidelines/Quality/Q1A_R2/Step4/Q1A_ R2__Guideline.pdf (accessed Oct 2018). Q11 Development And Manufacture of Drug Substances. Section V. www.fda.gov/downloads/ drugs/guidances/ucm261078.pdf (accessed Oct 2018).

24

30. Sperotto, E.; van Klink, G. P. M.; de Vries, J. G.; van Koten, G. Tetrahedron 2010, 66, 9009–9020. 31. Benyahya, S.; Monnier, F.; Taillefer, M.; Man, M. W. C.; Bied, C.; Ouazzani, F. Adv. Synth. Catal. 2008, 350, 2205–2208. 32. Liu, X.; Zhang, S. Synlett 2011, 2, 268–272. 33. Tsvelikhovsky, D.; Buchwald, S. L. J. Am. Chem. Soc. 2011, 133, 14228–14231. 34. Ruiz-Castillo, P.; Buchwald, S. L. Chem. Rev. 2016, 116, 12564–12649. 35. Stambuli, J. P.; Kuwano, R.; Hartwig, J. F. Angew Chem., Int. Ed. Engl. 2002, 41, 4746–4748. 36. Seechurn, C. C. C. J.; Sperger, T.; Scrase, T. G.; Schoenebeck, F.; Colacot, T. J. J. Am. Chem. Soc. 2017, 139, 5194–5200. 37. Paul, F.; Patt, J.; Hartwig, J. F. J. Am. Chem. Soc. 1994, 116, 5969–5970. 38. Guram, A. S.; Buchwald, S. L. J. Am. Chem. Soc. 1994, 116, 7901–7902. 39. Louie, J.; Hartwig, J. F. Tetrahedron Lett. 1995, 36, 3609–3612. 40. Guram, A. S.; Rennels, R. A.; Buchwald, S. L. Angew. Chem., Int. Ed. Engl. 1995, 34, 1348–1350. 41. Sunesson, Y.; Limé, E.; Nilsson Lill, S. O.; Meadows, R. E.; Norrby, P.-O. J. Org. Chem. 2014, 79, 11961–11969. 42. Driver, M. S.; Hartwig, J. F. J. Am. Chem. Soc. 1997, 119, 8232–8245. 43. Hartwig, J. F.; Richards, S.; Barañano, D.; Paul, F. J. Am. Chem. Soc. 1996, 118, 3626–3633. 44. Guram, A. S.; King, A. O.; Allen, J. G.; Wang, X.; Schenkel, L. B.; Chan, J.; Bunel, E. E.; Faul, M. M.; Larsen, R. D.; Martinelli, M. J.; Reider, P. J. Org. Lett. 2006, 8, 1787–1789. 45. Gassman, P. G.; Schenk, W. N. J. Org. Chem. 1977, 42, 918–920.

25

Chapter 2

Discovery and Development of Lorlatinib: A Macrocyclic Inhibitor of EML4-ALK for the Treatment of NSCLC Robert Dugger,1 Bryan Li,1 and Paul Richardson*,2 1Process Development, Medicinal Sciences, Chemical Research and Development,

Pfizer Inc., Groton, Connecticut 06340, United States 2Oncology Medicinal Chemistry, Medicine Design, Pfizer Inc.,

La Jolla, California 92122, United States *E-mail: [email protected].

This chapter describes the discovery and synthesis of lorlatinib, a selective EML4ALK inhibitor for the treatment of non-small cell lung cancers. We give a brief overview of the SAR behind the discovery of lorlatinib, and discuss early optimization efforts into the synthesis through macrolactamization of the amidebased macrocycles. In addition, we discuss rationale for developing an alternative synthesis of this class of macrocycles, and herein describe the initiation and optimization of a second-generation approach to access the API based through a Pd-mediated direct-arylation. Accelerated timelines for the project required enabling strategies to be implemented to facilitate the synthesis of preclinical supplies of the lead compound through the macrolactamization-based chemistry, while the expedited development of the commercial synthesis utilizing the direct ring closure is described.

Introduction One of the most common and lethal malignancies worldwide has long been lung cancer, with 2.1 million new cases and 1.8 million deaths in 2018. This represents 12% and 18% of new cancers and cancer mortality, respectively, as estimated by the International Agency for Research on Cancer of the World Health Organization (1). The majority of lung cancers (~90%) are non-small cell lung cancers (NSCLC), consisting of a number of subtypes driven by various activated oncogenes (2). Of these, anaplastic lymphoma kinase (ALK) is a tyrosine kinase receptor belonging to the insulin receptor superfamily. The discovery of echinoderm microtubule-associated protein-like 4 (EML4)ALK fusion gene in NSCLC led to ALK emerging as a novel drug target for cancer therapy (3). This was the first report of an oncogenic fusion gene in a solid tumor, with EML4-ALK being detected in approximately 5% of NSCLC patients (4). © 2019 American Chemical Society

Xalkori™ (crizotinib) was developed by Pfizer and approved by the FDA in 2011 after the majority of patients showed reduced tumor burden in early stage clinical trials (5). Crizotinib is currently approved in 70 countries for the treatment of locally advanced or metastatic NSCLC ALKpositive patients, and is also established as an inhibitor of both c-MET (the target for which it was originally developed) and ROS1 (c-ros oncogene 1), and has since been approved in NSCLC patients presenting gene rearrangements of both these proteins (6). Although crizotinib demonstrates initial robust efficacy in ALK-positive tumors, patients eventually develop resistance, leading to recurrence of disease (7). In addition to the up-regulation of alternative signaling pathways, the most common cause of resistance to crizotinib is mutation of the ALK-kinase domain, the most common of which is L1196M in the kinase gatekeeper region. This inhibits the binding of crizotinib through steric interference. A number of additional mutations have also been identified in the solvent-exposed region near the binding site. With respect to our next generation ALK-inhibitor program the gatekeeper mutation was chosen as our primary assay due to both its prevalence and its strong resistance to crizotinib. A number of other ALK-inhibitors have been approved for clinical use to overcome the issue of resistance to crizotinib. However, with our next generation program, an additional goal was to develop an inhibitor capable of accessing the CNS in order to treat brain cancers (8–10). Approximately 200,000 brain metastases (BM) are diagnosed annually in the United States, accounting for 20% of cancer mortality (11). From peripheral tumors, metastasis to the CNS occurs in as many as 40% of cases per year, with lung (46%) and breast (20%) representing the most prevalent tumors of origin (12). For almost half of ALK-positive NSCLC patients, the CNS represents the first site of progression for patients receiving an ALK-directed therapy (13). With regard to current treatment options, kinase inhibitors are typically not freely CNS penetrant. Although treatment of the peripheral tumor is effective, once the disease spreads to the CNS, rapid progression occurs, leading to a poor prognosis with a median survival of 2.5 months (14). The SAR and design strategies employed in the development of lorlatinib have previously been described in detail. Figure 1 highlights the key milestones in terms of the evolution of the compounds (15). As noted, crizotinib (1) loses 10-fold potency against the ALK L1196M gatekeeper mutation, although it was initially developed as an inhibitor of c-MET. Truncation and optimization of both the head and tail portions of the molecule’s interactions with the protein lead to the identification of the highly potent acyclic inhibitor, PF-06439015 (2) (16). However, a multidrug resistance ratio (MDR — experimentally determined as a measure of a compound’s permeability) of greater than 2.5 indicates that this molecule is likely to be a P-gp (P-glycoprotein) substrate and thus not brain penetrant. This is not surprising given both the high molecular weight as well as the number of hydrogen bond donors present in the molecule. Through modulation of both these features, it was possible to identify triazole compound 3, which met our target criteria in terms of MDR ratio, although it concomitantly showed a significant loss of potency. Over 750 acyclic inhibitors were evaluated, highlighting the difficulties in overlapping a favorable MDR ratio with the desired potency against both wild-type and L1196M ALK. Inspection of a number of crystal structures of these acyclic inhibitors indicated their U-shaped nature, with the head and tail portions in close proximity, suggesting that formation of a macrocyclic ring might not only “lock” the compound in a productive binding mode, but also enable additional favorable interactions to be realized through the linker with the target protein (17–20). After initially evaluating an ether-based linker, which proved to be too lipophilic, our first amidebased macrocycle (4) substantiated these hypotheses, providing a stable, highly potent inhibitor with the MDR ratio trending in a favorable direction. Further optimization of the tail piece specifically 28

for selectivity purposes led to the identification of aminopyridine lorlatinib (5) as well as its aminopyrazine analogue (6). Tropomyosin receptor kinase B (TrkB) was identified as a key protein to obtain selectivity against, as inhibition of this would lead to adverse CNS-based effects. The nitrile group of the pyrazole tail proved to be the optimal substituent for obtaining selectivity between TrkB and ALK. Although better potencies can be achieved against ALK by employing bulkier substituents at this position, the choice of the nitrile gives good selectivity with a minimal molecular weight gain. Both 5 and 6 were extensively profiled at the discovery stage. They were synthesized in multigram quantities in order to attain sufficient screening to progress lorlatinib (5) as the development lead.

Figure 1. Selected SAR highlights leading to the discovery of lorlatinib (5).

Discovery Chemistry for Lorlatinib (5) Disconnection Approach to Macrocyclic Inhibitors The key challenges faced by the program in accessing the lead compounds are summarized in Figure 2. While Mandolini and Illuminati noted this, demonstrating that ring formation was most difficult for 8- to 11-membered rings, the lead compounds in our series are 12-membered (21). It was not clear if a 12-membered ring constrained by a variety of rings and functional groups would cyclize in a facile manner. It is ironic that the chief benefit in terms of the macrocycles as drug targets being their highly rigid conformation (i.e., a reduction of the number of degrees of rotational freedom/ entropy) also causes the issues with their synthesis (22). Considering the possible bond disconnections to assemble the amide-linked macrocycles in particular lorlatinib (5), the molecule can be broken into three fragments. It was also important to our strategy at the discovery stage to try to develop as modular an approach as possible in order to efficiently access analogues. In the previous discussion, we noted the importance of the substitution pattern of the highly-engineered pyrazole (9) to optimize selectivity, and testing variations of this fragment proved to be central to our SAR studies. 29

Figure 2. Potential disconnections to assemble the macrocyclic ring. Previous work on the ether-linked series had evaluated a range of ring-closing reactions for compound synthesis. However, for the amide-based macrocycles, a more focused approach was undertaken with only three key bond disconnections considered for closure of the macrolactam ring. First of all, establishment of the stereochemistry of the methyl-bearing center utilizing either chemoor bio-catalysis, and formation of the ether bond of the aminopyridine fragment (disconnection d) had been well established in chemistry developed for the synthesis of crizotinib. As such, it made sense to exploit this approach to access this intermediate for lorlatinib (5). Although ether bond formation (disconnection c) through a Mitsunobu (23) or mesyl-transfer reaction (24) has been successfully achieved for ring closure in the ether-linked series, this proved to be a nonproductive approach with the introduction of the methyl-group at one of the reacting centers. Therefore, disconnecting to the aminopyridine (7) and the pyrazole (9) led us to two options in terms of order of steps for assembling the ring: closure through either disconnection a or b. Intuitively, macrolactamization (disconnection a) made more sense, as not only has this methodology been widely exploited for closing macrocycles but also aryl–aryl bond formation (disconnection b) through Suzuki coupling has been thoroughly investigated for the formation of the acyclic-compounds profiled within the program (14). Formation of the aryl–aryl bond to close the ring (disconnection b) is attractive at least from a hypothetical standpoint, particularly for the formation of analogues as this is potentially a shorter route (amidation followed by ring closure). The more conventional macrolactamization approach will most likely involve a number of protection/deprotection steps in order to access the correct precursor for ring formation. In addition to adding operations, one action would likely be ester hydrolysis under basic conditions; one must be cognizant herein of the susceptibility of the nitrile substituent to hydrolysis to the amide. Asymmetric Synthesis of the Chiral Aminopyridine Headpiece The initial approach to the required aminopyridine headpiece 17 is shown in Scheme 1. A number of aspects of this synthesis are worthy of discussion. While chiral supercritical fluid chromatography (SFC) was first used to separate the alcohol enantiomers of racemic 12, asymmetric reduction of the ketone 11 could be achieved using established borane-based reagent chemistry with the absolute configuration of the product being determined through vibrational circular dichroism (VCD) (25). This is in contrast to the case of crizotinib, where attempts to use borane-based reagents 30

either failed or led to poor enantioselectivities owing to the ortho, ortho-disubstituted nature of the acetophenone. Whereas a biocatalytic reduction is clearly favorable, particularly in terms of ease of workup, the stoichiometric diisopinocampheylborane chloride (DIP-Cl) approach offered the versatility of enabling access to either enantiomer of the alcohol in relatively high (96%) enantiomeric exces (ee) (26, 27). Furthermore, ready access to both enantiomers was important to enable an array of methods to be assayed for formation of the ether-bond, either with inversion or retention of stereochemistry. The latter is exemplified for the pyrazine-based macrocycle (6) as the ether bond was formed with retention of stereochemistry through initial reaction of the chiral alcohol with 2amino-3,5-dibromopyrazine, whereas for lorlatinib (5) an inversion was required.

Scheme 1. Initial synthesis of the aminopyridine headpiece (17). Returning to the synthesis of the pyridine headpiece 17, formation of the ether bond through an activation-displacement sequence is carried out in an identical manner to that of crizotinib. This was followed by Pd-mediated carbonylation to synthesize the ester 16. The order of events herein is important as carbonylation prior to ether bond-formation led to issues with lactone formation. Regioselective bromination of the aminopyridine (16) was achieved using one equivalent of NBS with careful temperature control to allow access to 17 on a 200 g scale. Synthesis of the Pyrazole Tailpiece Our first synthesis of the pyrazole 24 is shown in Scheme 2. Despite a number of obvious issues described below, this route was initially pursued to provide ~100 g quantities of 24 based on the aggressive timelines of the program. Although, as we will see, the pyrazole is easily accessible, initially the free N-H pyrazole was purchased. Methylation with dimethyl sulfate occurred as expected predominantly adjacent to the pyrazole for electronic reasons to give 18. The main issue with this synthesis was dibromination of both the ring and the methyl group positions to produce 19. The free radical bromination of the methyl group is difficult to control and leads to the dibromomethyl intermediate. Although this theoretically could be directed to the methylamine derivative 20 through the aldehyde and reductive amination, this added a step. Desired dibromide 19 was isolated through a chromatographic purification followed by crystallization. Displacement, Boc deprotection, hydrolysis, amide 31

formation, and dehydration all proceeded relatively smoothly to afford the nitrile 24 (28). Direct amidation of the ethyl ester 21 was possible. However, early attempts to optimize this reaction required the use of a sealed vessel to obtain a synthetically useful yield, and logistically this was considered too expensive a procedure for this transformation.

Scheme 2. Initial synthesis of the pyrazole-based tailpiece (24). A second approach (Scheme 3) was developed to access 500 g quantities of the pyrazole 24. A key change was the removal of the problematic bromination step, and the use of a cheaper starting material. Herein, a de novo synthesis of the pyrazole core (28) is carried out from acetone and diethyl oxalate, followed by condensation with hydrazine. Methylation as previously shown gave a 9:1 mixture of regioisomers in favor of the desired 18. The regioisomers, which differed significantly in polarity, could be separated by simple filtration through a silica plug. This represented the only chromatographic operation in this sequence. The two brominations were then carried out as discrete reactions and, although the free radical bromination is still relatively low-yielding, the desired product 32 could be isolated by crystallization. Direct amidation of the ester 29 has now been enabled with the dehydration carried out using POCl3 as opposed to TFAA to give 31. The endgame displacement/protection is the same, but as an alternative displacement with NaH/tert-butylmethylcarbamate was also demonstrated to be an option. The improved route utilized commodity chemicals to access 18, minimized separations, and allowed for a higher throughput of material.

Scheme 3. Modified synthesis of the pyrazole-based tailpiece (24). 32

Synthesis of the First Amide-Linked Macrocycle Compound 4 was the first amide-linked macrocycle made within the program (~ 2 months prior to the first synthesis of lorlatinib 5). Scheme 4 highlights the approaches taken to close the ring as well as the difficulties encountered in scaling a macrocyclic compound to obtain sufficient quantities (~500 mg) for a pharmacokinetic (PK) evaluation.

Scheme 4. Synthesis and initial scale-up of the first amide-based macrocycle (4). Two complementary sequences were explored for compound 4. In the first approach, the amidation was followed by macrocyclization via formation of the biaryl bond. Initial amide bond formation to provide 33 proceeded smoothly, although aryl–aryl bond formation through an in situ boronate formation/Suzuki coupling approach (which had proved relatively successful for the etherlinked macrocycles) led to poor yields of the desired product. Given this, we subsequently relied on the second approach, which involved biaryl bond formation followed by the more established macrolactamization step to close the ring. The one-pot in situ Suzuki coupling worked reasonably well to deliver ester 35 from the Bocprotected pyrazole 34 and aminopyridine 17. Functional group manipulation involving Bocdeprotection and ester hydrolysis provided the desired precursor, which could be cyclized by amide bond formation with HATU to provide 4. The chemistry did appear to be somewhat capricious in that scaling up by only a factor of 6 led to a dramatic attrition in yield. A surprisingly poor yield was realized in the aryl–aryl bond formation despite what appeared to be a relatively clean LC-MS profile. This suggested that there were unproductive background reactions going on that we were not seeing. However, it was considered satisfactory at the time. Considering the complexity of the molecules, this synthetic approach was deemed acceptable in the pursuit of our initial cohort of macrocyclic amides (leading to the eventual identification of lorlatinib 5), with the main bottleneck being the design and syntheses of the various tailpieces (exemplified by pyrazole 34) utilized. Initial Chemistry Optimization for Lorlatinib Within 2 months of the initial synthesis and identification of lorlatinib (5) as a potential lead compound for the program, we required to access gram quantities for in vivo efficacy and toxicity profiling. At this stage, a series of challenges were identified, specifically the low-yielding nature of the one-pot Suzuki coupling as well as the relatively large projected amounts of the custom intermediates (17 and 24) which would be required for a 5 g batch. For the Suzuki coupling, the base was 33

established as CsF in order to avoid issues with ester/nitile hydrolyses when NaOH was used (Scheme 5).

Scheme 5. Projected starting material amounts for initial scale-up of lorlatinib (5). With a tight timeline in hand for chemistry development, a judicious decision was made in terms of which of the chemistry steps to improve. Although the final three steps proceed in only 40% yield, this was deemed reasonable for a 12-membered macrocycle. The sequence of the ester hydrolysis and Boc deprotection steps had been reversed (vide infra), and crude material from these steps is taken directly into the macrolactamization procedure. The bigger issue involved the aryl–aryl bond formation in which we were relying on several processes to take place for the reaction to proceed, specifically formation of a boronate on a reaction partner, Pd-insertion, transmetalation, and coupling. Clearly numerous side reactions (such as deshalogenation, homocoupling, etc.) can be invoked here, leading to a difficult purification, and thus explaining the low yields observed. With this in mind, we decided to reevaluate this process as a two-step procedure, examining boronate formation and Suzuki coupling separately. We had to first determine which fragment (17, 24, or 37) was most effective in terms of being able to form and react as the boronate ester. In order to evaluate this step, a reaction screen was carried out looking at boronate formation for both the reaction partners (17 and 24) as well as for the bis-Boc protected compound 37. The inclusion of 37 is explained as it was believed that this would both modulate the rate of oxidative addition into the C–Br bond of the pyridine, and also potentially alleviate issues with Pdcomplexation to the substrates. The screen encompassed 72 reactions run overnight at 80 °C looking at 6 catalysts and 4 solvents with KOAc employed as the only base used in Suzuki–Miyaura borylations. The results of the screen (Figure 3) were informative, demonstrating that the pyrazole 24 did not form the boronate under any of the conditions evaluated. Although the MIDA-boronate of this fragment was synthesized using halogen-metal exchange and trapping, this added an additional layer of complexity. The aminopyridine 17 is a better substrate, and numerous conditions gave reasonable conversion to the boronate. However, by far the best substrate is the bis-Boc derivative 37, which gave > 90% yield. The same ligand that we had used thus far is shown to be the best for this transformation (CataCXium A). In addition, dioxane and toluene were identified as far superior solvents to methanol with regard to levels of conversion. We easily formed the boronate 38 from the bromide 37. A screen was performed to evaluate the Suzuki–Miyaura coupling of the boronate 38 with the pyrazole 24. The intention here was not only to identify the optimum conditions for this transformation, but also to understand the reason behind the poor isolated yields from the one-pot borylation/coupling reaction despite an apparently clean reaction profile by LC-MS. Typically, disparities such as this occur due to unknown side reactions leading to products that cannot be observed at the UV wavelength of the analysis (280 nm in this case). 34

Figure 3. Screening results for the Suzuki–Miyaura borylation. We hypothesized and found MS evidence to support that the pyrazole 24 was undergoing des-bromination at approximately the same ratio relative to reaction with the boronate 38, thus limiting the yield. To solve this problem, eight catalysts/three bases and four solvents were screened. As a result, there were reactions in which we observed a good correlation between the UV and mass response. From these experiments we determined that in contrast to the one-pot reaction, Pd(dppf)Cl2 was the best catalyst (Figure 4). Either NaOH or CsF appeared to be a good choice of base for the reaction, while the emergence of toluene as the best solvent potentially provided the opportunity to telescope the borylation/coupling steps.

Figure 4. Screening of the Suzuki aryl–aryl bond formation to form (39). While the aryl–aryl bond was a critical bond-forming process for optimization, it was also important to prudently select the conditions for the deprotection steps specifically herein the ester hydrolysis. The nitrile of the pyrazole moiety was susceptible to hydrolysis to the primary amide 35

under basic conditions. As the byproduct resulting from this was difficult to purge by crystallization, chromatography typically carried out at the end of the synthesis was required to remove it. In order to avoid nitrile hydrolysis, we screened the ester hydrolysis using five bases (NaOH, LiOH, KOH, Na2CO3, and K2CO3) in two solvents (MeOH, THF) with and without the addition of water. This revealed that although it was not possible to completely eliminate this side reaction, using KOH in MeOH at 50 °C gave the minimum amount of nitrile hydrolysis (< 10%) during the ester hydrolysis (Scheme 6). It was also discovered that if one carried out the ester hydrolysis prior to the Boc deprotection, the amount of the amide impurity increased; thus explaining the order of events ultimately chosen. Operationally, after the ester hydrolysis, the reaction is acidified, generating the soluble HCl salt and leading to precipitation of the majority of the inorganic salts. The solids were removed by filtration, and the filtrate concentrated to dryness. The residue (containing amino-acid 41) was dissolved in DMF and DIPEA, and the resulting solution was added slowly to the coupling reagent (HATU) in order to affect the macrolactamization and form lorlatinib (5).

Scheme 6. Deprotection sequence to form the macrolactamization precursor (41). Despite its versatility and widespread use, HATU is obviously not an ideal coupling reagent for macrolactamization as it is potentially explosive (29, 30). In addition, HATU-related byproducts, such as PF6 residues and TMU, can be difficult to both detect and remove. Given these issues, 38 commonly used coupling reagents were screened in order to identify alternatives to HATU as well as evaluate alternative solvents (Figure 5). As the Spotfire results depict, it appears that the results of the screen are extremely promising with both a range of coupling reagents and solvents appearing to be effective. When DCE was utilized (note that DCE is a Class 1 solvent and, as such, is not recommended for the final step of an API synthesis), the reactions were evaluated with and without HOAt as an additive. Drilling down into these results, COMU and T3P appeared to be feasible alternatives, although we were probably most excited to see EDCI emerge as a promising option. This reagent offered the potential to greatly simplify the workup. Subsequently, EDCI was evaluated in a series of experiments, and was shown to function relatively well, with the maximum yield obtained being 35%. The low yields were attributed to solubility issues when the solvent was switched from DMF, and highlighted a common obstacle in replicating microscale reaction screening results. These are typically run under highly dilute conditions, and need to be translated to a realistic concentration for preparative scale work. The workup for EDCI was simpler than for a HATU/COMU amidation, although yields further decreased on scaling up when the reaction formed a globular insoluble mass. Upon further development, this may lead to an improved protocol in avoiding the current requirement for slow addition of the reagent solution, with instead slow dissolution of material leading to reaction taking place in a controlled manner. However, at this time, the EDCI-mediated reactions were extremely sluggish and HATU was retained as the coupling reagent for now.

36

Figure 5. Screening of coupling reagents for the macrolactamization to form lorlatinib (5).

Scheme 7 summarizes all the optimization work as it was implemented for the synthesis of the initial multigram batch of lorlatinib (5). The yields from the original screens held up well with the key optimization being in the Pd-mediated aryl–aryl coupling to form 39. However, on closer inspection of this reaction, there were significant amounts (up to 20%) of homocoupled products in this reaction. The boronate ester 38 was isolated as a crystalline solid, thus enabling an ee upgrade from the DIP-Cl derived starting material. However, a concern of utilizing this as a starting material was the potential for it to be a genotoxic impurity. Aryl boronic acids have been found to be weakly mutagenic in microbial assays (31). Scavenging or more typically protodeboronation are used to remove these mutagens. Careful monitoring would be required during the synthesis, as controlling the residual levels of such boron-related species in APIs could become a regulatory requirement. The HATU-mediated ring closure of 41 was run under high dilution conditions with reverse syringe-pump addition of the substrate solution to the reaction. The reaction was exceedingly rapid, and was essentially finished upon the completion of the addition of the substrate. The workup, however, was highly labor-intensive, consisting of initial removal of DMF, taking up the residue in EtOAc followed by multiple base washes (5–7) to remove PF6 residues, and then column chromatography to remove the amide impurity formed through nitrile hydrolysis. A final trituration with water was then carried out to remove both residual solvents and TMU. A preliminary evaluation of salt formation was carried out at this time, with the hydrochloride identified as a promising final salt form. 37

Scheme 7. Initial scale-up of lorlatinib (5). The overall synthesis here provided a 33% yield over six steps compared to the initial route of 11% over four steps. While longer, the main advantage of this approach was the significant reduction in the amounts of starting materials required (Figure 6) to obtain the target amount of compound for testing.

Figure 6. Starting material amounts utilized for the initial scale-up of lorlatinib (5). Despite the fact that we had a synthesis in hand capable of providing gram quantities of lorlatinib (5), there were still issues that needed to be addressed: • The synthesis of the requisite starting materials was suboptimal in terms of efficiency. For the pyrazole 24, dimethyl sulfate was used for the methylation and a relatively low-yielding free-radical bromination was utilized. The syntheses developed thus far also provided a compound with the 4-bromo substituent in place on the pyrazole ring. An alternative approach (vide infra) at this point was being developed initially for cyclolactam candidate 6, which required the des-bromo compound at the analogous point. Although this could be achieved through Pd-mediated des-halogenation, this obviously added an additional step. It was also capricious with some reduction of the nitrile to the corresponding primary amine. For the aminopyridine (17), the DIP-Cl-mediated ketone reduction was effective in terms of both yield and ee. However, it did come with an arduous workup, and as such an alternative approach was desirable. Furthermore, several yields in the current sequence could be improved upon. • The homocoupling in the aryl–aryl bond formation step presented a challenge particularly as this reaction had undergone a degree of optimization already. Clearly a better understanding of this reaction was required, although an alternative approach to form the macrocycle might eliminate this issue. • The unwanted nitrile hydrolysis during the conversion of the ester 40 to the carboxylic acid 41 prior to ring closure was a real issue for the final purification. Further options to execute the desired ester hydrolysis without concomitant nitrile hydrolysis need to be examined. • Several alternatives to HATU have been identified in preliminary screening experiments. It was a question whether retaining HATU brought sufficient benefits to keep it and, if 38

so, carry out the requisite safety testing to see if this reagent could be utilized on scale. In parallel, we could also continue to evaluate alternatives that would circumvent the solubility difficulties encountered with EDCI. • The ring closure was run under highly dilute conditions with a slow addition of substrate 41 to reagent. Obviously these are not ideal either operationally or more importantly from a material throughput perspective. An alternative approach to macrocyclization through tethering/templating or developing a more-efficient ring-forming reaction would be the best solution to this. • Efficient purification of the final API was still problematic. • Salt forms had been briefly investigated; however, while the HCl salt appeared best, additional screening needed to be done. Initial Improvements in the Synthesis of N-Methylpyrazole A modified synthesis of the requisite pyrazole tailpiece 24, which was initially scaled to 20 kg, is shown in Scheme 8. This synthesis starts from readily available Boc-protected N-methylglycine (Boc-sarcosine) 42, and proceeded via initial activation and formation of the Weinreb amide. This was trapped with methylmagnesium bromide to generate the methyl ketone 43 in excellent yield. Condensation with dimethyloxalate led to the diketone 46, which was treated with methylhydrazine (thus avoiding the hazardous dimethyl sulfate) to provide predominantly the desired regioisomer of the pyrazole 47, which was purified by crystallization. Amminolysis and dehydration as described previously afforded the desired pyrazole 49, which can be elaborated to either the brominated compound 24 or deprotected to give the amine 50. This new route realized a significant cost-saving, and provided, as we will see, a common intermediate for both the macrolactamization and direct arylation-based approaches.

Scheme 8. Modular scalable syntheses of the pyrazole tailpieces. With regard to the synthesis of the aminopyridine (17) portion of the molecule at this time, the asymmetric reduction of the acetophenone moiety offered the best opportunity in terms of finding an improved method to achieve this. While initially investigating analogues within the macrocyclic series, we utilized a stoichiometric DIP-Cl based reduction. This reagent was mainly selected due to its ability to deliver either enantiomer of the desired alcohol with relatively high enantiopurities for subsequent elaboration to either the aminopyridine or aminopyrazine core through either clean inversion or retention at the chiral center, respectively. We focused our attention on the (S)enantiomer, which was likely to be required for the synthesis of lorlatinib (5).

39

The problem in adopting this approach was that we had already demonstrated that the asymmetric boron-based reduction had worked for the current substrate, wherein it had failed with respect to the ortho,ortho-disubstituted acetophenone required for crizotinib. However, we were drawn at this time specifically to the body of work that had been done on the development of a ketoreductase to achieve the reduction (32). Two biocatalytic systems were developed to reduce the acetophenone with high purity and ee. Both systems delivered only the S-enantiomer that proved compatible with the required inversion approach to forming the ether linkage. The enzymatic system that progressed into development for the successful reduction of 2,6dichloro-3-fluoroacetophenone (this is the ketone substrate required in the synthesis of crizotinib) was based on a lysate of 2,4-diketogluconic acid reductase (DkgA), derived from Proteus mirabilis recombinantly expressed in E. coli. Mutants of DkgA obtained by engineering the wild-type sequence were prepared, enabling the reduction to be carried out at substrate concentrations of > 100 g/L. The initial experiments indicated that the enzyme could tolerate a concentration of 50 mg/mL. But conversion stalled at 70%, with either product inhibition or solubility suspected of being a problem. Addition of IPA as a cosolvent (up to 15%) enhanced the activity, indicating solubility was the issue. In addition, IPA by its conversion to acetone was also capable of functioning as a cofactor for recycling NADPH, as opposed to the previous system which utilized glucose. This highlighted the importance in the development of biocatalytic reductions in optimizing both the enzymatic systems in synergy with the cofactor recycling here achieved using a commercial preparation of Lactobacillus brevis ADH from X-zyme. The original reaction was successfully scaled to 50 g/L, but when increased to 100 g/L, it stalled at 88% conversion. This led to a further round of optimization through bio-engineering, with 475 mutations being tested, and with 33 substitutions leading to enhanced activity. Solvent-exposed sites were mutated to a charged amino acid with the Qd11 mutant displaying 2.66 times the specific activity of wtDkgA. Most importantly within the context of the current discussion, the optimized biocatalytic system was demonstrated to have a broad substrate scope with compounds possessing a single orthosubstituent (11) appearing to be even better substrates. The desired product from the enzymatic transformation is easily isolated through extraction and concentration. As noted, this approach led only to the (S)-enantiomer, although asymmetric hydrogenation does potentially enable access to either enantiomer. This again was investigated during the development of crizotinib; and despite limited hits, Ru-mediated hydrogenation using the Naud family of ligands was identified as optimal with the ligands available in both orientations (33). Through sequential optimization, catalyst loadings were reduced while maintaining good conversions; and although the ee was not as high as obtained through the biocatalytic process, we demonstrated it could be upgraded through subsequent crystallization. This approach, although not investigated at this time, may represent a viable alternative for the asymmetric synthesis of the precursors to the macrocycles; however, there are concerns regarding the potential for deshalogenation of the iodide during the hydrogenation. With the enzymatic reduction in hand, the downstream chemistry steps were very similar to those that we have discussed previously. Optimization of the ether bond formation was achieved through telescoping the mesylate preparation utilizing a single solvent (acetone) for both reactions to give 15 (Scheme 9). The Pd-mediated carbonylation was followed by bromination at low temperature to yield 17. Following Boc protection, the conditions from the aforementioned

40

borylation-screen were employed to afford the crystalline boronate ester 38 as the starting material for the synthesis of lorlatinib (5).

Scheme 9. Modified synthesis of the aminopyridine headpiece (38). Initial Development of a Second-Generation Macrocycle Synthesis Given the relatively competitive landscape around the development of next-generation ALKinhibitors, one of the early initiatives within the program was to investigate shorter sequences to access the macrocyclic framework. This would not only expedite access to new analogues, but also potentially enable a synthesis of an array of such compounds in a combinatorial fashion for SAR examination. With the headpiece assembled, we could once again investigate the potential disconnections to access the molecules of interest, with an initial focus on the aminopyrazine portion as exemplified by 6 (Figure 7A). There were a number of options available to try different chemistry, and a series of observations from previous experiments on the project served as a driving force to pursue these.

Figure 7. Approaches to the pyrazine-based macrocycle (6). Experimental results supporting the pursuit of a direct arylation-based approach. Firstly, we asked would it be possible to generate the amide bond in a different fashion through a carboamidation-type process. With a suitable headpiece, and the desired selectivity potentially between two dissimilar halogens, this would enable us to hypothetically pursue a two-step (carboamidation/ring closure) approach to the macrocycles. This is shown in the intermolecular carboamidation demonstrated in Figure 7B. Selectivity is achieved with the iodide in aryl fluoride 41

11, while the bromide present in the pyrazole 51 is left intact. For construction of a macrocycle, we would likely need a halogen at C-5 of the pyrazine (or pyridine). The second question was how to achieve the ring closure with either a Suzuki-type reaction or direct arylation. In the case of the pyrazine series, formation and subsequent stability of a boronate adjacent to a nitrogen may prove to be problematic, and this made a CH-activation type reaction an attractive alternative (34). The viability of this approach was supported by studies within our acyclic ALK inhibitor series, wherein a thiazole underwent a direct arylation with a brominated aminopyridine to yield 53. Screening demonstrated the following conditions — CataCXium A/Pd(OAc)2, DMAc, KOAc, PivOH, 110 °C for 16 h — and led to approximately 30% conversion to the coupled compound (Figure 7C), with the addition of CuI increasing this to > 50%. However, reproduction of these conditions on preparative scale provided only a 16% isolated yield of the desired target 53. Extension of this approach to other acyclic analogues led to limited success. It was thought that tethering the amide, and carrying out this reaction in an intramolecular fashion, may lead to better results. Thus, the intermediate triazole (R = propyl) 55 was formed via an in situ “click” reaction performed in flow using a Cu reactor (Figure 7D). Subsequent ring closure via a Pd-mediated process led to macrocycle 56 in 30% yield (CataCXium A/Pd(OAc)2, DMAc, KOAc, 100 °C for 16 h). The similarity of conditions for the inter- and intramolecular direct arylation processes to yield 53 and 56, respectively, presented us with a starting point to test on the aminopyrazine if a suitable precursor could be accessed. An additional feature that may work in our favor is the presence of the adjacent nitrile substituent in both 5 and 6. The ability of the cyano-group to complex Pd and direct C–H activation type processes has been demonstrated most notably by Sun et al. (35, 36). This concept has been further reinforced here in our work on the acyclic series, wherein the successful thiazole 53 has a nitrile adjacent to the site of reaction. The preliminary results derived in evaluating a two-step approach to 6 are highlighted in Scheme 10. The desired headpiece 57 was accessed through a displacement reaction on 3,5dibromopyrazine, using the enantiomer of the benzylic alcohol 12 (this reaction proceeds with retention of stereochemistry). The synthesis of the two (X = H or Br) pyrazole-based tailpieces are described in Scheme 8. The first challenge to surmount was to achieve selectivity for the bromide over the iodide in the carboamidation of 57 with 9. This could be readily achieved, although it should be noted that when a similar carbonylation reaction was carried on dihalo 57 to access the corresponding methyl ester, it often led to a mixture of the mono- and bis-esters. As suspected, the intramolecular Suzuki coupling worked poorly. Pleasingly, a slight modification of the conditions from the previously successful direct arylation process for the triazole macrocycle 56 led to a reasonable yield of 6. However, there were issues with reproducibility, which we hoped to resolve while obtaining an increased yield through an exhaustive screening campaign. In hindsight, with regard to our investigations into this transformation, we were somewhat fortunate as the data from much of our screening effort was of limited utility owing to the choice of solvent. Our screen was split into the following four components: • 72 catalyst/ligand combinations using DMAc/KOAc at 100 °C. • CataCXium A in DMAc with 20 different bases at three temperatures (60, 80, and 100 °C). • 12 ligands (representative of all the major ligand families utilized in Pd-mediated reactions) in DMAc or dioxane with three bases. 42

• CataCXium A and KOAc examining 12 different solvents. Several conclusions were reached based on the outcome of this work. The most critical was that solvent was the key variable (Figure 8). The standard solvent DMAc and related solvents, such as DMF and DMSO, performed poorly in this reaction; whereas t-AmOH was the best (37). Primary and secondary alcohols also worked, although debromination to give 60 was a significant side reaction. This latter issue caused some early disconnects in interpretation of our data in that the desbrominated product 60 eluted with the desired product 6. Once this was recognized, the results could be adjusted for coelution using mass ion extraction to segregate the components of the peak. The other byproduct observed in the reaction was the des-cyano compound 61.

Scheme 10. Suzuki coupling vs direct arylation to form pyrazine-based macrocycle (6). Compound 6 was subsequently scaled to ~50 g quantities to supply material for toxicity studies as illustrated in Scheme 11. The requisite (R)-alcohol 62 is obtained through a DIP-Cl reduction and introduced into the headpiece through a SNAr-displacement reaction to give 57. Carboamidation with 50 and direct intramolecular arylation utilizing a stainless-steel pressure vessel led to 6. Purification was achieved through chromatography followed by slurrying the solid in aqueous methanol. The major impurity was the des-cyano compound 61, which required chromatography to purge. Use of a pressure vessel was potentially viewed as a limitation to this chemistry, although either switching the reaction conditions to a higher-boiling alcohol as solvent or investigation of a flow-based approach may alleviate this. The pyrazole 50 was synthesized as shown in Scheme 11. It is interesting to note that previous syntheses of this pyrazole had been through Pd-mediated des-halogenation of pyrazole 24 from Scheme 3. Care had to be taken to avoid reduction of the nitrile functionality with this approach. Despite the application of this chemistry to scaling 6, as well as the efficient synthesis of a series of related analogues based on a better selectivity profile, lorlatinib (5) was chosen as the development candidate at this time based on a better selectivity profile for ALK over TrkB, which was believed would mitigate against adverse effects within the CNS. 43

Figure 8. Solvent screening for the direct arylation to form (6).

Scheme 11. Direct arylation proof of principle. Gram scale synthesis of (6). As such, we looked to see if we could apply the direct arylation for cyclization within the aminopyridine series. The initial attempts to execute this chemistry are shown in Scheme 12. Compound 15 is accessed in three steps from readily available starting materials, and elaborated through carboamidation and bromination using NBS to provide the precursor aminopyridine 61. However, applying the conditions developed from the screen led to only the des-halogenation being detected. Protection of the amine as the bis-acetamide 62, followed by cyclization under slightly modified conditions (optimization of temperature and concentration) and acid-mediated deprotection, led to the desired macrocycle 5 in modest yield. Although the protection/deprotection sequence added steps to the overall reaction sequence (13% yield over 8 steps), it did have a secondary advantage of preventing chelation of Pd to the aminopyridine core, and also ensured that Pd was not used in the final API step, thus derisking the potential for contamination. 44

Scheme 12. Initial application of a direct arylation approach for the synthesis of lorlatinib (5).

Early Development Chemistry for Lorlatinib (5) Given the relatively competitive landscape, the program was moved into development with the intention to advance at an accelerated pace to supply lorlatinib (5) for regulatory toxicology and preclinical studies. With this goal in mind, the preparation of ~400 g was required. Although initial proof of concept of a direct arylation was in hand, we decided that a more valid strategy would be needed to meet the aggressive timelines. We carried out a systematic evaluation of the original macrolactamization approach to identify the synthetic shortcomings and enable suitable modifications of this chemistry to meet our delivery goals. Figure 9 provides a modified overview of the macrolactamization route, highlighting the issues (chromatography and HATU use) which we felt were critical to address in our exploratory development studies (38).

Figure 9. Potential issues with the macrolactamization route to address with the enabling studies. 45

The enabling efforts were aimed at seeking solutions to the following issues: 1. In the cross-coupling step, formation of the homocoupled byproducts (63). 2. During the methyl ester saponification, the formation of the amide impurity 65 arising from competitive hydrolysis of the nitrile group. 3. The lack of any crystalline intermediates before the final API-forming step necessitated the use of silica gel chromatography to purge impurities. 4. The use of the highly energetic HATU in the macrolactamization raised process safety concerns (39). We continued to utilize the the bis-Boc-protected derivative 38 in the cross-coupling reaction (40). This permitted us to reduce the catalyst loading of PdCl2(dppf) to ~ 3–5 mol% for producing to C–C coupling product 39. However, the homocoupled impurity 63 was still formed at significant levels despite strenuous efforts to exclude oxygen from the reaction system (41). After a brief reaction screening targeted to minimize this impurity formation, we discovered that the best method was to feed a toluene solution of 38 and 24 slowly to the mixture containing PdCl2(dppf) and CsF in toluene/water at reflux (42). This protocol successfully minimized the formation of 63 to less than 3% (UPLC area) during the scale-up. We determined that 63 could be purged effectively without resorting to chromatography here or downstream. Thus, the crude product 39 was carried into the Boc deprotection performed with gaseous HCl in iPrOAc (replacing the previously used HCl/dioxane). As the predicted pKa values of the products 64 and 40 were 3.8 ± 0.1 and 6.4 ± 0.1, respectively (ACD Labs), this pKa differential was sufficient to allow separation of the two by aqueous extraction under different pH conditions. Therefore, after the Boc-deprotection, the pH of the reaction mixture was adjusted to 4.5 to bring the ester 40 into the aqueous phase. Meanwhile, dimer impurity 64 remained in the organic iPrOAc phase. The pH of the separated product-rich aqueous phase was subsequently adjusted to 9–10, and submitted to an extractive workup with iPrOAc. This protocol resulted in complete purge of 64 from the isolated product. The product loss from the extractive manipulation was estimated to be less than 2%. After a solvent switch from iPrOAc to acetonitrile, 40 was subjected to the methyl ester hydrolysis. Potassium trimethylsilanoate (TMSOK) was the reagent of choice, as it gave a highly chemoselective hydrolysis and left the nitrile group intact (43, 44). In addition, product acid 41 directly precipitated out of the reaction mixture as a crystalline potassium salt, and was isolated in 66% yield in a three-step telescoped process without chromatography. The crystalline intermediate offered a convenient purging opportunity, as the product isolated was > 98% purity (UPLC, 210 nm). For the macrocyclization reaction, we turned to the use of COMU, as it was considered both a more process-friendly and efficient peptide-coupling reagent than HATU (45). Thus, a DMF solution of 41 was slowly added to a solution of COMU in DMF/THF at 35 °C. The slightly elevated reaction temperature provided for faster reaction kinetics, which kept the substrate concentration low. The reaction was aided by the large volume of solvents (50 L/kg total) and the slow addition employed with this protocol that minimized competitive intermolecular coupling to the dimer. Filtration of the reaction mixture through a silica gel pad removed N,N-dimethylmorpholine4-carboxamide, a byproduct of COMU. The free base form of 5 was converted to its acetic acid solvate (the initial development form) in a 46% yield over the last two steps (46). These process 46

improvements increased the overall yield (from 38, Scheme 13) to 30%, and resulted in the delivery of the preclinical batch (438 g) during the first scale-up campaign.

Scheme 13. Scale-up macrolactamization route to support preclinical studies. After the successful delivery of the preclinical batch, the first GMP campaign was immediately initiated. We encountered a reagent quality issue for the bulk TMSOK that was ordered, leading to the reoccurrence of a significant amount of nitrile hydrolysis byproduct 65. We discovered that this was due to the batch of TMSOK containing a high level of potassium hydroxide; the majority of commercial suppliers list TMSOK as a technical grade at approx. 90 – 92% purity. We discovered that the most effective purification protocol for TMSOK was to dissolve TMSOK in acetonitrile, and pass the resulting solution through a plug of neutral aluminum oxide. The resultant TMSOK solution could be employed directly for the reaction to give a clean hydrolysis of the methyl ester during the scale-up. Since the cyclization step suffered a relatively large yield loss (yield 46–56%), we conducted a further screening of coupling reagents, and found the reaction rate to be on the order of HATU > COMU > T3P > TPTU ≅ TSTU (47). A faster reaction could be expected to reduce the substrate concentration, and therefore lower the competitive intermolecular coupling impurities. We revisited the possibility of using HATU in our internal scale-up facility. In order to do this, a comprehensive series of process safety testing per UN guidelines was initiated on both solid HATU and its 20% w/w solution in DMF, with the results summarized in Table 1 (38). Table 1. Summary of HATU Test Resultsa Test

HATU

20% w/w HATU in DMF

Koenen Test – UN 1 (b)

Positive

Negative

Koenen Test – UN 2 (b)

Negative

Negative

UN Gap UN 2 (a)

Negative

Not tested

Time/Pressure UN 2 (c) (i)

Negative

Not tested

BAM Fallhammer–Test UN 3 (a) (ii)

Negative

Negative

BAM Friction Test – UN 3 (b) (i)

Negative

N/A

Flammability of Solids Test UN N.1

Negative

N/A

a Conducted at Chilworth.

47

With regard to these tests, the Koenen test utilizes a tube apparatus and assesses the sensitivity of a material to the effect of intense heat while under confinement. As noted, solid HATU had a positive result for Koenen Test UN 1 (b), but a negative outcome for Koenen Test UN 2 (b), demonstrating that the sample shows “some effect” upon heating under confinement but not a “violent effect.” Given the negative results from the UN Gap Test UN 2 (a) and the Time/Pressure Test UN 2 (c) (i), solid HATU can be rigorously excluded from classification as a UN Class 1 explosive. For the BAM Fallhammer Test UN 3 (a) (ii), which is used to assess the sensitivity of a sample to drop-weight impact, solid HATU showed an “explosion” at 60 J and decomposition at 50 J with no impact tests performed below this impact energy. The limiting impact energy observed of 60 J is considered to be a negative test result, as it is well above the positive test criteria of “explosion” at 2 J or less. The BAM Friction Test – UN 3 (b) (i) showed solid HATU to be insensitive to friction below a limiting load of 80 N, while the sample also passed the Solid Flammability Test and can therefore be excluded as a flammable solid of UN Class 4, Division 4.1. In addition to process safety test on solid HATU and its 20% w/w solution in DMF, thermal stability assessments using TSu (thermal screening units) of the reaction mixture at various stages and the workup process streams indicated that the reaction was safe to be scaled. As in the preclinical campaign, the solution of 41 in DMF was added over a course of 14 h to the reaction mixture containing HATU in DMF/EtOAc at 35 °C. After an aqueous workup, the crude product was treated with acetic acid to form the desired API form (acetic acid solvate). This crystallized directly from the reaction mixture in 99.2% purity with no single impurity greater than 0.2%. From this campaign, 1.55 kg of the API was obtained in 56% overall yield for the last two steps (cyclization and final form conversion); as compared to a 46% yield from the preclinical campaign where COMU was used. The use of HATU also represented a process improvement, as it allowed the elimination of the only silica gel operation in the sequence and it increased the overall yield from 30% to 37% (from 38). As macrolactam 5 showed promising results in first-in-patient (FIP) studies, the API demand to support clinical developments started to escalate. After assessing the first campaign route for its suitability to further scale-up, we recognized that it might be more sustainable to reverse the bond-formation order (amide first, then aryl–aryl cyclization). This would shorten the synthesis and achieve a potentially higher throughput. We thus revisited a direct arylation-based approach. This alternative route (Scheme 14) for the C–C coupling was one step shorter in the GMP endgame synthesis. Additionally it does not require the introduction of the boronate ester and the bromo groups for the coupling partners 38 and 24, respectively. While our safety group had vetted the use of HATU in our facility, this route also removed the concerns regarding the potentially highenergy nature of the reagent.

Scheme 14. Second-generation synthesis of lorlatinib (5). 48

Thus, we embarked on the development of the second-generation synthesis, with the acid 66 being readily accessed through saponification of the previously utilized methyl ester 37. Amidation of 66 with 50 using T3P provided 67 (48). A preliminary screen led to promising conditions (Pd(OAc)2, CataCXium A, and KOAc in t-amyl alcohol at reflux) for the C–C cross-coupling to give 68 as the penultimate intermediate (49). This was followed by deprotection of the Boc groups and conversion to the final form to give the API 5. Before implementing this route for the next campaign, we needed to address the robustness of the critical C–C cross-coupling step and manage the impurity profile of the final API. Intermolecular C–C bond formation as a competitive pathway is a major problem for macrocyclization reactions, and it is commonly managed by dilution. We found that reaction was optimized at ~35 L/kg (solvent/substrate 67), with comparable results being obtained through feeding the substrate solution slowly to the reaction mixture at reflux. Another challenge was the presence of a process impurity des-cyano 69 formed in the reaction at various levels (2–6%). This impurity was inefficiently purged when crude 68 was telescoped into the Boc deprotection for the final API isolation. The mechanism for its formation is not clear, although it is thought to be related to Pd-CN interaction (50, 51). Further high-throughput screening of the reaction conditions did not result in the exclusion of 69. To our delight, the cross-coupling product 68 could be isolated as a highly crystalline intermediate, and the process impurity 69 was rejected conveniently by recrystallization from ethyl acetate, ethanol, 2-propanol, or acetonitrile with high recovery. With the confidence in this process, we started the second GMP campaign from advanced intermediates diBoc 66 and pyrazole 50. The amidation reaction proceeded uneventfully using T3P as the coupling reagent. The intermediate product 67 was obtained crude as an oil, which was used directly in the following coupling step as a solution in t-amyl alcohol after a solvent exchange. The reaction was initiated with 20% of the substrate in the pot from the outset, as we found the practice beneficial in keeping the active palladium catalyst in the solution phase while the remainder of the substrate was fed into the batch over 5–6 h (52). The cyclization step was split into three batches for the initial scale-up, and it was disappointing to note that the first two batches stalled at ~70% and ~90% conversion, respectively. Resubjecting the recovered batch under the same conditions resulted in complete conversion. It is noteworthy that the quality of CataCXium A used in the production might have played a role in the stalling (53). The third batch was started with a 50% higher loading of the catalyst/ligand, and it reached completion uneventfully. The three batches were combined in the workup and isolation. The crude product was recrystallized from acetonitrile, which reduced the des-cyano impurity 69 to 0.31% from the original 6.0%. As 68 was determined to be the best stage for palladium residue control, it was treated with SiliaMetS® Thiol at reflux in ethyl acetate. The product isolated from ethyl acetate recrystallization was of 99.9% UPLC purity with 69 at 0.1% level and a palladium content of 17 ppm. 4.74 kg of 68 was obtained in 44% overall yield from 66 in two bond-forming steps featuring two recrystallizations. Finally, Boc deprotection was achieved with conc. HCl (aq). After an aqueous extractive workup, the solution of 5 in ethyl acetate was treated with acetic acid to give the final API as an acetic acid solvate (3.14 kg, 86% from 67, or 38% from 66 overall) in 99.90% UPLC purity with only a single impurity (70) detectable at 0.10%. The second-generation synthesis not only delivered

49

a substantially higher throughput, but also met tighter API acceptance criteria at the developmental stage. As we encountered robustness issues with the cyclization step on scale-up, we revisited the reaction and found the addition of 2,6-lutidine (2 equiv) and water (2 equiv) as additives were beneficial to the reproducibility of the reaction to form 68. On multigram scale runs, the in situ yield of the reaction could reach as high as 70%. The isolated yield was consistently around 50–54%, although the des-cyano impurity (69) still formed at ~3%. For the removal of the Boc groups, we envisioned the process could be further streamlined under thermal conditions in flow with a reaction screen conducted to demonstrate that the deprotection can be effected cleanly at 220 °C with a 20-min residence time in either tetrahydrofuran or 2methyltetrahydrofuran solutions containing 3–5% water (54). The reaction resulted in > 95% isolated yield of 5 on multigram scales.

Commercial Process for the Synthesis of Lorlatinib (5) The process illustrated in Scheme 14 successfully made several 10–20 kg batches for clinical supplies. However, the long linear synthesis of preparing the starting acid 66 was a concern. As lorlatinib (5) moved into later stage clinical trials and commercialization, the quantities required were greatly increased and a more efficient synthesis of the precursor for the macrolactamization 67 was required. After further consideration of the various possible bond-forming sequences and the chemistry that could be used for each, the decision was made to use the pieces shown in Figure 10. The lactone 71 would be ring-opened with amine 50. The resulting hydroxy group could be displaced with the hydroxypyridine 72 using a sulfonyl transfer SN2 reaction. This would provide the key intermediate 67 quickly without the numerous functional group manipulations required previously. The ring could then be closed as before with Pd catalyzed C–H insertion/cyclization, followed by removal of the Boc-protecting groups. If successful, this route would meet the goal of minimizing the number of functional group manipulations during the formation of the 12-membered ring.

Figure 10. Proposed starting materials for the commercial route. Approaches to the Chiral Lactone 71 The syntheses examined to this point demonstrated that the enantiomeric purity of lorlatinib (5) will be solely determined by the chiral purity of the benzylic position. As we now envisioned chiral lactone 71 delivering the stereocenter, we determined it could be accessed either through a carbonylation or lactonization reaction (55). This approach enabled us to evaluate a range of functional groups at the para-position to the fluoro group of the acetophenone for their commercial accessibility/cost, as well as their performance in a range of different asymmetric reductions. 50

Benzylic alcohol 12 (Scheme 9) provided the precursor in order to evaluate the carbonylationbased approach and can be accessed with high enantiopurity either through a chemical or enzymatic reduction of the commercially available acetophenone 11. As noted, the DIP-Cl based approach, although successful, proved challenging to scale, owing to the workup with diethanolamine. This was needed in order to break up the borate intermediate formed in the reaction mixture. Studies showed that this was critical to fragment this completely in order to avoid it negatively impacting the downstream chemistry. The removal of the subsequently formed diethanolamine borate complex was achieved through filtration, although this process was extremely slow, even with the addition of a filter aid. Finally, given the low melting point of 12 and the need to purge the α-pinene from the crude product, recrystallization had to be carried out under cryogenic conditions leading to a lower yield of 67%. Use of CBS catalyst as an alternative to the stoichiometric-based DIP-Cl reduction was demonstrated on laboratory scale and led to both good yield and ee. Unfortunately, it was deemed to be not as cost-efficient on industrial scale. In contrast, the 2,4-diketogluconic acid (DkgA) enzymebased process proved to be successful for preparing 12, and was scaled to 350 kg. The reaction was shown to reach > 99% conversion without the need to remove acetone from the system. The product was easily isolated through precipitation from the IPA/water reaction mixture, followed by a recrystallization under noncryogenic conditions. The desired lactone 71 could be accessed through a Pd-mediated CO insertion reaction, which was found to proceed smoothly using Pd(dppf)Cl2 as the catalyst under 50 psi CO at 80 °C (Scheme 15). Preforming the catalyst was also shown to be advantageous with the reaction rate moderated through the use of catalyst loadings as low as 0.05 mol%. Control of temperature was also critical in terms of both controlling the reaction rate while also ensuring the reaction reached completion. Additional dppf (0.025 mol%) proved to be beneficial in preventing degradation of the catalyst and the formation of palladium black. Interestingly, solvent played a role in terms of the enantiopurity of the product with alcohols leading to an erosion of up to 10% in chiral purity. This loss was believed to arise not from a deprotonation event, but from a Pd-mediated β-hydride elimination/reduction. This can be avoided by running the reaction in either THF or 2-MeTHF, with the latter preferred, owing to ease of extractive workup due to its immisicibility with water. The enzymatic reduction/carbonylation process was successfully scaled up with > 1000 kg of 71 produced through this sequence with both high yield (90%) and purity (> 99.8 HPLC area % and > 99.9% ee).

Scheme 15. Carbonylation-based approach to lactone 71. Despite the success of this carbonylation-based approach, several drivers still existed for finding an alternative synthetic route to access the key lactone intermediate 71. First of all, the route was not particularly cost-efficient. Although the acetophenone 11 was commercially available, it was synthesized in 4–6 steps from commodity chemicals. As such, the potential for further cost reductions for accessing 11 was limited.

51

The use of the chloro- or bromo-fluoro analogues of 11 would represent a possible alternative. However, these materials are likely to still be made in a similar manner to 11. In addition, more forcing conditions would likely be required to promote the Pd-mediated insertion for the carbonylation reaction due to the less active halogen. This step was also problematic as CO gas is poisonous and thus requires special handling on scale. This would limit the number of suppliers able to perform this chemistry. With this in mind, a second approach was developed in which the carbonyl group is already in place para to the fluoro-substituent, with the acetyl group being installed through a directed ortho-metalation. 4-Fluorobenzoyl chloride 73, which is commercially available on large scale, was chosen as a suitable starting material. The successful macrolactonization approach to access 71 from 73 is shown in Scheme 16.

Scheme 16. Asymmetric transfer hydrogenation/lactonization to form 71. The key to the success of this approach was the selection of the diisiopropyl amide optimal directing group (initially the 2-oxazoline was evaluated). The diisopropyl amide not only improved the ortho-selectivity and the ability of the metalation step to proceed to completion, but it also facilitated the downstream asymmetric reduction and cyclization steps. The ortho-lithiation was achieved using n-BuLi in THF at ₋70 °C, with reverse addition of the amide to the base being employed in order to suppress formation of the undesired meta-isomer. Although quenching with the Weinreb amide led directly to 76, the high cost of this reagent was not justified. Instead trapping the lithiated species with a THF solution of acetaldehyde, followed by TEMPO oxidation, led to the desired acetophenone 76 at much lower cost. Two alternative biocatalysts were identified for the successful reduction of 76 to provide the (S)-alcohol 77 on laboratory scale with both high yield and ee, but cost analysis demonstrated that an asymmetric hydrogenation process was more favorable. Initial screening of the Ru-mediated asymmetric hydrogenation indicated that DPEN-type ligands were uniquely effective, while both the BINAP and SEGPHOS-families of ligands led to poor conversions (< 10%). Solvent had some impact, with IPA being optimal with RuCl[(S,S)-TsDPEN](p-cyemene) selected as the best catalyst. It was hypothesized that the bulky diisopropylamide was a key contributor to the high enantioselectivity observed in the reaction. Upon scaling up, the limiting factor for the hydrogenation proved to be the batch size that could be accommodated in the high-pressure reactor. Thus, we pursued an alternative atmospheric pressure asymmetric transfer hydrogenation in which a combination of formic acid and triethylamine was the hydrogen source. The conditions for this 52

procedure were developed based on those for the high-pressure hydrogenation reaction. The catalyst proved to be even more active here, allowing the loading to be lowered to 0.25 mol%. The reaction was scaled to 125 kg, reaching completion within 2 h at 50 °C, producing the desired product 77 in 94% yield with > 99.9% ee. Lactonization of 77 to form 71 was effected through treatment with HCl. The reaction proceeded through a tetrahedral intermediate and the reaction rate was controlled by maintaining the temperature at 45–55 °C. Although the reaction typically did not reach full conversion (~95% complete after 18 h), the product precipitated from the reaction upon cooling. The purity of the isolated product was readily upgraded through a reslurry in EtOH/aqueous NaHCO3. A key feature throughout this route was the high degree of the crystallinity of all of the intermediates due to the presence of the diisopropylamide group. This greatly facilitated isolations and impurity purges. Preparation of Pyridine Tosylate 72 The preparation of the required tosylate 72 was straightforward (Scheme 17). Treatment of 78 with TsCl and TEA formed exclusively the O-tosylate 79. Weaker bases, such as pyridine, gave a mixture of N- and O-tosylation. Treatment of 79 with Boc2O produced 72 as a crystalline solid.

Scheme 17. Synthesis of derivatized aminopyridine 72. Endgame Synthesis With the requisite intermediates in hand, the stage was set for the key bond-forming steps to access lorlatinib 5 (Scheme 18). There are many examples in the literature of opening lactones with amines using aluminum chloride as a Lewis acid catalyst. We screened a number of solvents for the reaction and found that nonpolar solvents led to very slow reactions due to the low solubility of AlCl3 in such solvents. On the other hand, the high heat of dissolution of AlCl3 in many polar solvents was an alternative problem, especially on larger scale where heat removal is not as efficient as small scale. We initially used CH2Cl2 because it gave a good rate of reaction with no exothermicity problems. That was then replaced by a 2-MeTHF/toluene mixture, by slurrying AlCl3 in toluene then slowly adding 2-MeTHF. The toluene provided enough of a heat sink for heat removal, so that the exotherm during 2-MeTHF addition was controllable. Addition of lactone 71 to the suspension of AlCl3 in 2-MeTHF/toluene mixture was accompanied by a modest exotherm. Adding 50 to the reaction mixture gave no reaction until the addition of TEA to neutralize the HCl. This gave a quick reaction to form 80 with almost 100% conversion. Since 80 was not a crystalline solid, it was kept in solution for the next step. We initially attempted to convert the benzyl alcohol of 80 to its tosylate or mesylate and displace the leaving group with the hydroxypyridine moiety. Tosylation of 80 was extremely poor. Although mesylation worked better, it required a large excess of mesyl chloride and the reaction proved difficult to push to completion.

53

We then attempted a sulfonyl transfer process, as this type of reaction has been demonstrated on a number of azoles as well as phenols (56). Adding tosylate 72 to the solution of pyrazole 80 at low temperature followed by the addition of a solution of t-BuOK in THF initiated the tosyl transfer to the hydroxy group of 80. After the transfer is complete (usually 50 °C induced the alkylation phase of the process producing ether 67.

Scheme 18. Commercial route for the synthesis of lorlatinib (5). Early in development, attempts to crystallize 67 led only to amorphous material. Serendipitously, it was discovered that 67 formed a crystalline solvate with t-amyl alcohol, the solvent used in the next step. This crystallization was an important discovery as it allowed us to purify 67 and purge possible catalyst poisons before the critical palladium-mediated step. The overall yield for this two-step process was typically 65–70%. Previous screening work indicated that the combination of Pd(OAc)2 and CataCXium A demonstrated good activity for the cyclization chemistry. Unfortunately, CataCXium A is both proprietary and expensive. A screen with our library of ligands indicated that di(n-butyl)-tbutylphosphine would produce a catalyst with similar activity. However, di(n-butyl)-tbutylphosphine is highly oxygen sensitive and is commonly supplied as its HBF4 salt, which conveniently is a crystalline solid. It required neutralization with NaOtBu before heating it with Pd(OAc)2 to prepare the active catalyst. Addition of 67 and KOAc followed by heating induced the cyclization to 68 along with three impurities. The cyclic dimer and trimer occurred from intermolecular coupling followed by a ring closure, although these both fortunately purged well in the crystallization. The third was the aforementioned des-cyano impurity 69. This was to some extent controlled by reducing the amount of the catalyst. We discovered that 5 mol% was optimum in terms of providing a good balance of reaction rate while minimizing the level of the des-cyano impurity 69. Cyclic lactam 68 also formed crystalline solvates, which could be used as a means for further purification. Among the candidates, the acetonitrile solvate was chosen since it would have no effect on the next reaction and was shown to be stable to vacuum-drying. At the end of the reaction, the reaction mixture was washed with brine and water, azeotropically dried, and acetonitrile was added to isolate the solvate. The yield for this step was typically 65–70%. For the final deprotection, 68 was heated with HCl in aqueous methanol until both Boc groups were removed. After neutralization, lorlatinib (5) could be extracted with 2-MeTHF. The solution was azeotropically dried, n-heptane was added to reach supersaturation, the mixture was seeded with lorlatinib (5), and more n-heptane was added to complete the crystallization. Lorlatinib (5) was 54

typically isolated in 95% yield with this process having been scaled to produce > 120 kg of API at the current time.

Summary Lorlatinib (5, Lorbrena®) was approved in November 2018 for the treatment of NSCLC for patients who may have received a previous treatment with a tyrosine kinase inhibitor, such as crizotinib (1) (57). In addition, the agent had demonstrated a degree of efficacy on tumors within the brain. The macrocyclic framework and the accelerated timelines of the project posed numerous synthetic challenges to the synthesis chemists in both discovery and development. Established partnerships across lines enabled delivery milestones to be met initially through the macrolactamization route. Early route-scouting and the exhaustive use of reaction screening in both discovery and development enabled a direct-arylation-based approach to be identified to meet the demands for clinical supplies.

References 1. 2.

3.

4.

5.

6.

7.

Lung. World Health Organization. http://gco.iarc.fr/today/data/factsheets/cancers/15Lung-fact-sheet.pdf (accessed February 28, 2019). Ettinger, D. S.; Akerley, W.; Borghaei, H.; Chang, A. C.; Cheney, R. T.; Chirieac, L. R.; D′amico, T. A.; Demmy, T. L.; Govindan, R.; Grannis, F. W., Jr.; Grant, S. C.; Horn, L.; Jahan, T. M.; Komaki, R.; Kong, F.-M.; Kris, M. G.; Krug, L. M.; Lackner, R. P.; Lennes, I. T.; Loo, B. W., Jr.; Martins, R.; Otterson, G. A.; Patel, J. D.; Pinder-Schenck, M. C.; Pisters, K. M.; Reckamp, K.; Riely, G. J.; Rohren, E.; Shapiro, T. A.; Swanson, S. J.; Tauer, K.; Wood, D. E.; Yang, S. C.; Gregory, K.; Hughes, M. Non-Small Cell Lung Cancer. J. Natl. Compr. Canc. Netw. 2010, 8, 740–801. Soda, M.; Choi, Y. L.; Enomoto, M.; Takada, S.; Yamashita, Y.; Ishikawa, S.; Fujiwara, S.; Watanabe, H.; Kurashina, K.; Hatanaka, H.; Bando, M.; Ohno, S.; Ishikawa, Y.; Aburatani, H.; Niki, T.; Sohara, Y.; Sugiyama, Y.; Mano, H. Identification of the Transforming EML4ALK Fusion Gene in Non-Small-Cell Lung Cancer. Nature 2007, 448, 561–566. Solomon, B.; Wilner, K. D.; Shaw, A. T. Current Status of Targeted Therapy for Anaplastic Lymphoma Kinase-rearranged Non-Small-Cell Lung Cancer. Clin. Pharmacol. Ther. 2014, 95, 15–23. Shaw, A. T.; Yeap, B. Y.; Solomon, B. J.; Riely, G. J.; Gainor, J.; Engelman, J. A.; Shapiro, G. I.; Costa, D. B.; Ou, S. H.; Butaney, M.; Salgia, R.; Maki, R. G.; Varella-Garcia, M.; Doebele, R. C.; Bang, Y. J.; Kulig, K.; Selaru, P.; Tang, Y.; Wilner, K. D.; Kwak, E. L.; Clark, J. W.; Iafrate, A. J.; Camidge, D. R. Effect of Crizotinib on Overall Survival in Patients with Advanced NonSmall-Cell Lung Cancer Harbouring ALK Gene Rearrangement: A Retrospective Analysis. Lancet Oncol. 2011, 12, 1004–1012. Cui, J. J.; McTigue, M.; Kania, R.; Edwards, M. Case History: Xalkori™ (Crizotinib), a Potent and Selective Dual Inhibitor of Mesenchymal Epithelial Transition (MET) and Anaplastic Lymphoma Kinase (ALK) for Cancer Treatment. Annu. Rep. Med. Chem. 2013, 48, 421–434. Pao, W.; Chmielecki, J. Rational, Biologically Based Treatment of EGFR-Mutant Non-SmallCell Lung Cancer. Nat. Rev. Cancer 2010, 10, 760–764.

55

8.

9.

10.

11. 12.

13. 14. 15.

16.

Kinoshita, K.; Kobayashi, T.; Asoh, K.; Furuichi, N.; Ito, T.; Kawada, H.; Hara, S.; Ohwada, J.; Hattori, K.; Miyagi, T.; Hong, W. S.; Park, M. J.; Takanashi, K.; Tsukaguchi, T.; Sakamoto, H.; Tsukuda, T.; Oikawa, N. 9-Substituted 6,6-Dimethyl-11-oxo-6,11-Dihydro-5Hbenzo[b]carbazoles as Highly Selective and Potent Anaplastic Lymphoma Kinase Inhibitors. J. Med. Chem. 2011, 54, 6286–6294. Zhang, S.; Anjum, R.; Squillace, R.; Nadworny, S.; Zhou, T.; Keats, J.; Ning, Y.; Wardwell, S. D.; Miller, D.; Song, Y.; Eichinger, L.; Moran, L.; Huang, W. S.; Liu, S.; Zou, D.; Wang, Y.; Mohemmad, Q.; Jang, H. G.; Ye, E.; Narasimhan, N.; Wang, F.; Miret, J.; Zhu, X.; Clackson, T.; Dalgarno, D.; Shakespeare, W. C.; Rivera, V. M. The Potent ALK Inhibitor Brigatinib (AP26113) Overcomes Mechanisms of Resistance to First- and Second-Generation ALK Inhibitors in Preclinical Models. Clin. Cancer Res. 2016, 22, 5527–5538. Marsilje, T. H.; Pei, W.; Chen, B.; Lu, W.; Uno, T.; Jin, Y.; Jiang, T.; Kim, S.; Li, N.; Warmuth, M.; Sarkisova, Y.; Sun, F.; Steffy, A.; Pferdekamper, A. C.; Li, A. G.; Joseph, S. B.; Kim, Y.; Liu, B.; Tuntland, T.; Cui, X.; Gray, N. S.; Steensma, R.; Wan, Y.; Jiang, J.; Chopiuk, G.; Li, J.; Gordon, W. P.; Richmond, W.; Johnson, K.; Chang, J.; Groessl, T.; He, Y. Q.; Phimister, A.; Aycinena, A.; Lee, C. C.; Bursulaya, B.; Karanewsky, D. S.; Seidel, H. M.; Harris, J. L.; Michellys, P. Y. Synthesis, Structure-activity Relationships, and In Vivo Efficacy of the Novel Potent and Selective Anaplastic Lymphoma Kinase (ALK) Inhibitor 5-chloroN2-(2-isopropoxy-5-methyl-4-(piperidin-4-yl)phenyl)-N4-(2-(isopropylsulfonyl)phenylpyrimidine-2,4-diamine (LDK378) Currently in Phase 1 and Phase 2 Clinical Trials. J. Med. Chem. 2013, 56, 5675–5690. Chi, A.; Komaki, R. Treatment of Brain Metastasis from Lung Cancer. Cancers 2010, 2, 2100–2137. Weickhardt, A. J.; Scheier, B.; Burke, J. M.; Gan, G.; Bunn, P. A.; Aisner, D. L.; Gaspar, L. E.; Kavanagh, B. D.; Doebele, R. C.; Camidge, D. R. Local Ablative Therapy of Oligoprogressive Disease Prolongs Disease Control by Tyrosine Kinase Inhibitors in Oncogene-Addicted NonSmall-Cell Lung Cancer. J. Thorac. Oncol. 2012, 7, 1807–1814. Steeg, P. S.; Camphausen, K. A.; Smith, Q. R. Brain Metastases as Preventive and Therapeutic Targets. Nat. Rev. Cancer. 2011, 11, 352–363. Heffron, T. P. Small Molecule Kinase Inhibitors for the Treatment of Brain Cancer. J. Med. Chem. 2016, 59, 10030–10066. Johnson, T. W.; Richardson, P. F.; Bailey, S.; Brooun, A.; Burke, B. J.; Collins, M. R.; Cui, J. J.; Deal, J. G.; Deng, Y-L.; Dinh, D.; Engstrom, L. D.; He, M.; Hoffman, J.; Hoffman, R. L.; Huang, Q.; Kania, R. S.; Kath, J. C.; Lam, H.; Lam, J. L.; Le, P. T.; Lingardo, L.; Liu, W.; McTigue, M.; Palmer, C. L.; Sach, N. W.; Smeal, T.; Smith, G. L.; Stewart, A. E.; Timofeevski, S.; Zhu, H.; Zhu, J.; Zou, H. Y.; Edwards, M. P. Discovery of (10R)-7-Amino-12-fluoro2,10,16-trimethyl-15-oxo-10,15,16,17-tetrahydro-2H-8,4-(metheno)pyrazolo[4,3h][2,5,11]benzoxadiazacyclotetra decine-3-carbonitrile (PF-06463922), a Macrocyclic Inhibitor of Anaplastic Lymphoma Kinase (ALK) and c-ros Oncogene 1 (ROS1) with Preclinical Brain Exposure and Broad-Spectrum Potency against ALK-Resistant Mutations. J. Med. Chem. 2014, 57, 4720–4744. Huang, Q.; Johnson, T. W.; Bailey, S.; Brooun, A.; Bunker, K. D.; Burke, B. J.; Collins, M. R.; Cook, A. S.; Cui, J. J.; Dack, K. D.; Deal, J. G.; Deng, Y.-L.; Dinh, D.; Engstrom, L. D.; He, M.; Hoffman, J.; Hoffman, R. L.; Johnson, P. S.; Kania, R. S.; Lam, H.; Lam, J. L.; Le, P. 56

17. 18. 19. 20. 21. 22. 23. 24. 25. 26.

27.

28. 29. 30.

31.

32.

T.; Li, Q.; Lingardo, L.; Liu, W.; Lu, M. W.; McTigue, M.; Palmer, C. L.; Richardson, P. F.; Sach, N. W.; Shen, H.; Smeal, T.; Smith, G. L.; Stewart, A. E.; Timofeevski, S.; Tsaparikos, K.; Wang, H.; Zhu, H.; Zhu, J.; Zou, H. Y.; Edwards, M. P. Design of Potent and Selective Inhibitors to Overcome Clinical Anaplastic Lymphoma Kinase Mutations Resistant to Crizotinib. J. Med. Chem. 2014, 57, 1170–1187. Driggers, E. M.; Hale, S. P.; Lee, J.; Terrett, N. K. The Exploration of Macrocycles for Drug Discovery–An Underexploited Structural Class. Nat. Rev. Drug Discovery 2008, 7, 608–624. Giordanetto, F.; Kihlberg, J. Macrocyclic Drugs and Clinical Candidates: What Can Medicinal Chemists Learn from Their Properties? J. Med. Chem. 2014, 57, 278–295. Marsault, E.; Peterson, M. L. Macrocycles Are Great Cycles: Applications, Opportunities, and Challenges of Synthetic Macrocycles in Drug Discovery. J. Med. Chem. 2011, 54, 1961–2004. Vendeville, S.; Cummings, M. D. Synthetic Macrocycles in Small-Molecule Drug Discovery. Annu. Rep. Med. Chem. 2013, 48, 371–386. Illuminati, G.; Mandolini, L. Ring Closure Reactions of Bifunctional Chain Molecules. Acc. Chem. Res. 1981, 14, 95–102. Blankenstein, J.; Zhu, J. Conformation-Directed Macrocyclization Reactions. Eur. J. Org. Chem. 2005, 10, 1949–1964. Swamy, K. C. K.; Kumar, N. N. B.; Balaraman, E.; Kumar, K. V. P. P. Mitsunobu and Related Reactions: Advances and Applications. Chem. Rev. 2009, 109, 2551–2651. Sach, N. W.; Richter, D. T.; Cripps, S.; Tran-Dube, M.; Zhu, H.; Huang, B.; Cui, J.; Sutton, S. C. Synthesis of Aryl Ethers via a Sulfonyl Transfer Reaction. Org. Lett. 2012, 14, 3886–3889. He, Y.; Wang, B.; Dukor, R. K.; Nafie, L. A. Determination of Absolute Configuration of Chiral Molecules Using Vibrational Optical Activity: A Review. Appl. Spectrosc. 2011, 65, 699–723. Brown, H. C.; Chandrasekharan, J.; Ramachandran, P. V. Chiral Synthesis via Organoboranes. 14. Selective Reductions. 41. Diisopinocampheylchloroborane, an Exceptionally Efficient Chiral Reducing Agent. J. Am. Chem. Soc. 1988, 110, 1539–1546. Ramachandran, P. V.; Gong, B.; Brown, H. C. A Remarkable Inversion in Configuration of the Product Alcohols from the Asymmetric Reduction of Ortho-Hydroxyacetophenones with ΒChlorodiisopinocampheylborane. Tetrahedron Lett. 1994, 35, 2141–2144. Campagna, F.; Carotti, A.; Casini, G. A Convenient Synthesis of Nitriles from Primary Amides under Mild Conditions. Tetrahedron Lett. 1977, 18, 1813–1815. El-Faham, A.; Albericio, F. Peptide Coupling Reagents, More than a Letter Soup. Chem. Rev. 2011, 111, 6557–6602. Sperry, J. B.; Minteer, C. J.; Tao, J-Y.; Johnson, R.; Duzguner, R.; Hawksworth, M.; Oke, S.; Richardson, P. F.; Barnhart, R.; Bill, D. R.; Giusto, R. A; Weaver, J. D. Thermal Stability Assessment of Peptide Coupling Reagents Commonly Used in Pharmaceutical Manufacturing. Org. Process Res. Dev. 2018, 22, 1262–1275. Hansen, M. M.; Jolly, R. A.; Linder, R. J. Boronic Acids and Derivatives – Probing the Structure-Activity Relationships for Mutagenicity. Org. Process Res. Dev. 2015, 19, 1507–1516. Martinez, C. A.; Smogowicz, A.; Steflik, J. S.; Brown, M. S.; Midelfort, K. S.; Burns, M. P.; Wong, J. W. Practical Methods for Biocatalysis and Biotransformations 2 ; Whittall, J. ; Sutton, P. W. ; Eds.; John Wiley and Sons: Chichester, UK, 2012; pp. 121–124. 57

33. Kung, P-P.; Jones, R. A.; Richardson, P. Crizotinib (Xalkori): The First-in-Class ALK/ROS Inhibitor for Non-small Cell Lung Cancer. Innovative Drug ; Li, J-J.; Johnson, D. S.; Eds.; John Wiley and Sons, Hoboken, USA, 2016; pp. 119–155. 34. Ackermann, L.; Vicente, R.; Kapdi, A. R. Transition Metal Catalyzed Direct Arylation of (Hetero)Arenes by C-H Bond Cleavage. Angew. Chem., Int. Ed. 2009, 48, 9792–9826. 35. Du, B.; Jiang, X.; Sun, P. Palladium-Catalyzed Highly Selective Ortho-Halogenation (I, Br, Cl) of Arylnitriles via Sp2 C–H Bond Activation Using Cyano as Directing Group. J. Org. Chem. 2013, 78, 2786–2791. 36. Li, W.; Xu, Z.; Sun, P.; Jiang, X.; Fang, M. Synthesis of Biphenyl-2-Carbonitrile Derivatives via a Palladium-Catalyzed sp2 C–H Bond Activation Using Cyano as a Directing Group. Org. Lett. 2011, 13, 1286–1289. 37. Bensaid, S.; Laidaoui, N.; El Abed, D.; Kacimi, S.; Doucet, H. Palladium-Catalysed Direct Arylations of Heteroaromatics Using More Eco-Compatible Solvents Pentan-1-Ol Or 3Methylbutan-1-Ol. Tetrahedron Lett. 2011, 52, 1383–1387. 38. Li, B.; Barnhart, R. W.; Hoffman, J. E.; Nematalla, A.; Raggon, J.; Richardson, P.; Sach, N.; Weaver, J. Exploratory Process Development of Lorlatinib. Org. Process Res. Dev. 2018, 22, 1289–1293. 39. Patterson, D. E.; Powers, J. D.; LeBlanc, M.; Sharkey, T.; Boehler, E.; Irdam, E.; Osterhout, M. H. Development of a Practical Large-Scale Synthesis of Denagliptin Tosylate. Org. Process Res. Dev. 2009, 13, 900–906. 40. Our initial intention was to make the mono-Boc, but the reaction always gave a mixture of monoand bis-Boc compounds, so it was deemed more feasible to make the bis-Boc intermediate. 41. Miller, W. D.; Fray, A. H.; Quatroche, J. T.; Sturgill, C. D. Development of an Impurity Control Strategy Supporting Synthesis of LY451395. Org. Process Res. Dev. 2007, 11, 359–364. 42. The reaction system was evacuated and refilled with N2 three times to exclude oxygen before the addition of 38. 43. Marshall, J. A.; Adams, N. D. Total Synthesis of Bafilomycin V1: A Methanolysis Product of the Macrolide Bafilomycin C2. J. Org. Chem. 2002, 67, 733–740. 44. Lovrić, M.; Cepanec, I.; Litvić, M.; Bartolinčić, A.; Vinković, V. Scope and Limitations of Sodium and Potassium Trimethylsilanolate as Reagents for Conversion of Esters to Carboxylic Acids. Croat. Chem. Acta 2007, 80, 109–115. 45. El-Faham, A.; Subirós-Funosas, R.; Prohens, R.; Albericio, F. COMU: A Safer and More Effective Replacement for Benzotriazole‐Based Uronium Coupling Reagents. Chem. Eur. J. 2009, 15, 9404–9416. 46. Jensen, A. J.; Luthra, S.; Richardson, P. F. Preparation of solid forms of a macrocyclic kinase inhibitor. WO Patent 2014207606, December 31,2014. 47. The reaction rate was determined by quenching the reaction mixture with a large excess of pyrrolidine and comparing the ratio of the desired coupled product and the corresponding amide formed with pyrrolidine. 48. Dunetz, J. R.; Xiang, Y.; Baldwin, A.; Ringling, J. General and Scalable Amide Bond Formation with Epimerization-Prone Substrates Using T3P and Pyridine. Org Lett. 2011, 13, 5048–5051.

58

49. From all the bases screened, only acetate bases worked for this reaction, which seemed to be consistent with C–H functionalization reactions. For a Heck coupling, it was expected that it would work with a variety of other bases, based on literature precedents. 50. Ji, Y.; Plata, R. E.; Regens, C. S.; Hay, M.; Schmidt, M.; Razler, T.; Qiu, Y.; Geng, P.; Hsiao, Y.; Rosner, T.; Eastgate, M. D.; Blackmond, D. G. Mono-Oxidation of Bidentate Bisphosphines in Catalyst Activation: Kinetic and Mechanistic Studies of a Pd/XantphosCatalyzed C–H Functionalization. J. Am. Chem. Soc. 2015, 137, 13272–13281. 51. Suzuki, K.; Okuda, H. Synthesis of Fumaronitrile Complexes of Palladium(0) and Platinum(0) containing 1-diphenylphosphino-2-diphenylarsinoethane. Synth. React. Inorg. Metal-Org. Chem. 1973, 3, 369–374. 52. It was rationalized that the initial substrate/formed product would chelate with active catalyst to keep it in the solution phase. Formation of Pd(0) that precipitated out of the solution as palladium black was detrimental to the reaction. 53. At the time of the campaign, the particular lot of CataCXium A contained ~30% of the corresponding phosphine oxide impurity by 31P NMR. Although lab use test with the charge amount adjusted appeared to give the expected results, it remained unclear whether the phosphine oxide present had inhibited the catalyst turnover. Unfortunately, the timing of the campaign did not allow time to purchase fresh reagent with high quality. 54. Li, B.; Li, R.; Dorff, P.; McWilliams, C.; Guinn, R. M.; Guinness, S. M.; Han, L.; Wang, K.; Yu, S. Deprotection of N-Boc Groups under Continuous-Flow High Temperature Conditions. J. Org. Chem. 2019, 84, 4846–4855. 55. Duan, S.; Li, B.; Dugger, R. W.; Conway, B.; Kumar, R.; Martinez, C.; Makowski, T.; Pearson, R.; Olivier, M.; Colon-Cruz, R. Developing an Asymmetric Transfer Hydrogenation Process for (S)-5-Fluoro-methylisobenzofuran-1(3H)-one, a Key Intermediate to Lorlatinib. Org. Process Res. Dev. 2017, 21, 1340–1348and references therein. 56. Panteleev, J.; Maguire, R. J.; Kung, D. W. Alkylation of Nitrogen-Containing Heterocycles via In Situ Sulfonyl Transfer. Synlett. 2015, 26, 953–959. 57. FDA approves lorlatinib for second- or third-line treatment of ALK-positive metastatic NSCLC. U.S. Food and Drug Adminstration. https://www.fda.gov/drugs/fda-approveslorlatinib-second-or-third-line-treatment-alk-positive-metastatic-nsclc (accessed June 12, 2019).

59

Chapter 3

From Discovery to Market Readiness: The Research and Development of the β-Sparing Phosphatidylinositol 3-Kinase Inhibitor Taselisib Rémy Angelaud,*,1 Steve Staben,*,2 Timothy Heffron,2 Andreas Schuster,3 and Frédéric St-Jean1 1Department of Small Molecule Process Chemistry, Genentech,

South San Francisco, California 94080, United States 2Discovery Chemistry, Genentech, South San Francisco, California 94080, United States 3Small Molecules Technical Development PTDC-C, F. Hoffmann-La Roche Ltd.,

4070 Basel, Switzerland *E-mails: [email protected] (R.A.); [email protected] (S.S.).

Inhibition of the phosphatidylinositol 3-kinase (PI3K) signaling pathway, as an approach to the treatment of cancer, has garnered significant interest from the pharmaceutical industry, and many PI3K inhibitors have entered clinical study. Taselisib is an inhibitor of PI3Kα, PI3Kδ, and PI3Kγ with selectivity over the PI3Kβ isoform. Taselisib also exhibits enhanced potency against oncogenic mutant forms of PI3Kα. This chapter describes the medicinal chemistry efforts leading from a high-throughput screening (HTS) hit to the discovery of taselisib. Insights from crystal structures and physicochemical property improvements that led to the design of taselisib will be described. Additionally, the development and optimization leading to an efficient, robust, and environmentally friendly largescale manufacturing process of taselisib will be described.

The Discovery of GDC-0032 (Taselisib) Direct inhibition of mutant, oncogenic proteins remains one of the most successful strategies for improving outcomes in patients with well-characterized solid tumors. PIK3CA, the gene that encodes the catalytic p110α subunit of phosphatidylinositol 3-kinase-alpha (PI3Kα), is among the most frequently mutated genes in cancer. It is estimated that ~30% of women diagnosed with breast cancer harbor mutation in PIK3CA, usually in one of two “hotspots” (i.e., E542 or 545K or H1047R) (1). Over 15 years of preclinical and clinical research directed at targeting PI3Ks has led to many exciting advances. For instance, disclosed data from a Phase 3 trial of the PI3Kα-selective

© 2019 American Chemical Society

inhibitor, BYL-719, in combination with fulvestrant, for second line metastatic breast cancer patients, have underscored the promise of targeted therapies directed at this particular oncogene (2). From (High-Throughput Screening) HTS to Clinical Candidate In early 2007, we conducted a HTS of our internal library to identify inhibitors of recombinant PI3Kα using a fluorescence polarization (FP) assay. The goal of the screen was to identify a starting point for optimization that was structurally unrelated to our clinical, thienopyrimidine-based PI3K inhibitors that were advancing from our group [i.e., GDC-0941 or pictilisib (3) and GDC-0980 or apitolisib (4)]. We envisioned that a novel, structurally-distinct series could hedge potential (unpredicted) scaffold-associated risks of our leads and provide additional opportunities to target selectivity among the Class I PI3K isoforms. From this screen, a number of “hits” were selected for follow-up, one of which was compound 1. Approximately two years of structure and property-guided design effort resulted in optimization of 1 to GDC-0032. In 2009, GDC-0032 was nominated for early clinical development and entered Phase 1 clinical trials the following year. Figure 1 presents a schematic representation of the necessary key steps in the optimization process to transform HTS hit 1 to GDC-0032. In this portion of the chapter, we present a chronological overview of advances made by the medicinal chemistry team. Additional details are presented in a series of four publications (5–8).

Figure 1. Schematic representation of transformations that led from HTS hit 1 to taselisib (GDC-0032). In many ways, the early lead thienobenzopyran 2 (Table 1), generated by positional halogen scan from HTS hit 1, represented a poor starting point for optimization. Compound 2 had moderate activity in our PI3Kα FP assay [half maximal inhibitory concentration (IC50) = 97 nM], high lipophilicity relative to inhibitory activity (cLogD7.4 = 4.6, LLE = 2.4), and two structural alerts of high concern for Ames testing (thiophene and aniline). In our estimation, this lead necessitated around 4-log-improvements in potency, relative to lipophilicity, to get to our targeted property and potency range (cLogD7.4 = 1–3, PI3Kα pIC50 < 9, LLE > 6). Yet this series met the goals of our screen: (1) benzopyran chemical matter had not been reported for the PIK family; and (2) a cocrystal structure with p110g indicated vectors toward differences in primary structure that could potentially result in isoform-selective inhibitors. Thienobenzopyran 2 had properties consistent with its high lipophilicity, including poor stability in human liver microsomes (HLM Clhep = 18.7 mL/min/kg, Table 1). In addition, we 62

noted an issue with chemical stability, such that extended exposure in organic solvents (including dimethylsulfoxide) was leading to significant ring-opening of the pyran. A structure-guided ring expansion to the benzoxepin, as in 3, solved the chemical, but not metabolic, stability issues (HLM Clhep = 18.4 mL/min/kg) and provided a reliable 3–5-fold improvement in biochemical potency. Next, small polar substitution was examined on the 8-position of the thienobenzoxepin, in an attempt to improve potency through H-bonding interactions and block anticipated soft spots of metabolism. Potency improvements were realized, while metabolic stability remained a major issue. For example, the primary amide-substituted analogue 4 was ~15-fold more potent than 3, considerably less lipophilic (ΔcLogD7.4 = 1.1), but was predicted to have high hepatic clearance (HLM Clhep = 18.4 mL/min/kg). Table 1. Ring Expansion from Pyran to Oxepin Followed by Small, Polar 8-Substitution

A metabolite identification experiment of 4, in cultured human hepatocytes, indicated Ndemethylation and hydrolysis of the aniline amide as major metabolic pathways. The amide hydrolysis was consistent with our observation that significant turnover in vitro was observed even in the absence of enabling cytochrome P450 (CYP)-mediated clearance (HLM assay devoid of reduced nicotinamide adenine dinucleotide phosphate). Crystal structures of this class of inhibitors, including 2 in p110g (Figure 2, light grey), indicated a cis-orientation of the amide, a conformation that was predicted by computation and analysis of the Cambridge Structural Database. As such, simple replacement with alkyl amides was not well-tolerated (data not shown). We designed a series of heteroaromatic isosteric replacements for this N-methyl aniline amide with hopes of improving potency and metabolic stability further. 1,2,4-Triazoles, as in compound 5, provided the best balance of potency and metabolic stability (Table 2). Importantly, replacement of the amide with a 1,2,4-triazole allowed replacement of the arene with a smaller alkyl group, while maintaining the desired small-molecule conformation and a significant amount of potency. Compared to early lead 2, thienobenzoxepin 6, at an IC50 = 2.9 nM and cLogD7.4 = 3.6, possesses a >30-fold improvement in inhibitory potency, along with a 1.063

unit reduction in lipophilicity (ΔLLE = 2.4), while maintaining a nearly identical molecular weight. Consistent with improved properties and the replaced aniline amide, 6 had reduced turnover in HLM (Clhep = 10 mL/min/kg). Figure 2 shows an overlay of 6 (dark grey) with N-methyl aniline amide 2.

Figure 2. Isosteric replacement of the cis N-Me-aniline amide with a 1,2,4-triazole. (a) 1-Aryl versus 1alkyl substituted 1,2,4-triazoles. (b) Overlay of X-ray structure of triazole 6 with amide 2 in p110γ.

Table 2. Replacement of the N-Methyl Aniline Amide with a 1,2,4-Triazole

Our initial target candidate profile was to spare the β isoform of Class I PI3Ks to avoid some of the metabolic effects associated with PI3K-inhibition. Empirically, we discovered that large substitution at the 8-position gave analogs with maintained and improved inhibitory activity against PI3Kα while sparing the β isoform. For example, thienobenzoxepin 7 with 8-(N-hydroxyethyl)4-pyrazoyl substitution, possessed ~13-fold selectivity over the β isoform (Table 3). Compound 7 was still too lipophilic (cLogD7.4 = 4.4) and possessed an undesirable thiophene. Replacement 64

with a more polar thiazole, pyrazole (not shown), or imidazole was tolerated. Benzoxazepin 9 had a desirable range of potency and lipophilicity (LLE = 6.6) and further improved liver microsome stability (HLM Clhep = 8.6 mL/min/kg). Among all subseries of benzoxepin-based PI3K inhibitors, these imidazobenzoxazepins had superior bioavailability and reduced free clearance. As a result, they were selected for final optimization. Table 3. 8-Pyrazole Substitution and Isosteric Replacement of the Thiophene with Other 5Membered Heterocycles

We discovered that bulkier alkyl substituents on the pyrazole could further boost selectivity over the PI3Kβ isoform [e.g., azetidine 10, (Table 4)]. We later disclosed a structure-based hypothesis for this profile based on crystal structures disclosed subsequent to our results (4). During these same studies, we discovered that elaboration to what became known as GDC-0032 resulted in acceptable levels of selectivity between PI3Kα and PI3Kβ, but with much improved human liver microsomal stability and potency in cell lines harboring activating mutations of PI3Kα (Table 4). Table 4. 8-Pyrazole Substitution Can Affect Isoform Selectivity and Cellular Potency

65

The combination of excellent metabolic stability and cellular potency of GDC-0032 led us to characterize this molecule further. For example, we demonstrated that this compound could achieve excellent tumor growth inhibition in a tumor xenograft model driven by a PI3Kα-activating mutation (Figure 3). We also showed that GDC-0032 had excellent selectivity over other kinases including phosphatidylinositol 3-kinase-related kinase (PIKK) family members (6).

Figure 3. Dose-response curves of fitted MCF7-neo/HER2 tumor volumes in response to GDC-0032 treatment. Ultimately, we were able to demonstrate that the exceptional cellular potency of GDC-0032, against PI3Kα mutants, was attributable to its unique ability to selectively induce degradation of mutant PI3Kα [and not wild-type PI3Kα (9)]. This desirable mechanism of action was not observed with other clinical PI3K inhibitors. Owing to the overall preclinical package, including potency, selectivity, and projected human pharmacokinetics, GDC-0032 was selected for development. Discovery Chemistry Synthetic Route The discovery route to our imidazobenzoxepine inhibitors (Scheme 1) centered on the 9bromo-2-iodo intermediate 13, which was set up for late-stage diversification at the 2- and 9positions. 4-Bromo-2-hydroxybenzaldehyde (11) was treated with glyoxal in ammonia to provide an intermediate 2-arylimdazole. Treatment with 1,2-dibromoethane under basic conditions furnished the benzoxazepine core 12. Selective installation of an iodine at the 2-position was accomplished by diiodination (NIS) and selective monodeiodination (EtMgBr). Primary amide 14 was generated by a halogen-selective carbonylative amidation using hexamethyldisilazane (HMDS) as an ammonia equivalent. The 1-isopropyl-3-methyl-1,2,4-triazole was created through intermediacy of an Nacylamidine. Finally, Suzuki coupling delivered GDC-0032.

66

Scheme 1. Discovery chemistry synthesis of GDC-0032.

Development Work toward Commercial Manufacture of Taselisib As illustrated in Scheme 2, the discovery route via aminocarbonylation of the imidazole iodide 13 served as the basis for the development activities toward accessing the key intermediate bromocore 15. However, in order to avoid the use of pressurized CO and toxic N,Ndimethylacetamide dimethyl acetal (DMA-DMA) reagent on scale, a different route was developed using imidazole bromoiodide 13 (10). The new synthesis route used a Pd-catalyzed Negishi reaction to couple triazole 16 and 13. Triazole 16 was manufactured by alkylation of 17 with 2-iodopropane (11). The alkylation of 17 led to the formation of a 1.5:1 inseparable mixture of the desired product triazole 16 and triazole isomer 16′ (Scheme 2). Even though this route omitted the use of pressurized CO and toxic DMA-DMA, it suffered from the inefficient and resource-intensive route to produce key building block imidazole iodide 13, which resulted in a total process mass intensity (PMI) of 1334 to produce 15 [starting from 3 bromophenol, (12)]. To accommodate the need for a larger scale up to supply the clinical trials, the route to 15 was completely redesigned. A more efficient route to benzoxazepine 15 was developed to eliminate the need for 13. The strategy involved building the imidazole ring via a [3+2] condensation of α-chloroketone 18 with aryl amidine 19 [Figure 4, (10)]. 67

First, a more efficient approach to triazole 16 via N'-isopropyl acetamidrazone (IPAA, 20) was implemented by treating isopropyl hydrazine and acetamidine. This approach prevented the formation of the triazole isomer 16′ and provided the 1,2,4-triazole 16 in 52% yield (Scheme 3).

Scheme 2. Negishi route to 15.

Figure 4. The [3+2] imidazole ring formation approach to 15.

Scheme 3. Synthesis of triazole 16. Subsequent alkylation of lithiated 16 with the α-chloro Weinreb amide derivative 21 led to the α-chloroketone 18 (72% assay yield), which was used directly in the [3+2] condensation with aryl amidine 19 to provide the corresponding imidazole 22 in 66% overall yield (13). Next, 22 was converted to the benzoxazepine 15 in a one-pot reaction, using an efficient and highly regioselective alkylation and SNAr tandem reaction with ethylene carbonate and afforded, after crystallization, 15 in 69% overall yield and >99.9 high-performance liquid chromatography (HPLC) area (A)% purity 68

measured by HPLC (Scheme 4). This second generation [3+2] process allowed us to successfully deliver ~35 kg of 15. The [3+2] route created two major impurities (Scheme 4). First, the vinyl impurity 24 was formed by opening the oxazepine ring of the benzoxazepine 15 under the strongly basic reaction conditions during the SNAr. Second, the benzoxazepine isomer 25 was the result of nonselective Nalkylation of the imidazole ring in the previous step. Although the [3+2] route reduced the PMI from 1334 (Negishi route) to 452, it had some disadvantages, such as the use of unstable and genotoxic chloride reagents (21 and 18) and the formation of two impurities that were difficult to remove and required several crystallizations to be purged. Thus, an efficient alternative strategy to enable largescale manufacture of taselisib needed to be developed.

Scheme 4. [3+2] and one-pot alkylation and SNAr reactions to 15. Development of the Large-Scale Manufacture Route to Bromobenzoxazepine 15 The strategy for the large-scale manufacture route to key intermediate 15 was based on the formation of the benzoxazepine ring first (e.g., 26) from the readily available starting material 4bromo-2-fluorobenzonitrile (14). We intended to leverage the amidine chemistry and the [3+2] type condensation experience gained in the development of the previous route to construct the 1,2,4-triazole ring via the imidazole carboxylic acid 27 (Figure 5). Cyclic amidine 26 was made in two steps from 4-bromo-2-fluorobenzonitrile via a chemoselective SNAr reaction with ethanolamine to provide the HCl salt of aminoether 28 in 91% yield, followed by a Mg-mediated amidine cyclization (Scheme 5). During development, it was found that the cyclic amidine 26 could be produced using AlMe3 (14). For scale-up, the much safer, nonpyrophoric alternative Mg(OEt)2 was used. Initially, 26 was not isolated, but telescoped, into the 69

imidazole ring formation to produce 27 using bromopyruvic acid (BPA). In the final process, 26 was isolated as the HCl salt to avoid hydrolysis of the free-base cyclic amidine during a solvent swap.

Figure 5. Revised retrosynthetic analysis of 15.

Scheme 5. Mg-mediated cyclization to amidine 26. The amidine cyclization occurred faster using the HCl salt of 28 than the free base. Other metal alkoxides, such as LiOEt, NaOEt, and KOEt gave no conversion, nor did HCl in 1-propanol. From these experiments, it was concluded that the MgCl2 produced during the reaction was important, possibly as a Lewis acid activator of the nitrile group. Increasing the amount of Mg(OEt)2 from 1.8 equiv to 2.5 equiv had no impact on the reaction rate or product quality. Cyclic amidine 26•HCl was directly isolated as a solid by simply adding ~5 N HCl in 1-propanol to the crude reaction mixture. However, residual MeOH was found to have an impact on the recovery, the residual Mg salt content in the isolated product, and the filtration time. Thus, MeOH was removed first by distillation with continuous 2-methyltetrahydrofuran (2-MeTHF) addition and a proposed range of 4–21wt% of MeOH was tested to quantify these effects. The results are given in Table 5. Table 5. Crystallization of 26•HCl with Varying Amounts of Residual Methanol Entry Number Methanol content (%w/ (No.)a w)

Filtration time (min)

Yield (%)

Residue on ignition [(ROI) Assay (%w/w) , %w/w]

1

3.8

01

93.7

97.8

2.8

2

8.0

15

91.0

84.4

8.5

3

12.1

10

87.6

54.4

29.5

4

20.5

25

95.2

45.0

26.1

a Slurries stirred at 20 °C for 6 h.

It was clear, based on these results, that a low MeOH content (Entry 1, Table 5) produced better product quality with higher yield and shorter filtration time. The amount of propanolic HCl added to form the salt was also investigated (Table 6). A low stoichiometric amount of HCl (4.4 equiv, Entry 3, Table 6) resulted in a high ROI value compared to experiments that used more HCl. Since 4.4 equiv HCl were close to the minimum amount (4.2 70

equiv) needed to quench the remaining Mg(OEt)2 and form the HCl salt of 26, it was concluded that some excess HCl was needed for the process (6–7 equiv, Entry 1 and 4, Table 6). Table 6. Impact of HCl Stoichiometry on 26•HCl Quality HClb Equiv

Entry No.a

Yield (%)

ROI (wt%)

1

6.2

85.6

1.7

2

5.2

90.2

1.3

3

4.4

88.1

5.7

4

7.8

86.7

1.3

a Reactions performed at 20 °C and stirred for 22 h before filtration.

b 5–6 N HCl in 1-propanol.

The final conditions [Target and Proven Acceptable Ranges (PAR)] for the Mg-mediated formation of 26•HCl are summarized in Table 7. Table 7. Operating Ranges for Manufacture of 26•HCla Equivalents Material

Target

PAR

28•HCl (e.g., 74 kg)

1.0

0.95–1.05

MeOH (L/kg)

4.0

3.2–4.8

Mg(OEt)2

2.1

1.9–2.3

MeTHF (L/kg)

8.1

6.4–9.6

HCl in 1-Propanol (19.0–24.0 %w/w) 7.0

6.5–7.5

a Reaction temperature: 55–66 °C (reflux).

The next step in the process was the regioselective formation of imidazole 27. Amidine 26•HCl was subjected to a tandem alkylation and condensation sequence with BPA in the presence of 1,1,3,3-tetramethylguanidine (TMG) in N-methyl-2-pyrrolidone (NMP) at 50 °C which afforded imidazole 27 (15).

Scheme 6. BPA alkylation and condensation to imidazole 27. However, the compound obtained exhibited a dark purple tint, and it was later demonstrated that the color could be improved by addition of a diluted solution of BPA in 2-MeTHF (2 vol) and implementation of a carbon treatment (10 wt%) prior to the final acidification. Imidazole 27 was 71

isolated as a single isomer after acidification with aqueous HCl in 85% yield and >99.5 HPLC A% purity (Scheme 6). The robustness of the reaction was demonstrated by testing feasible ranges for BPA and TMG within 1.20–1.40 equiv and 4.00–5.00 equiv, respectively. The results are listed in Table 8. Table 8. Impact of BPA and TMG Stoichiometry on Quality of 27 TMG (equiv)

Entr No.a

BPA (equiv)

27 (A%)b

1

4.0

1.2

99.8

2

5.0

1.4

99.7

3

4.0

1.4

99.2

4

5.0

1.2

99.6

5

4.5

1.3

99.7

6

4.5

1.3

99.7

a 50 °C; BPA addition time 1 h.

b A% by HPLC.

In most experiments, 27 was obtained in high purity (>99.6 A%), except at the extreme limit of the range (Entry 3), with the lowest amount of TMG (4.0 equiv) and the highest amount of BPA (1.4 equiv). In that scenario, 27 was obtained in 99.2 HPLC A%, which was below the required specifications of >99.5 A%. Therefore, the operating range of TMG was set to 4.5–5.0 equiv and BPA was set to 1.2–1.4 equiv. The final conditions (Target and PAR) for this step are summarized in Table 9. Table 9. Operating Ranges for Manufacture of 27a Equivalents Material

Target

PAR

26•HCl (e.g., 61.9 kg)

1.00

0.95–1.05

NMP (L/kg)

5.10

4.00–6.00

TMG

4.75

4.50–5.00

BPA

1.30

1.20–1.40

MeTHF (L/kg)

1.60

1.20–2.00

Water (L/kg)

15.0

12.0–18.0

a Reaction temperature: 40–60 °C.

Subsequent transformation of 27 to form the 1,2,4-triazole derivative 15 was inspired by previous work (16). The one-pot through-process involved two separate steps: (1) activation of acid 27 with 1,1′-carbonyldiimidazole (CDI) to provide the amide 28; followed by (2) in situ chemoselective coupling with IPAA to furnish the desired 1,2,4-triazole ring after loss of water (Scheme 7). Bromobenzoxazepine 15 was then isolated by filtration after addition of water to the crude reaction mixture in 83% yield and >99.9 wt% assay. 72

Scheme 7. One-pot condensation of 20 and 27 to benzoxazepine 15. At the initial stage of development for this step, several bases (N,N-diisopropylethylamine, N-methylmorpholine, 1,8-diazabicyclo[5.4.0]undec-7-en, and K2CO3) and activation reagents [hexafluorophosphate azabenzotriazole tetramethyl uranium (HATU) and CDI] were tested in a series of solvents [tetrahydrofuran (THF), 2-MeTHF, dimethylformamide, N,Ndimethylacetamide, NMP, NEP, t-amyl alcohol, and isoamyl alcohol]. The combination of K2CO3 and CDI in NMP gave the cleanest conversion to product and was therefore chosen for further development. Residual water in 27 was shown to have a negative influence on the completion of the activation of the starting material. Batches of 27 containing up to 2.4 wt% of water were tested, leading to incomplete activation and about 10–15% reduction in yield compared to batches with low water content (200 kg of this key intermediate as a single regioisomer in ~55% overall yield and with very high purity (99.9% assay). This shorter and less complex route to 15 was more efficient and sustainable compared to the 73

previous longer routes, which is also indicated by the significantly lower PMI of 219. A comparison of the synthesis lengths (longest synthesis chain) and the PMI factors of the three routes to 15 is given in Figure 7.

Figure 6. Interaction profiles with CDI, 20, and K2CO3.

Scheme 8. One-pot formation of 20•HCl.

Figure 7. PMI comparison of the three manufacturing routes to 15.

74

Endgame Optimization to Taselisib The endgame process to taselisib was achieved in three parts. The first is a one-pot, three-step reaction sequence including: (1) Miyaura borylation of 29 to obtain 30; (2) Suzuki coupling of boronate ester 30 with 15 to afford 32; and (3) saponification of ester 32 to provide carboxylic acid 33 (Scheme 9). The second part is the amidation of acid 33 to obtain crude taselisib (Scheme 10). The third part is the recrystallization of the crude product to provide pure taselisib with the required quality attributes, such as the desired polymorphic form and PSD.

Scheme 9. Endgame process to acid 33. Various Pd catalysts such as Pd(t-Bu3P)2, Pd(Cy3P)2Cl2, Pd(dtbpf)2Cl2, Pd(dppf)Cl2, Pd(Amphos)2Cl2, and precatalysts Pd(XPhos) (G1 and G2) were screened for the optimization of the Miyaura borylation reaction using conditions similar to those reported in the literature (18–22). From the screen, only the Buchwald Pd(XPhos) G1 and G2 precatalyst systems with bis(pinacolato)diboron (B2Pin2) in EtOH afforded complete conversion to the desired boronic ester 30 (14, 23–26). After further optimization, the reaction was performed with 1.25 equiv of (B2Pin2) and 1.25 equiv of AcOK in EtOH at 77 °C. One hundred percent conversion of 29 to 30 could be achieved with only 0.3 mol% of precatalyst Pd(XPhos) G2 and 0.6 mol% of extra XPhos ligand. The subsequent Suzuki coupling to form 32 was conducted using LiOH as base and bromobenzoaxepine 15 as a coupling partner. To minimize borylation and homocoupling of 15 due to excess (B2Pin2), reagent from the borylation step, the aqueous LiOH solution (1 M), was added in two portions at 77 °C before (0.3 equiv) and after (1.7 equiv) the addition of 15 to deactivate (B2Pin2). For the saponification of ester 32 to acid 33, an additional amount of LiOH aqueous solution (3.5 M, 3.40 equiv) was added. The reaction was performed at reduced temperature (55 °C) to avoid corrosion to the glass-lined reaction vessel under the strongly basic conditions. After completion of the saponification reaction, most of the EtOH was removed by vacuum distillation, and after filtration of the Li and borate salts, the resulting aqueous layer was extracted with n-PrOAc. This extraction was very efficient in removing all nonacidic organic by-products generated during the tandem borylation and Suzuki reactions. 75

Finally, acid 33 was crystallized by direct acidification of the aq layer with an aqueous solution of H2SO4 (~20 wt%) to pH 3.1, the isoelectric point of acid 33 (27). Using this one-pot, three-step process, 33 was isolated in 75% overall yield with very high purity (>99.9 HPLC A%). The final conditions (Target and PAR) for this step are summarized in Table 10. Table 10. Operating Ranges for Manufacture of 33 Equivalents Material

Target

PAR

KOAc

1.20

1.15–1.25

B2Pin2

1.20

1.15–1.25

Pyrazole 29 (e.g., 16.4 kg)

1.00

0.98–1.03

EtOH (L/kg)

10.24

9.23–11.28

XPhos G2 Pd (mol%)

0.30

0.29–0.32

XPhos (mol%)

0.60

0.57–0.63

1 M LiOH

0.3 – 1.7

1.9–2.1

15

0.90

0.88–0.93

THF (L/kg)

5.7

4.0–7.5

3.5 M LiOH

3.40

3.23–3.57

n-PrOAc (L/kg)

13.5

10.8–16.2

1-Propanol (L/kg)

4.5

3.5–5.5

For the final amidation reaction, carboxylic acid 33 was preactivated with CDI in THF at 20 °C. The resultant acyl imidazole was then added to a THF solution containing 5–7 N NH3 in MeOH to provide crude taselisib (Scheme 10).

Scheme 10. Endgame process to taselisib. The three most important parameters for this process were: • CDI equivalents • NH3 equivalents • NH3 concentration in MeOH

76

These parameters were tested in a half-factorial DOE. The amount of CDI used for the reaction can influence the amount of residual 33 in the reaction mixture and product, whereas the other two parameters for NH3 can influence the formation of the methyl ester impurity 34, formed via a side reaction between activated 33 and MeOH. The parameter range for the different reagents was set based on prior experience from development as follows: (1) CDI, 1.3–1.4 equiv; (2) NH3, 3.0–4.0 equiv; and (3) NH3 in MeOH, 5.5–6.9 mol/L. The results are presented in Table 11. Table 11. Half-Factorial DOE on Amidation to Taselisiba Entry No.

CDI (equiv)

NH3 (equiv)

NH3/Methanol (mol/L)

Yield (%)

Taselisib (A%)

33 (A%)

34 (A%)

1

1.4

3.5

6.25

94.9

99.4

0.13

0.15

2

1.5

3.0

6.90

95.7

99.5

0.05

0.21

3

1.5

4.0

5.50

95.5

99.5

0.05

0.22

4

1.3

4.0

6.90

95.7

99.6

0.05

0.14

5

1.3

3.0

5.50

96.1

99.5

0.06

0.16

6

1.4

3.5

6.25

95.6

99.5

0.05

0.15

a Reactions performed at 20 °C.

Overall, good quality product was consistently achieved in high yield at all parameter ranges tested and content of residual 33 and formation of Me ester impurity 34 were not influenced by the three tested parameters. A carbon treatment by filtration through an activated charcoal filter cartridge was implemented during the final crystallization of taselisib to control the amount of residual Pd (and color) in the active pharmaceutical ingredient (API). The final crystallization was performed in 3-methyl-1butanol (isoamyl alcohol) to produce the correct polymorph form and required a minimum temperature of 95 °C to solubilize the API, which was too high given that the stability of the cartridge material was limited to 65 °C. Therefore, crude taselisib was first dissolved in 1-propanol/water (70:30 w/w), then the filter cartridge treatment was performed, and finally the solvent exchange to isoamyl alcohol was carried out. This procedure allowed for lowering the temperature of the filtration to 60 °C. As shown in Table 12, neither crystallization nor treatment with activated charcoal alone depleted Pd sufficiently. The residual Pd content remained close to the specified limit of 10 parts per million (ppm). However, a combination of activated charcoal treatment in 1-propanol/water and crystallization from isoamyl alcohol reproducibly reduced the residual Pd to 97% purity. 99

Figure 13. Process flow diagram of semicontinuous oxidation process. With 15 readily available, we evaluated incorporation of the piperidine 9. The pendant aryl fluoride of 15 allowed for a smooth SNAr reaction at 0 to 25 °C to cleanly give 10 (Scheme 5). Following reaction completion, solvent exchange enabled an aqueous phase split and subsequent precipitation to provide 10 in 77 to 86% yield and >96% purity. Subjecting the organic phase to an aqueous work-up was crucial for controlling fluoride ion and residual amine, since 9 could be a competitive nucleophile for chloride 12 in the next stage.

Scheme 5. Synthesis of 10 by SNAr displacement. In order to facilitate carrying 10 forward to the linker-drug 13, we required an activated form of the benzylic alcohol 11. After investigating several options, including the corresponding bromide and acetate and more exotic deoxygenation conditions, we settled on utilizing benzylic chloride 12. However, we determined that it would be better to isolate the material, instead of telescope it through, as had been done in the discovery route, due to a complex impurity profile that the telescoping maneuver delivered. Isolation from N-methyl-2-pyrrolidone (NMP) solvent required identifying conditions that would remove the compound from solution, without degrading the sensitive material. Organic antisolvents delivered only sticky solids that were difficult to filter, so we turned to precipitation through the use of water. While the benzylic chloride was labile to aqueous conditions, cooling the reaction mixture to 0 °C and dosing cold antisolvent allowed for precipitation of a nicely handled solid. Further, washing to remove the remaining water with a sequence of ethyl acetate (EtOAc), acetonitrile, and methyl tert-butyl ether delivered a stable form of 12. 100

Scheme 6. Formation of crude 13 via alkylation. The last chemical step required coupling amine 10 with 12 to provide the crude PABQ (Scheme 6). Screening experiments revealed that NaI was not a necessary catalytic mediator for the nucleophilic displacement (Table 2). However, additional acid-spiking reactions indicated that trace HCl from the background hydrolysis of 12 did inhibit conversion. Inclusion of N,Ndiisopropylethylamine to trap the acid facilitated a 1:1 stoichiometry of 10 and 12. Yet, equimolar base caused the formation of numerous rifamycin ring degradants. Table 2. Crude Linker-Drug Formation from 10 and 12a Entry

Additive, equiv

T (°C)

Solvent

equiv 10

A% 13b

1c

NaI, 0.1

DMF

55

1.0

68.0

2

N/A

NMP

60

0.9

82.3

3

HCl, 0.1

NMP

60

0.9

75.1

4

HCl, 0.5

NMP

60

0.9

39.4

5

DIPEA, 1.0

NMP

60

1.0

74.1

6d

DIPEA, 0.1

NMP

60

1.0

85.8

a All

reactions at 0.5 g, 0.54 mmol scale of 6, unless indicated. b Determined by HPLC analysis. c 0.04 g, 0.04 mmol scale. d The crude reaction contained 3.2 A% 10. Following precipitation with EtOAc, the crude salt was isolated with 90.0 A% HPLC purity and 1.6 A% HPLC 10.

In contrast, trace base (~10 mol%) facilitated smooth conversion without the degradation sideproducts. In concert, these modifications enabled clean alkylation without a large amount of impurities that interfered with downstream purification. Isolation of the crude PABQ as an amorphous solid on >1 kg scale was achieved by precipitation with EtOAc (86–88% yield and 86–90 A% HPLC). Next, the dark blue solid was purified by low-pressure chromatography over DIAION HP20SS resin with formic acid as the eluent modifier, followed by lyophilization to deliver just under 101

1 kg of the formate salt of vc-dmDNA31 13 in 75–76% yield and >95 A% HPLC purity (43). The production achieved the desired linker-drug attributes of a specified purity and impurity control and a single counter-ion to facilitate the required dissolution for conjugation. With this delivery, conjugation to form >30 kg of the TAC was made possible, reflective of the molar mass difference between the small molecule linker-drug and the large molecule antibody.

Scheme 7. Final process chemistry route. In conclusion, we have designed and demonstrated a novel linker-antibiotic 13 and a robust route to this complex and delicate molecular construct. The chemistry involved exploring a series of potent bactericidal compounds, along with a practical functional group, to connect the linkerdrug to the antibody. Through this effort, tertiary amines were established as viable functionality to produce PABQs. These ionic species were shown to be stable in circulation, yet cleavable by the desired cathepsin to release the drug in the desired environment to selectively kill S. aureus. Process work navigated a number of stability and chromatic challenges in reactions, work-ups, and chromatography. The final synthetic route (Scheme 7) takes advantage of an efficient semicontinuous oxidative condensation to 15. A mild SNAr reaction to 10 and isolation of benzylic chloride 12 set the stage for an expedient formation of the quaternary ammonium species in the target on kg-scale. The dark blue linker-drug 13 was obtained in 29% yield over 4 steps—a throughput improvement of 102

100-fold over the initial route—to support the first anti-S. aureus TAC construct in the clinic. This chapter captures the main impact of small molecule discovery and development of the first antibodyantibiotic construct, by explaining the linker-drug evolution and production to enable studies that benefit patients.

Acknowledgments The authors would like to thank Ifat Abramovich, Remy Angelaud, David Askin, Stephan Bachmann, Fritz Bliss, Eric Brown, Nicolas Burki, Alan Deese, Serena Fantasia, John Flygare, Francis Gosselin, Jason Gruenhagen, Christine Gu, Wouter Hazenbos, Hilda Hernandez, Michael Jansen, Janice Kim, Stephan Lauper, Christian Lautz, Sophie Lehar, Yi Li, Xin Linghu, Kelly Loyet, Sanjeev Mariathasan, Colin Medley, Barbara Müller, Laura Murray, Nadja Neubauer, Summer Park, Helga Raab, Sebastian Rieth, Nathaniel L. Segraves, C. Gregory Sowell, Leanna Staben, Sarah Stowers, Man Wah Tan, Shaoning Wang, Nicholas Wong, Min Xu, Donghong Yan, Andreas Zogg, and Ruth Zuniga for their contributions.

References 1. 2. 3. 4. 5. 6. 7. 8. 9.

10. 11. 12. 13. 14. 15.

Rosen, W. Miracle Cure: The Creation of Antibiotics and the Birth of Modern Medicine; Penguin Books: New York, 2018. Aminov, R. I. A Brief History of the Antibiotic Era: Lessons Learned and Challenges for the Future. Frontiers in Microbiology 2010, 1, 1–7. McKenna, M. Superbug: The Fatal Menace of MRSA; Free Press: New York, 2010. Hall, W.; McDonnell, A.; O’Neill, J. Superbugs: An Arms Race Against Bacteria; Harvard University Press: Cambridge, MA, 2018. Fabbretti, A.; Gualerzi, C. O.; Brandi, L. How to Cope with the Quest for New Antibiotics. FEBS Lett. 2011, 585, 1673–1681. Bax, R.; Griffin, D. Antibiotic Resistance; In Handbook of Experimental Pharmacology; Coates, A. R. M. , Ed.; Springer: Heidelberg, 2012; Vol. 211, pp 1–12. Wright, P. M.; Seiple, I. B.; Myers, A. G. The Evolving Role of Chemical Synthesis in Antibacterial Drug Discovery. Angew. Chem., Int. Ed. Engl. 2014, 53, 8840–8869. Ventola, C. L. The Antibiotic Resistance Crisis – Part 1: Causes and Threats. Pharmacy and Therapeutics 2015, 40, 277–283. Talkington, K.; Shore, C.; Kothari, P. A Scientific Roadmap for Antibiotic Discovery. http://www. pewtrusts.org/~/media/assets/2016/05/ascientificroadmapforantibioticdiscovery.pdf (accessed May 11, 2019). Friedman, N. D.; Temkin, E.; Carmeli, Y. The Negative Impact of Antibiotic Resistance. Clin. Microbiol. Infect. 2016, 22, 416–422. Chambers, H. F.; DeLeo, F. R. Waves of Resistance: Staphylococcus aureus in the Antibiotic Era. Nat. Rev. Microbiol. 2009, 7, 629–641. Cooper, M. A.; Shlaes, D. Fix the Antibiotics Pipeline. Nature 2011, 472, 32. Nathan, C.; Cars, O. Antibiotic Resistance – Problems, Progress, and Prospects. N. Engl. J. Med. 2014, 371, 1761–1763. Blaskovich, M. A. T.; Zuegg, J.; Elliott, A. G.; Cooper, M. A. Helping Chemists Discover New Antibiotics. ACS Infect. Dis. 2015, 1, 285–287. National Action Plan for Combating Antibiotic-Resistant Bacteria. https://obamawhitehouse.archives.gov/ sites/default/files/docs/national_action_plan_for_combating_antibotic-resistant_bacteria.pdf (accessed May 11, 2019).

103

16. Declaration by the Pharmaceutical, Biotechnology and Diagnostics Industries on Combating Antimicrobial Resistance. https://amr-review.org/sites/default/files/Industry_Declaration_on_Combating_ Antimicrobial_Resistance_UPDATED%20SIGNATORIES_MAY_2016.pdf (accessed May 11, 2019). 17. Gale, J.; Fourcade, H. There’s Big Money Again in Saving Humanity with Antibiotics. https://www. bloomberg.com/news/articles/2016-06-30/superbugs-and-subsidies-draw-big-pharma-back-toantibiotics (accessed May 11, 2019). 18. Mantaj, J.; Jackson, P. J. M.; Rahman, K. M.; Thurston, D. E. From Anthramycin to Pyrrolobenzodiazepine (PBD)-Containing Antibody-Drug Conjugates (ADCs). Angew. Chem., Int. Ed. Engl. 2017, 56, 462–488. 19. Chari, R. V. J. Expanding the Reach of Antibody-Drug Conjugates. ACS Med. Chem. Lett. 2016, 7, 974–976. 20. Ledford, H. Weaponized Antibodies Use New Tricks to Fight Cancer. Nature 2016, 540, 19–20. 21. Ghosh, M.; Miller, P. A.; Mollmann, U.; Claypool, W. D.; Schroeder, V. A.; Wolter, W. R.; Suckow, M.; Yu, H.; Li, S.; Huang, W.; Zajicek, J.; Miller, M. J. Targeted Antibiotic Delivery: Selective Siderophore Conjugation with Daptomycin Confers Potent Activity Against Multidrug Resistant Acinetobacter baumannii both In Vitro and In Vivo. J. Med. Chem. 2017, 60, 4577–4583. 22. Brandish, P. E.; Garbaccio, R. M.; Kern, J.; Liang, L.; Shah, S.; Zaller, D.; Beck, A.; Gately, D.; Knudsen, N.; Manibusan, A.; Wang, J.; Sun, Y. Patent WO 2017062271 A2, 2017. 23. Lehar, S. M.; Pillow, T.; Xu, M.; Staben, L.; Kajihara, K. K.; Vandlen, R.; DePalatis, L.; Raab, H.; Hazenbos, W. L.; Morisaki, J. H.; Kim, J.; Park, S.; Darwish, M.; Lee, B.-C.; Hernandez, H.; Loyet, K. M.; Lupardus, P.; Fong, R.; Yan, D.; Chalouni, C.; Luis, E.; Khalfin, Y.; Plise, E.; Cheong, J.; Lyssikatos, J. P.; Strandh, M.; Koefoed, K.; Andersen, P. S.; Flygare, J. A.; Tan, M. W.; Brown, E. J.; Mariathasan, S. Novel Antibody-Antibiotic Conjugate Eliminates Intracellular S. aureus. Nature 2015, 527, 323–328. 24. Thwaites, G. E.; Gant, V. Are Bloodstream Leukocytes Trojan Horses for the Metastasis of Staphylococcus aureus? Nat. Rev. Microbiol. 2011, 9, 215–222. 25. The mAb developed for our target was engineered to have two cysteine residues in specific sites to enable a drug-antibody ratio (DAR) of two. These Genentech-engineered mAbs have been named THIOMABTM to account for the two defined cysteine thiol functional groups available for conjugation. See: Junutula, J. R.; Raab, H.; Clark, S.; Bhakta, S.; Leipold, D. D.; Weir, S.; Chen, Y.; Simpson, M.; Tsai, S. P.; Dennis, M. S.; Lu, Y.; Meng, Y. G.; Ng, C.; Yang, J.; Lee, C. C.; Duenas, E.; Gorrell, J.; Katta, V.; Kim, A.; McDorman, K.; Flagella, K.; Venook, R.; Ross, S.; Spencer, S. D.; Lee Wong, W.; Lowman, H. B.; Vandlen, R.; Sliwkowski, M. X.; Scheller, R. H.; Polakis, P.; Mallet, W. Site-Specific Conjugation of a Cytotoxic Drug to an Antibody Improves the Therapeutic Index. Nat. Biotechnol. 2008, 26, 925–932. 26. Shen, B.-Q.; Xu, K.; Liu, L.; Raab, H.; Bhakta, S.; Kenrick, M.; Parsons-Reponte, K. L.; Tien, J.; Yu, S.-F.; Mai, E.; Li, D.; Tibbitts, J.; Baudys, J.; Saad, O. M.; Scales, S. J.; McDonald, P. J.; Hass, P. E.; Eigenbrot, C.; Nguyen, T.; Solis, W. A.; Fuji, R. N.; Flagella, K. M.; Patel, D.; Spencer, S. D.; Khawli, L. A.; Ebens, A.; Wong, W. L.; Vandlen, R.; Kaur, S.; Sliwkowski, M. X.; Scheller, R. H.; Polakis, P.; Junutula, J. R. Conjugation Site Modulates the In Vivo Stability and Therapeutic Activity of AntibodyDrug Conjugates. Nat. Biotechnol. 2012, 30, 184–189. 27. Vollmar, B. S.; Wei, B.; Ohri, R.; Zhou, J.; He, J.; Yu, S.-F.; Leipold, D.; Cosino, E.; Yee, S.; FourieO’Donohue, A.; Li, G.; Phillips, G. L.; Kozak, K. R.; Kamath, A.; Xu, K.; Lee, G.; Lazar, G. A.; Erickson, H. K. Attachment Site Cysteine Thiol pKa is a Key Driver for Site-Dependent Stability of THIOMAB Antibody-Drug Conjugates. Bioconjugate Chem. 2017, 28, 2538–2548. 28. Prashad, A. S.; Nolting, B.; Patel, V.; Xu, A.; Arve, B.; Letendre, L. From R&D to Clinical Supplies. Org. Process Res. Dev. 2017, 21, 590–600.

104

29. Yamane, T.; Hashizume, T.; Yamashita, K.; Konishi, E.; Hosoe, K.; Hidaka, T.; Watanabe, K.; Kawaharada, H.; Yamamoto, T.; Kuze, F. Synthesis and Biological Activity of 3′-hydroxy-5′aminobenzoxazinorifamycin Derivatives. Chem. Pharm. Bull. 1993, 41, 148–155. 30. Holdiness, M. R. A Review of the Redman Syndrome and Rifampicin Overdosage. Med. Toxicol. Adverse Drug Exper. 1989, 4, 444–451. 31. The Dictionary of Modern Medicine; Segen, J. C., Ed.; The Parthenon Publishing Group Inc.: Park Ridge, NJ, 2012; pp 515. 32. 2-Amino-3-((tert-butyldimethylsilyl)oxy)phenol was utilized in the original work since the unprotected aminoresorcinol is unstable as a free base. 33. Jabri, S. Y.; Overman, L. E. Enantioselective Total Synthesis of Plectosphaeroic Acid B. J. Am. Chem. Soc. 2013, 135, 4231–4234. 34. Ruiz-Castillo, P.; Buchwald, S. L. Applications of Palladium-Catalyzed C-N Cross-Coupling Reactions. Chem. Rev. 2016, 116, 12564–12649. 35. After our work started, 5-chloro-aminoresorcinol was reported. See: Yang, X.; Shan, G.; Rao, Y. Synthesis of 2-aminophenols and Heterocycles by Ru-catalyzed C–H Mono- and Dihydroxylation. Org. Lett. 2013, 15, 2334–2337. 36. Several degradation products are formed during the reaction. 37. Kaizer, J.; Csonka, R.; Speier, G. TEMPO-Initiated Oxidation of 2-aminophenol to 2aminophenoxazin-3-one. J. Mol. Catal. A: Chem. 2002, 180, 91–96. 38. Reactions run under N2 inertion with no external oxidants consistently gave baseline conversion to 15. 39. Kono, Y. Oxygen Enhancement of Bactericidal Activity of Rifamycin SV on Escherichia coli and Aerobic Oxidation of Rifamycin SV to Rifamycin S Catalyzed by Manganous Ions: The Role of Superoxide. J. Biochem. 1982, 91, 381–395. 40. Kono, Y.; Sugiura, Y. Electron Spin Resonance Studies on the Oxidation of Rifamycin SV Catalyzed by Metal Ions. J. Biochem. 1982, 91, 397–401. 41. Saez, G. T.; Valls, V.; Cabedo, H.; Iradi, A.; Bannister, W. H.; Bannister, J. V. Effect of Metal Ion Catalyzed Oxidation of Rifamycin SV on Cell Viability and Metabolic Performance of Isolated Rat Hepatocytes. Biochim. Biophys. Acta, Mol. Cell Res. 1991, 1092, 326–335. 42. Degradation of 14 results in insoluble material forming in the reaction. 43. Linghu, X.; Segraves, N. L.; Gavish-Abramovich, I.; Wong, N.; Müller, B.; Neubauer, N.; Fantasia, S.; Rieth, S.; Bachmann, S.; Jansen, M.; Sowell, C. G.; Askin, D.; Koenig, S. G.; Gosselin, F. Highly Efficient Synthesis of a Staphylococcus aureus-Targeting Payload, Enabling the First Antibody-Antibiotic Conjugate. Chem. − Eur. J. 2018, 24, 2837–2840.

105

Chapter 5

The Discovery of the Nav1.7 Inhibitor GDC-0276 and Development of an Efficient Large-Scale Synthesis Andreas Stumpf,*,1 Daniel Sutherlin,*,2 Christoph M. Dehnhardt,3,1 and Rémy Angelaud1 1Department of Small Molecule Process Chemistry, Genentech, Inc.,

1 DNA Way, South San Francisco, California 94080, United States 2Department of Discovery Chemistry, Genentech, Inc., 1 DNA Way, South San Francisco, California 94080, United States 3Medicinal Chemistry, Xenon Pharmaceuticals, 3650 Gilmore Way Burnaby, BC, Canada V5G 4W8 *E-mails: [email protected] (A.S.); [email protected] (D.S.).

The voltage-gated ion channel Nav1.7 has been identified as an important mediator of pain signals and has emerged as a promising therapeutic target for the treatment of chronic pain. GDC-0276 is a potent inhibitor of the channel that demonstrated pharmacokinetic properties and pharmacological activity that warranted further exploration in the clinic, progressing to Phase I studies. Following a summary of the key medicinal chemistry discoveries that led to GDC0276 and its initial synthesis, the process chemistry development that enabled its advancement to clinical studies is described in detail.

The Discovery of the Nav1.7 Inhibitor GDC-0276 Chronic pain is estimated to affect 20% of the U.S. population, resulting in an estimated direct and indirect economic burden of about $600 billion (1). While many pain therapies are well known, they tend to be associated with side effects that limit their dosage and, as a consequence, their efficacy, especially when used to treat chronic pain. As an example, opioids can lead to respiratory depression, constipation, somnolence, and other undesired adverse effects. In addition, opioids are associated with other challenges relating to tolerance and serious addiction. Recently, opioid addiction cases have risen to an epidemic level in the United States, with reported opioid overdoserelated deaths exceeding deaths from vehicle accidents. All of these shortcomings contribute to a strong desire to identify safer, nonaddictive, and more effective pain treatment drugs (2). Voltage-gated sodium channels (Nav's) constitute a family of nine isoforms named Nav1.1 to Nav1.9. Nav1.7, a member of this family, has been strongly linked to pain through human genetics. Mutations that activate the channel have been linked to pain syndromes where individuals sense extreme pain in response to normal stimuli. Additionally, mutations that effectively deactivate the © 2019 American Chemical Society

Nav1.7 channel have been identified in individuals who do not experience pain, even in response to normally very painful events (3–6). The Nav1.7 channel, expressed almost exclusively in the nerve cells, is a membrane-spanning protein that functions by sensing a change in voltage between the inside and outside of the cell. This process is triggered by nociceptors found at the nerve termini that are capable of initiating cellular depolarization in response to interactions with painful stimuli (temperature, chemical, physical). Once a threshold is reached, the channel opens and initiates an action potential that cascades through the nerve, in effect, transmitting pain signals to the spinal cord (7, 8). Therefore, selective inhibition of Nav1.7 represents a viable target for the treatment of pain. Efforts to safely inhibit Nav1.7 were boosted by the discovery of compounds that selectively inhibit this channel relative to the eight other Nav isoforms that are expressed in a variety of tissues throughout the body. Arylsulfonamides represented by 1 were shown to potently inhibit Nav1.7 while completely sparing Nav1.5, a channel that is critical to heart function (Figure 1) (9–11). This level of selectivity was notable relative to previously established inhibitors that have similar levels of affinity across Nav isoforms and was found to be the result of binding to the fourth voltage sensing domain (VSD) of the channel, a site that is unique from the typical binding site of historical Nav blockers (the highly conserved pore region). Binding of these molecules to the VSD effectively locks the channel in an inactivated state (12).

Figure 1. Structure of arylsulfonamide 1 and acylsulfonamide 2. In addition to the arylsulfonamide class, acylsulfonamides represented by 2 were also reported to be exquisitely selective for inhibition of Nav1.7 relative to Nav1.5. Although the early acylsulfonamides were on average less potent compared to known arylsulfonamides, they tended to have relatively lower molecular weights and lower calculated logPs. Combined, these two trends contributed to overall more favorable drug-like properties for the acylsulfonamide class, thus we chose to explore this class further in our pursuit of selective Nav1.7 inhibitors. Several assays were developed and used to evaluate the therapeutic potential of Nav1.7 inhibitors (13). Direct binding to the VSD of Nav1.7 was determined through a radiolabeled ligand binding assay. A plasma shift assay was used to rapidly measure the impact of plasma protein binding (PPB) on potency and was expressed as a ratio between two cell-based assays, one containing a fraction of plasma proteins and one lacking plasma proteins. To predict human pharmacokinetics (PK), we measured the stability of molecules when incubated with human liver microsomes (HLMs) and projected a hepatic clearance from those results. Lastly, a pharmacokinetic/pharmacodynamic (PK/ PD) assay in a novel transgenic mouse was employed, which reported activity on the human channel, expressed and selectively activated in the mouse. Acylsulfonamide 2 was inactive in our PK/PD assay at concentrations as high as 200 μM despite a reasonable potency of 36 nM in our binding assay (Table 1). This lack of activity was attributed to very high PPB of >99.9% and reflected in the plasma shift assay. Related compounds were designed by increasing the number of sp3 carbons in the molecule, a strategy expected to reduce PPB. Adamantane derivative 3 reduced the plasma shift ratio, had similar potency compared to 2, and had 108

a behavioral response in our PK/PD assay with an EC50 of 35 nM, despite being >99.9% plasma protein bound similar to 2. A 10-fold improvement in intrinsic potency that directly translated to improved PK/PD EC50s was realized by replacing the aryl Cl in 3 with the slightly larger cyclopropyl group in 4. These steps in the right direction were unfortunately offset by significant erosion in the metabolic stability. This was measured by the projected clearance in HLM going from a stable 4 mg/mL/kg in the chloroaryl derivative 2, to 19 mg/mL/kg for both adamantane derivatives 3 and 4, approximating the liver blood flow rate in humans. This lack of stability was attributed to increased oxidation of the adamantyl group by cytochrome p450s. Attempts to alter this metabolic route by blocking sites of oxidation with fluorines was successful, represented by 5 and 6, but addition of the more polar functionality was met with a meaningful reduction in potency. Moderate stabilities were obtained by making alterations in the arylsulfonamide portion of the molecule, a change that was anticipated to affect the recognition of the compound by cytochrome p450s. Ultimately, the sulfonylurea 7 (Table 1) was found to have the best balance of potency and HLM stability while maintaining an EC50 in our PK/PD assay in the low single-digit μM range. Additional safety and efficacy data collected for 7 and other related molecules contributed to the selection of the compound for further development. Table 1. Structure-Activity Relationship (SAR) for Selected Nav1.7 Inhibitors

The synthetic route used to prepare 7 initiated with 3-chloro-4,6-difluorobenzoic acid (8) (Scheme 1) was representative of the route used for the synthesis of all other compounds outlined in Table 1. Following protection of 8 by preparing the t-butyl ester 9, 1-adamantane methanol was used in an SNAr reaction with the ester to prepare compound 10. A palladium-mediated cross-coupling between aryl chloride 10 and cyclopropylboronic acid (CPBA) followed by t-butyl deprotection with trifluoroacetic acid (TFA) provided acid 12. 1,1’-Carbonyldiimidazole (CDI) was then used to couple the sulfonylurea 16, prepared in three steps from chlorosulfonyl isocyanate (CSI) 13, to acid 12 to yield 7. 109

Scheme 1. Discovery chemistry synthesis of GDC-0276.

Development of an Efficient Large-Scale Synthesis of GDC-0276 The initial discovery chemistry synthesis of GDC-0276 (Scheme 1) suffered from several shortcomings that needed to be addressed before scaling up to support early clinical studies. The main concerns with this route were • The formation of a significant amount of adamantyl ester 17 (~14 A%) during SNAr reaction of 9 with 1-adamantanemethanol (Scheme 2), which required purification by column chromatography; • The high catalyst loading (5 mol%) employed in the Suzuki–Miyaura cross-coupling between 10 and CPBA; • Labor-intensive workups including several extractions, solvent exchanges, and triturations.

Scheme 2. Synthesis of intermediate 10. 110

The synthesis provided room to further optimize stoichiometry and reaction conditions. Other objectives were to control the purge of impurities throughout the synthesis, isolate all intermediates and the active pharmaceutical ingredient (API) directly from reaction mixtures by crystallization, and obtain the desired crystal form of the API.

Development of First-Generation Synthesis First we attempted to reduce the amount of adamantyl ester 17 formed during the SNAr reaction of 9 with 1-adamananemethanol. Various bases (Cs2CO3, K2CO3, t-BuOK, LiHMDS) and solvents (DMF, DMSO, THF, THF/DMF, THF/DMSO) were screened to assess their impact on the selectivity of the SNAr reaction. None of the combinations tested showed an improved reaction profile compared to the original conditions used in the discovery chemistry. Therefore the original protocol was retained, and 8.3 kg of 10 was obtained after trituration with MeOH in 32% yield and 86 HPLC area %. Although the side-product adamantyl ester 17 was present in 14 HPLC area %, the undesired regioisomer 18 was almost completely purged to 98.9 area % by HPLC. 117

Scheme 7. Suzuki–Miyaura cross-coupling of 31 with CPBA and phase transfer catalyzed (PTC) hydrolysis of 26. New Sulfamide Synthesis The next objective was to reduce the formation of side products in the API that required an additional treatment with activated carbon. It was postulated that these impurities originated from precursors that were generated by the azetidine ring opening during the deprotection of Boc sulfamide 15 with TFA. Using a Cbz group instead of the Boc-protecting group in 34 would allow mild deprotection by hydrogenolysis in nonacidic conditions. Thus replacing t-butyl alcohol with benzyl alcohol during the sulfamide synthesis provided the corresponding Cbz-sulfamide 34 (Scheme 8). CSI 13 was treated with benzyl alcohol in DCM and NEt3 to provide zwitterion 33 and directly quenched with azetidine•HCl to give Cbz-protected 34 (97.4 A% HPLC area %). In order to further upgrade purity, crude 34 was recrystallized from EtOAc, acetone, and n-heptane to give 34 in 68% yield and 99.8 HPLC area %.

Scheme 8. New synthesis of sulfamide 16. Subsequently, the hydrogenolysis of 34 was examined using Pd/C catalyst and H2 in a variety of solvents (Table 6). Reactions were complete in most solvents within 4 h. Sulfamide 16 precipitated from reaction mixtures except when MeOH and acetone (Table 6, entries 4 and 6) were employed. Maintaining the product in solution allowed removal of the catalyst by filtration. When using MeOH, the catalyst loading could be reduced from 6 mol% down to 1 mol% while achieving a similar conversion of >99% and a cleaner reaction profile (90.2 by HPLC area %, Table 6, entries 5 and 7). After filtration to remove the catalyst, cyclopentyl methyl ether antisolvent was added to the resulting filtrate to crystallize 16, which was then isolated by filtration. The new optimized process delivered 48 kg of 16 in 87% yield and 99.9 GC area % purity. A sample from this material was brought forward 118

into the API amidation (see below), and in this case impurities 22 and 23 were not detected in the crude API by HPLC. Table 6. Hydrogenolysis of 34 in Various Solvents

Final Amidation and Crystallization of API Finally, the CDI-activated amidation of 12 with sulfamide 16 and the following direct isolation of the API from the reaction mixture were optimized. Toluene needed to be replaced with another solvent because it resulted in the isolation of the API as an undesired crystal solvate (see earlier). From a polymorph screen, EtOAc and iPrOAc were identified to give the desired polymorph Form I. Table 7. Amidation of 12. Reproduced from reference (15). Copyright 2019, American Chemical Society.

119

Using various bases, 12 was activated with CDI in EtOAc and the resultant acyl imidazole was treated with sulfamide 16 (Table 7). The use of DBU, DIPEA, or NaHCO3 generated mostly side products (Table 7, entries 1, 2, and 6). K3PO4 was identified as the best base in terms of reaction profile (Table 7, entry 6), and the use of iPrOAc as solvent produced the highest conversion (>99 HPLC area %) (Table 7, entry 7). However, attempting to isolate the API from the mixture after filtration, washing with aqueous HCl, and adding heptane as antisolvent resulted in only an ~85% recovery. In order to increase the yield, the solvent was swapped to 1-propanol, and 108 kg of GDC-0276 was isolated using the previously described conditions as the desired form I in 88% yield and >99.7 A% by HPLC (no single impurity >0.05 A%).

Conclusion The acylsulfonylurea GDC-0276 was identified as a potent and efficacious inhibitor of Nav1.7 and was chosen for clinical development. The compound was discovered by making modifications to the acyl sulfonamide class of VSD4 binders that increased the intrinsic potency on the target and also increased the free fraction available to interact with the channel as evidenced by a reduced ratio in a protein shift cell-based assay. The overall metabolic stability of lead molecules was improved by increasing the size of the acidic sulfonamide, ultimately resulting in the clinical candidate. A new synthesis of GDC-0276 has been developed via an eight-step chemical sequence starting with 1-chloro-2,4-difluorobenzene (24) (Scheme 9), and has delivered >100 kg of API for clinical trials in excellent purity in 64% overall yield. The route toward penultimate intermediate 12 has been completely redesigned and resulted in a three-fold increase of yield over the first-generation route. It features a highly regioselective SNAr of 24 and electrophilic bromination of 27, followed by subsequent palladium-catalyzed methoxycarbonylation of aryl bromide 25 to efficiently furnish methyl ester 31. Methyl ester 31 was taken into the Suzuki–Miyaura cross-coupling with CPBA and then directly telescoped into the PTC ester hydrolysis to provide 12. The synthesis of sulfamide 16 has also been modified to proceed through a Cbz instead of a Boc protecting group. The new protocol then allowed deprotection under hydrogenolysis and avoided side products that required labor intensive purification in the first-generation route. The final CDI activated amidation of 12 with sulfamide 16 has been further improved and was telescoped into the crystallization of GDC-0276 from 1-propanol to give the API as the desired polymorphic form in excellent purity and in 64% overall yield. This constituted a five-fold improvement over the old route.

120

Scheme 9. Optimized second-generation synthesis of GDC-0276.

References 1. 2. 3. 4.

5.

Gaskin, D. J.; Richard, P. The Economic Costs of Pain in the United States. J. Pain 2012, 13, 715–724. Preventable Deaths. Injury Facts. National Safety Council. https://injuryfacts.nsc.org/all-injuries/ preventable-death-overview/odds-of-dying/ (accessed June 10, 2019). Dib-Hajj, S. D.; Yang, Y.; Black, J. A.; Waxman, S. G. The Nav1.7 Sodium Channel: From Molecule to Man. Nat. Rev. Neuro. 2013, 14, 49–62. Cox, J. J.; Reimann, F.; Nicholas, A. K.; Thornton, G.; Roberts, E.; Springell, K.; Karbani, G.; Jafri, H.; Mannan, J.; Raashid, Y.; Al-Gazali, L.; Hamamy, H.; Valente, E. M.; Gorman, S.; Williams, R.; McHale, D. P.; Wood, J. N.; Gribble, F. M.; Woods, C. G. An SCN9A Channelopathy Causes Congenital Inability to Experience Pain. Nature 2006, 444, 894–898. Eberhardt, M.; Nakajima, J.; Klinger, A. B.; Neacsu, C.; Huhne, K.; O’Reilly, A. O.; Kist, A. M.; Lampe, A. K.; Fischer, K.; Gibson, J.; Nau, C.; Winterpacht, A.; Lampert, A. Inherited Pain: Sodium Channel Nav1.7 A1632T Mutation Causes Erythromelalgia Due to a Shift of Fast Inactivation. J. Biol. Chem. 2014, 289, 1971–1980.

121

6.

7. 8. 9.

10. 11. 12.

13.

14.

15.

16. 17.

Choi, J. S.; Boralevi, F.; Brissaud, O.; Sánchez-Martín, J.; Te Morsche, R. H.; Dib-Hajj, S. D.; Drenth, J. P.; Waxman, S. G. Paroxysmal Extreme Pain Disorder: A Molecular Lesion of Peripheral Neurons. Nat. Rev. Neurol. 2011, 7, 51–55. Wood, J. N.; Boorman, J. P.; Okuse, K.; Baker, M. D. Voltage‐Gated Sodium Channels and Pain Pathways. J. Neurobiol. 2004, 61, 55–71. Gold, M. S.; Gebhart, G. F. Nociceptor Sensitization in Pain Pathogenesis. Nat. Rev. Med. 2010, 16, 12481257. Swain, N. A.; Batchelor, D.; Beaudoin, S.; Bechle, B. M.; Bradley, P. A.; Brown, A. D.; Brown, B.; Butcher, K. J.; Butt, R. P.; Chapman, M. L.; Denton, S.; Ellis, D.; Galan, S. R. G.; Gaulier, S. M.; Greener, B. S.; de Groot, M. J.; Glossop, M. S.; Gurrell, I. K.; Hannam, J.; Johnson, M. S.; Lin, Z.; Markworth, C. J.; Marron, B. E.; Millan, D. S.; Nakagawa, S.; Pike, A.; Printzenhoff, D.; Rawson, D. J.; Ransley, S. J.; Reister, S. M.; Sasaki, K.; Storer, R. I.; Stupple, P. A.; West, C. W. Discovery of Clinical Candidate 4-[2-(5-Amino-1H-pyrazol-4-yl)-4-chlorophenoxy]-5-chloro-2fluoro-N-1,3-thiazol-4-ylbenzenesulfonamide (PF-05089771): Design and Optimization of Diaryl Ether Aryl Sulfonamides as Selective Inhibitors of Nav1.7. J. Med. Chem. 2017, 60, 7029–7042. Tfelt-Hansen, J.; Winkel, B. G.; Grunnet, M.; Jespersen, T. Inherited Cardiac Diseases Caused by Mutations in the Nav1.5 Sodium Channel. J. Cardiovasc. Electrophysiol. 2010, 21, 107–115. McKerrall, S. J.; Sutherlin, D. P. Nav1.7 Inhibitors for the Treatment of Chronic Pain. Bioorg. Med. Chem. Lett. 2018, 28, 31413149. Ahuja, S.; Mukund, S.; Deng, L.; Khakh, K.; Chang, E.; Ho, H.; Shriver, S.; Young, C.; Lin, S.; Johnson, J. P.; Wu, P.; Li, J.; Coons, M.; Tam, C.; Brillantes, B.; Sampang, H.; Mortara, K.; Bowman, K. K.; Clark, K. R.; Estevez, A.; Xie, Z.; Verschoof, H.; Grimwood, M.; Dehnhardt, C.; Andrez, J.-C.; Focken, T.; Sutherlin, D. P.; Safina, B. S.; Starovasnik, M. A.; Ortwine, D. F.; Franke, Y.; Cohen, C. J.; Hackos, D. H.; Koth, C. M.; Payandeh, J. Structural Basis of Nav1.7 Inhibition by an Isoform-Selective Small-Molecule Antagonist. Science 2015, 350, aac5464. Sun, S.; Jia, Q.; Zenova, A. Y.; Wilson, M. S.; Chowdhury, S.; Focken, T.; Li, J.; Decker, S.; Grimwood, M. E.; Andrez, J. C.; Hemeon, I.; Sheng, T.; Chen, C. A.; White, A.; Hackos, D. H.; Deng, L.; Bankar, G.; Khakh, K.; Chang, E.; Kwan, R.; Lin, S.; Nelkenbrecher, K.; Sellers, B. D.; DiPasquale, A. G.; Chang, J.; Pang, J.; Sojo, L.; Lindgren, A.; Waldbrook, M.; Xie, Z.; Young, C.; Johnson, J. P.; Robinette, C. L.; Cohen, C. J.; Safina, B. S.; Sutherlin, D.P.; Ortwine, D. F.; Dehnhardt, C. M. Identification of Selective Acyl Sulfonamide-Cycloalkylether Inhibitors of the Voltage Gated Sodium Channel Nav 1.7 with Potent Analgesic Activity. J. Med. Chem. 2019, 62, 908927DOI: 10.1021/acs.jmedchem.8b01621. Kinzel, T; Zhang, Y.; Buchwald, S. L. A New Palladium Precatalyst Allows for the Fast Suzuki−Miyaura Coupling Reactions of Unstable Polyfluorophenyl and 2-Heteroaryl Boronic Acids. J. Am. Chem. Soc. 2010, 132, 1407314075. Stumpf, A.; Cheng, Z. K.; Beaudry, D.; Angelaud, R.; Gosselin, F. Improved Synthesis of the Nav1.7 Inhibitor GDC-0276 via a Highly Regioselective SNAr Reaction. Org. Process Res. Dev. 2019DOI: 10.1021/acs.oprd.9b00082. Burgess, E. M.; Penton, H. R.; Taylor, E. A. Thermal Reactions of Alkyl N-Carbomethoxysulfamate Esters. J. Org. Chem. 1973, 38, 2631. Winum, J-Y.; Toupet, L.; Barragan, V.; Dewynter, G.; Montero, J-L. N-(tert-Butoxycarbonyl)N-[4-(dimethylazaniumylidene)-1,4-dihydropyridin-1-ylsulfonyl]azanide: A New Sulfamoylating Agent. Structure and Reactivity toward Amines. Org. Lett. 2001, 3, 22412243.

122

18. Although DCM is not a desired reaction solvent, its use was acceptable at this stage of development. For selecting sustainable solvents: Prat, D.; Pardigon, O.; Flemming, H-W.; Letest, S.; Ducandas, V.; Isnard, P.; Guntrum, E.; Senac, T.; Ruisseau, S.; Cruciani, P.; Hosek, P. Sanofi’s Solvent Selection Guide: A Step Toward More Sustainable Processes. Org. Process Res. Dev. 2013, 17, 1517–1525. 19. Diorazio, L. J.; Hose, D. R. J.; Adlington, N. K. Toward a More Holistic Framework for Solvent Selection. Org. Process Res. Dev. 2016, 20, 760–773. 20. Aq. citric acid is added as antisolvent to reduce solubility of 15 and purge residual NEt3. 21. Chakravarty, P. A.; DiPasquale, G.; Stumpf, A.; Lubach, J. W.; Nagapudi, K. Understanding Phase Behavior of Nearly Energetically Equivalent Polymorphs To Achieve Controlled Crystallization for a Nav1.7 Pain Inhibitor Compound. Mol. Pharmaceutics 2018, 15, 50725080. 22. Starting material 25 is available in bulk and about four-fold less expensive than starting material 9 based on molar ratio. 23. Ouellet, S. G.; Bernardi, A.; Angelaud, R.; O’Shea, P. D. Regioselective SNAr Reactions of Substituted Difluorobenzene Derivatives: Practical Synthesis of Fluoroaryl Ethers and Substituted Resorcinols. Tetrahedron Lett. 2009, 50, 37763779. 24. When the reaction was performed at 50 °C, significant amounts of a by-product, presumably the tbutoxide adduct, was formed. 25. Water content in THF used for the pilot plant batches was controlled ( 99:1 dr. The chiral auxiliary was removed under anhydrous acidic conditions (HCl, dioxane, MeOH, dichloromethane (DCM)), resulting in an 85% yield after solvent removal and crystallization from MTBE/dioxane. In this step, methanol was used as an additive resulting in formation of methyl tertbutylsulfinate ester (t-BuSO2Me) during the cleavage step, as measured by quantitative NMR and mass spectrometry (MS) techniques. DCM was required to enhance solubility. Establishing mass balance was necessary to confirm that iso-butylene was not produced during this step, which could present a safety risk. Distillative removal of MeOH and DCM prior to MTBE addition (until < 500 ppm each, measured by gas chromatography (GC)), proved essential to high yield and consistency during the crystallization. The final process proceeded in 48% yield over three steps and afforded 1.04 kg of amine 13 hydrochloride salt as a pale yellow crystalline solid in 99.7% chiral purity and 100 wt % potency (Scheme 5). This was sufficient to support all candidate selection activities allowing for selection of AMG 333 as a clinical candidate. The process was also successfully transferred to a manufacturing partner to generate further lots (3.3 kg and 8.4 kg) of amine-hydrochloride 13 to support subsequent GMP manufacturing campaign of AMG 333.

Scheme 5. Process chemistry first delivery of amine 13.

AMG 333 Process Development for First-in-Human Clinical Studies Discovery team efforts led to the promotion of AMG 333 as a clinical candidate. In preparation for manufacture of AMG 333 for clinical supplies, the process chemistry team performed an evaluation of the medicinal chemistry route. The primary focus of this assessment was to determine whether the synthesis could provide AMG 333: 134

1. With acceptable quality attributes for clinical study (including chemical and chiral purity, acceptable levels of process impurities, physical properties including particle size and crystal form, etc.); 2. With process safety characteristics that were compatible with standard manufacturing equipment; 3. With robustness and predictable yield to meet toxicology and clinical demands; and 4. In a timely manner to initiate clinical studies. Research areas targeted enhancing process understanding and development to address any gaps to meet these key criteria. The enabling synthetic route had been used to successfully manufacture AMG 333 on ca. 600 g scale and included a convergent assembly of the target in five synthetic steps from commercially available raw materials (Scheme 6). For Phase 1 regulatory filing, the team designated chiral amine 13 and commercially available carboxylic acid 17 as appropriate API Starting Materials. Strategically, pushback from regulatory agencies on these API Starting Materials could be addressed in regulatory updates. Through the Phase 2 clinical program, this starting material designation was endorsed by regulatory agencies. In analogy to the medicinal chemistry approach, assembly of these fragments could be performed through amide-bond formation, followed by ester hydrolysis to yield AMG 333. Crystallization of penultimate 18 was targeted as an opportunity to upgrade purity prior to the final step.

Scheme 6. Medicinal Chemistry scale-up route (600 g). Development would focus on process optimization of GMP steps because their performance would have a direct impact on the quality of the drug substance. As part of developing the final drug substance step, the crystal forms of AMG 333 were also characterized. A single batch would 135

be manufactured for use in both GLP-Tox and First-in-Human (FIH) studies, to ensure that drug substance impurities would be qualified in relevant toxicological models. Discovery lots of AMG 333 were generally obtained as a single crystalline form as measured by powder X-ray diffraction, defined as freebase anhydrous Form 1. AMG 333 displayed no significant polymorphism, although a hydrate form was characterized, which converted rapidly to the freebase anhydrous Form 1 upon slurrying in various organic solvents (esters, hydrocarbons, alcohols) or mixed organic or aqueous solvents. AMG 333 Form 1 showed no form conversion or mass change after dynamic vapor sorption analysis and was not considered hygroscopic. The physical properties of freebase anhydrous Form 1 provided flexibility of final crystallization solvents for isolation of AMG 333. The pKa values of AMG 333 are < 1.75 (pyridine-H, site not defined), 2.8 (CO2H), and > 11.2 (amide-NH) as measured by capillary electrophoresis (32, 33). AMG 333 Form 1 displays low solubility in water (0.11 mg/mL at pH 5.7), although solubility is highly pH-dependent. Preliminary evidence suggested that AMG 333 could racemize after extended time at pH 14, although this characteristic was poorly understood. While salt (Mg2+ and NH4+) and co-crystal (nicotinamide) forms were prepared, the physical properties of these backup forms provided no advantage over the freebase form, and freebase anhydrous Form 1 was used in all phases of development. The melting point of AMG 333 Form 1 is 165–166 °C, and differential scanning calorimetry (DSC) analysis shows a distinct melt/recrystallization/melt pattern (Figure 4). The second melt (ca. 214 °C) was confirmed to be racemic AMG 333 (defined as racemate Form 2), indicating that a thermal racemization of AMG 333 occurred in liquid form above 170 °C. Follow-up studies confirmed that thermal racemization of AMG 333 was possible at temperatures as low as 100 °C in aprotic solvents (dimethylsulfoxide (DMSO), 2-MeTHF, isopropyl acetate (IPAc), etc.), although the racemization rate was significantly slower in protic solvents (IPA, 2-BuOH, MeOH, H2O).

Figure 4. DSC analysis of AMG 333 freebase anhydrous form 1. AMG 333 crystallization experiments in aqueous IPA using nongrinding conditions (i.e., overhead stirrers instead of stir bars) resulted in birefringent crystalline rods with a typical length of 100–500 μm. Based on projected clinical dose ranges (3–300 mg), a target particle size criterion 136

was set at d90 < 50 μm to ensure content uniformity in drug product processing. Because of ease of operation and high product recovery, wet milling was used to engineer AMG 333 drug substance to an acceptable particle size. With an ample supply of API starting material amine hydrochloride 13 available (vide supra), the first planned GMP step for the synthesis of AMG 333 was the amide bond formation between amine 13 and acid 17. Several amide-coupling reagents were screened for this purpose, resulting in several unexpected side products (Scheme 7). 1,1′-Carbonyldiimidazole cleanly activated acid 17 to produce acyl imidazolide 19, however residual imidazole reacted with amine 13 to produce impurity 20. The lability of the benzhydrylamine also prevented application of thionyl chloride or oxalyl chloride as activators for acid 17, as side product ketone 22 was observed (~25 Area%) upon attempted reaction with in situ formed acyl chloride 21, after aqueous workup. Propylphosphonic anhydride (T3P) produced impurity imide 23, which could not be removed through crystallization from various solvent systems.

Scheme 7. Primary impurities observed, alternate amide-formation conditions. The impurity profile resulting from HATU/DIPEA-mediated amidation was superior to alternative conditions and was applied to scale-up (34, 35). Because of the exothermic nature of this reaction, the medicinal chemistry procedure involved portionwise addition of solid HATU to an ice-cooled stirring mixture of amine-hydrochloride 13, carboxylic acid 17, and DIPEA in N,Ndimethylformamide (DMF), at a rate to keep the internal temperature below 20 °C. This portionwise addition of solid HATU reagent was deemed challenging to scale. A control experiment was performed in the absence of DIPEA, in which the HATU-promoted amidation proceeded to only 4% conversion without any measurable racemization of aminehydrochloride 13. Based on these results, dropwise addition of liquid DIPEA to a mixture of amine 13, carboxylic acid 17, and HATU allowed dose control of conversion and reaction exotherm (Scheme 8). 137

After reaction completion, amide 18 was crystallized by addition of water (0.4 L/kg), seeding, and then continued water addition to complete the crystallization. Isolation from DMF/water completely removed intrinsic impurities from the HATU reagent including tetramethylurea (monitored by GC), hexafluorophosphoric acid (monitored by IC and 19F NMR spectroscopy analysis), and 1-hydroxy-7-azabenzotriazole (HOAt, monitored by HPLC). Side products from decomposition of amine 13, which were present using other amide-coupling techniques, were not observed in isolated solids or mother liquors.

Scheme 8. HATU-mediated amidation conditions on 2 kg scale. The main impurity in development batches was bis-amide 24, whose formation was dependent on bis-acid content in raw material 17. Bis-amide 24 was qualified in AMG 333 Tox batches up to 0.51%, and control was established in future deliveries by setting a bis-acid specification (≤2.0 %) in raw material 17. The first GMP campaign was performed on 2.05 kg scale of amide 18, and the amidation process supported the synthesis through the lifetime of the program. Ester hydrolysis of penultimate 18 was initially performed in the discovery group by biphasic reaction in 2-MeTHF and aqueous sodium hydroxide. Although this was successful on ~600 g scale, stress experiments revealed that AMG 333 was prone to partial racemization after prolonged stirring at the end of the reaction (~4% enantiomer over 7 h reaction time, Table 2). Since the rate of this biphasic reaction is subject to mass transfer effects, we anticipated that increases in reaction scale might lead to variable process performance. Alternative single-phase saponification conditions using potassium trimethylsilanolate (KOTMS) in 2-MeTHF (36) led to simultaneous conversion and racemization. A successful alternative was realized using NaOH in the protic solvent mixture IPA/water (Table 2, Entries 4 and 5). Although these conditions affected a reactive dissolution, the reaction rate displayed first-order kinetics, and the chiral purity of AMG 333 remained unchanged for extended hold times (Figure 5). The retention of chiral purity in protic solvents and lack of chiral stability under basic aprotic conditions proved consistent across several reagent/solvent combinations. A brief study of the thermal stability of AMG 333 and related analogues provided some further insight into the racemization (Figure 6). Structurally related compounds bearing pyridyl-carboxyl (25) or pyridinone (26) functional groups were racemized at 120 °C (butyl acetate (BuOAc) solvent, completion within 12 h). In contrast, substrates lacking these substitutions on the pyridine (ester 138

4, picoline 27) showed negligible racemization. Addition of exogenous acid (acetic acid (AcOH) or 2-picolinic acid, 5 equiv) did not promote racemization at 120 °C, suggesting that the position of the acidic proton within the substrate plays a role. The initial rate kinetics of AMG 333 thermal racemization displayed second-order dependency on starting material (AMG 333). Table 2. Conversion of Ester 18 to AMG 333 (9) Under Basic Conditionsa

Figure 5. First-order formation of AMG 333 and chiral purity during reaction. Inset illustrates first-order reaction rate kinetics. 139

Figure 6. Thermal racemization studies of AMG 333 analogues (in BuOAc, 120 °C, reaction time 12 h). Based on this structure–activity and kinetic analysis, a working mechanistic hypothesis of this thermal racemization involves substrate preorganization into a bimolecular hydrogen-bonding system leading to productive racemization. This hypothesis is further supported by additional control experiments. Whereas 2-picolinic acid could not promote racemization of ester 8, AMG 333 (9) could selectively promote racemization (Scheme 9), although only in aprotic solvents. If thermal and base-promoted racemization have similar requirements, then the presence of protic solvents (IPA, MeOH, H2O) may interfere with this template-directed hydrogen bonding, thereby shutting down the racemization pathway.

Scheme 9. Template-assisted racemization effects observed in ester 18. In the first GMP campaign, the optimized ester hydrolysis conditions performed well, resulting in high retention of stereochemistry in the AMG 333 product. For this campaign, there were concerns that direct acid-driven crystallization would result in some amorphous content and potential entrainment of inorganics (NaCl, sodium citrate, etc.) in the product. Therefore, the workup procedure involved inverse quench of the reaction stream onto a biphasic mixture of 2MeTHF and aqueous HCl, followed by a phase-cut and extraction into aqueous 2-MeTHF, and finally, water washes to remove inorganics. The 2-MeTHF solution was subjected to a distillative 140

solvent switch to yield an IPA solution, which was seeded with AMG 333 Form 1. Water was subsequently added to induce further crystallization. To ensure the desired particle size, the final slurry was circulated from the reactor through an IKA Magic Mill wet mill using a Medium/Fine/Fine rotor/stator configuration. The end point of particle size reduction was monitored by in situ focused beam reflectance measurement as a For Information Only test, and by offline microscopy as an inprocess test. The milling end point was established within six batch turnovers (2 L/min, 60 minutes total) in line with development batches, and the batch was milled for a total of 12 batch turnovers (Figure 7).

Figure 7. Microscopy used to monitor AMG 333 particle size reduction through wet milling (a) before wet milling, (b) after 6 batch turnovers, and (c) after 12 batch turnovers. Following wet milling, the batch was filtered and washed with aqueous IPA to produce 1.49 kg of AMG 333 in 78% isolated yield as the desired freebase anhydrous Form 1, with 98.1 A% purity and 99.89% enantiomeric purity (Scheme 10). Particle size distribution analysis of the dry solids met the predefined criteria of d90 < 50 μm (d90 = 42.3 μm, wet method).

Scheme 10. First GMP synthesis of AMG 333. The first GMP manufacture of AMG 333 was performed internally to provide maximum timeline flexibility. The GMP campaign was completed 17 days after advancement of AMG 333 as a clinical compound, producing 1.49 kg of API. The first GMP batch of AMG 333 was used for both GLP-Tox and FIH clinical studies. A number of impurities were identified in the batch and were qualified through these toxicological studies (Table 3). Preliminary process characterization studies identified the origins of these impurities. Strategies were implemented to control levels of each impurity in future deliveries. The program was terminated prior to finalization of process characterization and definition of a commercial specification. 141

Table 3. Qualified Impurities in First GMP AMG 333 Lot

Second-Generation AMG 333 Crystallization and Additional GMP Campaigns The initiation of FIH trials allowed the process development team time to evaluate the performance of the chemical process. Phase 2 clinical projections indicated that annual drug substance demand (clinic and tox) would increase from 1 kg to >20 kg over the next 2 years. This increased material demand required the transfer of the GMP steps to an external partner for drug substance manufacture. The first internal GMP campaign of AMG 333 featured a fit-for-purpose crystallization of the API that provided acceptable impurity control while producing the desired crystal form and particle size. However, subsequent studies revealed the need to further improve the robustness of the API workup and crystallization to support tech transfer activities. In the first GMP campaign, the final step of AMG 333 manufacture involved basic ester hydrolysis conditions in IPA/water. The product was extracted into 2-MeTHF, distilled to produce a saturated solution in IPA, crystallized by seeding from IPA, and water was added to complete the crystallization. In subsequent form screening studies, a new metastable IPA solvate form was discovered that proved thermodynamically stable in pure IPA (Figure 8). This indicated that under the crystallization conditions employed in the first GMP campaign, an IPA solvate could have formed during the seeding step. Analysis of the in situ focused beam reflectance measurement spectra from this campaign indicated that this form was not produced at levels at which it could be detected. However, based on these findings, it was determined that a new crystallization procedure should 142

be developed, including seeding at 95% conversion at the end of DIPEA addition and proceeded to >99% conversion within 30 minutes (passed in-process limit: ≤1.0%). Following a polish filtration, water (0.4 L/kg) was added to reach the seed point and then 1 wt % seed was charged as a slurry in 0.10 L/kg of 40% H2O/DMF. The seed held at 20 °C and was allowed to grow for 2 hours, resulting in a robust mobile seed bed. Additional antisolvent (1.6 L/kg of water) was added over 2 h, and the reaction was stirred overnight to complete the crystallization. Prior to filtration, the supernatant concentration was measured to be 11 mg/mL (passed inprocess limit: 50% I) only 2 kinases (including TrkA) in an Invitrogen panel of 40 biochemical kinase assays, giving an improved Kinase Gini score of 0.82 (Figure 4). The TrkA co-crystal structure of desamino pyrrolopyrimidine 3, a close-in analogue of 2 in which the p-Cl phenyl group was replaced with a p-CN phenyl, is shown in Figure 5. Des-amino pyrrolopyrimidines such as 2 and 3 were shown to suffer from significant aldehyde oxidase (AO)–mediated oxidation at the weakly acidic aromatic C-H between the pyrimidyl nitrogen atoms. While methods of scaling in vitro intrinsic clearance data to predict the in vivo clearance of compounds are well-established for CYP P450-dominated clearance, similar methods have not yet been established for drugs metabolized by AO. As a result, the impact of having AO-mediated metabolism cannot be fully established until the compound has reached the clinic (33, 34). 158

Figure 4. Overview of strategies to improve Absorption Distribution Metabolism and Excretion and kinase selectivity properties of 1. Hence, we sought to remove the AO liability in the pyrrolopyrimidine series, and a number of medicinal chemistry design strategies were investigated in order to achieve this objective. One of the strategies explored was the addition of an –NH2 “blocking group” at the metabolically labile 2position. This approach delivered compounds such as 4, with comparable or improved potency and LipE. Compound 4 was a potent and selective Trk kinase inhibitor, with a high P-gp ER. As with other 2-aminopyrrolopyrimidines in this series, Met ID studies with human S9 in the absence of NADP confirmed no AO-mediated oxidation on the 2-aminopyrimidyl ring. The logD of 4 was high (4.2), likely contributing to the relatively poor solubility of 12 µM. Pleasingly, 4 was comparable (in fact slightly superior) to des-amino pyrrolopyrimidines 2 and 3 in terms of kinase selectivity, even though there was the potential for an additional H-bond with the conserved hinge region. However, as exemplified in Figure 7, the –NH2 group of the 2aminopyrrolopyrimidine hinge binder made an imperfect hydrogen bonding interaction with the backbone carbonyl of hinge residue Met592. Due to the sub-optimal geometry of the donor–acceptor-donor arrangement (O…H-N bond angle is 122° cf. an ideal O…H-N bond angle of 180°) the additional hydrogen bond was not likely to contribute significantly to the free energy of binding, and hence did not erode kinase selectivity (35).

Figure 5. Crystal structure of 3 bound to TrkA. Key protein-ligand binding interactions are outlined. The circle highlights the hydrogen bond interactions made by a conserved water molecule bridging the main chain

159

carbonyl Glu590 and the ketone oxygen atom of compound 3. Reproduced with permission from reference (11). Copyright 2016 American Chemical Society. Candidate Seeking In order to design 2-aminopyrrolopyrimidine analogues with enhanced solubility and an improved LipE profile, compounds of reduced logD were required. An analysis of available Trk co-crystal structures suggested polar groups would be tolerated at the N-isopropyl motif of 2aminopyrrolopyrimidine compounds such as 4, as this region of the binding site is highly solvent accessible. Pleasingly, addition of a solubilising hydroxymethylene group to the N-isopropyl motif generated compounds such as 5 that retained Trk potency, increased LipE because of a lower logD, and effected an improved solubility profile. In addition, replacement of the chlorophenyl ring with a lower logD pyridyl motif delivered ligands such as PF-06273340, with further optimized LipE and solubility profiles (Figure 6).

Figure 6. Overview of strategies to improve Absorption Distribution Metabolism and Excretion properties of 4 and deliver PF-06273340. As highlighted in Figure 6, PF-06273340 was a highly potent pan-Trk inhibitor, with an excellent LipE profile. PF-06273340 had low metabolic turnover in hHeps (AO oxidation was not detected on the amino pyrrolopyrimidine substituent in human S9) and was a good substrate for the efflux transporter P-gp (ER = 36). The aqueous solubility of PF-06273340 was 131 µM, much improved over previous analogues, and the compound exhibited high kinase selectivity (Gini score of 0.92). PF-06273340 was profiled in a series of in vitro safety assays, showing little cytotoxicity in THLE and HepG2 cell lines (IC50 >42 μM and >300 μM, respectively) and was evaluated for broader pharmacological activity in a panel of receptors, ion channels, and enzymes. In this broad panel, all IC50/Ki values were >10 μM except for COX-1 (IC50 = 2.7 μM) and dopamine transporter assays (Ki = 5.2 μM) and PDEs 4D, 5A, 7B, 8B, and 11 (54–89% inhibition at 10 μM). Figure 7 shows the crystal structure of PF-06273340 bound to TrkA kinase, and highlights key protein-ligand interactions. PF-06273340 adopted a DFG-out binding mode; the 2-amino pyrrolopyrimidine hinge binding motif makes a hydrogen bonding interaction to the backbone N-H of Met592 via a pyrimidyl nitrogen. A “through water” hydrogen bond network exists between main chain carbonyl Glu590 and the ketone oxygen atom of PF-06273340. An additional hydrogen bond existed between the –NH2 group of the 2aminopyrrolopyrimidine hinge binder and the backbone carbonyl of hinge residue A592, but with sub-optimal geometry (O…H-N bond angle, 122°). The central pyridyl group of PF-06273340 160

made a π-stacking interaction with the gatekeeper (Phe589) and Phe669 of the DFG triad. The carbonyl oxygen of the amide group made a hydrogen bond interaction with the backbone N-H of Asp 668 from the DFG triad and the amide N-H group made a hydrogen bond to a conserved water molecule, which further interacted with catalytic lysine (Lys544) and Glu560 from the alpha-C helix. The lipophilic back pocket accommodated the chloropyridyl group. Brain availability experiments with PF-06273340 in rat confirmed significant peripheral restriction, with Bru/Plu = 0.026. In addition, no measurable TrkB or TrkC occupancy was detected from brain lysates using the ActivX biotinylated acyl phosphate ADP probe. PF-06273340 was assessed in the ultraviolet irradiation induced hyperalgesia model of inflammatory pain in rodents. Statistically significant efficacy was observed at unbound plasma concentrations of ≥1 × TrkA IC50, with maximal efficacy at an unbound plasma concentration of ~10 × TrkA IC50. The safety profile of PF-06273340 was assessed in a 14 day rat toxicology study. Overall PF-06273340 was well tolerated up to 1000 mg/kg/day and was selected as a candidate for clinical development.

Figure 7. Crystal structure of PF-06273340 bound to TrkA. Key protein-ligand binding interactions are outlined. Reproduced with permission from reference (11). Copyright 2016 American Chemical Society.

Medicinal Chemistry Synthesis of PF-06273340 Within the Pfizer organization, the initial scale-up work for toxicology studies is carried out in the medicinal chemistry synthesis teams. To facilitate effective delivery of batches of up to hundreds of grams, close partnership between the medicinal chemistry synthesis team and the early process chemistry team is essential. This close cooperation should ideally ensure that the route used by the medicinal chemistry team to synthesize the batches used for initial toxicology studies is very close in nature to the route used by the process chemistry team for the synthesis of the initial regulatory batches; the development of a robust scalable route at this early stage is designed to ensure that the delivery of bulk compound to key studies occurs with speed and without major issues. At the time that PF-06273340 was being scaled-up, this process was fairly new, and work on this project helped define the process used on other projects. As the pan-Trk project was seen as a key part of the pain portfolio, the teams were challenged to move forward to scale-up as quickly as possible, and interactions between the medicinal chemistry team and the process chemistry team occurred regularly from an early stage on the project. To ensure that the project delivered with speed, the medicinal chemistry synthesis team needed to balance the requirements of identifying routes capable of scaling compounds to hundreds of grams in a short time frame with the need to rapidly establish

161

structure-activity relationships (SAR) within the series. Balancing these differing requirements has remained a challenge to this day, as projects always wish to move forward as quickly as possible. For the pan-Trk project, work quickly focused in on the pyrrolopyrimidine structures exemplified by the final candidate PF-06273340. A retrosynthesis of this compound is shown in Scheme 1. The chloroiodo pyrrolopyrimidine 7 proved to be a flexible intermediate, allowing rapid examination of the SAR across 3 different vectors of the molecule. At the same time that the SAR was being generated, the feasibility of this strategy for scale-up was also being considered. A number of methods to directly build the ring system of chloroiodopyrrolopyrimidine 7 were investigated concurrently. Pragmatically, starting from the commercially available chloropyrrolopyrimidine 8 was highly attractive, as the iodination-alkylation sequence proved to be highly effective. There were a range of options to convert the chloro group to an amine and the iodo to a ketone (all of which were investigated during the SAR generation).

Scheme 1. Retrosynthesis of PF-06273340. As the SAR was generated, it became clear that there were 3 compounds in play for scale-up. Work on all 3 was optimized in the project team, working on the assumption that the medicinal chemistry team would need to synthesize around 30–40 g of each. The early process chemistry group would then quickly need to make around 60 g and then, soon thereafter, 1 kg batches. Once it became clear that PF-06273340 was the scale-up candidate, the medicinal chemistry team worked with input from the process chemistry team on refining the route for scale-up and generating around 35 g of material. For facilitating the scale-up of PF-06273340 at speed by the medicinal chemistry team, chromatography would be removed from reaction steps where possible, but the primary focus of the optimization would be identifying the optimal step sequence and maximizing yields of key reactions. It is interesting to note that in subsequent years, thinking in this area has evolved. Far greater emphasis is now placed on removing chromatography at this initial scale-up point, rather than addressing it in subsequent campaigns.

Scheme 2. Synthesis of pyrrolopyrimidine 11.

162

In terms of the defining a scalable approach to PF-06273340, a number of challenges in the route needed to be addressed. Iodination of chloropyrrolopyrimidine 8 proved straightforward and scalable (the use of DMF, while not optimal for scale-up campaigns due to toxicity concerns, could be eliminated in later scale-up campaigns). However, a scalable method to incorporate the dimethylated alcohol side chain onto the pyrrolopyrimidine was challenging to identify and proved key to defining an effective scale-up route (Scheme 2). Alkylation of the pyrrolopyrimidine had generally proven to be straightforward, but a direct SN2 alkylation with a tertiary halide was not feasible and so incorporation of the dimethylated side chain required some optimization. A key breakthrough in this area was the discovery that alkylation with dimethylated bromoester 9 was successful (presumably, bromoester 9 is sterically less encumbered than a tertiary halide as well as being electronically activated towards SN2 reaction by the presence of the carbonyl next to the CBr bond), and this could then be followed by reduction. The initial alkylation proved optimal with cesium carbonate as base and with potassium iodide as additive to increase the overall rate of the reaction. The seemingly straightforward reduction of the ester to alcohol 6 proved challenging on scale, with a range of typical reducing agents failing to give good levels of product. A number of options were identified for this step via a reaction optimization screen. This revealed that hydrolysis to the acid, conversion to the CDI ester, and reduction with sodium borohydride proved to be effective. However, direct reduction of the ester to the alcohol was ultimately superior, and 2 scalable methods, super-hydride or DIBAL-H, were identified. The resulting alcohol crashed out on work-up for either and could be crystallized from ethanol to give clean material in yields of around 70%. Subsequent protection of the alcohol as the silyl ether 11 proved necessary for subsequent manipulation of the iodide. The other key challenge in the sequence was the incorporation of the amino-pyridyl ketone motif into the molecule to access ketone 13 (Scheme 3). Fairly early in the SAR generation, it had been established that the benzophenone imine was a robust protecting group for the amino group on the pyridine. While palladium-mediated carbonylation and cyanation of silyl ether 11 were both viable on scale, subsequent reaction with an organometallic reagent was then required to generate the ketone. Direct metalation of the iodide of the pyrrolopyrimidine using iso-propylmagnesium chloride was carried out at 0°C over 1 h. The resulting Grignard species reacted with the Weinreb amide 12 (synthesized from 5-bromonicotinic acid or 5-aminonicotinic acid as discussed later—see Scheme 7) to give the desired ketone 13 directly. For the initial scale-up work targeting 35 g of final compound, the decision was made to utilize chromatography to purify the material due to time constraints. The medicinal chemistry team established that the purity of the solid product could be enhanced through trituration in n-heptane/ ethyl acetate but in consultation with the process chemistry team, it was agreed that optimizing purification procedures would be addressed in later scale-up batches. With ketone 13 in hand, amination of the chloropyrimidine moiety could be carried out using an SNAr reaction with 2,4-dimethoxybenzylamine (DMBA) (14) in the presence of DMAP in dioxane. The use of dioxane was not ideal due to toxicity concerns on scale, but additional optimization of the reaction would be carried out in the background as the scale-up proceeded. Based on work with other compounds, there was concern about the feasibility of getting a completely selective amidation reaction to occur, and to mitigate this risk, selective amine deprotection was utilized. The benzophenone imine group could be deprotected cleanly in the presence of the DMB group using citric acid to produce the DMB protected compound 15, which 163

was purified by chromatography. Subsequent amide coupling using T3P proved high yielding and both the DMB and silyl protecting groups were subsequently removed using TFA. Slurrying of the crude API in ethanol produced a high purity crystalline material. We ultimately synthesized about 70 g of PF-06273340 in two batches of 35 g each using this route (thus overdelivering on the targeted 35 g). With the finalized route in hand, the hand-off to the process chemistry group then occurred and a review of the process used in the medicinal chemistry group identified the key improvers that could be addressed in subsequent campaigns.

Scheme 3. Completion of PF-06273340 medicinal chemistry synthesis.

Early Stage Process Chemistry Development Early development needs for the project were assessed with regard to bulk demands to meet study timelines and strategies for immediate supplies of PF-06273340. At this point, medicinal chemistry had provided the initial 35 g bulk request, but further demands for 60 g and a 1 kg regulatory toxicology batch were tasked to Chemical Research and Development. Hence, efforts started with evaluation of the original medicinal chemistry route (Scheme 4). This was made more facile due to the close partnership between the 2 groups in the design of the initial route. While the medicinal chemistry route could be used effectively to provide the early research demands of PF06273340, a critical assessment of this route identified the following concerns that needed to be addressed before immediate scale-up of the synthesis: • Seven linear steps (although some were multiple transformations) with no immediate opportunity for convergent synthesis evident. • The reduction of the ester to the corresponding alcohol using DIBAL-H, which is a nonpreferred reductant for large-scale manufacture. • A number of different protecting groups were used throughout, which would limit throughput. • A capricious Grignard exchange reaction existed which routinely failed to initiate and gave significant amounts of de-halogenation. • Nondesirable solvents such as DMF and dioxane were used. • Multiple column chromatography purifications were required to obtain the desired purity at key steps of the synthesis. 164

Although it would not be ideal as the eventual commercial route for manufacture, it was felt that with some optimization and further process understanding, the original medicinal chemistry route could be used to provide early supplies of API up to at least a number of kilograms. In order to limit the number of steps run internally, we decided to outsource a number of early synthetic intermediates. Based on our knowledge of the existing chemistry and intermediates, we selected the chloroiodopyrrolopyrimidine ester 10 and pyridine 16 (both are stable and highly crystalline materials) for external manufacture.

Scheme 4. Medicinal Chemistry route to PF-06273340. Manufacture of the Exploratory Toxicology Batch of API There was an immediate need to support a 1 kg Exploratory Toxicology batch, and this would need to be done through an in-house synthesis. Initial pilot runs in the laboratory did not identify any concerns for immediate scale-up. However, during a laboratory large-scale alkylation of chloroiodopyrrolopyrimidine 7 with bromoester 9, a colourless liquid collected within the condenser return U-bend, which tended to solidify on standing to a plastic-like material. Given this finding, we were unable to complete the alkylation step in the pilot plant, and as an alternative, we had to conduct a series of 14 preparative experiments to produce a total of 400 g in the large-scale laboratory. 165

Iodination of Chloropyrrolopyrimidine 8 The iodination of chloropyrrolopyrimidine 8 with NIS in DMF was assessed in the laboratory and was determined to be amenable for immediate safe scale-up. As such, 4 batches of chloroiodopyrrolopyrimidine 7 were prepared using the conditions developed by medicinal chemistry. This furnished a total of 5.6 kg of 7 with good purity and an average yield of 84%. Alkylation of Chloroiodopyrrolopyrimidine 7 with 9 To Produce Chloroiodopyrrolopyrimidine Ester 10 A total of 5.4 kg of chloroiodopyrrolopyrimidine 7 was processed across 14 batches and afforded a total of 6.1 kg of chloroiodopyrrolopyrimidine ester 10 averaging a yield of 85%. During these reactions, we cautiously monitored for the formation of the colourless liquid seen previously in our equipment train. Analysis of the colourless liquid identified it to be a 7:3 mixture of esters 9 and 17 (Figure 8), which were believed to subsequently polymerize to the plastic-like material observed in the U-bend. Despite our best efforts, it was not possible to determine when and why this mixture would condense. But as it would occur in some reactions and not others, we postulated it may result due to varying reaction temperatures and rates of nitrogen sweep. However, fortunately the material was not present as a contaminant in our product and we were still able to progress through the remainder of our synthesis.

Figure 8. Esters 9 and 17 identified in equipment train (7:3 ratio) during alkylation of compound 7 with 9. Reduction of Chloroiodopyrrolopyrimidine Ester 10 with DIBAL-H The reduction of chloroiodopyrrolopyrimidine ester 10 to the alcohol 6 was achieved with DIBAL-H in THF, which, albeit not ideal, was preferred to super-hydride from a large scale handling perspective. Despite smaller scale reactions only affording modest yields, it was pleasing that larger scale reactions on >1 kg scale afforded excellent quality material in yields of >85%. This was believed to be due to better crystallization techniques giving rise to much better purge of aluminium byproducts on scale and the subsequent removal of chromatography. A total of 5 batches of chloropyrrolopyrimidine alcohol 6 were synthesized in >95% purity with an average yield of 90%. Protection of Chloroiodopyrrolopyrimidine Alcohol 6 with TBDMSCl Early lab assessment of the protection of chloroiodopyrrolopyrimidine alcohol 6 identified a number of concerns. First, the product from the previous step contained variable levels of water (up to 3.7%) and this compromised the performance of the reaction due to TBDMSCl hydrolysis. To allow for this diminution, the equivalents of TBDMSCl were increased to 1.6 equivalents to tolerate the variability in water levels. Second, as a consequence of the greater TBDMSCl charge, there resulted significant contamination with TBDMSOH of the isolated product oil following work-up and distillation. This material sublimed or distilled during the distillation used to remove the t-butyl methyl ether solvent and was evident as a white solid on the condenser coil. Co-distillation with toluene and further 166

distillation afforded an oil that, having only trace levels of TBDMSOH, was successfully utilized in the subsequent step. Preparation of Grignard 18 from Iodo 11 and Reaction with Weinreb Amide 12 The preparation of Grignard reagent 18 (see Scheme 5) from compound 11 had routinely caused problems in the lab from no initiation occurring on a number of occasions. The success of this reaction was visually evident by the formation of a “blood-red” solution confirming metal exchange was successful. From further process understanding it was found that levels of water in smaller lab reactions were quenching metalated intermediate 18 and consequently giving significant amounts of des-iodinated impurity 19.

Scheme 5. Preparation of Grignard reagent 18 with key des-iodinated impurity 19. A total of 4 large-scale Grignard reactions were completed on up to 1.7 kg scale of iodo silyl ether 11 to provide 8.8 kg of crude 13, which was purified by chromatography in 3 batches in the pilot plant to afford 4.6 kg of clean keto-imine 13 in an overall yield of 57%. Despite our best efforts in the laboratory, this material had never been isolated as a solid previously. But following distillation to remove solvent post-chromatography, a solid precipitated. This was resuspended in n-heptane and filtered to afford a free flowing granular orange solid. Further analysis by HPLC and microscopy indicated this material to be of >95% purity and crystalline, hence offering a possible purification point for future batches without the need for chromatography. SNAr Amination of Keto-Imine 13 with Amine 14 and Subsequent Benzophenone Imine Hydrolysis to Keto-Amine 15 Although SNAr amination of keto-imine 13 used 7 equivalents of DMBA and was performed in the undesirable solvent 1,4-dioxane, it consistently provided moderate to high yields of the desired product (following chromatography). As such, this reaction was scaled up without further optimization. A total of 4.4 kg of 13 was processed to afford 5.5 kg crude keto-amine 15, which gave greater than 3.0 kg in an overall yield of 72% as an off-white solid following chromatography. This was the first evidence of the desired intermediate residing as a solid, as previous lab efforts had always isolated an oil. This gave promise that future campaigns may be able to avoid chromatography at this stage and make use of crystallization to purify. Preparation of 15 to PF-06273340 Due to time constraints, no optimization of the process was conducted on this step. It had been shown to reliably furnish the required compound in high yield, albeit with a number of processing challenges. Thus, amidation of keto-amine 15 with pyridine acetic acid 16 was achieved via T3P 167

activation in the presence of TEA on 2 kg scale. The reaction was quenched with saturated NaHCO3 solution and concentrated to a foam before TFA was added to complete the deprotection of the DMB and TBDMS protecting groups. The solution was concentrated to remove TFA and replaced with methanol. The removal of an insoluble DMB-related polymeric material was achieved by filtration. The mixture was made basic by addition of solid potassium carbonate to cleave a bis-amide impurity. The resulting solution was added to vigorously stirred water to precipitate the product as an orange solid. Following filtration, 1.3 kg of crude PF-06273340 were obtained in a yield of 81%.

Further Process Chemistry Development for Multi-Kilogram Scale-Up The amount of API required to support regulatory toxicology studies and initial clinical studies was 5 kg. Following this delivery, a second 19 kg campaign was requested for further program progression. Over the course of time and reassessment of program needs, the second API delivery amount was significantly reduced down to 3 kg. At the time the process enabling was carried out, 19 kg was still the target. The first API delivery was produced in a kilo lab setting, while the second delivery started in the pilot plant but then transitioned to the kilo lab once the API demand decreased. In the end, 4.4 kg of API was made in the first campaign followed by 3.2 kg of API in the second campaign. The commentary below describes the learnings over the course of enabling the chemistry to make PF-06273340 on kilogram scale. In order to maintain the early phase momentum of API delivery and project progression, the established synthetic sequence was deemed fit-for-purpose and was conserved in the kilogram campaigns. However, significant effort was still dedicated to increasing process understanding. Optimization was performed at each step to ensure robustness of the chemistry upon scale-up. With the process used to manufacture the exploratory toxicology supplies of API as the starting point, the main synthetic issues to address were avoidance of silica gel chromatography purification for intermediates 13 and 15 as well as removing several concentrate-to-dryness operations throughout the synthesis. Also, a more robust endgame was needed since the existing process telescoped the final 2 steps together and did not offer an opportunity to obtain high purity penultimate intermediate to set the stage for control of final API purity. Outsourcing/Starting Material Strategy Since the longest linear sequence to make API from the readily commercially available raw material 6-aminouracil was 11 steps, we decided to outsource the first 6 steps to produce chloroiodopyrrolopyrimidine 6 (Scheme 6). The downstream intermediate silyl ether 11 was an oil, and the processes to make and isolate the downstream intermediates 13 and 15 were still underdeveloped (chromatographic purification), so ordering 6 made the most sense. While the manufacture of chloroiodopyrrolopyrimidine 6 was in progress, internal efforts focused on optimizing the downstream steps all the way to API (5 steps). The outsourcing vendor delivered 15 kg and 67 kg of 6 in 98.7–99.1% purity to fund the first and second API deliveries, respectively. The bulk synthesis of Weinreb amide 12 was also outsourced, and deliveries of 15 kg and 49 kg were made to supply the first and second campaigns, respectively (Scheme 7). Since 12 is an oil, both deliveries were made as 56–61 wt % solutions in 3:1 (THF to toluene). The first 15 kg delivery started from 5-bromonicotinic acid and produced 12 by conversion to the Weinreb amide followed by Buchwald amination with benzophenone imine.

168

Scheme 6. Outsourced synthesis of chloroiodopyrrolopyrimidine 6. However, silica gel chromatography was required to purify the final product since the Buchwald amination process produced numerous impurities that were difficult to purge. For the 49 kg scaleup of 12, the process was refined by starting from 5-aminonicotinic acid instead of 5-bromonicotinic acid to make the Weinreb amide and then adding benzophenone imine to the same pot to effect an exchange reaction to produce Weinreb amide-imine 12. Silica gel chromatography was still required for purification but was much more efficient since fewer impurities were formed starting from the alternative starting material.

Scheme 7. Outsourced syntheses of Weinreb amide 12.

The last key custom starting material was pyridyl acetic acid 16 with deliveries of 4.8 kg and 11.7 kg made to supply the first and second API campaigns, respectively. The synthesis (Scheme 8) started from 2-bromo-5-chloropyridine and upon reaction with diethyl malonate produced a mixture of products due to partial decarboxylation of the initially formed product. This is inconsequential, since upon ester hydrolysis, both intermediates produce the desired pyridine acetic acid 16. The overall yield was 38% for the 2 steps. Unbeknownst to us at the time, the batch of 16 delivered for the first API manufacture contained a small amount of malonic acid which led to one of the main impurities in the final API (see step 5 discussion below). For the second API campaign, we set a specification of ≤0.1% for residual malonic acid to avoid forming this impurity in the future. Note: 16 is known to decompose upon prolonged storage at room temperature (presumably via decarboxylation), despite being a solid, and was stored at −20°C until use. 169

Scheme 8. Outsourced synthesis of pyridine acetic acid 16.

In-House API Delivery Strategy The in-house portion of API synthesis was 5 steps starting from intermediate 6. The following section provides details of the process enabling for each step along with the composite experience from the 2 bulk campaign runs. Step 1: TBS Protection to Form Silyl Ether 11 The step 1 process optimization focused on decreasing the equivalents of TBSCl so that removal of residual TBSOH at the end of the process was easier via azeotropic distillation with toluene. We also wanted to isolate 11 as a toluene concentrate that could be diluted with THF and taken directly into step 2. The slightly modified optimized process used 2.0 equivalents of imidazole and 1.25 equivalents of TBSCl (Scheme 9). The TBSCl was added as a DMF solution to the slurry of starting material and imidazole in DMF. Only 90% conversion was observed when using 1.1 equivalents of TBSCl, and warming the reaction did not increase the conversion to 11. Once complete, the reaction was quenched with 1.0 M aqueous NaOH in place of NaHCO3 to avoid off-gassing concerns with aqueous side streams. The remaining work-up was the same as the original process incorporating a MTBE back extraction of the aqueous phase and then 3 water washes of the combined organic phase. The solvent was then exchanged from MTBE to toluene and then further chases were conducted with toluene until the residual TBSOH and water levels were each ≤0.5% in the toluene concentrate. The actual level of TBSOH detected by GC was 0.4 wt % and the water content was ≤0.03 wt % in the step 1 product toluene concentrate. A sample was then taken to dryness in the lab and quantitative proton nuclear magnetic resonance spectroscopy was run to determine potency of the toluene concentrate of iodo silyl ether 11 (24 wt %). The calculated isolated yield was 97%.

Scheme 9. Step 1 TBS protection to prepare silyl ether 11. 170

Step 2: Acylation Reaction with Grignard 18 The main issues identified for step 2 were an increased level of des-iodo product 19 upon scaleup which lowered the isolated yield (57% on scale vs. 75% in lab) and the added requirement for silica gel chromatography purification of the resulting crude product. Both step 2 starting materials were oils and, as such, both had to be used as solutions on scale. Iodo silyl ether 11 was a 24 wt % solution in toluene and Weinreb amide 12 was a 56–61 wt % solution in THF/toluene. In order to explore the formation of the reduction product, we ran the reaction as described, but also with alternative sets of conditions (e.g., lower temperatures, adding the iPrMgCl to a mixture of 7 and chloropyrrolopyrimidine 8, inverse addition of anion intermediate to the Weinreb amide, more equivalents of Weinreb amide), and in each case the reaction worked well as long as the water and TBSOH content was low. When using dry starting material solutions, the level of observed reduction product 19 was 99:1 post crystallization) and hydrogenolytic deprotection to chiral pyridylamine 65 in 98% ee in only five steps (compared with seven in the previous route) and good overall yield.

Figure 7. 6-Azaindazole CCR1 antagonists.

Scheme 6. Asymmetric benzylamine moiety synthesis route. 197

Compound 56b showed excellent functional activity in an assay assessing CCR1-driven chemotaxis of THP-1 cells (IC50 = 0.3 nM) that translated into excellent potency in a human whole blood CCR1 receptor internalization assay that had also been used in the clinic with 55 (Figure 7). Compound 56b was among the most potent CCR1 antagonists, including published antagonists and the clinical candidates in Figure 1 profiled in our in-house assays. Although the species selectivity with approximately 100-fold reduced potency against rodent CCR1 remained, the compound was so potent that mouse CCR1 activity was substantial (mouse CCR1 IC50 = 44 nM). We were able to profile the compound in a murine model of arthritis. In a mouse collagen antibody-induced arthritis model, the compounds displayed complete inhibition of disease progression as well as histological changes in the paw joints with 100 mg/kg bid. The plasma levels of 56b were approximately at mouse CCR1 IC90 levels. Compound 56b maintained attractive druglike properties. The crystalline thermodynamic solubility of the most stable polymorph was determined to be 20 μg/mL. The compound retained a molecular mass under 500 Da while having a slightly more lipophilic logP of 2.9. The compound showed no detectable off-target activities, no inhibition of cytochrome P450-metabolizing enzymes, and no inhibition of the hERG channel at physiologically relevant concentrations. Excellent metabolic stability and promising preclinical pharmacokinetic properties were observed. In comparison with the initial indazole 26, optimization had led to an additional 100-fold increase in potency, a 100-fold reduced lipophilicity, and a >100-fold increase in solubility. Based on preclinical data, the human dose was predicted at 1.25 mg qd. The compound was advanced into preclinical safety profiling. Unfortunately, 56b showed an unexpected and unpredicted substantial toxicity in dog. Dramatic decline in peripheral leukocytes, caused by sudden bone marrow ablation, was observed at all doses > 5 mg/kg starting at day 8. No therapeutic window could be identified. Compound 56b caused centromere disruption in a human lymphocyte chromosomal aberration test and was classified as an aneugen in an in vivo micronucleus test. The compound could not be advanced into clinical testing and was discontinued. To avoid the added complexity of the asymmetric synthesis as well as to avoid the unexplained detrimental effects in preclinical safety studies, a compound that behaved more like 55 but maintained the potency increases observed with 56b was desired. Disubstitution in the benzylic position typically led to reduced potency (33). For example, the gem-dimethyl analogue 59 was slightly less potent (CCR1 Ca flux IC50 = 12 nM) than was 55. However, once the two alkyl groups were tied into a small cycloalkane, some potency was regained. The cyclopropyl analogue 60 and the cyclobutyl analogue 61 showed acceptable CCR1 activity (CCR1 Ca flux IC50 = 1.8 nM and 2.0 nM, respectively) while maintaining aqueous solubility (kinetic solubility >50 μg/mL). A modified synthesis of the cycloalkyl benzylamines was developed. 2-Bromoisonicotinic acid (66) was reduced to the corresponding benzylalcohol, which was converted to the corresponding mesylate and subjected to nucleophilic substitution to the nitrile 67, resulting in the one-carbon homologation of 66 (Scheme 7). Double alkylation with 1,2-dibromoethane followed by hydrolysis of the nitrile resulted in the formation of the cyclopropane carboxylic acid 68, which was then subjected to a Curtius rearrangement at high temperature to give the Boc-protected amine 69. Introduction of the sulfonyl group and deprotection provided the desired cyclopropyl benzylamine 70. Cyclopropyl benzylamine-substituted 6-azaindazole 60 was profiled in advanced cellular assays (Figure 7). It was 5- to 10-fold more potent than the initial clinical compound 55. Compound 60 198

effectively inhibited CCR1-driven chemotaxis of THP-1 cells (IC50 = 2.0 nM) that translated into excellent potency in a human whole blood CCR1 receptor internalization assay. The polarity and molecular mass were essentially identical to the properties of 56b. The compound maintained the overall attractive, established profile of the 6-azaindazole with regard to off-target activity, drug–drug interaction potential, metabolic stability, and pharmacokinetic properties. The compound was predicted to have an 18 h human half-life with an efficacious dose of 58 mg qd. Most important, 60 did not show any of the findings observed with 56b in the pre-clinical toxicology studies.

Scheme 7. Cyclopropyl benzylamine synthesis route. In order to compare the performance of our candidate with competitors’ compounds, a quantitative systems pharmacology (QSP) model was developed (Figure 8). Prior knowledge from in-house programs and publicly available data for CCX354 (4) was used to establish a semimechanistic PK/PD relationship between drug exposure and target engagement. The model included important components needed to describe the differences between CCR1 antagonists such as plasma protein binding, along with reversible and competitive binding of drug and ligand. Constant system components such as target, ligand, and plasma protein synthesis and degradation were also included. There is distribution of drug between the central (plasma) compartment and the periphery as well as between the central and disease (joint) compartment. We simulated a headto-head comparison of 60 with CCX-354 (4) to compare exposure profiles and estimate target engagement in the plasma. Both molecules were dosed once daily (60 at 58 mg and CCX-354 [4] at 200 mg). Pharmacokinetic and binding parameters were modified and separate simulations were run for each molecule. As depicted in Figure 9, simulations predicted a lower Cmax for 60 compared with CCX-354 (4), but because of low plasma protein binding and a higher potency for CCR1, 60 provided a more desirable target engagement profile. Modeling results supported the strategic decision to move this molecule into early clinical development. Amide 60 was advanced into preclinical development as the third and final candidate from the CCR1 antagonist program. The development of a scalable large-scale synthesis is described in the remainder of this report. Unfortunately, clinical development for 60 was halted before the initiation of first-in-human trials based on the aforementioned clinical results for BMS-817339. However, the compound represents one of the best CCR1 antagonists in the field with regard to potency, selectivity, and druglike properties. Amide 60 is now available to the scientific community for basic research on the Boehringer Ingelheim Open Innovation Portal OpnMe [www.opnme.com]. 199

Figure 8. Full model including disease, central, and periphery compartments. The numbers represent model rate parameters.

Figure 9. Simulated concentration of drug (left) and simulated target occupancy (right) after dosing to steady state with 60 and CCX-354 (4) qd at 58 mg and 200 mg, respectively.

CMC Development of CCR1 Antagonists Following the discovery of 55 and subsequent evolution toward selection of amide 60 as a selective and potent CCR1 antagonist, CMC efforts focused on the scale-up and formulation activities required to support clinical development (Figure 10). To support this work under an aggressive timeline, large quantities of the drug substance with high chemical purity were required prior to Phase I. With regard to requested supply and delivery timelines, active pharmaceutical ingredient (API) demand is typically dependent on the stage of the project. In most cases, delivery of the first kilogram to support early preclinical studies (typically a 4-week toxicological study in 1 or 2 species) requires a primary focus on speed. As a program progresses, chemical development focuses on the design and implementation of a potential manufacturing route that strives to incorporate or maximize key attributes including quality, regulatory, safety, brevity, scalability, cost, and green chemistry. The factors above must also be considered in the context of the intrinsic properties of the drug substance,

200

especially regarding safety and bioavailability. Key challenges associated with the chemical development of this API included the following: • Solubility: Solubility was extremely low in most organic solvents, leading to purification challenges during final polish filtration and recrystallization. Low pKa (~2.5) provides limited opportunity for salt formation (only three observed, none stable), and cocrystallization was not observed after conducting a thorough screen. • Solid form landscape: The drug substance freebase can exist as one of five known polymorphs including one hydrate and four anhydrous forms that exhibit melting points of 200 ± 10 °C. Controlling the generation of the thermodynamic Form I chosen for development is made more difficult by low solubility and impact of impurity profile on form purity. • Regulatory: Several synthetic intermediates and impurities were flagged as potentially genotoxic impurities (pGTIs), requiring tight control at levels 99.5A%). Moreover, the use of aqueous HCl allowed for digestion of any solid inorganics and suppression of alkyl chlorides that would need control in the final API (2-chloropropane and p-methoxybenzyl chloride were not detected above 20 ppm prior to amide coupling). It must be noted that the corresponding benzyl carbamate could not be deprotected under these conditions (in lieu of failed hydrogenation)—more forcing conditions (H2SO4, 2-butanol, 110 °C) were required that resulted in many low-level impurities.

Scheme 30. Mild acidic carbamate deprotection en route to cyclopropylamine 70·HCl.

Initial Endgame—Amide Coupling Synthesis of 60·H2O The final chemical step en route to a scalable process toward 60 relied on an amide coupling between amine 70 and acid 71, a tactic unchanged from the Discovery approach (Scheme 11). However, the smaller-scale conditions utilized the uronium/guanidinium salt HATU, which was not preferred for use upon scale-up because of its: 1. Prohibitively high cost; 2. Large molecular weight contributing to large waste generation and environmental factor (EF, kg waste/kg product); 3. Safety concerns regarding high energy embedded in the triazole ring system; and 220

4. Worker safety concerns surrounding the cytotoxic tetramethylurea byproduct (91). Considering the lack of epimerizable protons and high stability of the starting materials, process research investigated the use of standard amide coupling conditions with a focus on scalability and chemical purity of the amide product. Solvent choice was limited because of the low solubility of 60 and acid 71, contributing to a potentially difficult purification scenario depending on the reaction profile and coupling byproducts. A screen of several polar aprotic solvents determined that NMP provided the highest reagent and product solubility. Further screening of acid chloride/anhydride-mediated approaches identified propanephosphonic anhydride (T3P) as an efficient coupling reagent that was preferred because of its ready availability on scale as a 50 wt % solution in many solvents, reasonable cost, and water-soluble phosphate byproducts (92). Initial conditions utilized equimolar amounts of acid 71 and amine 70, N-methylmorpholine (NMM, 6 equiv) as base, T3P (1.5 equiv), and NMP (4 V) at 65 °C (Scheme 31). A large excess of base was required in order to push the reaction rate. However, the reaction was still found to stall at ~7A% acid 71, even when excess amine 70 was charged. By the addition of water at this point, precipitation and isolation of API monohydrate 60·H2O (3.8 wt % H2O) was achieved in 80–90% yield and >98.5A% purity, with the major impurity initially identified as residual acid 71. At this early stage of development, this purity was deemed acceptable considering the planned use of the material for a 4-week toxicological study. The presence of process impurities at low levels was preferred for qualification. Under these conditions, it is also noteworthy that confirmed GTI azaindazole 76 was effectively purged to levels 99.8A% purity as the desired Form I (Scheme 33). However, further scale-up of this procedure using 60·H2O prepared from the T3P coupling procedure (>98.5A%, presumably containing acid 71) provided API predominantly as Form I yet contaminated with 5–10% Form IV as determined by XRPD and DSC analysis. A reslurry of this material in MEK at 60 °C could then achieve full conversion to Form I in about 92% recovery.

Scheme 33. Initial, two-stage crystallization procedure for 60 Form I. The observation that the Form I and IV mixtures would not convert into Form I in MEK before the first isolation, yet did so after the reslurry, was perplexing because this phenomenon had not been observed during early screening. Our hypothesis was that the first isolation removed low-level impurities that favored Form IV formation. To further understand the impact of impurity profile on the Form I/IV ratio, the mother liquor from the first recrystallization was concentrated and analyzed. Impurity enrichment up to 7A% was observed (Scheme 34); however, the UV spectrum of this peak differed slightly from the presumed impurity acid 71. Further investigation using a combination of 1H NMR spectroscopy and LCMS techniques confirmed the identity of the impurity as morpholine amide 107, presumably formed via coupling of NMM and 71 followed by demethylation because of the large excess of NMM or residual morpholine in the bulk NMM in the prior chemical step. The identity was further confirmed 223

by independent synthesis of an authentic standard and comparison to both the crude amide coupling reaction mixture and crystallization mother liquor.

Scheme 34. Discovery of morpholine amide impurity 107. There are an extremely limited number of examples in the literature illustrating the impact of low-level impurities on polymorph control (as opposed to supramolecular structure-induced crystallization) during API crystallization (93). In this particular case, the root cause for the difficulty in producing Form I as a pure phase is the close energy barrier between Form I and Form IV coupled with the mostly flat, sp2 nature and structural similarity of the API and impurity 107. To further understand the packing of Forms I and Form IV in the crystal lattice, single crystals of each form were obtained (Figure 11). Both forms had the same triclinic P1 space group with slightly different bond angles and rotation, indicating a very small energy difference between the two forms. Furthermore, Form I had a smaller unit cell volume and stronger intermolecular potentials, confirming that Form I was the more stable form at lower temperatures and that Forms I and IV were enantiotropically related.

Figure 11. Single crystal X-ray structures of API Forms I and IV illustrating a low energy barrier for conversion. To further understand the impact of 107 on form conversion, different batches of 60 with and without this impurity were evaluated under the same crystallization conditions using MEK/nheptane (Table 6). Batches containing ~1% 107 provided mixtures of Forms I and IV (entries 1 and 2). When 60 containing no 107 (vide infra) was recrystallized, only Form I was observed (entry 3). Furthermore, when 60 containing 4.5% 107 was recrystallized (pure material spiked into 60), Form

224

IV was obtained exclusively. These results underscored the need for development of an effective amide coupling procedure that would provide 60 in high purity with no residual 107. Table 6. Impact of Impurity 107 on Ratio of Form I versus Form IV Entry

107 in Crude 60a

Observed crystal formb

1

0.7 A%

Form I and IV

2

~0.9 A%

Form I and IV

3

0 A%

Form I

4

~4.5 A%a

Form IV

a Impurity was charged into lot for comparison purpose.

b All batches were seeded with pure Form I.

Optimization and Mass Intensification of Amide Coupling En Route to High-Purity API 60·H2O Following identification of impurity 107 and its impact on polymorph control during recrystallization, the obvious solution for removing this impurity was changing the base used during amide coupling. As expected, replacement of NMM with NEt3 or DIPEA completely inhibited formation of 107. Because 107 was previously formed in ~7A% and was previously postulated to be unreacted acid 71, further optimization of the protocol with respect to reagent ratio and solvent were performed. It was found that 1.1 equiv of amine 70, 5 equiv of NEt3, and 1.25 equiv T3P in NMP (2 V) afforded full conversion of acid 71 to 60 after heating to 65 °C for 3 h with no impurities generated during the process (Scheme 35). To further optimize processing and achieve 60·H2O with consistent water content (also found to be a critical parameter for the subsequent crystallization), the batch was seeded with 60·H2O crystals, followed by water addition (6 V) over 1 h at 65 °C and one round of thermal cycling (65 → 25 → 65 → 25 °C) to achieve crystal growth via Ostwald ripening. After cooling to ambient temperature, a fast filtration afforded 60·H2O as large crystals (>200 μm) in 96% yield and an excellent purity of >99.5A%.

Scheme 35. Amide coupling optimization. A comparison of the first- and second-generation processes is illustrated in Table 7. The new process resulted in a substantially large increase in throughput, yield, and purity, while simultaneously reducing waste and increasing overall quality of the crude API prior to the critical recrystallization step. 225

Table 7. Comparison of First- and Second-Generation Amide Coupling Protocols

Second-Generation API Recrystallization With high purity 60·H2O available, efforts were focused on modification of the API recrystallization step. In order to perform a more practical polish filtration, achieve form purity, and improve the recovery, additional solvent systems were evaluated. DMSO was identified as a potential cosolvent with MEK for API dissolution at a higher concentration prior to polish filtration, and IPA was chosen to replace n-heptane as antisolvent to provide higher recovery while maintaining the desired form purity. Thus, a modified procedure was developed wherein 60·H2O in MEK (10 V) and DMSO (1 V) was heated to 65 °C to effect dissolution. Following a polish filtration, the solution was then cooled to 60 °C and seeded with Form I seed crystals. After cooling to 20 °C over 4 h, IPA (9 V) was added over 4 h. After filtration and drying, API 60 was obtained in high chemical purity (>99.95A%) in 93% recovery as the desired Form I (Table 8). The optimized procedure was executed on 40 kg scale and provided a marked improvement in overall purity, plant throughput, and waste reduction compared with the original process. Table 8. Comparison of First- and Second-Generation Recrystallization Procedures

226

Summary of Continuous-Flow-Based Pilot Plant Route The optimized synthesis of 60 utilizing a continuous-flow Curtius rearrangement for cyclopropylamine installation is shown below in Scheme 36. Key highlights of the synthesis include: • A metal-free, streamlined three-step synthesis of acid 71 in high overall yield and purity. • A sequential, three-step telescopic SNAr synthesis of cyclopropylpyridyl acid 98 with high throughput, yield, and purity (40 kg prepared in one week via two-shift plant manufacture). • Development of a mechanistically guided, continuous-flow Curtius rearrangement utilizing in-house constructed flow reactors and PAT for online reaction monitoring that was safely executed in a cGMP pilot plant environment. • An amide coupling providing the API monohydrate in extremely high chemical purity with minimal waste. • A highly controlled crystallization process that could achieve specific conversion to the desired Form I while navigating through a difficult polymorph landscape.

Scheme 36. Robust pilot plant, continuous-flow GMP synthesis of 60. Four campaigns were successfully executed in 98% purity with a complete E-factor (cEF) of 14. This example of a tandem, semicontinuous Curtius rearrangement and isocyanate–acid coupling synthesis of an amide starting from two unsymmetrical carboxylic acids is, to our knowledge, the first reported, especially for use in manufacture of an investigational API. Additionally, the synergistic union of these tactics into a single executable process obviates two batch operations (carbamate deprotection and amide coupling), lowers the overall cost dramatically, and markedly increases the overall efficiency and greenness of the API synthesis. Unfortunately, the program was strategically terminated before a kilogram-scale process could be developed because of lack of target efficacy based on a competitor’s CCR1 program.

Scheme 38. Direct, semicontinuous synthesis of 60 via Curtius rearrangement and acid isocyanate coupling.

Environmental (Greenness) Assessment via Green Aspiration Level (GAL) Analysis In order to understand the positive and powerful impact of continuous processing on the synthesis of 60, overall process greenness was analyzed by Roschangar’s Green Aspiration Level (GAL) methodology, recently adopted by the American Chemical Society Green Chemistry Working Group (108, 109). For each synthetic Curtius-based route (semibatch process, CSTR, flow, and direct Curtius-amide synthesis), process complexity and ideality were utilized to determine the GAL (110). Then, using the overall complete environmental factor (cEF, kginput/kg product-1) for each route, the relative process greenness (RPG), relative process improvement (RPI), relative complexity improvement (RCI, no change), and overall process improvement (PI) were calculated (Table 10). Based on a cEF of 26 kg/kg for an average “productive” process step in the pharmaceutical industry, the GAL for 60 is calculated to be 208 kg waste per kg API (108). Although plagued with impurities and a low yield for rearrangement, notably, the semibatch Curtius route is 44% greener than the industrial average for a commercial API of similar complexity. When moving to the first-generation, semicontinuous CSTR-based process, an overall process that is 116% greener than the industrial average is achieved. Although the second-generation flow synthesis did not improve RPG substantially, the safety, scalability, and overall throughput were significantly improved. 229

However, a large efficiency increase was observed for the third-generation direct, sequential Curtius amide synthesis because of the removal of a non-bond-forming concession (deprotection) step, overall cEF, and greenness. This enhancement resulted in a continuous-flow-based process 160% greener than the industrial average as well as 58% more efficient than the original semibatch process, demonstrating how incorporation of continuous-flow technology can enable a safe, scalable, high-yielding, and environmentally friendly synthesis of a pharmaceutically relevant molecule. Table 10. Environmental GAL Analysis of CCR1 Synthesis Routes Route

Complexity

Ideality (%)

GALa

cEFb

RPGc (%)

RPId (%)

PIe (%)

Batch Curtius

8

80

208

144

144

—

—

CSTR

8

80

208

96

216

72

36

PMB Flow

8

80

208

95

218

74

37

Flow Amide

8

90

208

78

266

122

61

Aspiration Level. b Complete Environmental factor (kg waste/kg API). Greenness. d Relative Process Improvement. e Overall Process Intensity. a Green

c Relative

Process

Conclusions In conclusion, mechanistic guidance via PAT and in-house process development of continuousflow technology allowed for the execution of a concise and scalable synthesis of CCR1 antagonist 60. After the efficient and parallel construction of two high-purity carboxylic acid coupling partners, a safe, robust, high-throughput, and green continuous-flow Curtius rearrangement was achieved with cyclopropylamine synthesis in high yield and quality on large scale. Following successful execution using a CSTR for acyl azide generation, the use of PMBOH allowed for scale-up (40 kg) of a fully continuous synthesis of a highly crystalline PMB-carbamate easily unmasked under mild acidic conditions without the need for metal-based hydrogenation. Additionally, critical mechanistic understanding of each process parameter allowed for accomplishment of the first example of direct amide synthesis semicontinuous Curtius rearrangement and acid–isocyanate coupling as the key step in a highly efficient and an environmentally friendly API synthesis.

Acknowledgments We are grateful for all the support from our BI R&D colleagues during execution of this program. Additionally, we wish to acknowledge the Royal Society of Chemistry, specifically the journal Green Chemistry, for prior publication of the flow-based manufacturing route and permission to adapt several schemes for use in this manuscript.

References 1.

Godessart, N. Chemokine receptors—Attractive targets for drug discovery. Ann. N. Y. Acad. Sci. 2005, 1051, 647–657.

230

2. 3.

4.

5.

6.

7. 8.

9. 10. 11. 12.

13. 14.

15.

Horuk, R.; Ribeiro, S. The clinical potential of chemokine receptor antagonists. Pharmacol. Ther. 2005, 107, 44–58. Plater-Zyberk, C.; Hoogewerf, A. J.; Proudfoot, A. E. I.; Power, C. A.; Wells, T. N. C. Effect of a CC chemokine receptor antagonist on collagen induced arthritis in DBA/1 mice. Immunol. Lett. 1997, 57, 117–120. Revesz, L.; Bollbuck, B.; Buhl, T.; Eder, J.; Esser, R.; Feifel, R.; Heng, R.; Hiestand, P.; JachezDemange, B.; Loetscher, P.; Sparrer, H.; Schlapbach, A.; Waelchli, R. Novel CCR1 antagonists with oral activity in the mouse collagen induced arthritis. Bioorg. Med. Chem. Lett. 2005, 15, 5160–5164. Horuk, R.; Shurey, S.; Ng, H. P.; May, K.; Bauman, J. G.; Islam, I.; Ghannam, A.; Buckman, B.; Wei, G. P.; Xu, W.; Liang, M.; Rosser, M.; Dunning, L.; Hesselgesser, J.; Snider, R. M.; Morrissey, M. M.; Perez, H. D.; Green, C. CCR1-specific non-peptide antagonist: efficacy in a rabbit allograft rejection model. Immunol. Lett. 2001, 76, 193–201. Amat, M.; Benjamin, C. F.; Williams, L. M.; Prats, N.; Terricabras, E.; Beleta, J.; Kunkel, S. L.; Godessart, N. Pharmacological blockade of CCR1 ameliorates murine arthritis and alters cytokine networks in vivo. Br. J. Pharmacol. 2006, 149, 666–675. Horuk, R. Promiscuous drugs as therapeutics for chemokine receptors. Expert Rev. Mol. Med. 2009, 11, 1–16. Gladue, P. R.; Cole, S. H.; Roach, M. L.; Tylaska, L. A.; Nelson, R. T.; Shepard, R. M.; McNeish, J. D.; Ogborne, K. T.; Neote, K. S. The human specific CCR1 antagonist CP481,715 inhibits cell infiltration and inflammatory responses in human CCR1 transgenic mice. J. Immunol. 2006, 176, 3141–3148. Gladue, R.; Zwillich, S.; Clucas, A.; Brown, M. CCR1 antagonists for the treatment of autoimmune diseases. Curr. Opin. Invest. Drugs 2004, 5, 499–504. Szczucinski, A.; Losy, J. Chemokines and chemokine receptors in multiple sclerosis. Potential targets for new therapies. Acta. Neurol. Scand. 2007, 115, 137–146. Ubogu, E. E.; Cossoy, M. B.; Ransohoff, R. M. The expression and function of chemokines involved in CNS inflammation. Trends Pharmacol. Sci. 2006, 27, 48–55. Liang, M.; Mallari, C.; Rosser, M.; Ng, H. P.; May, K.; Monahan, S.; Bauman, J. G.; Islam, I.; Ghannam, A.; Buckman, B.; Shaw, K.; Wei, G.-P.; Xu, W.; Zhao, Z.; Ho, E.; Shen, J.; Oanh, H.; Subramanyam, B.; Vergona, R.; Taub, D.; Dunning, L.; Harvey, S.; Snider, R. M.; Hesselgesser, J.; Morrissey, M. M.; Perez, D.; Horuk, R. Identification and characterization of a potent, selective, and orally active antagonist of the CC chemokine receptor 1. J. Biol. Chem. 2000, 275, 19000–19008. Horuk, R. BX471: A CCR1 antagonist with anti-inflammatory activity in man. Mini-Rev. Med. Chem. 2005, 5, 791–804. Zipp, F.; Hartung, H. P.; Hillert, J.; Schimrigk, S.; Trebst, C.; Stangel, M.; Infante-Duarte, C.; Jakobs, P.; Wolf, C.; Sandbrink, R.; Pohl, C.; Filippi, M. Blockade of chemokine signaling in patients with multiple sclerosis. Neurology 2006, 67, 1880–1883. Vergunst, C. E.; Gerlag, D. M.; von Moltke, L.; Karol, M.; Wyant, T.; Chi, X.; Matzkin, E.; Leach, T.; Tak, P. P. MLN3897 plus methotrexate in patients with rheumatoid arthritis: safety, efficacy, pharmacokinetics, and pharmacodynamics of an oral CCR1 antagonist in a phase

231

16.

17.

18. 19.

20.

21.

22. 23.

24.

25.

26.

27.

IIa, double-blind, placebo-controlled, randomized, proof-of-concept study. Arthritis Rheum. 2009, 60, 3572–3581. Pusalkar, S.; Plesescu, M.; Milton, M.; Balani, S. K.; Chowdhury, S.; Prakash, S. Metabolism, excretion and pharmacokinetics of MLN3897, a CCR1 antagonist, in humans. Drug Metab. Lett. 2016, 10, 22–37. Dairaghi, D. J.; Zhang, P.; Wang, Y.; Seitz, L. C.; Johnson, D. A.; Miao, S.; Ertl, L. S.; Zeng, Y.; Powers, J. P.; Pennell, A. M.; Bekker, P.; Schall, T. J.; Jaen, J. C. Pharmacokinetic and pharmacodynamic evaluation of the novel CCR1 antagonist CCX354 in healthy human subjects: Implications for selection of clinical dose. Clin. Pharmaco. Ther. 2011, 89, 726–734. Pease, J. E. CCR1 antagonism for the treatment of inflammatory diseases: focus on CCX-354. Drugs of the Future 2012, 37, 735–739. Pennell, A. M. K.; Aggen, J. B.; Sen, S.; Chen, W.; Xu, Y.; Sullivan, E.; Li, L.; Greenman, K.; Charvat, T.; Hansen, D.; Dairaghi, D. J.; Wright, J. J. K.; Zhang, P. 1(4-Phenylpiperazin-1-yl)2-(1H-pyrazol-1-yl)ethanones as novel CCR1 antagonists. Biorg. Med. Chem. Lett. 2013, 23, 1228–1231. Tak, P. P.; Balanescu, A.; Tseluyko, V.; Bojin, S.; Drescher, E.; Dairaghi, D.; Miao, S.; Marchesin, V.; Jaen, J.; Schall, T. J.; Bekker, P. Chemokine receptor CCR1 antagonist CCX354-C treatment for rheumatoid arthritis: CARAT-2, a randomized, placebo controlled clinical trial. Ann. Rheum. Dis. 2012, 72, 337–344. Santella, J. B., III; Gardner, D. S.; Duncia, J. V.; Wu, H.; Dhar, M.; Cavallaro, C.; Tebben, A. J.; Carter, P. H.; Barrish, J. C.; Yarde, M.; Briceno, S. W.; Cvijic, M. E.; Grafstrom, R. R.; Liu, R.; Patel, S. R.; Watson, A. J.; Yang, G.; Rose, A. V.; Vickery, R. D.; Caceres-Cortes, J.; Caporuscio, C.; Camac, D. M.; Khan, J. A.; An, Y.; Foster, W. R.; Davies, P.; Hynes, J., Jr. Discovery of the CCR1 antagonist, BMS-817399, for the treatment of rheumatoid arthritis. J. Med. Chem. 2014, 57, 7550–7564. Zhang, P.; Dairaghi, D. J.; Jaen, J. C.; Powers, J. P. Recent advances in the discovery and development of CCR1 antagonists. Annu. Rep. Med. Chem. 2013, 48, 133–147. Lionakis, M. S.; Albert, N. D.; Swamydas, M.; Lee, C.-C. R.; Loetscher, P.; Kontoyiannis, D. P. Pharmacological blockade of the chemokine receptor CCR1 protects mice from systemic candidiasis of hematogenous origin. Antimicrob. Agents Chemother. 2017, 61, e02365-16. Conroy, M. J.; Galvin, K. C.; Kavanagh, M. E.; Mongan, A. M.; Doyle, S. L.; Gilmartin, N.; O’Farrelly, C.; Reynolds, J. V.; Lysaght, J. CCR1 antagonism attenuates T cell trafficking to omentum and liver in obesity-associated cancer. Immunol. Cell Biol. 2016, 94, 531–537. Kholodnyuk, I.; Rudevica, Z.; Leonciks, A.; Ehlin-Henriksson, B.; Kashuba, E. Expression of the chemokine receptors CCR1 and CCR2B is up-regulated in peripheral blood B cells upon EBV infection and in established lynphoblastoid cell lines. Virology 2017, 512, 1–7. Cook, B. N.; Harcken, C.; Lee, T. W-H.; Liu, P.; Mao, C., Lord, J.; Mao, W.; Raudenbush, B. C.; Razavi, H.; Sarko, C. R.; Swinamer, A. D. Pyrazole compounds as CCR1 antagonists. PCT Int. Appl. WO2009137338, 2009. Carter, P. H.; Hynes, J. N-aryl pyrazoles, indazoles and azaindazoles as antagonists of CC chemokine receptor 1: patent cooperation treaty applications WO2010/036632, WO2009/ 134666 and WO2009/137338. Expert Opin. Ther. Patents 2010, 20, 1609–1618.

232

28. Harcken, C.; Sarko, C.; Mao, C.; Lord, J.; Raudenbush, B.; Razavi, H.; Liu, P.; Swinamer, A.; Disalvo, D.; Lee, T.; Lin, S.; Kukulka, A.; Grbic, H.; Patel, M.; Patel, M.; Fletcher, K.; Joseph, D.; White, D.; Amodeo, L.; Berg, K.; Brown, M.; Thomson, D. T. Discovery and optimization of pyrazole amides as antagonists of CCR1. Bioorg. Med. Chem. Lett. 2019, 29, 435–440. 29. Disalvo, D.; Kuzmich, D.; Mao, C.; Razavi, H.; Sarko, C.; Swinamer, A. D.; Thomson, D.; Zhang, Q. Indazoles compounds as CCR1 receptor antagonist. PCT Int. Appl. WO2009134666, 2009. 30. Harcken, C.; Kuzmich, D.; Cook, B.; Mao, C.; Disalvo, D.; Razavi, H.; Swinamer, A.; Liu, P.; Zhang, Q.; Kukulka, A.; Skow, D.; Patel, M.; Patel, M.; Sherry, T.; Joseph, D.; Smith, D.; Canfield, M..; Souza, D.; Bogdanffy, M.; Berg, K.; Brown, M. Identification of novel azaindazole CCR1 antagonist clinical candidates. Bioorg. Med. Chem. Lett. 2019, 29, 441–448. 31. Betageri, R.; Cook, B.; Disalvo, D.; Harcken, C.; Kuzmich, D.; Liu, P.; Lord, J.; Mao, C.; Razavi, H. Pyrazolopiperidine compounds as CCR1 receptor antagonists. PCT Int. Appl. WO2012087782, 2012. 32. Cook, B.; DiSalvo, D.; Harcken, C.; Kuzmich, D.; Liu, P.; Lord, J.; Mao, C.; Raudenbush, B.; Razavi, H.; Reeves, J.; Song, J.; Swinamer, A.; Fandrick, D.; Lee, T.; Neu, J.; Tan, Z. Azaindazole compounds as CCR1 receptor antagonists. U.S. Patent 7,879,873, 2011. 33. Cook, B.; Kuzmich, D.; Mao, C.; Razavi, H. Indazole and pyrazolopyridine compounds as CCR1 receptor antagonists. USPTO Appl. US20120270870, 2012. 34. ICH M9 Concept Paper on Biopharmaceutical Classification System-Based Biowaivers, 2018. 35. Marsini, M. A.; Buono, F. G.; Lorenz, J. C.; Yang, B.-S.; Reeves, J. T.; Sidhu, K.; Sarvestani, M.; Tan, Z.; Zhang, Y.; Li, N.; Lee, H.; Brazzillo, J.; Nummy, L. J.; Chung, J. C.; Luvaga, I. K.; Narayanan, B. A.; Wei, X.; Song, J. J.; Roschangar, F.; Yee, N. K.; Senanayake, C. H. Development of a concise, scalable synthesis of a CCR1 antagonist utilizing a continuous flow Curtius rearrangement. Green Chem. 2017, 19, 1454–1461. 36. Gu, Y. G.; Bayburt, E. K. Synthesis of 4-alkyl-3,5-dibromo‑, 3-bromo-4,5-dialkyl- and 3,4,5trialkylpyridines via sequential metalation and metal-halogen exchange of 3,5dibromopyridine. Tetrahedron 1996, 37, 2565–2568. 37. Abarbri, M.; Dehmel, F.; Knochel, P. Bromine-magnesium-exchange as a general tool for the preparation of polyfunctional aryl and heteroaryl magnesium-reagents. Tetrahedron Lett. 1999, 40, 7449–7453. 38. Bertus, P.; Szymoniak, J. A direct synthesis of 1-aryl- and 1-alkenylcyclopropylamines from aryl and alkenyl nitriles. J. Org. Chem. 2003, 68, 7133–7136. 39. Wiedemann, S.; Frank, D.; Winsel, H.; de Meijere, A. Primary 1-arylcyclopropylamines from aryl cyanides with diethylzinc and titanium alkoxides. Org. Lett. 2003, 5, 753–755. 40. Li, W.; Gao, J. J.; Lorenz, J. C.; Xu, J.; Johnson, J.; Ma, S.; Lee, H.; Grinberg, N.; Busacca, C. A.; Lu, B.; Senanayake, C. H. Process development and pilot-plant synthesis of (s)-tert-butyl 1-oxo-1-(1-(pyridin-2-yl)cyclopropylamino)propan-2-ylcarbamate: studies on the scale-up of Kulinkovich–Szymoniak cyclopropanation. Org. Process Res. Dev. 2012, 16, 836–839. 41. Kulinkovich, O. G.; de Meijere, A. 1,n-dicarbanionic titanium intermediates from monocarbanionic organometallics and their application in organic synthesis. Chem. Rev. 2000, 100, 2789–2834.

233

42. Reeves, J. T.; Tan, Z.; Reeves, D. C.; Song, J. J.; Han, Z. S.; Xu, Y.; Tang, W.; Yang, B.-S.; Razavi, H.; Harcken, C.; Kuzmich, D.; Mahaney, P. E.; Lee, H.; Busacca, C. A.; Senanayake, C. H. Development of an enantioselective hydrogenation route to (s)-1-(2(methylsulfonyl)pyridin-4-yl)propan-1-amine. Org. Process Res. Dev. 2014, 18, 904–911. 43. Burk, M. J.; Casy, G.; Johnson, N. B. A three-step procedure for asymmetric catalytic reductive amidation of ketones. J. Org. Chem. 1998, 63, 6084–6085. 44. Zhu, G.; Casalnuovo, A. L.; Zhang, X. Practical syntheses of β-amino alcohols via asymmetric catalytic hydrogenation. J. Org. Chem. 1998, 63, 8100–8101. 45. Tang, W.; Capacci, A.; Sarvestani, M.; Wei, X.; Yee, N. K.; Senanayake, C. H. A facile and practical synthesis of n-acetyl enamides. J. Org. Chem. 2009, 74, 9528–9530. 46. Chen, J.; Zhang, W.; Geng, H.; Li, W.; Hou, G.; Lei, A.; Zhang, X. Efficient synthesis of chiral β-arylisopropylamines by using catalytic asymmetric hydrogenation. Angew. Chem., Int. Ed. 2009, 48, 800–802. 47. Li, W.; Rodriguez, S.; Duran, A.; Sun, X.; Tang, W.; Premasiri, A.; Wang, J.; Sidhu, K.; Patel, N. D.; Savoie, J.; Qu, B.; Lee, H.; Haddad, N.; Lorenz, J. C.; Nummy, L.; Hossain, A.; Yee, N.; Lu, B.; Senanayake, C. H. The p-chiral phosphane ligand (MeO-BIBOP) for efficient and practical large-scale rh-catalyzed asymmetric hydrogenation of n-acetyl enamides with high tons. Org. Process Res. Dev. 2013, 17, 1061–1065. 48. Koenig, S. G.; Vandenbossche, C. P.; Zhao, H.; Mousaw, P.; Singh, S. P.; Bakale, R. P. A facile deprotection of secondary acetamides. Org. Lett. 2009, 11, 433–436. 49. Caron, S.; Vazquez, E.; Wojcik, J. M. Preparation of tertiary benzylic nitriles from aryl fluorides. J. Am. Chem. Soc. 1999, 122, 712–713. 50. Klapars, A.; Campos, K. R.; Jensen, M. S.; McLaughlin, M.; Chung, J. L. Y.; Cvetovich, R. J.; Chen, C. Mild and practical method for the α-arylation of nitriles with heteroaryl halides. J. Org. Chem. 2005, 70, 10186–10189. 51. Dunn, J. M. M.; Kuethe, J. T.; Orr, R. K.; Tudge, M.; Campeau, L.-C. Development of a palladium-catalyzed α-arylation of cyclopropyl nitriles. Org. Lett. 2014, 16, 6314–6317. 52. Chang, R. K.; Di Marco, C. N.; Pitts, D. R.; Greshock, T. J.; Kuduk, S. D. Preparation of 4-heteroaryl-4-cyanopiperidines via SNAr substitution reactions. Tetrahedron Lett. 2009, 50, 6303–6306. 53. Wei, X.; Shu, C.; Haddad, N.; Zeng, X.; Patel, N. D.; Tan, Z.; Liu, J.; Lee, H.; Shen, S.; Campbell, S.; Varsolona, R. J.; Busacca, C. A.; Hossein, A.; Yee, N. K.; Senanayake, C. H. A highly convergent and efficient synthesis of a macrocyclic hepatitis c virus protease inhibitor BI 201302. Org. Lett. 2013, 15, 1016–1019. 54. Maloney, K. M.; Kuethe, J. T.; Linn, K. A practical, one-pot synthesis of sulfonylated pyridines. Org. Lett. 2007, 11, 913–916. 55. ICH M7(R1) - Assessment and control of DNA reactive (mutagenic) impurities in pharmaceuticals to limit potential carcinogenic risk, 2017. 56. Zhang, L.; Kauffman, G. S.; Pesti, J. A.; Yin, J. Rearrangement of N-α-protected L-asparagines with iodosobenzene diacetate. A practical route to β-amino-L-alanine derivatives. J. Org. Chem. 1997, 62, 6918–6920.

234

57. Beliaev, A.; Wahnon, J.; Russo, D. Process research for multikilogram production of etamicastat: A novel dopamine β-hydroxylase inhibitor. Org. Process Res. Dev. 2012, 16, 704–709. 58. Senanayake, C. H.; Fredenburgh, L. E.; Reamer, R. A.; Larsen, R. D.; Verhoeven, T. R.; Reider, P. J. Nature of N-bromosuccinimide in basic media: The true oxidizing species in the Hofmann rearrangement. J. Am. Chem. Soc. 1994, 116, 7947–7948. 59. Kaufhold, M.; Kleemiss, W.; Feld. M. Multi-step process for the preparation of the cyclopropylamine from γ-butyrolactone. European Patent EP970943, 2000. 60. Zagulyaeva, A. A.; Banek, C. T.; Yusubov, M. S.; Zhdankin, V. V. Hofmann rearrangement of carboxamides mediated by hypervalent iodine species generated in situ from iodobenzene and oxone: reaction scope and limitations. Org. Lett. 2010, 12, 4644–4647. 61. am Ende, D. J.; DeVries, K. M.; Clifford, P. J.; Brenek, S. J. A calorimetric investigation to safely scale-up a Curtius rearrangement of acryloyl azide. Org. Process Res. Dev. 1998, 2, 382–392. 62. Mittendorf, J.; Benet-Buchholz, J.; Fey, P.; Mohrs, K.-H. Efficient asymmetric synthesis of βamino acid BAY 10-8888/PLD-118, a novel antifungal for the treatment of yeast infections. Synthesis 2003, 136–140. 63. Yue, T.-Y.; McLeod, D. D.; Albertson, K. B.; Beck, S. R.; Deerberg, J.; Fortunak, J. M.; Nugent, W. A.; Radesca, L. A.; Tang, L.; Xiang, C. D. Stereoselective process for a CCR3 antagonist. Org. Process Res. Dev. 2006, 10, 262–271. 64. Varie, D. L.; Beck, C.; Borders, S. K.; Brady, M. D.; Cronin, J. S.; Ditsworth, T. K.; Hay, D. A.; Hoard, D. W.; Hoying, R. C.; Linder, R. J.; Miller, R. D.; Moher, E. D.; Remacle, J. R.; Rieck, J. A.; Anderson, D. D.; Dodson, P. N.; Forst, M. B.; Pierson, D. A.; Turpin, J. A. Design, development, and scale-up of a selective meso-epoxide desymmetrization process. Org. Process Res. Dev. 2007, 11, 546–559. 65. Tang, W.; Wei, X.; Yee, N. K.; Patel, N.; Lee, H.; Savoie, J.; Senanayake, C. H. A practical asymmetric synthesis of isopropyl (1r,2s)-dehydrocoronamate. Org. Process Res. Dev. 2011, 15, 1207–1211. 66. Grongsgaard, P.; Bulger, P. G.; Wallace, D. J.; Tan, L.; Chen, Q.; Dolman, S. J.; Nyrop, J.; Hoerner, R. S.; Weisel, M.; Arredondo, J.; Itoh, T.; Xie, C.; Wen, X.; Zhao, D.; Muzzio, D. J.; Bassan, E. M.; Shultz, C. S. Convergent, kilogram scale synthesis of an AKT kinase inhibitor. Org. Process Res. Dev. 2012, 16, 1069–1081. 67. Shioiri, T.; Ninomiya, K.; Yamada, S. Diphenylphosphoryl azide. New convenient reagent for a modified Curtius reaction and for peptide synthesis. J. Am. Chem. Soc. 1972, 94, 6203–6205. 68. Tudhope, S. R.; Bellamy, J. A.; Ball, A.; Rajasekar, D.; Azadi-Ardakani, M.; Meera, H. S.; Gnanadeepam, J. M.; Saigenesh, R.; Gibson, F.; He, L.; Behrens, C. H.; Underiner, G.; Marfurt, J.; Favre, N. Development of a large-scale synthetic route to manufacture (−)huperzine A. Org. Process Res. Dev. 2012, 16, 635–642. 69. Honda, I.; Shimonishi, Y.; Sakakibara, S. tert-Amyloxycarbonyl as a new protecting group in peptide synthesis. IV. Synthesis and use of tert-amyl azidoformate. Bull. Chem. Soc. Jpn. 1967, 40, 2415–2418. 70. Dieckmann, W.; Breest, F. Notiz über das verhalten von carbonsäuren gegen phenylisocyanat. Ber. 1906, 39, 3052–3055.

235

71. Naegeli, C.; Tyabji, A. Über den umsatz aromatischer isocyansäure-ester mit organischen säuren. I. Theorie und anwendung der reaktion für die präparative darstellung von säureanhydriden. Helv. Chim. Acta. 1934, 17, 931–957. 72. Goldschmidt, S.; Wick, M. Über peptid-ynthesen I. Liebigs Ann. Chem. 1952, 575, 217–231. 73. Fry, A. A tracer study of the reaction of isocyanates with carboxylic acids. J. Am. Chem. Soc. 1953, 75, 2686–2688. 74. Sorenson, W. R. Reaction of an isocyanate and a carboxylic acid in dimethyl sulfoxide. J. Org. Chem. 1959, 24, 978–980. 75. Neumann, W.; Fischer, P. Carbodiimide aus isocyanaten. Angew. Chem. 1962, 74, 801–806. 76. Davies, I. W.; Welch, C. J. Looking forward in pharmaceutical process chemistry. Science 2009, 325, 701–704. 77. Yoshida, J.-I.; Takahashi, Y.; Nagaki, A. Flash chemistry: Flow chemistry that cannot be done in batch. Chem. Commun. 2013, 49, 9896–9904. 78. Webb, D.; Jamison, T. F. Continuous flow multi-step organic synthesis. Chem. Sci. 2010, 1, 675–680. 79. Malet-Sanz, L.; Susanne, F. Continuous flow synthesis. A pharma perspective. J. Med. Chem. 2012, 55, 4062–4098. 80. Mascia, S.; Heider, P. L.; Zhang, H.; Lakerveld, R.; Benyahia, B.; Barton, P. I.; Braatz, R. D.; Cooney, C. L.; Evans, J. M. B.; Jamison, T. F.; Jensen, K. F.; Myerson, A. S.; Trout, B. L. End-to-end continuous manufacturing of pharmaceuticals: integrated synthesis, purification, and final dosage formation. Angew. Chem., Int. Ed. 2013, 52, 12359–12363. 81. Gutmann, B.; Cantillo, D.; Kappe, C. O. Continuous-flow technology—a tool for the safe manufacturing of active pharmaceutical ingredients. Angew. Chem., Int. Ed. 2015, 54, 6688–6728. 82. Correia, C. A.; Gilmore, K.; McQuade, D. T.; Seeberger, P. H. A concise flow synthesis of efavirenz. Angew. Chem., Int. Ed. 2015, 54, 4945–4948. 83. Sahoo, H. R.; Kralj, J. G.; Jensen, K. F. Multistep continuous-flow microchemical synthesis involving multiple reactions and separations. Angew. Chem., Int. Ed. 2007, 46, 5704–5708. 84. Baumann, M.; Baxendale, I. R.; Ley, S. V.; Nikbin, N.; Smith, C. D.; Tierney, J. P. A modular flow reactor for performing Curtius rearrangements as a continuous flow process. Org. Biomol. Chem. 2008, 6, 1577–1586. 85. Carter, C. F.; Lange, H.; Ley, S. V.; Baxendale, I. R.; Wittkamp, B.; Goode, J. G.; Gaunt, N. L. ReactIR flow cell: a new analytical tool for continuous flow chemical processing. Org. Process Res. Dev. 2010, 14, 393–404. 86. Rumi, L.; Pfleger, C.; Spurr, P.; Klinkhammer, U.; Bannwarth, W. Adaptation of an exothermic and acyl azide-involving synthesis sequence to microreactor technology. Org. Process Res. Dev. 2009, 13, 747–750. 87. Xiang, Y.; Lucas, J.; VanAlsten, J.; Li, B.; Preston, B.; Lovdahl, M.; Hayward, C. Using process analytical technology (pat) tools to support flow chemistry development and production. Am. Pharmaceut. Rev. 2012, 15 (3), 56. 88. Deerberg, J.; Prasad, S. J.; Sfouggatakis, C.; Eastgate, M. D.; Fan, Y.; Chidambaran, R.; Sharma, P.; Li, L.; Schild, R.; Muslehiddinoglu, J.; Chung, H.-J.; Leung, S.; Rosso, V. Stereoselective bulk synthesis of CCR2 antagonist BMS-741672: Assembly of an all-cis 236

(S,R,R)-1,2,4-triaminocyclohexane (TACH) core via sequential heterogeneous asymmetric hydrogenations. Org. Proc. Res. Dev. 2016, 20, 1949–1966. 89. Chen, S.-T.; Wang, K.-T. A new synthesis of O-benzyl-L-serine. Synthesis 1989, 36–37. 90. Kedrowski, B. Synthesis of orthogonally protected (R)- and (S)-2-methylcysteine via an enzymatic desymmetrization and Curtius rearrangement. J. Org. Chem. 2003, 68, 5403–5406. 91. Medley, C. D.; Chetwyn, N. Byproducts of commonly used coupling reagents: Origin, toxicological evaluation and methods for determination. Am. Pharmaceut. Rev. 2012, 56. 92. Dunetz, J. R.; Xiang, Y.; Baldwin, A.; Ringling, J. General and scalable amide bond formation with epimerization-prone substrates using T3P and pyridine. Org. Lett. 2011, 13, 5048–5051. 93. McLachlan, H.; Ni, X. On the effect of added impurity on crystal purity of urea in an oscillatory baffled crystallizer and a stirred tank crystallizer. J. Crys. Growth 2016, 442, 81–88. 94. Hofle, G.; Steglich, W.; Vorbruggen, H. 4-Dialkylaminopyridines as highly active acylation catalysts. Angew. Chem., Int. Ed. 1978, 17, 569–583. 95. Sheldon, R. A. E factors, green chemistry and catalysis: An odyssey. Chem. Commun. 2008, 3352–3365. 96. Anastas, P. T.; Kirchhoff, M. M. Origins, current status, and future challenges of green chemistry. Acc. Chem. Res. 2002, 35, 686–694. 97. Trost, B. M. The atom economy—a search for synthetic efficiency. Science 1991, 254, 1471–1477. 98. Schafer, G.; Matthey, C.; Bode, J. W. Facile synthesis of sterically hindered and electrondeficient secondary amides from isocyanates. Angew. Chem., Int. Ed. 2012, 51, 9173–9175. 99. Schafer, G.; Bode, J. W. Synthesis of sterically hindered n-acylated amino acids from Ncarboxyanhydrides. Org. Lett. 2014, 16, 1526–1529. 100. Sasaki, K.; Crich, D. Facile amide bond formation from carboxylic acids and isocyanates. Org. Lett. 2011, 13, 2256–2259. 101. Schuemacher, A. C.; Hoffmann, R. W. Isocyanates and carboxylic acids in the presence of 4-dimethylaminopyridine (DMAP): A mild and efficient synthesis of amides. Synthesis 2001, 243–246. 102. Blagbrough, I. S.; Mackenzie, N. E.; Ortiz, A. I. The condensation reaction between isocyanates and carboxylic acids. A practical synthesis of substituted amides and anilides. Tetrahedron Lett. 1986, 27, 1251–1254. 103. Motoki, S.; Saito, T.; Kagami, H. Reaction of isocyanates with carboxylic acids and thiocarboxylic acids. Bull. Chem. Soc. Jpn. 1974, 47, 775–776. 104. Pfister, J. R.; Wymann, W. E. A useful variant of the Curtius reaction. Synthesis 1983, 38–40. 105. Dunetz, J. R.; Magano, J.; Weisenberger, G. A. Large-scale applications of amide coupling reagents for the synthesis of pharmaceuticals. Org. Process Res. Dev. 2016, 20, 140–177. 106. El-Faham, A.; Albericio, F. Peptide coupling reagents, more than a letter soup. Chem. Rev. 2011, 111, 6557–6602. 107. Valeur, E.; Bradley, M. Amide bond formation: Beyond the myth of coupling reagents. Chem. Soc. Rev. 2009, 38, 606–631. 108. Roschangar, F.; Sheldon, R. A.; Senanayake, C. H. Overcoming barriers to green chemistry in the pharmaceutical industry – the Green Aspiration Level™ concept. Green Chem. 2015, 17, 752–768. 237

109. Roschangar, F.; Colberg, J.; Dunn, P. J.; Gallou, F.; Hayler, J. D.; Koenig, S. G.; Kopach, M. E.; Mergelsberg, I.; Tucker, J. L.; Sheldon, R. A.; Senanayake, C. H. A deeper shade of green: Inspiring sustainable drug manufacturing. Green Chem. 2017, 19, 281–285. 110. Gaich, T.; Baran, P. S. Aiming for the ideal synthesis. J. Org. Chem. 2010, 75, 4657–4673.

238

Chapter 9

Discovery and Development of Non-Covalent, Reversible Bruton’s Tyrosine Kinase Inhibitor Fenebrutinib (GDC-0853) James J. Crawford*,1 and Haiming Zhang*,2 1Deparment of Discovery Chemistry and Genentech, Inc., 1 DNA Way,

South San Francisco, California 94080, United States 2Deparment of Small Molecule Process Chemistry, Genentech, Inc., 1 DNA Way,

South San Francisco, California 94080, United States *E-mails: [email protected] (J.J.C.); [email protected] (H.Z.).

In this chapter, we describe the discovery and synthesis of fenebrutinib (GDC0853), a selective, noncovalent, reversible Bruton’s tyrosine kinase (Btk) inhibitor for the treatment of autoimmune diseases such as rheumatoid arthritis and systemic lupus erythematosus (SLE). The biological rationale for targeting Btk is presented, followed by a brief overview of the medicinal chemistry efforts that led to the discovery of fenebrutinib, as well the first synthetic route used in the discovery stage. We then describe our phase-appropriate approach to address key limitations of the synthesis based on the stage of development and describe an efficient manufacturing process to support mid- to late-stage clinical developments.

Introduction Bruton’s tyrosine kinase (Btk) is a nonreceptor cytoplasmic tyrosine kinase that plays a central role in immune pathway signaling through both the B-cell receptor in B cells (1, 2) and the Fc receptors in myeloid cells (2). Clinical-stage and FDA-approved Btk inhibitors have demonstrated clinical efficacy in a range of B-cell malignancies including chronic lymphocytic leukemia, mantle cell lymphoma, and Waldenström’s macroglobulinemia (3–6). In addition, Btk inhibitors have shown activity in preclinical models of autoimmune diseases that are thought to be driven by pathogenic myeloid or B cells, such as collagen or adjuvant-induced arthritis as well as lupus (7–10). The lion’s share of Btk inhibitors that have been disclosed bind covalently and irreversibly with residuecysteine-481 in the ATP binding site of Btk (11, 12). At the outset of our program, our strategy was to develop noncovalent Btk inhibitors. This was due to our goal of bringing a Btk therapy to bear in the autoimmune/inflammatory arena, wherein stringent safety requirements represent a significant obstacle. Any potential idiosyncratic toxicities arising from covalent inhibitors might not be tolerated. Herein we describe the discovery and process development of fenebrutinib © 2019 American Chemical Society

(GDC-0853, 1, Figure 1), a potent and selective noncovalent Btk inhibitor currently in clinical development (13).

Figure 1. Fenebrutinib (1), a reversible Btk inhibitor currently in clinical development.

Discovery of Fenebrutinib Our early discovery efforts on developing noncovalent Btk inhibitors led to the identification of an early clinical lead, GDC-0834 (2), in addition to key milestone compounds in the shape of latergeneration inhibitors such as G-278 (3, Figure 2) (14–16).

Figure 2. First- and second-generation Btk inhibitors, GDC-0834 and G-278. While G-278 (3) represented a significant step in inhibitor development in terms of its potency and pharmacokinetic properties, it and other compounds from the same chemical series suffered from dose-limiting toxicities in both rat and dog pilot toxicity studies. Perhaps most significantly, in the Sprague Dawley rat strain, an unusual pancreatic toxicity occurred with this and other Btk inhibitors. Through extensive in vitro and in vivo efforts, we are able to demonstrate that these Btkrelated findings in the Sprague Dawley rat strain were the result of a species-specific, strain-variable, on-target effect of Btk inhibition. This was corroborated by a histologic evaluation of untreated Btk KO Sprague Dawley rats that resulted in the presence of the same pancreatic lesions (17, 18). However, for the purposes of medicinal chemistry optimization, the findings in dog liver became our focus. There, hepatotoxicity was observed in the form of increased alanine aminotransferase and aspartate aminotransferase liver enzymes, as well as perivascular mixed-cell infiltrates, Kupffer cell hypertrophy, and hepatocellular degeneration. As a result, hepatotoxicity was the primary issue that we sought to eliminate in next-generation analogues. Leveraging a cell viability assay in cryopreserved human hepatocytes, coupled to an in silico model, we prioritized our compound design goals to focus on optimizing the balance between 240

potency and lipophilicity (13). In terms of synthesis, our approach was to disconnect the molecule at the aryl-aryl bond to fragments 4 and 5, in turn made via a Suzuki–Miyaura coupling, which can be further broken into its constitutive parts, 6–10, which can be assembled via C–N coupling reactions (Scheme 1).

Scheme 1. Retrosynthetic analysis of Genentech Btk inhibitors for the purposes of SAR exploration. With the goal of reducing overall lipophilicity, we investigated replacing the aryl ring in the core with a heterocycle. Despite this being a relatively conservative structural change, it proved to be pivotal in the identification of fenebrutinib. The simplest such substitution, a pyridine, was successful. While all three possible pyridyl analogues reduced the logD7.4 by at least one log unit, it was the pyridine substituted with the H3 motif at the 2-position, 11, that offered the best balance of biochemical and whole blood potency, metabolic stability, and physical properties (Table 1). Turning our attention to target potency, we hypothesized that this might be improved by betterfilling the H3 selectivity pocket. As previously reported (13), this is formed when inhibitors from our chemical series trap an inactive form of Btk, such that Tyr551 from the activation loop moves 18Å from its position in apo-Btk structures, resulting in significant potency and kinase selectivity improvements. The tricyclic ring system in G-278 (3) features a terminal cyclohexyl motif. It appeared, later confirmed by potency data, that a dimethyl-substituted cyclopentyl replacement might better fill the available space (Figure 3). Due to the utilization of the H3 binding pocket, our Btk inhibitors, including fenebrutinib, retained potent inhibition of the ibrutinib-resistant Cys481Ser mutant of Btk (13, 19). Typical ATPmimetic covalent Btk inhibitors like ibrutinib were rendered reversible and lost enough potency in the context of the Cys481Ser Btk mutant that even increasing the dose could not overcome this resistance mutation. Retention of this beneficial mutant activity of fenebrutinib enabled its progress into clinical trials in hematological diseases to determine its clinical activity in patients, including some harboring the Cys481Ser Btk mutation (20). While a raft of H3 motifs was prepared, for the sake of brevity it is instructive to look at just two matched pairs; in this instance from thiophene- and pyrrole-containing tricyclic ring systems (Table 2). In both cases an improvement in whole blood potency was observed; in this case measuring the inhibition of up-regulation of the early activation marker CD69 on B cells after overnight anti-IgM stimulation. In the case of the pyrrole-containing compound, the resultant pyrrolopiperazinone 14 offered improved whole blood potency (IC50 = 15.4 nM vs 36.5 nM for G-278) without eroding metabolic stability in human liver microsomes leading to an increase in logD.

241

Table 1. Pyridyl for Phenyl Substitution SAR

Figure 3. Overlay of X-ray structures in the H3 pocket of Btk comparing the tricyclic motif from G-278 (3) to that of fenebrutinib (1). 242

Table 2. Selectivity Pocket, H3 SAR

Having optimized the H3 and aryl linker sections, the two portions of the molecule that remained were the aminopyridone hinge binding motif and the partially solvent-exposed H2 group. While efforts to improve the former were broadly unfruitful, at least in the context of closely-related analogues, work around the H2 group represented a significant portion of our SAR investigations. Many analogues were made and tested in this region with a view to replacing the oxetane, all the way to interchanging the entire three-ring motif. However, given the beneficial effects of this moiety on solubility and pharmacokinetic properties, it was a relatively modest change of C-alkylation of the piperazine ring that proved to be successful. Our goal was twofold: to improve van der Waals contacts with the edge of the protein while also influencing the conformation of the ring, seeking both enthalpic and entropic gains (21). Indeed, molecular modeling suggested that adding a 2-Me substituent on the piperidine ring imparts a restricted rotation relative to the unsubstituted analogue, favoring a more out-of-plane orientation of the piperidine ring (13). 243

The result of the change was a twofold improvement in whole blood CD69 potency in moving from 14 to fenebrutinib (1). In addition, we discovered that in some cases our inhibitors exhibited an increase in potency after preincubation with the enzyme. This was the case with fenebrutinib (1), which shifted sevenfold from an IC50 of 2.9 nM without preincubation to 0.4 nM after a two-hour preincubation.

Pharmacology and in Vivo Assessment of Fenebrutinib In a jump dilution experiment (Figure 4), Btk was preincubated with inhibitors 1, 14, 17, and 18 and then diluted into the assay mixture. There was a clear lag in product formation, suggesting that inhibitors with an (S)-Me (1) or (S)-Et (18) piperazine dissociate extremely slowly from Btk. In the case of fenebrutinib (1), where the balance of Btk affinity and lipophilicity resulted in improved whole blood potency, the residence time with Btk at 30 °C was 16.1 h versus 4.4 h for the unsubstituted analogue, 14.

Figure 4. Btk inhibitor H2-region SAR and jump dilution experiment.

Fenebrutinib (1) is a highly potent and selective Btk inhibitor. We evaluated its cellular potency via inhibition of anti-IgM-induced Btk phosphorylation (IC50 = 3.1 nM), anti-IgM- or CD40Lstimulated proliferation in B cells (IC50s of 1.2 and 1.4 nM, respectively), and in human monocytes where it inhibited FcγR-dependent TNFα production in human monocytes with an IC50 of 1.3 nM. When compared with a panel of other clinical-stage Btk inhibitors, each tested at 1 µM against a broad panel of human kinase biochemical assays, fenebrutinib (1) was the most Btk-selective molecule we tested, with >100-fold selectivity against 286 kinases tested (Figure 5) (13). When tested for off-target activity, fenebrutinib (1) had a clean pharmacologic profile, was negative in both in vitro Ames and MNT assays, and had an IC50 > 30 μM in an in vitro hERG assay. With promising in vitro pharmacology, coupled with excellent in vitro and in vivo pharmacokinetic properties, we proceeded to assess the in vivo efficacy of fenebrutinib (1) in a B-cell and the myeloid cell-dependent inflammatory arthritis model (13). Rats with developing collageninduced arthritis (22) were dosed orally for 16 days at a range of doses of either once (QD) or twice daily. In both cases, fenebrutinib (1) dose-dependently reduced ankle thickness. The efficacy data for the higher bar QD dosing regimen, in this case daily ankle diameter measurements, as well as inhibitor plasma concentrations, are shown in Figure 6. 244

Figure 5. Visualization of the kinase selectivity of fenebrutinib (1) versus a panel of clinical-stage Btk inhibitors against off-target kinases. IC50 values were determined against kinases that were inhibited by >50% at 1 µM. Kinases against which inhibitors did not achieve 50% inhibition at 1 µM, or for which the determined IC50 was greater than 1 µM, are noted with IC50 plotted at 1000 nM. We also assessed the ankle diameter expressed as area under the curve (AUC) as an efficacy parameter. The AUC was significantly reduced toward the normal range for rats treated QD with 16 mg/kg (99% reduction), 4 mg/kg (89%), 1 mg/kg (71%), and 0.25 mg/kg (26%), compared with the disease controls. Efficacy was measured relative to both vehicle and a positive control, in this case dexamethasone. Doses of 1, 4, and 16 mg/kg QD maintained plasma concentrations above the rat whole blood pBtk potency (IC50 = 9 nM, IC70 = 27 nM, IC90 = 135 nM, Figure 6B), for 245

a minimum of 12 h of the 24 h dosing period. In turn, doses of 1 mg/kg or greater were associated with plasma concentrations that, at a minimum, exceeded the rat whole blood pBtk IC70 (27 nM) for approximately 12 h in a 24 h period. Combined with the impressive in vivo efficacy results with our noncovalent Btk inhibitors in mouse models of systemic lupus erythematosus (SLE) (10), an excellent preclinical pharmacologic, pharmacokinetic, and in vitro safety profile, we were encouraged to progress the compound into tolerability studies.

Figure 6. Efficacy (A) and plasma concentrations (B) of fenebrutinib (GDC-0853, 1) in a rat type II collagen-induced arthritis model wherein female Lewis rats with developing collagen-induced arthritis (n = 10 per group) were dosed orally QD for 16 days at the indicated doses of 1, or the reference compound dexamethasone (0.05 mg/kg QD). A: Daily ankle diameter measurements are shown as mean ± SEM and were significantly (by ANOVA) reduced toward normal for all drug-treated rats as compared with the respective disease control. B: Each point represents a mean concentration from three rats. Dotted lines denote the mean IC50 = 9 nM, IC70 = 27 nM, and IC90 = 135 nM levels, as determined by inhibition of Btk Y223 autophosphorylation in rat whole blood. As previously mentioned, the most sensitive preclinical species for our chemical series had consistently proven to be the dog. In the case of fenebrutinib (1), it was gratifying to see that the no-observed-adverse-effect level was >80-fold higher than the targeted efficacious exposure; that is, exceeding the IC70 concentration for 12 h (in this case, from the human whole blood CD69 assay). This improvement was in line with the lack of cytotoxicity observed in primary human hepatocytes, where 1 had an IC50 >300 μM. During tolerability studies, we discovered that in Sprague Dawley rats, fenebrutinib (1) and other structurally distinct Btk inhibitors were associated with islet-centric pancreatic lesions at pharmacologically relevant doses. As chronicled elsewhere, we carried out an 246

exhaustive investigation involving evaluation of strain and species sensitivity differences, Btk knockout mice, and literature reports of humans with XLA mutations, and concluded that the fenebrutinib-related pancreas findings in the Sprague Dawley strain were the result of a rat-specific, strain-variable, on-target effect of Btk inhibition that is not relevant for humans (17). Given the previous observations with GDC-0834 (2), perhaps one of the greatest hurdles for the team, and certainly one of the key questions posed, was whether the pharmacokinetic profile in humans would mimic the excellent preclinical profile both in vitro and cross-species in vivo. In a double‐blinded, randomized, and placebo‐controlled Phase 1 healthy volunteer study, we were pleased to observe dose-proportional increases in exposure as shown in Figure 7A (23). Plasma concentrations peaked 1–3 h after oral administration, and steady‐state half‐life ranged from 4.2 to 9.9 h. Multiple independent assays demonstrated dose‐dependent Btk target engagement, and doses of 15 mg QD or greater afforded coverage of our preclinical target exposure for efficacy as assessed by inhibition of Btk autophosphorylation in whole blood (Figure 7B). Overall, the safety and PK/ PD data in Phase 1 supported the evaluation of fenebrutinib (1) in autoimmune diseases, and it is currently in Phase 2 clinical trials for autoimmune and inflammatory indications (24–26).

Figure 7. (A) Plasma concentration of fenebrutinib (1) in single-ascending dose studies from 0.5 mg to 600 mg QD. (B) Ratio of phospho-Btk to total Btk versus time for doses from 0.5 mg to 600 mg QD and placebo versus time.

Discovery Chemistry Synthesis of Fenebrutinib Throughout our medicinal chemistry efforts, we optimized the synthetic chemistry and the route such that we were able to routinely prepare tens of grams for efficacy and pilot toxicity studies. At the same time, we worked closely with our process chemistry colleagues to share insights and experience, while enhancing both the route and the synthesis of key intermediates. In the first instance, fenebrutinib (1) was synthesized as shown in Schemes 2 and 3. With the goal of divergence for SAR exploration, our synthetic scheme was split into two halves. Scheme 2 shows the synthesis of the left portion of the molecule, pyrrolopiperazinone 26. Wittig olefination of chlorovinyl aldehyde 19 gave diene 20, which upon treatment with sodium azide and subsequent denitrogenative cyclization upon heating afforded pyrrole 21. 247

The completed H3 motif, tricyclic lactam 24, was then finalized through an efficient sequence (58% yield over three steps): N-alkylation with bromoacetonitrile, nitrile reduction, and basemediated saponification/lactam cyclization. Completing this portion of the synthesis, the desired H3-linker pyrrolopiperazinone 26 was then prepared following palladium-catalyzed C–N coupling with bromopyridine aldehyde 25 in 67% yield.

Scheme 2. Synthesis of the left-hand portion of fenebrutinib, intermediate 26.

The synthesis was completed as shown in Scheme 3. Aminopyridine 27 was assembled via C–N coupling and subsequent nitro reduction, steps that will be covered in more detail (vide infra). A second coupling, this time with 3,5-dibromo-1-methylpyridin-2(1H)-one 9, provided the bulk of the right-hand portion of fenebrutinib, 28, in 61% yield. Removal of Boc and subsequent reductive amination with oxetan-3-one (30) in the presence of ZnCl2 completed the assembly of the key fragment, 31. What remained was the activation and union of the two halo intermediates, 26 and 31. This was achieved by borylation of 31 and subsequent Suzuki–Miyaura coupling, also possible in one pot, with H3-linker aldehyde 26 to give penultimate intermediate 33. Simple reduction to the alcohol with sodium borohydride in methanol at room temperature completed the gram-scale synthesis of fenebrutinib (1).

248

Scheme 3. Completion of the fenebrutinib synthesis (1).

First-Generation Process Chemistry Route Once fenebrutinib (1) was selected as a development candidate, the process chemistry team quickly turned its attention to developing a phase-appropriate and scalable synthesis to support our fast-moving toxicological and early-phase clinical studies. Retrosynthetically, we were in agreement with the discovery strategy in assembling fenebrutinib (1). Thus fenebrutinib (1) could be derived by a late-stage reduction from the penultimate aldehyde 33, which could be synthesized by a Suzuki–Miyaura coupling of pyrrolopiperazinone 26 and tetracyclic boronate 32. Compounds 26 and 32 could be generated by a palladium-catalyzed regioselective C–N coupling of tricyclic lactam 24 and 2,4-dichloronicotinaldehyde (34), and a palladium-catalyzed borylation of tetracyclic bromide 31, respectively (Scheme 4). 2,4-Dichloronicotinaldehyde (34) was judiciously selected over 2-bromo-4-chloronicotinaldehyde (25) because of the latter’s high price (ca. $3,000/kg at 50 kg scale) and issues with large-scale availability . We embarked on our first-generation fenebrutinib (1) process with the specific goal of developing a scalable and high-yielding process to tricyclic lactam 24, which is the bottleneck of the original synthesis, and optimizing and modifying the rest of the discovery synthesis as much as possible while still meeting tight delivery timelines. We envisioned that tricyclic lactam 24 could be synthesized by a base-mediated annulation of piperazin-2-one (35) and 2-chloro-4,4dimethylcyclopent-1-ene-1-carbaldehyde (19), which could be prepared from 3-methylcyclopent2-en-1-one (36) via a Grignard conjugate addition and a Vilsmeier-Haack chloroaldehyde formation (Scheme 5). 249

Scheme 4. Retrosynthetic analysis of fenebrutinib (1).

Scheme 5. Retrosynthetic analysis of tricyclic lactam 24. Our process development commenced with the methylmagnesium chloride conjugate addition to 3-methylcyclopent-2-en-1-one (36). We discovered that in the presence of 20 mol % of CuCl, the Grignard conjugate addition to 3-methylcyclopent-2-en-1-one (36) proceeded smoothly, producing a 43% yield of the desired 3,3-dimethylcyclopentanone (37) (Scheme 6). The subsequent Vilsmeier-Haack reaction was readily achieved by employing 3,3-dimethylcyclopentanone (37) and POCl3 in DMF. The relatively unstable Vilsmeier-Haack product was telescoped into the annulation reaction with piperazin-2-one. The reaction using N-methylmorpholine as the base in NMP at 115 °C afforded the desired product tricyclic lactam 24 in 51% yield (Scheme 6) (27). Although still low in overall yield (ca. 22%), this process allowed us to quickly access tens of kilograms of tricyclic lactam 24 in only three steps. This also dramatically accelerated our development of downstream chemistry. Because of the high price and unavailability of 2-bromo-4-chloronicotinaldehyde (25) employed in the discovery chemistry synthesis, we decided to pursue a cheaper alternative: 2,4dichloronicotinaldehyde (34) as the C–N coupling partner of tricyclic lactam 24. 2,4Dichloronicotinaldehyde (34) was readily prepared by treating commercially available 2,4dichloropyridine (38) with lithium diisopropylamide (LDA) at −70 °C to generate the 250

corresponding organolithium intermediate, which was then quenched with DMF to afford the desired 2,4-dichloronicotinaldehyde (34) in 70% yield (eq 1).

Scheme 6. First-generation synthesis of tricyclic lactam 24.

Our strategy for the first-generation process for tetracyclic bromide 31 was to improve the process without changing the synthetic route. Thus the chiral starting material (S)-2methylpiperazine (39) was first protected as N-Boc-(S)-2-methylpiperazine (40) in 77% yield. A palladium-catalyzed C–N coupling of 40 and 5-bromo-2-nitropyridine (41) provided an 83% yield of the desired coupling product 42. The nitro group of C–N coupling product 42 was subsequently reduced using Na2S to afford the 2-aminopyridine 27 in 96% yield (Scheme 7).

Scheme 7. First-generation synthesis of tetracyclic bromide 31.

251

Concurrently, 3,5-dibromo-1-methylpyridin-2-one (9) was prepared in 59% yield over two steps via N-bromosuccinimide bromination of 2-hydroxypyridine (43) followed by N-methylation of the dibromination product 3,5-dibromopyridin-2-one (44) using methyl tosylate (Scheme 7). About 5% O-methylated impurity was observed in the N-methylation reaction mixture and was purged to 95% conversion of the reaction to form 26, albeit with the formation of 5% of the C–N coupling regioisomer 45 (Figure 8). The reaction mixture was subsequently quenched with water and the crude pyridinylpyrrolopyrazine 26 was isolated by filtration. The crude product was slurried in THF/H2O at 65 °C, cooled to 10 °C, and then filtered to afford purified 26 in 65% isolated yield and >98% HPLC purity (Scheme 8). The regioisomer 45 was controlled to 1000 ppm). Due to time constraints, little optimization work was conducted on the Pd-catalyzed borylation step. Thus tetracyclic bromide 31 was treated with 3.0 equiv of bis(pinacolato)diboron (B2pin2) in the presence of 7 mol % Pd2(dba)3, 7 mol % XPhos, and 2.0 equiv KOAc in dioxane at 70 °C to afford a 60% corrected yield of crude boronate 32 after filtration and solvent exchange first to EtOAc then MTBE. A significant amount of product (>20%) was lost during isolation due to its relatively high solubility in a mixture of EtOAc/MTBE solvents. Based on HPLC and quantitative NMR analyses, the crude product only had ca. 70 wt% assay purity and contained des-bromo impurity 46 (ca. 4 area %), the dimer impurity 47 (ca. 3 area %), residual B2pin2 (ca.12 wt%), and MTBE (ca. 6 wt%) as major impurities (Figure 8). Nevertheless, the crude boronate 32 was carried directly to the subsequent Suzuki–Miyaura cross-coupling with C–N coupling product 26 under conditions employing 1 mol % PdCl2(dppf), 2.0 equiv of K3PO4•H2O in THF/H2O at 60 °C. Upon completion (>99% conversion of 26) of the reaction as determined by HPLC analysis, the Suzuki–Miyaura reaction mixture (ca. 80% assay yield of penultimate product 33) was filtered to remove any insoluble components and carried to the aldehyde reduction reaction directly. After a solvent exchange from THF to MeOH, the mixture was then charged with solid NaBH4 portionwise to reduce aldehyde 2 to the final product fenebrutinib (1). During the reduction, clumping of the reaction mixture occurred in one of our campaigns. Most likely this was due to limited solubility and uncontrolled crystallization of fenebrutinib (1) in MeOH, which resulted in very slow filtration of the fine crystals. The crude fenebrutinib (1) was isolated and 252

dissolved in DCM, followed by a palladium scavenger treatment using MP-TMT and Si-Thiol. An aqueous oxalic acid wash was conducted to purge the dimer impurity 48 formed in the borylation reaction.

Scheme 8. First-generation endgame.

Figure 8. Impurities observed in the Pd-catalyzed C–N coupling and borylation reactions.

253

The final crystallization of fenebrutinib (1) occurred from solvent exchange of DCM to MeOH and generated fenebrutinib (1) as the desired polymorph in 45% yield in 99.8 HPLC area % purity, 99.9% assay containing only 3 ppm Pd (Scheme 8). This fit-for-purpose first-generation process, although far from optimal, allowed us to produce multiple kilograms of GMP API to support our fastmoving Phase 1 and Phase 2a clinical studies.

Second-Generation Process Chemistry Route Upon evaluating the first-generation process for producing fenebrutinib (1), we realized that there were multiple areas that needed to improve significantly to ensure the process was suitable for late-stage development. These were to 1. 2. 3. 4. 5. 6. 7. 8.

Further improve the overall process to tricyclic lactam 24 to obtain a higher yield; Eliminate cryogenic conditions for the 2,4-dichloronicotinaldehyde (34) formation; Develop a more efficient process for preparing tetracyclic bromide 31; Improve the palladium-catalyzed C–N coupling of 24 and 34 to minimize the formation of regioisomer 45 and solve the slow filtration issue; Optimize the reaction conditions of palladium-catalyzed borylation of 31 to minimize impurity formation and streamline the isolation process to reduce mother liquor loss; Optimize the isolation of the Suzuki coupling product 33 to control process impurities and residual palladium; Select a suitable solvent and reducing agent for the reduction of penultimate aldehyde 33 to fenebrutinib (1) to eliminate the clumping/slow filtration issue; Eliminate expensive metal scavengers and develop a crystallization process to control form and particle size distribution (PSD) in the final step.

Tricyclic Lactam 24 In the addition of methyl Grignard to 3-methylcyclopent-2-en-1-one (36), we observed the formation of two major impurities, identified by GC-MS analysis as 48 and 49 (Scheme 9).We hypothesized that the Grignard adduct 50 could undergo further reaction with both the starting 3methylcyclopent-2-en-1-one (36) and the product 3,3-dimethylcyclopentanone (37) to produce these two impurities. Thus, perhaps an in situ trapping of the Grignard adduct 50 could minimize the formation of these impurities (Scheme 9). In practice, the addition of TMSCl to the Grignard reaction to trap the enolate intermediate 50 afforded the silyl ether product 51 in ca. 83% assay yield. This was subsequently solvent-exchanged to PhMe and cleaved using a substoichiometric amount of POCl3/H2O. The successive addition of DMF and POCl3 furnished the Vilsmeier-Haack product 19, which was charged directly to a DMF solution of piperazin-2-one (35) and i-Pr2NEt at 115 °C, producing a 45% yield of the desired tricyclic product 24 over four steps. Overall, with this improved second-generation process we were able to manufacture >300 kg of tricyclic lactam product 24 in excellent quality (>99.5 HPLC area % with 99.5% purity on >200 kg scale. Synthesis of Tetracyclic Bromide 31 Next, we examined a number of possible bond formation strategies to construct tetracyclic bromide 31 from four building blocks: (S)-2-methylpiperazine (39), 5-chloro-2-nitropyridine (52), oxetan-3-one (30), and 3,5-dibromo-1-methylpyridin-2-one (9). We identified the most strategically and scientifically sound approach, which was selected for further process optimization and development (Scheme 12). In summary, chiral Boc-piperazine 40 was readily obtained in 86% yield by a selective Boc protection of (S)-2-methylpiperazine (39). Subsequent palladium-catalyzed Buchwald–Hartwig amination of 40 and 5-chloro-2-nitropyridine (52) in the presence of 1.5 mol % Pd(OAc)2, 1.5 mol % BINAP, and 2.5 equiv K3PO4 in PhMe at 85 °C proceeded uneventfully to produce Boc-protected pyridinylpiperazine 42, which was telescoped into Boc deprotection, to produce pyridinylpiperazine 53 in 73% yield over two steps. Compound 53 then underwent a reductive amination with oxetan-3-one using NaBH(OAc)3 as the reducing agent to successfully give a 92% yield of the desired tricyclic 54. Hydrogenation of 54 with palladium on carbon-generated aminopyridine 55 as a solution in MeOH/PhMe after filtration which was carried directly to the next reaction. The Buchwald–Hartwig amination of 55 and 3,5dibromo-1-methylpyridin-2-one (9) employing 2 mol % Pd(OAc)2, 4 mol % Xantphos, and 1.5 equiv of K2CO3 at 105 °C furnished tetracyclic bromide 31 in 76% yield over two steps with >99.5 HPLC area % purity. Thus we accomplished the convergent synthesis of tetracyclic bromide 31 in eight steps from commercial starting materials (six linear steps) in ca. 44% yield (Scheme 12) and successfully produced >300 kg of 31 for downstream process development and manufacturing of fenebrutinib (1). Highly Regioselective C–N Coupling of 24 and 34 We set a goal to develop a highly regioselective Pd-catalyzed Buchwald–Hartwig amination of tricyclic lactam 24 and 2,4-dichloronicotinaldehyde (34) to control the regioisomeric impurity 45 and to address the issues of slow filtration and high residual palladium. To accomplish this, we resorted to catalyst screening. Extensive catalyst screening for the Pd-catalyzed Buchwald–Hartwig amination of 24 and 34 via high-throughput experimentation revealed a more regioselective and cost-effective catalyst system based on Pd(OAc)2 and 1,1′-bis(diphenylphosphino)ferrocene (dppf). The original highthroughput experimentation conditions were further validated and optimized, which led to the best seen conditions employing 1.5 equiv of 34, 2 mol % Pd(OAc)2, 4 mol % dppf, and 1.5 equiv K2CO3 in refluxing THF. 256

Scheme 12. Process development of tetracyclic bromide 31.

Scheme 13. Process development of C–N coupling to produce 26. Under the optimal conditions, regioisomer 45 was reduced to ca. 2 area % at reaction completion. A direct isolation of C–N coupling product 26 was achieved in 84% yield by simply charging H2O to the reaction mixture, followed by aging, filtration, and wash (Scheme 13, A). Although we still observed ca. 10% loss in the mother liquor, regioisomer 45 in the isolated product 26 was markedly lower at 0.6 area %. Furthermore, the residual Pd was also controlled to ca. 100 ppm by this process. We further discovered that lowering the stoichiometry of 2,4-dichloronicotinaldehyde 34 from 1.5 equiv to 1.1 equiv surprisingly generated less regioisomer 45. Interestingly, a sharp reduction of the regioisomer 45 was observed toward the end of the reaction (Figure 9), which was presumably caused by the competing consumption of regioisomer 45 by starting material 24 to form bis-coupling impurity 56. Since impurity 56 had a significantly different structure and solubility profile to those 257

of product 26, it was readily purged to 100 kg scale) of the desired C–N coupling product 26 (Scheme 13, B). The new process generated readily filterable crystalline 26, thus drastically increasing the filtration rate during isolation, and was successfully implemented in >100 kg production.

Figure 9. C–N coupling reaction of 24 and 34 (1.1 equiv). Borylation of Tetracyclic Bromide 31 Pd-catalyzed borylation of tetracyclic bromide 31 with bis(pinacolato)diboron (B2pin2) to generate boronate 32 was investigated next. We confirmed that Pd2(dba)3/XPhos was a highly effective catalyst system to complete the borylation reaction in the presence of KOAc in THF at 65 °C. Further optimization of the stoichiometry of catalyst/ligand, B2pin2 and KOAc, and solvent volume led to the optimal conditions for the borylation reaction, which employed 0.25 mol % Pd2(dba)3, 0.6 mol % XPhos, 1.5 equiv B2pin2, and 2 equiv KOAc in THF (10 L/kg) at 65 °C (Scheme 14). Under these conditions, the borylation reaction of 31 consistently afforded >99% conversion with 100 kg scale to generate a high quality of boronate 32 with >99.5 HPLC area % purity and 170, 99% purity (45)) to supply both early analogue synthesis and the upcoming kg scale manufacture of API. As this work was underway, further optimization was conducted to determine if safer reaction conditions for the azide cycloaddition could be employed. Lewis acid–promoted tetrazole formation by nitrile-azide cycloaddition presented an interesting option with milder conditions for avoiding formation of hydrazoic acid. Zinc salts were particularly effective, and evaluation of ZnO and ZnBr2 showed good reactivity at a neutral reaction pH (46, 47). The reaction mixture was pH 7.5, and the pKa of HN3 is 4.6 (48). After further optimization, it was discovered that the reaction could be conducted in aqueous iPrOH mixtures at 80 °C leading to a significant decrease in the amount of hydrazoic acid detected in the head space of the reactor. However, compared to the previous approach, this reaction was much more sluggish, requiring 48 h to reach 95% conversion. This process was demonstrated at multigram scale, providing product in 80–85% yield. While this Zn-catalyzed method provided an attractive alternative to the acidic processes, the initial manufacturing campaign scaled up the acidic process without Zn additives because the reaction was faster and, overall, a safe and effective way to measure and control HN3 levels in place. During analogue preparation, scalemic prodrug fragment 42 was produced via base-mediated alkylation of tetrazole 38 using 1-chloroethyl ethylcarbonate 39 (Scheme 9) (49). The alkylation was neither regioselective nor stereoselective, but individual enantiomers of either N-1 or N-2 279

regioisomers (40 and 41) were accessible after chromatographic separation. 1-Chloroethyl ethylcarbonate (39) can be converted into the bromo or iodo variants, and these were also screened in this reaction (50–52). The bromo analogue performed similarly to the chloro analogue (~3:1 ratio of N-2:N-1), but the iodo analogue provided better reactivity at a lower temperature and an improvement in the ratio of regioisomers (~1:1).

Scheme 9. Nonselective alkylation of 38 and isolation of 42. Unfortunately, upon attempted scale-up of 1-iodoethyl ethylcarbonate for further evaluation, decomposition of the carbonate was detected, resulting in pressurization of the storage vessels. Thermal stability data of 1-iodoethyl ethylcarbonate displayed a low onset temperature of decomposition (60 °C, releasing 363 J/g). Due to the challenges with manufacturing and storing this material and the potential for pressure build-updue to decomposition,, this was not pursued further. Optimization of the reaction using 1.2 equiv of ethylcarbonate 39 by screening base, temperature, and solvents provided a ~2.3:1 ratio of N-2:N-1 regioisomers (40, 41) and complete consumption of the starting material in 3–5 h using 1.2 equiv of triethylamine and 3 mL/g acetonitrile at reflux. To access 4 for early development studies, this alkylation approach was used for the first kg scale manufacture of 4, accessing ~2 kg of intermediate 42 after chromatographic separation of regioisomers and enantiomers. For other prodrugs under consideration during the discovery process, methodology to access an asymmetric prodrug using a chiral, nonracemic DMAP-derived catalyst 43 was developed. The tetrazole would form a hemiaminal with acetaldehyde that was then trapped with diethyl dicarbonate to form the carbonate prodrug 41 (Scheme 10). This was highly effective for accessing the N2–modified tetrazole 41 with enantiomeric enrichment that could be used for the synthesis of N-2 tetrazole prodrug analogs; however, it did not provide the regioisomer needed for 4 (18). Interestingly, when 1H-tetrazole was used in this reaction, a ~6.5:1 ratio of N-1:N-2 alkylated products 44 and 45 was formed, favoring the N-1 regioisomer (Scheme 11). The improved N1 regioselectivity was intriguing and prompted an evaluation of the tetrazole-pyrazole bond disconnection as an alternative synthetic strategy.

280

Scheme 10. Asymmetric formation of N-2–substituted tetrazole 41 using chiral DMAP catalyst 43.

Scheme 11. Synthesis of prodrug modified tetrazoles 44, 45, and 46 from 1H-tetrazole. Unfortunately, coupling 44 and 5-iodo-1-methyl-1H-pyrazole (48) (used as a model system) proved to be a significant challenge (Scheme 12). Few examples exist in the literature of crosscoupling via C-H activation pathways using tetrazoles. Additionally, the low onset temperature and the high energy of 44 (1576 J/g, onset at 128 °C) would limit the range of acceptable, safe reaction temperatures. Bromination of 44 to form 46 was possible using 1,3-dibromo-5,5-dimethylhydantoin in AcOH (Scheme 11). This compound also exhibited a low decomposition temperature and high energy release (1294 J/g, onset at 104 °C). Screening the cross-coupling with 46 and 47 or 48 was also unsuccessful. Predominantly, loss of prodrug or complete disappearance of the tetrazole core was observed. Studies suggested that requisite metallated tetrazole intermediate may be prone to decomposition, so this approach was not pursued further (53) (Scheme 12).

Scheme 12. Attempted cross-coupling of pyrazoles 47 or 48 with tetrazoles 44 or 46. Other common synthetic approaches to access N-1–substituted tetrazoles typically involve a [3+2] cycloaddition of an azide and an activated secondary amide derivative (e.g., thioamide, imidoyl triflate, or imidoyl chloride) (54–57). To access 40, amide 49 would be required. This proved to be challenging and attempts to alkylate 36 using 39 resulted in what appeared to be dioxazinones 50 or 51, as evidenced by mass spectrometry. Due to the observed prodrug instability under these conditions and the requirement for nucleophilic azide for tetrazole formation, this strategy to access 40 was abandoned (Scheme 13). 281

Scheme 13. Access to N-1–substituted tetrazole 40 via amide 49. Adapted from reference (49). Copyright 2019. American Chemical Society. Another strategy for accessing the desired regioisomer was “exhaustive alkylation” (58). This involved first installing a temporary blocking group at the N-2 position (Scheme 14, 52). Next, alkylation at the N-1 position with the hemiaminal carbonate would generate a tetrazolium cation 54, and selective cleavage of the N-2 blocking group would then provide the desired N-1–substituted tetrazole 40. A series of N-2 blocking groups was screened, including silyl-protecting groups (TIPS or TBS) and carbon-based protecting groups (SEM, t-Bu, trityl, or PMB). In all cases, challenges were met through either formation of the desired N-2 regioisomer, the alkylation to install the prodrug, or the removal of the blocking group.

Scheme 14. Exhaustive alkylation strategy. An interesting variation of this approach uses tin to block the N-2 position of the tetrazole. Exposure of 5-methyl-1H-tetrazole to bis(tributyltin) oxide followed by neat methyl iodide was reported to yield the N-1–methylated tetrazole as a single regioisomer (59, 60). This approach was attempted by using 38 and 1-bromoethyl ethylcarbonate (53); interestingly, the reaction resulted in a ~80:20 ratio of N-1 to N-2 regioisomers. The use of ethylcarbonate 39 did not result in the formation of product 40. Further study of this reaction was completed to understand the basis for this selectivity. First, NMR studies on the addition of bis(tributyltin)oxide to 38 resulted in a ~0.2-ppm shift in the NMR signal for the C-3 pyrazole proton. This change in the chemical shift was consistent with the NMR spectrum of the N-2–substituted tin-tetrazole complex 55 that was definitively accessed via reaction of cyanopyrazole 37 with tributyltin azide (Scheme 15). 282

Scheme 15. Formation of stannylated tetrazole 55. Tetrazole 38 reacted with bis(tributyltin) oxide in a variety of solvents including MTBE and acetonitrile. In MTBE, tetrazole 38 had low solubility, but, upon addition of the tin reagent, all solids dissolved. When concentrated, a viscous oil resulted and then solidified to a crystalline solid over time. A crystal structure of this solid was obtained, showing a polymeric lattice of 38 and tributyltin, with tin atoms bridging the N-1 and N-3 positions of the tetrazoles (Scheme 16).

Scheme 16. Polymeric tributyltin-tetrazole array. Adapted from reference (49). Copyright 2019. American Chemical Society. To further understand the reactive nature of the tin-tetrazolyl polymeric species, computational studies were undertaken. Cosmotherm calculations of the dissociation of the tin-tetrazole bond predicted that the desired form was likely the sterically less hindered N-2–substituted tetrazole 55 in a 98:2 ratio (Scheme 17).

Scheme 17. Dissociation of a polymeric tributyltin-tetrazole complex predicted to favor monomeric 55. Adapted from reference (49). Copyright 2019. American Chemical Society. Preliminary transition state modeling of the N-2–stannylated tetrazole 55 with 1-bromoethyl ethylcarbonate 53 was then conducted using a Gaussian 09 program (61) at DFT/B3LYP/Lanl2DZ (for Br, I, and Sn)/6-31G(d) level of theory in the gas phase (Scheme 18). The transition state energies for alkylation at the N-1, N-3, and N-4 positions were calculated, and the pathway of

283

reactivity at the N-4 position (leading to the N-1–alkylated product) was favored by ~2.1 kcal/mol over the pathway leading to the N-2–alkylated product.

Scheme 18. Transition state modeling of the reaction of 55 with 53. Adapted from reference (49). Copyright 2019. American Chemical Society. From a scale-up perspective, the reaction was operationally simple and performed well. A slurry of 38 in MTBE was treated with bis(tributyl)tin oxide, thus forming a solution. 1-Bromoethyl ethylcarbonate 53 was then added, and the mixture was stirred at room temperature until the reaction reached ~85–90% conversion. The use of stoichiometric tin was a major concern due to the need to control tin impurities in the API. Early dose predictions estimated that studies could reach ~4 g/day. While ICH guidelines for oral exposure to elemental impurities list tin as 6 mg/day, it was determined that this was likely in reference to inorganic tin salts and not organotin residues. Based on guidelines for organotin residues in food, a limit of 16 μg/g of 4 was determined to be acceptable, setting the specification for residual tin in the API to 4 ppm. A rigorous work-up of the reaction was completed to control tin impurities in the isolated solids. When the reaction was deemed complete by the NMR (~85% conversion), the reaction was partitioned between acetonitrile and n-heptane to remove nonpolar tin residues. The resulting acetonitrile solution was then treated with a 9% solution of aqueous KF, converting the reaction byproduct tributyltin bromide or any remaining tributyltin oxide species to a polymeric, insoluble tributyltin fluoride solid that could be removed by filtration. The final acetonitrile solution was concentrated and, fortuitously, the desired N-1 regioisomer 40 crystallized out in >99% purity (UPLC) and a 60% isolated yield (Scheme 19). Isolated solids typically contained 98% ee through hydrolysis of the ethyl carbonate prodrug of the undesired N-1 enantiomer. Additionally, early samples of material contained mixtures of the N-1 and N-2 regioisomers 40 and 41. The enzymeselectively hydrolyzed both enantiomers of the N-2 regioisomer 41 and the undesired N-1 enantiomer, which effectively accessed the desired enantiomer from a racemic, nonregioselective starting point (Scheme 20). As an interesting note, the CAL-B enzyme used for this process was also effective for hydrolyzing the undesired enantiomer of the tetrazole species 44 and 46. 284

Scheme 19. Tin–mediated tetrazole alkylation with 53.

Scheme 20. Enzymatic hydrolysis of the prodrug to provide enantiopure 42. After identying this enzyme through small scale screening, further optimization was undertaken to prepare for a scale-up of the reaction. Organic cosolvents were screened, and MTBE was chosen to enable downstream isolation of the product. The CAL-B enzyme exhibited some instability in MTBE, so an immobilized CAL-B enzyme formulation (Novozyme 435) with superior stability in the biphasic reaction mixture was identified. The biocatalytic resolution scaled well. The reaction was performed in a biphasic mixture of MTBE and a phosphate buffer with pH 7 at 42 °C. Novozyme 435 (40 wt%) was used, and the substrate was charged at 50 g/L. The reaction required 60–65 h to reach the target 98% conversion. The long reaction time was due to the decreased concentration of the substrate (undesired enantiomer of the N-1 regioisomer) as the reaction progressed. Upon completion, the aqueous layer was removed, and then the immobilized enzyme was removed by filtration, which left an MTBE solution of 42. This was exchanged into EtOH, seeded, and crystallized further by adding water (Scheme 21). The reaction was performed on an 8-kg scale and the product was isolated in a 42–45% yield with 97-100% ee and 99% conversion with minimal byproduct formation. Isolation of 59 from this reaction mixture was initially a challenge. For the first scale-up campaign, silica gel chromatography was required to remove residual palladium and other organic impurities to allow crystallization of the product. To avoid chromatography for the second campaign, a seeded crystallization requiring significant volumes of antisolvent was developed. After the reaction was completed, it was initially quenched by adding aqueous NH4Cl. MTBE was then added, and the aqueous layer was removed. The organic layer was concentrated, and the solvent was exchanged to 2MeTHF (1000 ppm to 99% purity (UPLC), ~0.1% of the undesired diastereomer, 99% purity (UPLC). The first kg scale manufacture of 4 produced 1.4 kg, and the second manufacture, carried out less than 12 months from the initial delivery, increased the scale to deliver 13.6 kg. The development of a new approach to 42 as well as further process understanding and optimization of the final bondforming steps were instrumental in this rapid scale-up to meet clinical demands for 4. 289

Conclusion We have described our path to small molecule PCSK9 inhibitors starting from screening hit 1 to orally-active small molecule inhibitors 2 or 3 to clinical candidate 4. The mode of action was determined to be inhibition of PCSK9 mRNA translation. The development of key synthetic methodologies allowed for the expansion of the SAR for the series that contained a hindered tertiary amide. The efficient exploration of alternative head groups led to a change from isoquinoline to 3chloropyridine. Changes to the tail group allowed for the identification of more selective compounds as assessed by ribosome profiling. Off-target cardiovascular effects (e.g., hERG inhibition) of the basic leads prompted exploration of zwitterionic leads. A liver-targeted approach produced the prodrug compound 4 that was suitable for clinical evaluation.

Scheme 26. Synthetic route from tetrazole 38 to API 4, scaled to manufacture over 13.6 kg. Adapted from reference (49). Copyright 2019. American Chemical Society. PCSK9 inhibitor 4 presented numerous synthetic challenges, centered on the preparation and handling of a high energy tetrazole moiety containing a sensitive, stereochemically defined hemiaminal carbonate prodrug. Due to the short timelines and the potential need for large quantities of API to supply early studies, enabling first and second generation routes to 42 and further optimization of the final steps provided access to multi-kg quantities of API. The first kg scale delivery of API yielded 1.4 kg, and rapid implementation of the tin-mediated regioselective alkylation and enzymatic prodrug resolution improved access to intermediate 42 (Scheme 26), thus enabling the manufacture of 13.6 kg of API. This new route doubled throughput to key intermediate 42 and 290

eliminated the requirement for chromatographic separation. Despite the Herculean effort by the team to progress 4 into the clinic, business decisions halted work beyond Phase 1 clinical trials (63).

Acknowledgments The authors would like to thank the countless number of contributors and collaborators who participated in the Pfizer PCSK9 small molecule inhibitor project from its inception through its Phase 1 clinical trials.

References 1.

Park, S. W.; Moon, Y. A.; Horton, J. D. Post-Transcriptional Regulation of Low Density Lipoprotein Receptor Protein by Proprotein Convertase Subtilisin/Kexin Type 9a in Mouse Liver. J. Biol.Chem. 2004, 279, 50630–50638. 2. Cunningham, D.; Danley, D. E.; Geoghegan, K. F.; Griffor, M. C.; Hawkins, J. L.; Subashi, T. A.; Varghese, A. H.; Ammirati, M. J.; Culp, J. S.; Hoth, L. R.; Mansour, M. N.; McGrath, K. M.; Seddon, A. P.; Shenolikar, S.; Stutzman-Engwall, K. J.; Warren, L. C.; Xia, D.; Qiu, X. Structural and Biophysical Studies of PCSK9 and Its Mutants Linked to Familial Hypercholesterolemia. Nat. Struct. Mol. Biol. 2007, 14, 413–419. 3. Piper, D. E.; Jackson, S.; Liu, Q.; Romanow, W. G.; Shetterly, S.; Thibault, S. T.; Shan, B.; Walker, N. P. C. The Crystal Structure of PCSK9: A Regulator of Plasma LDL-Cholesterol. Structure 2007, 15, 545–552. 4. Lo Surdo, P.; Bottomley, M. J.; Calzetta, A.; Settembre, E. C.; Cirillo, A.; Pandit, S.; Ni, Y. G.; Hubbard, B.; Sitlani, A.; Carfí, A. Mechanistic Implications for LDL Receptor Degradation from the PCSK9/LDLR Structure at Neutral pH. EMBO Rep. 2011, 12, 1300–1305. 5. Cohen, J. C.; Boerwinkle, E.; Mosley, T. H.; Hobbs, H. H. Sequence Variations in PCSK9, Low LDL, and Protection Against Coronary Heart Disease. New Engl. J. Med. 2006, 354, 1264–1272. 6. Giugliano, R. P.; Sabatine, M. S. Are PCSK9 Inhibitors the Next Breakthrough in the Cardiovascular Field? J. Am. Coll. Cardiol. 2015, 65, 2638–2651. 7. van Poelgeest, E. P.; Swart, R. M.; Betjes, M. G. H.; Moerland, M.; Weening, J. J.; Tessier, Y.; Hodges, M. R.; Levin, A. A.; Burggraaf, J. Acute Kidney Injury During Therapy with an Antisense Oligonucleotide Directed Against PCSK9. Am. J. Kidney Dis. 2013, 62, 796–800. 8. Fitzgerald, K.; Frank-Kamenetsky, M.; Shulga-Morskaya, S.; Liebow, A.; Bettencourt, B. R.; Sutherland, J. E.; Hutabarat, R. M.; Clausen, V. A.; Karsten, V.; Cehelsky, J.; Nochur, S. V.; Kotelianski, V.; Horton, J.; Mant, T.; Chiesa, J.; Ritter, J.; Munisamy, M.; Vaishnaw, A. K.; Gollob, J. A.; Simon, A. Effect of an RNA Interference Drug on the Synthesis of Proprotein Convertase Subtilisin/Kexin Type 9 (PCSK9) and the Concentration of Serum LDL Cholesterol in Healthy Volunteers: A Randomised, Single-Blind, Placebo-Controlled, Phase 1 Trial. Lancet 2014, 383, 60–68. 9. Pettersen, D.; Fjellström, O. Small Molecule Modulators of PCSK9 – A Literature and Patent Overview. Bioorganic Med. Chem. Lett. 2018, 28, 1155–1160. 10. Xu, S.; Luo, S.; Zhu, Z.; Xu, J. Small Molecules as Inhibitors of PCSK9: Current Status and Future Challenges. Eur. J. Med. Chem. 2019, 162, 212–233. 291

11. Pingali, H.; Kalapatapu, V. V. M. S.; Makadia, P.; Jain, M. R. Compounds for the Treatment of Dyslipidemia and Related Diseases. Patent WO2011051961, May 5, 2011. 12. McNutt, M. C.; Lagace, T. A.; Horton, J. D. Catalytic Activity is Not Required for Secreted PCSK9 to Reduce Low Density Lipoprotein Receptors in HepG2 Cells. J. Biol. Chem. 2007, 282, 20799–20803. 13. Li, J.; Tumanut, C.; Gavigan, J. A.; Huang, W. J.; Hampton, E. N.; Tumanut, R.; Suen, K. F.; Trauger, J. W.; Spraggon, G.; Lesley, S. A.; Liau, G.; Yowe, D.; Harris, J. L. Secreted PCSK9 Promotes LDL Receptor Degradation Independently of Proteolytic Activity. Biochem. J. 2007, 406, 203–207. 14. Schroeder, C. I.; Swedberg, J. E.; Withka, J. M.; Rosengren, K. J.; Akcan, M.; Clayton, D. J.; Daly, N. L.; Cheneval, O.; Borzilleri, K. A.; Griffor, M.; Stock, I.; Colless, B.; Walsh, P.; Sunderland, P.; Reyes, A.; Dullea, R.; Ammirati, M.; Liu, S.; McClure, K. F.; Tu, M.; Bhattacharya, S. K.; Liras, S.; Price, D. A.; Craik, D. J. Design and Synthesis of Truncated EGF-A Peptides That Restore LDL-R Recycling in the Presence of PCSK9 In Vitro. Chem. Biol. 2014, 21, 284–294. 15. Zhang, Y.; Eigenbrot, C.; Zhou, L.; Shia, S.; Li, W.; Quan, C.; Tom, J.; Moran, P.; Di Lello, P.; Skelton, N. J.; Kong-Beltran, M.; Peterson, A.; Kirchhofer, D. Identification of a Small Peptide That Inhibits PCSK9 Protein Binding to the Low Density Lipoprotein Receptor. J. Biol. Chem. 2014, 289, 942–955. 16. Taechalertpaisarn, J.; Zhao, B.; Liang, X.; Burgess, K. Small Molecule Inhibitors of the PCSK9·LDLR Interaction. J. Am. Chem. Soc. 2018, 140, 3242–3249. 17. Zhang, Y.; Ultsch, M.; Skelton, N. J.; Burdick, D. J.; Beresini, M. H.; Li, W.; Kong-Beltran, M.; Peterson, A.; Quinn, J.; Chiu, C.; Wu, Y.; Shia, S.; Moran, P.; Di Lello, P.; Eigenbrot, C.; Kirchhofer, D. Discovery of a Cryptic Peptide-Binding Site on PCSK9 and Design of Antagonists. Nat. Struct. Mol. Biol. 2017, 24, 848–856. 18. Petersen, D. N.; Hawkins, J.; Ruangsiriluk, W.; Stevens, K. A.; Maguire, B. A.; O’Connell, T. N.; Rocke, B. N.; Boehm, M.; Ruggeri, R. B.; Rolph, T.; Hepworth, D.; Loria, P. M.; Carpino, P. A. A Small-Molecule Anti-Secretagogue of PCSK9 Targets the 80S Ribosome to Inhibit PCSK9 Protein Translation. Cell Chem. Biol. 2016, 23, 1362–1371. 19. Lintner, N. G.; McClure, K. F.; Petersen, D.; Londregan, A. T.; Piotrowski, D. W.; Wei, L.; Xiao, J.; Bolt, M.; Loria, P. M.; Maguire, B.; Geoghegan, K. F.; Huang, A.; Rolph, T.; Liras, S.; Doudna, J. A.; Dullea, R. G.; Cate, J. H. D. Selective Stalling of Human Translation Through Small-Molecule Engagement of the Ribosome Nascent Chain. PLoS Biol. 2017, 15, e2001882. 20. Londregan, A. T.; Wei, L.; Xiao, J.; Lintner, N. G.; Petersen, D.; Dullea, R. G.; McClure, K. F.; Bolt, M. W.; Warmus, J. S.; Coffey, S. B.; Limberakis, C.; Genovino, J.; Thuma, B. A.; Hesp, K. D.; Aspnes, G. E.; Reidich, B.; Salatto, C. T.; Chabot, J. R.; Cate, J. H. D.; Liras, S.; Piotrowski, D. W. Small Molecule Proprotein Convertase Subtilisin/Kexin Type 9 (PCSK9) Inhibitors: Hit to Lead Optimization of Systemic Agents. J. Med. Chem. 2018, 61, 5704–5718. 21. McClure, K. F.; Piotrowski, D. W.; Petersen, D.; Wei, L.; Xiao, J.; Londregan, A. T.; Kamlet, A. S.; Dechert-Schmitt, A. M.; Raymer, B.; Ruggeri, R. B.; Canterbury, D.; Limberakis, C.; Liras, S.; DaSilva-Jardine, P.; Dullea, R. G.; Loria, P. M.; Reidich, B.; Salatto, C. T.; Eng, H.; Kimoto, E.; Atkinson, K.; King-Ahmad, A.; Scott, D.; Beaumont, K.; Chabot, J. R.; Bolt, M. W.; Maresca, K.; Dahl, K.; Arakawa, R.; Takano, A.; Halldin, C. Liver-Targeted Small292

22. 23. 24. 25.

26.

27.

28. 29.

30. 31. 32.

33. 34. 35.

36.

37.

Molecule Inhibitors of Proprotein Convertase Subtilisin/Kexin Type 9 Synthesis. Angew. Chem., Int. Ed. 2017, 56, 16218–16222. Abramovitch, R. A.; Singer, G. M. Direct Alkyl and Aryl Amination of Heteroaromatic Nitrogen Compounds. J. Am. Chem. Soc. 1969, 91, 5672–5673. Abramovitch, R. A.; Singer, G. M. Direct Acylamination of Pyridine 1-Oxides. J. Org. Chem. 1974, 39, 1795–1802. Abramovitch, R. A.; Rogers, R. B. Direct Acylamination of 3-Substituted Pyridine-1-Oxides. Directive Effect of the Substituent. J. Org. Chem. 1974, 39, 1802–1807. Abramovitch, R. A.; Pilski, J.; Konitz, A.; Tomasik, P. Direct Acylamination of Pyridine 1Oxide with N-Phenylarenimidoyl Chlorides and Fluorides. J. Org. Chem. 1983, 48, 4391–4393. Manley, P. J.; Bilodeau, M. T. A Mild Method for the Formation and In Situ Reaction of Imidoyl Chlorides: Conversion of Pyridine-1-Oxides to 2-Aminopyridine Amides. Org. Lett. 2002, 4, 3127–3129. Couturier, M.; Caron, L.; Tumidajski, S.; Jones, K.; White, T. D. Mild and Direct Conversion of Quinoline N-Oxides to 2-Amidoquinolines with Primary Amides. Org. Lett. 2006, 8, 1929–1932. Londregan, A. T.; Jennings, S.; Wei, L. General and Mild Preparation of 2-Aminopyridines. Org. Lett. 2010, 12, 5254–5257. Londregan, A. T.; Storer, G.; Wooten, C.; Yang, X.; Warmus, J. An Improved Amide Coupling Procedure for the Synthesis of N-(Pyridin-2-yl)amides. Tetrahedron Lett. 2009, 50, 1986–1988. Londregan, A. T.; Piotrowski, D. W.; Xiao, J. Rapid and Selective In Situ Reduction of Pyridine-N-Oxides with TetraHydroxydiboron. Synlett 2013, 24, 2695–2700. Balicki, R. TiCl4/NaI - A Novel, Efficient Reagent for Mild Reduction of the N - O Bond in Amine N-Oxides and Nitrones. Chem. Ber. 1990, 123, 647–648. All procedures performed on animals in this study were in accordance with established guidelines and regulations and were reviewed and approved by Pfizer Institutional Animal Care and Use Committee. Pfizer animal care facilities that supported this work are fully accredited by AAALAC International. Waring, M. J.; Johnstone, C. A Quantitative Assessment of hERG Liability as a Function of Lipophilicity. Bioorganic Med. Chem. Lett. 2007, 17, 1759–1764. Pfefferkorn, J. A. Strategies for the Design of Hepatoselective Glucokinase Activators to Treat Type 2 Diabetes. Expert Opin. Drug. Discov. 2013, 8, 319–330. Ming, X.; Knight, B. M.; Thakker, D. R. Vectorial Transport of Fexofenadine across Caco-2 Cells: Involvement of Apical Uptake and Basolateral Efflux Transporters. Mol. Pharm. 2011, 8, 1677–1686. Piotrowski, D. W.; Kamlet, A. S.; Dechert-Schmitt, A. M.; Yan, J.; Brandt, T. A.; Xiao, J.; Wei, L.; Barrila, M. T. Regio- and Enantioselective Synthesis of Azole Hemiaminal Esters by Lewis Base Catalyzed Dynamic Kinetic Resolution. J. Am. Chem. Soc. 2016, 138, 4818–4823. Akin, A.; Barrila, M. T.; Brandt, T. A.; Dechert-Schmitt, A. M. R.; Dube, P.; Ford, D. D.; Kamlet, A. S.; Limberakis, C.; Pearsall, A.; Piotrowski, D. W.; Quinn, B.; Rothstein, S.; Salan, J.; Wei, L.; Xiao, J. A Scalable Route for the Regio- and Enantioselective Preparation of a 293

38.

39.

40. 41. 42. 43.

44. 45. 46. 47.

48. 49.

50. 51. 52.

53.

Tetrazole Prodrug: Application to the Multi-Gram-Scale Synthesis of a PCSK9 Inhibitor. Org. Process Res. Dev. 2017, 21, 1990–2000. Trapa, P. E.; Beaumont, K.; Atkinson, K.; Eng, H.; King-Ahmad, A.; Scott, D. O.; Maurer, T. S.; Di, L. In Vitro–In Vivo Extrapolation of Intestinal Availability for Carboxylesterase Substrates Using Portal Vein–Cannulated Monkey. J. Pharm. Sci. 2017, 106, 898–905. De Bruyn, T.; Chatterjee, S.; Fattah, S.; Keemink, J.; Nicolaï, J.; Augustijns, P.; Annaert, P. Sandwich-Cultured Hepatocytes: Utility for In Vitro Exploration of Hepatobiliary Drug Disposition and Drug-Induced Hepatotoxicity. Expert Opin. Drug Metab. Toxicol. 2013, 9, 589–616. Taketani, M.; Shii, M.; Ohura, K.; Ninomiya, S.; Imai, T. Carboxylesterase in the Liver and Small Intestine of Experimental Animals and Human. Life Sci. 2007, 81, 924–932. Darout, E.; McClure, K. F.; Piotrowski, D.; Raymer, B. Substituted Amide Compounds. Patent WO2016055901, April 14, 2016. Zhang, K.; Wang, X.; Hao, L.; Zhao, Z.; Han, C. Dynamic Observation of 18F-FDG Uptake After Oral Administration in a Healthy Subject. J. Nucl. Med. Technol. 2013, 41, 78–80. Wiss, J.; Fleury, C.; Onken, U. Safety Improvement of Chemical Processes Involving Azides by Online Monitoring of the Hydrazoic Acid Concentration. Org. Process Res. Dev. 2006, 10, 349–353. Bräse, S.; Gil, C.; Knepper, K.; Zimmermann, V. Organic Azides: An Exploding Diversity of a Unique Class of Compounds. Angew. Chem., Int. Ed. 2005, 44, 5188–5240. UPLC area % purity. All other references to purity will be UPLC area % unless otherwise mentioned. Demko, Z. P.; Sharpless, K. B. Preparation of 5-Substituted 1H-Tetrazoles from Nitriles in Water. J. Org. Chem. 2001, 66, 7945–7950. Girardin, M.; Dolman, S. J.; Lauzon, S.; Ouellet, S. G.; Hughes, G.; Fernandez, P.; Zhou, G.; O’Shea, P. D. Development of a Practical Synthesis of Stearoyl-CoA Desaturase (SCD1) Inhibitor MK-8245. Org. Process Res. Dev. 2011, 15, 1073–1080. Patnaik, P. Handbook of Inorganic Chemical Compounds; McGraw Hill: New York, 2001. Akin, A.; Barrila, M. T.; Brandt, T. A.; Brennan, J.; Henegar, K. E.; Hoagland, S.; Kumar, R.; Magano, J.; McInturff, E. L.; Nematalla, A.; Piotrowski, D. W.; Haitsma, J. V.; Wei, L.; Xiao, J.; Yu, S. Overcoming the Challenges of Making a Single Enantiomer N-1 Substituted Tetrazole Prodrug Using a Tin-mediated Alkylation and Enzymatic Resolution. Org. Process Res. Dev. 2019, 23, 1167–1177. DOI: 10.1021/acs.oprd.9b00104. Iyer, R. P.; Yu, D.; Ho, N. H.; Agrawal, S. Synthesis of Iodoalkylacylates and Their Use in the Preparation of S-Alkyl Phosphorothiolates. Synth. Commun. 1995, 25, 2739–2749. Senet, J.-P.; Sennyey, G.; Wooden, G. P. Novel Chloride to Bromide Exchange: Preparation of 1-Bhomoalkyl Carbonates. Synth. Commun. 1988, 18, 1525–1530. Prybylski, J.; Thiele, N. A.; Sloan, K. B. Regioselective Synthesis of 2-O-Acyl-3-O-(1acyloxyalkyl) Prodrugs of 5,6-Isopropylidene-l-Ascorbic Acid. Tetrahedron Lett. 2016, 57, 1619–1621. Špulák, M.; Lubojacký, R.; Šenel, P.; Kuneš, J.; Pour, M. Direct C−H Arylation and Alkenylation of 1-Substituted Tetrazoles: Phosphine as Stabilizing Factor. J. Org. Chem. 2010, 75, 241–244. 294

54. Santella, J. B.; Gardner, D. S.; Yao, W.; Shi, C.; Reddy, P.; Tebben, A. J.; DeLucca, G. V.; Wacker, D. A.; Watson, P. S.; Welch, P. K.; Wadman, E. A.; Davies, P.; Solomon, K. A.; Graden, D. M.; Yeleswaram, S.; Mandlekar, S.; Kariv, I.; Decicco, C. P.; Ko, S. S.; Carter, P. H.; Duncia, J. V. From Rigid Cyclic Templates to Conformationally Stabilized Acyclic Scaffolds. Part I: The Discovery of CCR3 Antagonist Development Candidate BMS-639623 with Picomolar Inhibition Potency Against Eosinophil Chemotaxis. Bioorganic Med. Chem. Lett. 2008, 18, 576–585. 55. Sidduri, A.; Tilley, J. W.; Hull, K.; Ping Lou, J.; Kaplan, G.; Sheffron, A.; Chen, L.; Campbell, R.; Guthrie, R.; Huang, T.-N.; Huby, N.; Rowan, K.; Schwinge, V.; Renzetti, L. M. NCycloalkanoyl-l-Phenylalanine Derivatives as VCAM/VLA-4 Antagonists. Bioorganic Med. Chem. Lett. 2002, 12, 2475–2478. 56. Quast, H.; Bieber, L.; Meichsner, G. Photochemical Formation of Heteromethylenecyclopropanes. 16. 1,4,5-Substituted Tetrazolium Salts Through Methylation of 1,5-Substituted Tetrazoles and [3 + 2] Cycloaddition of Alkyl Azides to Nitrilium Ions. Liebigs Ann. Chem. 1987, 469–475. 57. Himo, F.; Demko, Z. P.; Noodleman, L.; Sharpless, K. B. Mechanisms of Tetrazole Formation by Addition of Azide to Nitriles. J. Am. Chem. Soc. 2002, 124, 12210–12216. 58. Koren, A. O.; Gaponik, P. N.; Ivashkevich, O. A.; Kovalyova, T. B. A New Route to 1Alkyltetrazoles: Via 2-Tert-Butyltetrazoles. Mendeleev Commun. 1995, 5, 10–11. 59. Sisido, K.; Nabika, K.; Isida, T.; Kozima, S. Formation of Organotin-Nitrogen Bonds III. NTrialkyltin-5-Substituted Tetrazoles. J. Organomet. Chem. 1971, 33, 337–346. 60. Isida, T.; Akiyama, T.; Nabika, K.; Sisido, K.; Kozima, S. The Formation of Tin-Nitrogen Bonds. V. The Selective 1-Substitution Reaction of Tetrazoles by the Reaction of 5-Substituted 2-(Tri-n-Butylstannyl)tetrazoles with Methyl Iodide, Methyl p-Toluenesulfonate, Dimethyl Sulfate, and Ethyl Bromoacetate. Bull. Chem. Soc. Jpn. 1973, 46, 2176–2180. 61. Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Mennucci, B.; Petersson, G. A.; Nakatsuji, H.; Caricato, M.; Li, X.; Hratchian, H. P.; Izmaylov, A. F.; Bloino, J.; Zheng, G.; Sonnenberg, J. L.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; Honda, Y.; Kitao, O.; Nakai, H.; Vreven, T.; Montgomery, Jr., J. A.; Peralta, J. E.; Ogliaro, F.; Bearpark, M.; Heyd, J. J.; Brothers, E.; Kudin, K. N.; Staroverov, V. N.; Keith, T.; Kobayashi, R.; Normand, J.; Raghavachari, K.; Rendell, A.; Burant, J. C.; Iyengar, S. S.; Tomasi, J.; Cossi, M.; Rega, N.; Millam, J. M.; Klene, M.; Knox, J. E.; Cross, J. B.; Bakken, V.; Adamo, C.; Jaramillo, J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli, C.; Ochterski, J. W.; Martin, R. L.; Morokuma, K.; Zakrzewski, V. G.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Dapprich, S.; Daniels, A. D.; Farkas, O.; Foresman, J. B.; Ortiz, J. V.; Cioslowski, J.; Fox, D. J. Gaussian 09, Revision D.01; Gaussian, Inc.: Wallingford, CT, 2013. 62. Mennen, S. M.; Alhambra, C.; Allen, C. L.; Barberis, M.; Berritt, S.; Brandt, T. A.; Campbell, A. D.; Castañon, J.; Cherney, A. H.; Christensen, M.; Damon, D. B.; de Diego, J. E.; GarciaCerrada, S.; Garcia-Losada, P.; Haro, R.; Janey, J. M.; Leitch, D. C.; Li, L.; Liu, F.; Lobben, P. C.; MacMillan, D. W. C.; Magano, J.; McInturff, E. L.; Monfette, S.; Post, R. J.; Schultz, D. M.; Sitter, B. J.; Stevens, J. M.; Strambeanu, I. I.; Twilton, J.; Wang, K.; Zajac, M. A. The Evolution of High-Throughput Experimentation in Pharmaceutical Development, and 295

Perspectives on the Future. Org. Process Res. Dev. 2019, 23, 1213–1242. DOI: 10.1021/ acs.oprd.9b00140. 63. Merrill, J. Pfizer Dashes Hope For A PCSK9 Pill. Scrip Intelligence [Online] 2016, No. 3815. https://scrip.pharmaintelligence.informa.com/-/media/Supporting-Documents/ScripIssue-PDFs/2016/scrip_3815.pdf?la=en.

296

Chapter 11

Rational Design to Large-Scale Synthesis: Development of GSK8175 for the Treatment of Hepatitis C Virus Infection Andrew J. Peat*,1 and Shiping Xie2 1Medicine Design and GlaxoSmithKline, 1250 S. Collegeville Rd.,

Collegeville, Pennsylvania 19426, United States 2API Chemistry, GlaxoSmithKline, 1250 S. Collegeville Rd.,

Collegeville, Pennsylvania 19426, United States *E-mail: [email protected].

GSK8175 is a second-generation nonnucleoside RNA-dependent RNA polymerase (NS5B) inhibitor for the treatment of hepatitis C virus (HCV) infection. The unique boron-containing pharmacophore imparts low nanomolar activity against HCV genotypes (GT) 1–6 replicons and GT 1 replicons bearing the 316N/Y mutations that impart resistance to a related clinical asset, HCV796. Importantly, the benzoxaborole analog GSK8175 demonstrated a long halflife (60–63 h) in human volunteers and a robust drop in plasma viral RNA levels in HCV infected subjects. The rational design of GSK8175 and the evolution of synthetic approaches to the complex oxaboryl heterocycle on kilogram scale are described. Access to these large quantities was enabled through several process improvements that reduced the benzofuran synthesis from 10 to 5 isolations and delivered intermediates on 100+ kg scale. Another key improvement was the identification of a unique Cu catalyst that enabled N-arylation of a sulfonamide via Chan-Lam coupling with a (methylsulfonamido)benzofuran-3-carboxylic intermediate. These improvements decreased the original 18-stage synthesis to 11 stages starting from the same β-keto ester and was used to prepare >3 kg of GSK8175 in 100 kg of key intermediates and >4 kg of API.

Discovery of Pan-GT HCV Inhibitors Bearing Boronic Acids Discovery of Boron-Containing Clinical Assets Compound 1 binds to the “palm” region of NS5B, an allosteric site within the RNA-dependent RNA-polymerase enzyme required for transcribing viral RNA (Figure 2a). This is the viral enzyme targeted by the HCV therapy sofosbuvir, a nucleoside analog that binds to the catalytic active site. An X-ray crystal structure of 1 bound in the palm-site pocket of NS5B had been published and revealed 299

a wealth of structural information to guide compound design (11). Importantly, the structure highlights the proximity of the inhibitor to serine and tyrosine residues present in the binding pocket and to the elongating RNA template strand (Figures 2b and 2c, respectively) (12). We quickly became interested in the opportunity to build interactions directly to the numerous hydroxyl groups present on the protein side chains or on the ribose sugars of viral RNA.

Figure 2. Structure of the HCV viral NS5B protein with HCV-796 (1) bound to the palm-site allosteric pocket. (a) The NS5B “fingers,” “thumb,” and “palm” and the elongating viral RNA template are denoted in variable shades. (b) Close-up of compound 1 in the palm-site pocket with the RNA removed, highlighting multiple hydroxy-bearing residues with which a boronic acid could interact (11). (c) Model showing the proximity of compound 1 in the palm-site pocket to the elongating RNA template strand, highlighting an opportunity for a boronic acid to interact with a hydroxyl group of the ribose sugar backbone (12). Nucleophilic hydroxyl groups of serine/threonine and ribose have been shown to form reversible covalent bonds with boronic acid-containing inhibitors when positioned appropriately within a binding pocket (13). The X-ray crystal structures of Velcade® (4, bortezomib), approved by the Food and Drug Administration in 2003 for the treatment of multiple myeloma, revealed a direct covalent bond between boron and a threonine hydroxyl of the β5-subunit of the 2S6 proteosome (5, Figure 3) (14). In a related fashion, the topical antifungal drug Kerydin® (6, tavaborole) has been observed to form a covalent adduct with the diol contained on a ribose sugar (7) (15). Lewis acidic boron is uniquely capable of creating such bonds with bioactive macromolecules due to its empty p-orbital. The fact that bortezomib is given systemically via parenteral administration (intravenous or subcutaneous) and successfully completed clinical trials suggests that boron-based therapeutics can achieve sufficient safety margins and risk/benefit ratios. Recognizing these facts, we speculated that boron could form the foundation of a key design element in our follow-on approach. Armed with structural insight and precedent for covalently targeting nucleophilic residues, we initiated an exploratory lead optimization effort. Our initial hypothesis was predicated on the idea that an appropriately placed boron warhead would form a covalent interaction within the binding pocket that would improve potency against the polymorphic viruses resistant to clinical asset 1. Based on the proximity of serine 407 and the RNA strand, we sought to test our hypothesis by incorporating boronic acids off the sulfonamide nitrogen of 1. Construction of the desired boronic acids was envisioned to proceed via N-alkylation of the sulfonamide of 15 (Scheme 1). Fortunately, the synthesis of the requisite intermediate had been described in reports pertaining to 1 (16). A similar reaction sequence with modest optimization afforded the desired intermediate in 10 steps from β-keto ester 8 in 12% overall yield. 300

Figure 3. Examples of marketed boron-based therapeutics that form covalent adducts with the hydroxyl of threonine (bortezomib, 4) and the diol of a ribose sugar (tavaborole, 6).

The synthesis started with a ZnCl2 catalyzed Nenitzescu reaction between ethyl 4fluorobenzoylacetate 8 and 1,4-benzoquinone to form 5-hydroxybenzofuran 9 in 39% yield (17). Alkylation of the phenol followed by nitration ortho to the ether gave 10 in 85% yield for the two steps. The isopropyl protecting group was removed with BCl3, and the phenol was converted to mesylate 11, which underwent Suzuki cross-coupling with cyclopropylboronic acid to afford 12 in high yield. Hydrogen in the presence of Pd/C cleanly reduced the nitro group and the aniline was treated with MsCl to give the bismesylated product 13. Although the monomesylate analog was desired, a mixture of mono- and bis-products resulted, thus requiring addition of excess MsCl to ensure complete conversion to 13. Saponification of the ester with KOH also cleaved one of the mesylate groups to give 14, which was treated with MeNH2 and hexafluorophosphate azabenztriazole tetramethyl uronium (HATU) to deliver 15. With multigram quantities of 15 in hand, alkylation of the sulfonamide nitrogen introduced alkene, alkyne, or aryl halide functional groups that were subsequently borylated. In this manner, a diverse set of linear and cyclic alkyl boronic acids, as well as aryl boronic acids, were rapidly prepared (16–18, Table 1) (18). We were excited to see that boronic acids installed at this position were more active than the reference compound 1 and exhibited single-digit nanomolar activity versus GT1b (Table 1) and GT1a (data not shown) replicons. Our goal however, was to improve activity versus polymorphic viruses resistant to 1; therefore, compounds were assessed against GT1b replicons bearing the 316N mutation. Unlike the wildtype replicon, the 316N assay proved to be more discriminating in terms of activity, with many analogs showing a dramatic drop in potency (18). However, boronic acids 16–18 performed well against this difficult mutation. For example, a small five-fold loss in potency was observed between wild type (WT) and 316N GT1b replicons for benzyl boronic acid analog 18, while the analogous benzyl analog lacking a boron warhead (19) showed a 25-fold drop in activity. Furthermore, replacement of the boronic acid with conventional acid isosteres such as carboxylate (20), phosphate, sulfate, or sulfonamide groups (not shown) failed to retain activity (18). The data generated in the 316N assay served to confirm our hypothesis that resistance to compound 1 could be overcome by installing a boron warhead. 301

Scheme 1. Synthesis of late-stage benzofuran intermediate 15.

The linear boronic acid 16 and cyclic oxaborolane 17 displayed good potency versus wildtype (GT1a/1b) and 316N (GT1b) replicons. Unfortunately, the in vivo pharmacokinetic (PK) profile of 16 showed high clearance in rat (intravenous Cl = 25 mL/min/kg) with the boronic acid readily oxidized to produce 1 as a metabolite (18). Oxaborolane 17 showed a significantly lower rat clearance (6.6 mL/min/kg), indicating that alkyl boronic acids of this type are stable to oxidation in vivo (18). Although an improvement in activity was noted for GT1b mutations (316N/Y/F), a 100-fold loss in potency was observed for the GT1a 316Y replicon (relative to wildtype). The poor activity was a concern as this mutation was observed in patients treated with 1 as a monotherapy. Structureactivity relationships (SAR) developed for the cyclic oxaborolane failed to improve activity versus the GT1a 316Y replicon. Therefore, our efforts focused on benzyl boronic acid 18. A strong preference for electron-withdrawing groups (EWGs) ortho to the boronic acid of 18 was shown to improve replicon potency. This effect was most pronounced for replicons bearing mutations at position 316. For example, replacing an ortho-hydrogen with fluorine (18 vs 2) resulted in a ~40-fold improvement in EC50 versus the GT1a 316Y replicon. For the first time, we had succeeded in identifying a compound with the desired antiviral profile—low nanomolar potency against replicons resistant to the HCV-796 and pan-GT activity (Figure 4) (19).

302

Table 1. HCV Replicon Data for N-Alkylated Sulfonamides 16–20

Figure 4. Clinical asset 2 (GSK5852) demonstrates low nanomolar activity (EC50) across HCV GTs 1–5. GSK5852 (2) displayed acceptable PK in all preclinical species (mouse, rat, dog) except monkey and was selected as a candidate in 2010 for clinical development to treat HCV infection (18). Unfortunately, the compound’s time in the clinic was short-lived due to a plasma half-life that was much shorter than predicted by preclinical allometric scaling (t1/2 ~ 5 h vs. 15 h, respectively) (20). As a result, the daily dose anticipated for efficacy rose significantly from 35 mg q.d. (once daily) to 420 mg b.i.d. (twice daily). Analysis of the plasma samples identified a metabolite (15) which displayed a long half-life of 45 h. It is worth noting that 15 is also a reported human metabolite of HCV-796 (1), which was associated with drug induced liver injury in a Phase 2 study. 303

Backup Strategy and Discovery of 3 (GSK8175) Bearing an Oxaboryl Heterocycle A backup program was initiated to remove the liabilities associated with 2. Based on the presence of metabolite 15 in the plasma samples from humans dosed with 2, it appeared that benzylic oxidation followed by C-N bond cleavage was the reason for rapid clearance (21). Ideas were proposed to block benzylic oxidation, which included adding substituents to the benzylic carbon and tethering to the aryl boronic acid ring (Strategy A) or the sulfonamide (Strategy B) (Figure 5). The latter approach was in part successful; however, new issues were introduced. Ultimately, the simplest strategy involving excision of the benzylic carbon was pursued (Strategy C) (22).

Figure 5. Strategic approaches A-C to reduce the propensity for benzylic oxidation associated with N-benzyl boronic acid 2. Two synthetic routes were explored to access the desired N-phenylboronic acid 21 from intermediates 15 or 22. The desired compound 21 required appendage of the phenyl ring to the sulfonamide nitrogen which was envisioned via two approaches: (1) direct N-arylation of sulfonamide intermediate 15, or (2) a two-step process involving Pd-catalyzed C-N coupling of 22 followed by mesylation of the bisarylaniline (Figure 5). To our surprise, the second approach was problematic due to the mesylation step, which we could not get to proceed beyond ~60% conversion (23). Therefore, we focused on the first approach, which had the added benefit that kilogram quantities of 15 had been stockpiled as part of the clinical campaigns for 2. Conversion of 15 to compound 21 was initially attempted via Chan-Lam coupling with aryl boronic acids (Table 2). Despite few reports of N-arylation of sulfonamides, both meta- and parabromophenyl boronic acids underwent Cu-mediated C-N bond formation with the 2°-sulfonamide of 15 to provide the desired N-phenyl derivatives 23 and 24 in low yields (39 and 26%, respectively) (24). Unfortunately, addition of a second EWG (e.g., Cl for compound 25) failed to give any isolated product. This problem was circumvented to some degree by employing more electron-rich aryl boronic acids 26 and 27. Attempts were also made to go directly to the functionalized boronic acid product (compound 28), but again, no product was observed. In general, Chan-Lam couplings with 15 were 304

slow, often requiring several days of stirring, and efforts to force the reaction with higher temperature led to loss of the mesylate group. The long reaction times and poor yields of the Cham-Lam approach prompted an exploration for alternative means of N-arylation. Nucleophilic aromatic substitution (SNAr) was well aligned with our desire to incorporate additional EWGs, ideally adjacent to the boron pharmacophore. The poor nucleophilicity of aryl sulfonamide 15 necessitated the use of highly activated electrophilic aromatic partners and in practice was limited to fluoro-nitrobenzenes (Table 3). Table 2. Substrate Scope of Chan-Lam Coupling of Aryl Boronic Acids with Intermediate 15

As expected, substrates with EWGs adjacent to the nitro facilitated fluoro displacement (29–31), which was also complimentary to the substrate scope observed for the Chan-Lam coupling (25). Interestingly, 1,3-difluoro-4-nitrobenzene 32 gave poor yield of the product, likely due to competing displacement of the undesired fluoro group. Fluoro-nitrobenzenes with electrondonating substituents were not viable partners for this transformation (e.g., 33). The nitro group served a dual role—it increased the SNAr reactivity of the fluorobenzene substrate while providing a convenient handle for elaboration to the boronic acid (Scheme 2). The nitro group of 29 was readily reduced to the corresponding aniline, then subsequently converted to the aryl bromide via a Sandmeyer reaction. Borylation of the arylbromide intermediate was achieved with bis(pinacolato)diboron (B2pin2) using Pd-catalysis. The pinacol ester was removed via oxidative cleavage with periodate to liberate the free boronic acid 34 (25). In the case of the 3-CO2Me derivative 30, an opportunity existed to introduce a benzoxaborole ring (Scheme 3). The nitro group was reduced to the aniline which was then treated with N305

chlorosuccinimide to afford the tetra-substituted benzene 35. Sandmeyer conditions converted the aniline to the arylbromide and LiBH4 reduction of the methyl ester gave the corresponding alcohol. Protection with a methoxymethyl (MOM) group gave 36 as a tan solid in 68% yield over the three steps. Borylation using our standard protocol followed by acid cleavage of the MOM group liberated the benzyl alcohol, which rapidly cyclized on the boronic ester, displacing the pinacol group, and forming the benzoxaborole analog 3. Table 3. Substrate Scope of SNAr Coupling of Fluoronitrobenzenes with Intermediate 15

Scheme 2. Conversion of nitrobenzene 29 to the corresponding phenylboronic acid 34. In removing the benzylic carbon, we recognized that the boron pharmacophore would shift significantly within the binding pocket relative to 2; therefore, it was particularly gratifying to discover that N-phenyl analog 34 retained activity against GT1a and 1b wildtype replicons (Table 4). More exciting was the low nanomolar potency against the resistant 316N/Y replicons. In general, all analogs with small EWGs adjacent to the boronic acid performed well against the 316N/Y replicons with chloro-derivative 34 being one of the most potent analogs identified in any series. The promising antiviral profile and novelty of the N-phenyl series prompted further expansion of the SAR study. The derivative 3 represented a novel structurally unique benzoxaborole 306

pharmacophore that offered diversity away from the corresponding boronic acid as well as a potential for enhanced chemical and metabolic stability (Table 4) (26). Once again, we were pleased to see a novel chemotype with potent activity in the four replicon assays (Table 4).

Scheme 3. Synthesis (Route A) of the benzoxaborole subunit of clinical asset 3. Table 4. HCV Replicon Data for N-Phenyl Boronic Acids 3 and 34 Relative to Earlier Clinical Assets 1 and 2

A major objective of our HCV discovery program was activity against the major GTs of HCV. At the time, very few compounds or mechanisms-of-action were capable of achieving this milestone. Although 1 demonstrated that pan-genotypic activity was possible for palm-site inhibitors of NS5B, it was unclear if this profile would translate to other compounds in the class given a lack of SAR 307

and poor sequence homology within the NS5B allosteric binding pockets (27). It was therefore a tremendous relief to observe that both benzyl boronic acid 2 and N-benzoxaborole 3 had EC50 < 10 nM against HCV GT 2a, 2b, 3a, 4a, and 5a replicons (Figure 6) (28). The activity at GT2a is perhaps the most impressive given this has been a historically difficult GT to inhibit as evidenced by the loss of activity noted for compound 1.

Figure 6. Comparison of antiviral potency of clinical assets 1–3 against the major HCV GTs (replicon assay). With low nanomolar potency against the major HCV GTs and the 316N/Y polymorphs resistant to 1, two major challenges had been addressed. However, a key hurdle remained. Our backup strategy was founded on the hypothesis that eliminating the benzylic oxidation associated with 2 would decrease clearance in preclinical species, prevent formation of metabolite 15, and ultimately translate into improved human PK and therapeutic dose. Once again, we were pleased to see that the PK of 3 was fantastic across all preclinical species studied (Table 5) (22). Relative to 2, compound 3 showed a significant improvement in clearance that did not exceed 13% of hepatic blood flow in any of the species tested. We were particularly pleased that the monkey clearance of 3 was >8-fold lower than 2, as post hoc analysis had highlighted that monkey was the most accurate of the preclinical species in predicting the poor human PK of 2. Importantly, there was no in vivo or in vitro evidence of metabolite 15 being formed from any of the backup compounds. Table 5. In Vivo Plasma or Blood Clearance (Cl) of 2 and 3 in Preclinical Species In Vivo Clearance [mL/min/kg] (Cl as a Percentage of Liver Blood Flow for the Species)

Compd

Mouse

Rat

Dog

Cynomolgus

2

1.4 (1%)

29 (37%)

11 (19%)

48 (~100%)

3

0.5 (0.6%)

6.9 (9%)

1.2 (2%)

5.7 (13%)

With all the backup program objectives achieved, 3 was progressed toward clinical development which necessitated access to more material. The route developed by the medicinal chemistry group (Scheme 3) was sufficient to prepare 120 g of 3 in support of 14-day rat and dog dose-range finding studies. More impressive was the fact that the 120 g was prepared within five months of 3 first being made. This shortened timeline reflects our recognition that 3 was likely to be a progressable molecule and the need to move quickly in this highly competitive landscape. To accelerate further, the team 308

decided to skip one month Good Laboratory Practice (GLP) toxicity studies and proceed directly to studies of three month duration which necessitated the scale-up to >3 kg as quickly as possible.

Process Development of Clinical Asset 3 (GSK8175) Objectives and Strategy for Efficient Synthesis of 3 (GSK8175) on Scale As further development of 3 (GSK8175) required a large supply of the drug substance to support formulation development and clinical studies, API Chemistry made a substantial effort to develop a more efficient synthesis. There were several drawbacks of the original synthesis (Route A; Schemes 1 and 3) utilized by Medicinal Chemistry. • First, the preparation of benzofuran-3-carboxamide 15 from β-keto ester 8 required ten isolations (Scheme 1). • Second, the introduction of the benzoxaborole ring from the SNAr product 30 to the API 3 was long, requiring seven isolations (stages), several of which contained process safety issues such as the chlorination with N-chlorosuccinimide (NCS) in N,Ndimethylformamide (DMF), Sandmeyer reaction with a diazonium salt, and use of highly carcinogenic MOM chloride (Scheme 3). • Third, the overall length of 18 linear stages from β-keto ester 8 to the target 3 was considered inefficient despite generally good to excellent yields for individual stages. The API Chemistry group was engaged to address these three drawbacks through the design of a convergent coupling reaction between the highly functionalized benzofuran 15 and several substituted phenyl moieties. A concise synthesis of the individual coupling partners and removal of the unsafe chemistry would also be required for a manufacturing route. The retrosynthesis for potential new routes is shown in Figure 7. The plan would require a shorter alternative synthesis of amide 15 starting from the reaction of a trihalobenzene and β-keto ester 8, or its amide analog 37.

Figure 7. Plan for a more convergent synthesis of 3 by coupling of benzofuran 15 with a phenyl moiety of variable degree of functionalization. A large number of phenyl moieties, with several examples shown in Figure 7, could potentially be used to couple with 15, ranging from a highly functionalized one with the benzoxaborole prebuilt (shown on the left on the lower scale), to a significantly less-functionalized coupling partner such as 309

methyl 3-fluoro-6-nitrobenzoate (shown on the right on this scale), which was used in the Medicinal Chemistry synthesis of 29 as shown in Scheme 2. The coupling of the phenyl moieties to the hindered methanesulfonamide 15 through the activation point (-X) is known to be a challenge, particularly when so many functionalities would have to be tolerated. Meeting this challenge would turn out to be a major endeavor for the process development. The results from the process research and development group, including significant method development and two new synthetic routes suitable for scale-up, will be described in the following section. New Synthesis of Benzofuran-3-carboxamide 15 through an Unexpected Enamine Lactone Pathway Our initial effort for an alternative synthesis of benzofuran-3-carboxamide 15 is shown in Scheme 4 (29). Coupling of commercially available ethyl 4-fluorobenzoylacetate 8 and 1,4-dibromo2-fluoro-5-nitrobenzene 38 in the presence of K2CO3 led to a 2-phenyl-3-ketoester intermediate (30). Following addition of a catalytic amount of CuI in the same pot, enol tautomer 39 underwent ring closure to give the highly functionalized ethyl 5-bromobenzofuran-3-carboxylate core 40. A Suzuki coupling with cyclopropyl boronic acid, catalyzed by palladium acetate and bis[(2diphenylphosphino)phenyl] ether (DPEPhos), provided ethyl 5-cyclopropylbenzofuran 12. The DPEPhos ligand was selected among many ligands screened because it caused the least amount of a byproduct arising from debromination. The sequence from β-keto ester 8 to benzofuran-3carboxylate 12 was only two stages compared to 6 stages shown in Scheme 1 for the original synthesis (route A). Synthesis of 12 as shown has been scaled up in 5000 L reactors several times. Specifically, the SNAr/CuI catalyzed preparation of 40 from 8 was run in 86–172 kg scale with yield ranging from 73–77%, slightly below the demonstration runs due to the higher difficulty to purge residual Cu and need of reslurry to remove Cu. The Suzuki reaction from 40 to 12 was run six times from 53–110 kg scale in yield range of 85–92%, generally higher than in the demonstration runs. The overall yield from 8 to 12 was about 70%. Analogous chemistry for the benzofuran formation with keto amide analog 37, with the structure shown in Figure 7 and enzymatically prepared from keto ester 8 and methylamine, was not as clean as with 8 for unknown reasons. If successful, it would have further reduced the linear synthetic sequence an additional stage.

Scheme 4. Expedient synthesis of benzofuran-3-carboxylate 12 and attempted conversion to amide 41 with methylamine. 310

For the preparation of the methyl amide 41, ethyl ester 12 was treated with 33 wt % MeNH2 in EtOH at room temperature. This mixing led to a clear solution that over 40 min gave rise to a substantial amount of crystalline solids. The purported methyl amide 41, highly crystalline and with the expected molecular weight and number of protons by 1H nuclear magnetic resonance (NMR) spectroscopy, was isolated in 76% yield by simple filtration of the reaction mixture. The crystalline product was then carried forward for the rest of the synthetic sequence to 2, the previous lead compound in the project. However, the sequence of chemistry led to a product which had slightly shorter retention time in high-performance liquid chromatography (HPLC) compared to the target product despite possessing the same molecular weight as 2 (29). Somehow we had produced an isomer of 2. With that discrepancy, 12, prepared from the new synthesis, was converted to amide 15 in four steps following the known sequence shown in Scheme 1. This confirmed the structure assignment for 12 from the new route and provided a 6-stage synthesis to 15, compared to the original 10 stages from the same keto ester 8. While the scaleup of the chemistry shown in Scheme 4 was being carried out, further development of the chemistry continued. Ester 12 was exposed to 33% MeNH2/EtOH as before, but the reaction was inadvertently allowed to continue for three weeks whereupon crystals were generated as expected. To our surprise, the isolated crystals from the three-week reaction bore no resemblance to the one isolated from the previous reaction with MeNH2/EtOH conducted over 40 min when analyzed by 1H NMR spectroscopy. The newly isolated product turned out to be the previously targeted amide 41, confirmed by a two-step process: hydrolysis of the ethyl ester 12 followed by amide formation with MeNH2, HATU, and Et3N. The unknown product from the initial exposure of 12 to MeNH2/EtOH was analyzed by NMR correlation experiments nuclear overhauser effect spectroscopy (NOESY) and heteronuclear multiple bond correlation (HMBC), which pointed to an enamine lactone structure represented by 42 (Scheme 5). The NH proton at 9.69 ppm for 42 relative to 8.51 ppm for 41 was presumably due to a potential NH hydrogen bonding with the lactone carbonyl oxygen. The (Z)-configuration of the enamine explained the shielding effect from the 4-fluorophenyl ring on one of the benzofuran aryl protons (shifted from 8.32 ppm for 41 to 5.09 ppm for 42) and one of the cyclopropyl CH2 group (shifted from 0.73 ppm to 0.07 ppm). This structure was subsequently confirmed by single crystal Xray analysis (Figure 8).

Scheme 5. Synthesis of amide 41 from ester 12 via rearrangement of enamine lactone 42.

311

It was clear by then that the enamine lactone 42 first forms upon treatment of ester 12 with MeNH2/EtOH at room temperature. If the crystalline solid was not isolated from MeNH2/EtOH, it rearranged to amide 41 over three weeks at room temperature or in one to two days with heating. In scaleup to 205 g, 12 was directly converted to 41 of 99.3% purity as a crystalline solid in 86% yield by direct filtration of the reaction mixture after three days at 50–55°C. The isolated amide 41 was stable but, if submitted to fresh MeNH2/EtOH and heated, it partially reversed to enamine lactone 42. No further reversal from 42 to ester 12 was observed. The equilibrium of enamine lactone 42 and amide 41 in MeNH2/EtOH was apparently biased towards 41, likely due to lower solubiity of 41. Although 42 could be isolated before it went to 41, we chose to let the reaction go longer and isolate 41 in up to 86% yield in scaleup. On occasions when enamine lactone 42 was indeed isolated, it was then treated with MeNH2/EtOH at 60°C for 18 h to give amide 41 in high yield. A reaction pathway from ester 12 to amide 41 via enamine lactone 42 was proposed through a series of transient intermediates and facilitated by H bonding throughout (29).

Figure 8. A view of two molecules of 42 from the crystal structure, highlighting the hydrogen bonding (dashed lines) and the orientation of the 4-fluorophenyl ring, shielding methylene protons of the cyclopropyl group and a methine proton of the fused ring system. Reproduced with permission from reference (29). Copyright 2018 American Chemical Society. Second Generation Synthesis of GSK8175 with 13 Linear Stages (Route B) The synthesis of 3 (GSK8175) from Discovery shown in Schemes 1 and 3 was considered the first generation synthesis or Route A. Our second generation or Route B synthesis, combining the new synthesis of benzofuran-3-caroboxamide 15 (vide supra), and the original coupling with methyl 3-fluoro-6-nitrobenzoate using an SNAr reaction described in Scheme 3, is shown in Scheme 6. This route was 13 stages long compared to the original 18 stages from the same keto ester 8. This chemistry was used to deliver over 4 kg of GSK8175 to support Phase 1 clinical trials. Despite high efficiency in the synthesis of the benzofuran core as shown in the upper box, the new route offered only moderately improved convergence, while still retaining some unsafe stages following the SNAr coupling (stage 6). A more dramatic change in the second half of the synthesis, shown in the lower box, would be required to provide a suitably efficient synthesis for the final manufacturing route. 312

Scheme 6. Route B synthesis of GSK8175: 13 stages.

Efforts towards More Convergent Synthesis of GSK8175 (Route C) The overall strategy laid out in Figure 7 for the synthesis of 3 (GSK8175) required the synthesis of the highly functionalized benzoxaborole 48, activated and suitable for coupling with the mesylamine 15. Our most ambitious effort in this regard is shown in Scheme 7 (28).

Scheme 7. Synthesis of a fully functionalized benzoxaborole 48 and coupling with amide 15 under typical Chan-Lam conditions. Bromo aniline 43 was converted to dibromide 44 followed by bis-Miyaura borylation to give bisBpin intermediate 45. A three-step one-pot process provided 48: benzylic bromination of 45 to give bromomethyl derivative 46, hydrolysis by KOH to obtain benzyl alcohol 47, and mild acidification 313

with 1 N HCl. The benzylic OH of 47 displaced the pinacol at the o-Bpin to give the benzoxaborole 48. This set the stage for the coupling of 48 with amide 15 under typical Chan-Lam conditions (31). Although we observed the initial formation of target product 3, the benzoxaborole decomposed under the reaction conditions due to Cu catalyzed deborylation. The unstability of 3 was further confirmed by subjecting pure 3 to the Chan-Lam reaction conditions, resulting in a rapid fromation of 49, the des-boron analog of 3. In an effort to salvage the approach of using the prebuilt boron moiety 48 for the synthsis of 3, we selectively hydrolyzed the top Bpin of bis-Bpin compound 45 to form the mixed Bpin/boronic acid 50, as shown in Scheme 8 (28). The selectivity of the hydrolysis was achieved because of the significantly higher steric hinderance for the bottom Bpin. Crude 50 was used without purification to couple with either benzofuran carboxylic acid 15 or its N-methyl amide 51 in the presence of Cu(OAc)2 (31). To our surprise, both reactions were relatively clean and achieved ~50% conversion to products 52 and 53 respectively on the first attempt as determined by HPLC and liquid chromatography–mass spectrometry analysis of the reaction mixtures. The boronic acid moiety was significantly more reactive and also less sterically hindered than the bottom Bpin. A sample of product 53 was isolated. However, the functionalization of the methyl group, following the sequence shown in Scheme 7 to build the oxaborolane ring, failed because of several undesired side reactions for the rest of the structure of 53 under the bromination conditions with N-bromosuccinimide (NBS) and azobisisobutyronitrile (AIBN).

Scheme 8. Preparation of mixed Bpin-boronic acid 50 and Chan-Lam coupling with benzofurans 15 and 51.

C-H Activation for Highly Functionalized Phenyl Boronic Ester 55 Suitable for Chan-Lam Reactions In our continued search for an activated and highly functionalized phenyl moiety to couple with a benzofuran moiety, no fewer than 15 phenyl moieties were synthesized and tested for the coupling with Pd or Cu catalysis as well as the more conventional SNAr reactions. In the end, the C-H activation of commercially available 54 under typical Hartwig conditions (32), shown in Scheme 9, proved to be extremely successful. The Ir-catalyzed reaction was not only very selective with only the most open position activated, but also used very low loading of the catalyst with 0.2 mol % of the Ir 314

dimer. The product 55 was crystalline and, when subjected to typical Chan-Lam reaction conditions with Cu(OAc)2 (31), showed sufficient potential for further optimization. From gram scale to kilogram scale, there was no scalability issue for the C-H activation to prepare dihalide 55. Initially, the C-H activation was catalyzed by methoxy(cyclooctadiene) is iridium (I) dimer but was subsequently replaced with the much less expensive chloro Ir-catalyst (Scheme 9). The product was extracted into methyl tert-butyl ether (MTBE) and crystallized out readily upon solvent exchange to MeCN to provide 55 of >99% purity at kilogram scale, greatly facilitating the reaction screening for the subsequent Chan-Lam coupling.

Scheme 9. Efficient C-H activation to prepare boronate 55.

Chan-Lam Reaction of Boronate 55: Search for the Magic Catalyst With boronic ester 55 now readily available, screening for coupling with either the benzofuran3-carboxamide 15 or the carboxylic acid 51 was carried out (33). It quickly became clear that the acid 51 was a better substrate as it was more reactive than 15 and any residual 51 could be purged by basic aqueous extractive workup. There was also a potential to isolate the product as a salt of the carboxylic acid, an attractive purification scenario for improving the synthesis. The results from the initial screening are summarized in Scheme 10 for the four components of the Chan-Lam reaction: Cu source, base, solvent, and oxidant. The criteria for screening took into consideration the conversion (ratio of product 56 to 51) and impurity profile (des-borylation, oxidative degradation and homo-coupling). The best combination is highlighted in Scheme 10 in bold, namely, [Cu(MeCN)4]PF6, Et3N, MeCN and air as the oxidant (33). This combination was selected for further optimization with kinetics, process safety, isolation of product, and other development criteria considered. [Cu(MeCN)4]OTf and N-methylpiperidine also performed well. An important aspect of the reaction shown in Scheme 10 was that the boronate was used directly, as opposed to the more typical and generally considered more active boronic acid for the Chan-Lam reactions. In our tests, the boronic acid analog as well as the Molander aryltrifluoroborate salt derived from 55 were not efficient for the coupling with 51, due to significantly more deborylation, oxidation to phenol, and homocoupling to the bi-phenyl impurity. Subsequently, [Cu(MeCN)4]PF6 and N-methylpiperidine in MeCN/air was developed as a methodology for room temperature Chan-Lam N-arylation of N-aryl sulfonamides, with a very broad substrate scope through a productive industry-academia collaboration (34). 315

Scheme 10. Screening for four components of Chan-Lam coupling of 51 and 55 with the best conditions highlighted in bold. Optimization of Chan-Lam Reaction of 51 and 55 Considering Kinetics and Process Safety The relatively low volume of solvents (6–10 volumes relative to weight of 51) in the initial screening gave a heterogenous reaction. While there was positive order of reaction rate to 51 and 55 producing 56, saturation kinetics in Cu were observed. In the optimization phase, there was a 13-fold rate increase when a clear solution was achieved with 20 volumes of MeCN (33). From 0–0.6 equiv of Cu, nearly first order behavior was observed with the new reaction concentration. There was no substantial rate increase from 0.6–1.2 equiv, as shown by the intial rate line with round dots and the solution yield line of 56 with the triangles in Figure 9.

Figure 9. Initial rate (dotted with circles) and the solution yield of 56 (dotted with triangles) of reaction of 51 and 55 versus equivalent of [Cu(MeCN)4]PF6. Reproduced by permission from reference (33). Copyright 2019 American Chemical Society. In a manufacturing setting, use of air in the presence of flammable solvents such as MeCN is considered a safety risk. As a result, further kinetics studies with 5–8% oxygen with the balance being nitrogen were carried out. A modest decrease of the initial reaction rate was observed, as shown by 316

the line dotted with the smaller squares in Figure 10 with 5% oxygen as the oxidant. The kinetics were very similar to that using 20% oxygen, suggesting that lower oxygen levels would be acceptable.

Figure 10. Initial rate of reaction of 51 and 55 vs. time with 5% O2. Reproduced by permission from reference (33). Copyright 2019 American Chemical Society. Further optimization was carried out (33). The resultant ~2:1 optimum ratio of carboxylic acid 51 to [Cu(MeCN)4]PF6 led to the hypothesis of a Cu(II) carboxylate complex with the carboxylic group of 51 shown in Figure 11. A green homogeneous reaction solution gradually formed as oxygen was introduced and Cu(II) was regenerated in the presence of oxygen. To probe the possibility of forming a 2:1 complex between the carboxylate of 51 and Cu(II), we measured the solubility of 51 in a 4:1 mixture of MeCN and Et3N with increasing amounts of [Cu(MeCN)4]PF6 (Figure 11, right). As clearly shown by the data, the solubility of 51 increases linearly with the amount of Cu present with a slope of 2, consistent with the proposed formulation of a 2:1 complex.

Figure 11. Left: hypothesized complex of Cu(II) and 51 in the presence of MeCN/Et3N. Right: correlation between the solubility of 51 and the amount of added [Cu(MeCN)4]PF6, consistent with a 2:1 stoichiometry. The optimized Chan-Lam reactions routinely gave conversions higher than 90% in the presence of air, with a slightly lower conversion obtained when 5–8% of oxygen was utilized. It should be noted that the Discovery team explored the Chan-Lam reactions for preparation of various analogs of 3 for SAR studies. The results, as shown in Table 2 in the Discovery section of the current chapter, showed very limited substrate scope with yields ranging 0–56% for the 6 substrates listed. They contrasted with the isolated yields of 50–95% for a wide range of substrates with [Cu(MeCN)4]PF6, underlining the importance of the catalyst selection and optimization of 317

reaction conditions in process development (33, 34). Additionally, [Cu(MeCN)4]PF6 was suitable for both Bpin boronates and boronic acids. Third Generation Synthesis of GSK8175 with 11 Linear Stages (Route C) The final synthesis of target compound 3, our third generation or Route C, is shown in Scheme 11, with the end game post the Chan-Lam coupling reaction shown in the second half of the scheme (33).

Scheme 11. Eleven-Stage synthesis (Route C) of 3 (GSK8175) from keto ester 8. The benzofuran core 40 derived from keto ester 8 was converted to 57 after the Suzuki reaction and hydrogenolysis. Mesylation and ester hydrolysis led to the benzofuran carboxylic acid 51, the preferred substrate for the key Chan-Lam reaction. The Chan-Lam reaction, stage 6, was carried out on a 1-L scale prior to transfer to manufacturing for planned delivery of 100 kg of 3 (GSK8175). Several features are worth noting post Chan-Lam reaction of 51 and 55. The product was isolated as the crystalline salt 58 by treating the solution of the crude carboxylic acid 56 with methylamine in tetrahydrofuran (THF). The salt was converted to the methyl amide 59 with TBTU, along with a small excess of methylamine HCl salt to achieve full conversion to 59. Reduction of the ester followed by protection of the crude alcohol gave 318

tetrahydropyran (THP) ether 60. The use of DHP for the protection was a huge improvement for process safety compared to the use of MOM-Cl in the first-generation synthesis (Route A). Miyaura borylation from bromide 60 to the aryl boronate 61 was hugely improved using cyclohexyldiphenylphospine as the ligand and K2CO3 as the base in the presence of a catalytic amount of pivalic acid. Significant screening and kinetics studies were carried out to reach the optimized conditions shown (33). Finally, hydrolysis of the THP ether under mild HCl conditions was followed by instantaneous cyclization to complete the synthesis of 3. The convergent synthesis shown has only 11 linear stages starting from the same β-keto ester 8 as the two longer previous routes. Route C was selected as the manufacturing route to supply future drug substance for the project.

Conclusion The rational design strategy based on reversible covalent interactions between a boron warhead and the viral enzyme NS5B afforded two clinical assets 2 (GSK5852) and 3 (GSK8175). The boronic acid functionality was critical for achieving low nanomolar activity versus HCV GT1–6 viruses and GT1 replicons bearing a 316N/Y mutation. The benzylic oxidation of 2 resulted in high clearance and short half-life (~5 h) in human subjects. Removal of the benzylic carbon via direct sulfonamideN-phenyl substitution (i.e. 3) improved the half-life to 60–63 h in human volunteers. Extensive process research and development reduced the original 18-stage synthesis of 3 (Route A) to 13 stages (Route B) and 11 stages (Route C) starting from the same β-keto ester. Route B has a much shorter and elegant synthesis of the benzofuran carboxamide compared to Route A, while Route C improves over Route B with the [Cu(MeCN)4]PF6 enabled Chan-Lam coupling for higher convergence. Over 4 kg of the API 3 was synthesized from Route B. While the key carboxylic acid intermediate 51 was synthesized in ~100 kg scale several times for an overall delivery of >300 kg of 51 for use in Route C, the Chan-Lam reaction and the subsequent stages have not been scaled up in the plant as of this writing. For the Medicinal Chemistry, the rapid process improvements and use of late stage intermediates from the scale-up of 2 provided access to over 3 kg of API 3 within 10 months of first synthesis in the medicinal chemistry laboratories. With a robust route and access to large quantities of API, dosing in humans (FTIH) was achieved in