Applications in forensic proteomics: protein identification and profiling 9780841236431, 0841236437

Citation preview

1339

PROTEINS IN BIOFORENSICS AND BIOSECURITY Proteomics is a mature research tool in the life sciences, and it can be a powerful addition to the forensic science toolbox. This work presents several areas in which proteomics was used to answer forensic questions. Illustrating current applications of proteomic methods, this work introduces opportunities for proteomics to answer compelling questions in forensic science and biosecurity. These “case studies” will be valuable to both practicing forensic scientists and researchers developing proteomics methods.

PUBLISHED BY THE

B I O L O G I C A L

VOLUME 1339

APPLICATIONS IN FORENSIC PROTEOMICS PROTEIN IDENTIFICATION AND PROFILING

ACS SYMPOSIUM SERIES

ACS SYMPOSIUM SERIES

APPLICATIONS IN FORENSIC PROTEOMICS

PROTEIN IDENTIFICATION AND PROFILING

American Chemical Society SPONSORED BY THE

ACS Division of Analytical Chemistry

MERKLEY

E. D. MERKLEY

Applications in Forensic Proteomics: Protein Identification and Profiling

ACS SYMPOSIUM SERIES 1339

Applications in Forensic Proteomics: Protein Identification and Profiling Eric D. Merkley, Editor Pacific Northwest National Laboratory Richland, Washington

Sponsored by the ACS Division of Analytical Chemistry

American Chemical Society, Washington, DC

Library of Congress Cataloging-in-Publication Data Library of Congress Cataloging in Publication Control Number: 2019046052

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 1. Introduction to Forensic Proteomics .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  Eric D. Merkley

1

2. A Proteomics Tutorial .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  Eric D. Merkley, Brooke L. D. Kaiser, and Helen Kreuzer

9

3. Proteomic Sample Preparation Techniques: Toward Forensic Proteomic Applications .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  29 Carrie Nicora, Marina Gritsenko, Anna Lipton, Karen L. Wahl, and Kristin E. Burnum-Johnson 4. NextGen Serology: Leveraging Mass Spectrometry for Protein-Based Human Body Fluid Identification .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  47 Heather E. McKiernan, Catherine O. Brown, Luciano Chaves Arantes, Phillip B. Danielson, and Kevin M. Legg 5. Informatics Approaches to Forensic Body Fluid Identification by Proteomic Mass Spectrometry .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  81 Wenke Liu, Erin Butler, Heyi Yang, David Fenyö, and Donald Siegel 6. Fingermarks as a New Proteomic Specimen: State of the Art and Perspective of In Situ Proteomics.  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  91 Simona Francese and Cristina Russo 7. Human Identification Using Genetically Variant Peptides in Biological Forensic Evidence .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  107 Fanny Chu, Katelyn E. Mason, Deon S. Anex, Phillip H. Paul, and Bradley R. Hart 8. Proteomics in the Analysis of Forensic, Archaeological, and Paleontological Bone .  . . . .  125 Michael Buckley 9. Proteomics for Microbial Forensics .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  143 Eric D. Merkley 10. ISO 17025 Accreditation of Method-Based Mass Spectrometry for Bioforensic Analyses .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  161 Stephen R. Cendrowski and Alaine M. Garrett 11. Unambiguous Identification of Ricin and Abrin with Advanced Mass Spectrometric Assays .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  175 Suzanne R. Kalb and François Becher

vii

12. Challenges in the Development of Reference Materials for Protein Toxins .  . . . . . . . . . . . . . . . . . . .  185 R. Zeleny, A. Rummel, D. Jansson, and B. G. Dorner 13. The Statistical Defensibility of Forensic Proteomics .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  203 Kristin H. Jarman and Eric D. Merkley Editor’s Biography .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  229 Indexes Author Index .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  233 Subject Index .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  235

viii

Chapter 1

Introduction to Forensic Proteomics Eric D. Merkley* Chemical and Biological Signature Sciences Group, Pacific Northwest National Laboratory, Richland, Washington 99352, United States *E-mail: [email protected].

Proteomics is formally defined as the study of the proteome—the set of all proteins expressed in a cell, tissue, or organism—and its changes with changing environmental conditions. In practice, the term proteomics has come mean the technologies and techniques used to identify and quantitate (in relative or absolute terms) a very large number of proteins (hundreds to thousands) in a single analysis. This chapter introduces the ways in which proteomics analysis can be applied to biological forensics.

The Success of Proteomics in the Biological Sciences Mass spectrometry-based proteomics is an extremely successful and rapidly growing research tool in the biological sciences. Proteins are the molecular machines that carry out the functions of life: enzymes that catalyze chemical reactions, receptors that allow cells to sense their environment, and scaffolds that give cells and tissues their structure. Cataloging the types and numbers of proteins in a system—its proteome—is thus a window into the biological state, responses, and functions of that system. Proteomics approaches have helped to uncover disease mechanisms, elucidate biological regulation programs, characterize microbial communities, and more. Proteomics methods are potentially applicable any time that information about the proteins contained in a sample is sought. Proteomics has been very successful in the life sciences. Since 2010, there have been over 3000 proteomics publications per year on average (from a search of the Clarivate Analytics Web of Science database, executed on May 13, 2019, using the search terms “proteom*” and “mass spectrom*”). These publications span a range of basic science topics, including cell and molecular biology, microbiology, immunology, cancer and other human disease research, and plant sciences. Proteomics is also a rapidly developing field. A significant number of those publications also deal with advances in mass spectrometric instrumentation and improved statistical and bioinformatics algorithms. (I will use the term proteomics to refer to the collection of technologies, methods and approaches that center around the use of mass spectrometry to identify and measure the abundance of proteins. I do not include other uses of the term, such as to refer to protein microarrays, or even as a synonym for protein science in general.) The power of proteomics to characterize biological systems comes from three things: © 2019 American Chemical Society

(1) The combination of wide applicability and exquisite specificity. Many methods for the detection of biomolecules rely on the specificity of biomolecular interactions. Immunological methods use antibodies that react with a specific protein. Polymerase chain reaction (PCR) requires specific primers to amplify a DNA region of interest. These methods can provide high specificity (that is, high confidence that the detected molecules are the intended ones), but often new antibodies or primers must be developed and tested for each new application or assay. In principle, mass spectrometry can elucidate the sequences of peptides from any protein source, without expensive and time-consuming testing and development of antibodies or primers, thus combining breadth and specificity. In this respect, proteomics is similar to next-generation genomic sequencing. (2) The ability to provide a global picture of the biochemical and physiological state of a system. As an analytical method, proteomics is highly multiplexed, meaning that hundreds to thousands of analytes can be measured at once. This ability is critical to gaining a holistic understanding of biological systems, but applies equally well to understanding the protein content and potential organisms of origin of an unknown case sample. (3) Relevance to biological function. Finally, since proteins are the functional molecules of many biological processes, proteomics provides a way to look at what a biological system or sample is actually doing. DNA provides the instructions, but it is largely proteins that carry out the instructions. The different information provided by protein and DNA analysis can be likened to a Shakespearean play. If DNA provides the script, proteins are the set, the stage crew, and the actors. Reading King Lear can be entertaining, inspiring, and thoughtprovoking. It will certainly leave you in no doubt as to which play you are reading. But it is not the same experience as seeing a well-executed live production of the play. In a similar way, DNA analysis identifies the organisms in question and reveals their potential, but proteomics measures how that genetic potential is realized under given conditions.

Why Analyze Proteins? In biological forensics, DNA is king. A host of studies have demonstrated the effectiveness of DNA typing for human identification (1, 2). Genotyping (e.g., by PCR-based methods) (3) and genomics (sequencing (4, 5)) approaches for microbial species and strain characterization are also well established. The power of next-generation sequencing promises to bring a new level of specificity in addition to inferred phenotypes such as hair, skin, and eye color, and perhaps even facial features for humans (6), and virulence factors for microbes. For these purposes, DNA techniques will always be a crucial component of the biological forensics toolbox. So why bother with protein analysis at all? There are in fact a number of common circumstances in forensics and related fields in which protein analysis in general, and proteomics in particular, can provide information that DNA analysis cannot. Some of these examples come from central applications in forensics as traditionally defined, that is, criminal-justice forensics (7, 8). Other come from related areas that also deal with the analysis and identification of unknown samples, such as biodefense and homeland security (9), paleontology (10), cultural heritage/archaeology (11, 12), sports anti-doping (13), species identification in food quality assurance (14), and detection of bacterial antibiotic resistance phenotypes in clinical medicine (15). I will use the term forensic proteomics in this larger sense throughout this chapter. Some samples that do not contain DNA, or that contain DNA that cannot be analyzed (because of degradation or interferents), nevertheless contain protein sufficient for informative analysis, for example proteinaceous binding materials in paint or objects of art or cultural heritage, or purified proteins. In some cases the protein itself is the biomolecule of interest, such as protein toxins, protein 2

pharmaceuticals, or protein hormones used in sports doping. In other cases, DNA analysis might not answer all forensically relevant questions. DNA methods excel in answering the question “Which individual contributed this sample?” but cannot by themselves answer the question “Which tissue or biological fluid did this sample come from?” Similarly, DNA sequencing can easily answer the question “Which organism does this sample come from?”, but protein expression data can additionally illuminate the conditions under which a sample was produced, pointing to methods of production. And of course, confirmatory and orthogonal evidence always has value in forensics. Although the discipline of forensic proteomics is still in its infancy, there are numerous published studies which show the utility of proteomics in many areas, including protein toxin detection and identification (16), human individualization (17), tissue and body fluid identification (18), detection of banned performance-enhancing proteinaceous substances in sports (19, 20), and discrimination of laboratory-adapted and wild strains of potentially pathogenic bacteria (21, 22).

Forensic Proteomics Application Areas Fundamentals As the title of this book suggests, the aim is to give a flavor of the various areas of forensics, broadly defined, in which proteomics is poised to make an important impact. The book begins with a tutorial on proteomics methods (Chapter 2) which focuses on two key enabling technologies of proteomics: mass spectrometry and bioinformatics. Chapter 2 is designed to give students and beginners a sufficient foundation to understand subsequent chapters. In particular, the differences between targeted and untargeted proteomics and the rationale of peptide identification by database searching are emphasized. Chapter 3 covers sample preparation for proteomics, focusing on bottomup (peptide-based) proteomics. Many potential problems with proteomics data arise from problems with the sample or the sample preparation methods, so this chapter covers a critical area. Two of the coauthors of this chapter (Carrie Nicora and Marina Gritsenko of Pacific Northwest National Laboratory) are proteomics sample preparation specialists who have prepared thousands of samples of diverse types for proteomics analysis. Human Forensic Proteomics The next four chapters cover topics in human forensic proteomics, that is, proteomics of humanderived forensic samples. Chapters 4 and 5 deal with forensic serology, the identification of bodily fluids. This application is a leading example of a case where DNA analysis cannot provide the desired information: the identity, not of the individual, but of the fluid--blood, semen, saliva, etc. Current methods include a series of presumptive tests for various biofluids markers, often requiring that a sample be tested multiple times, and, because these markers are not always perfectly specific, sometimes still not offering a conclusive answer. Chapter 4, contributed by Philip Danielson of Denver University and coworkers, describes a universal forensic serology assay based on targeted proteomics. In Chapter 5, David Fenyő (a luminary in the world of computational proteomics) and Donald Siegel (Chief Scientist of the Office of the Chief Medical Examiner, New York City Department of Health) delve into detail in distinguishing two closely related biofluids with significant forensic implications, namely peripheral blood and menstrual blood. Chapter 6 deals with the intersection of proteomics and another forensic science area: fingerprints. In Chapter 6, Simona Francese and coworkers review the biochemical and other information that is available from mass spectrometric analysis of fingerprints. Chapter 7 covers proteomics methods to identify individuals 3

from proteins containing single amino acid polymorphisms. Many human proteins, including those found in hair (17) and bone (23) are present in human populations as multiple alleles that differ in their amino acid sequences. Dr. Bradley Hart, former director of the Forensic Science Center at Lawrence Livermore National Laboratory, and coworkers, will describe the analysis of genetically variable proteins and their use in identifying individuals and ethnic groups. Nontraditional Forensics As mentioned earlier, a number of disciplines share goals and challenges with forensics narrowly defined. Such challenges include the characterization of uncharacterized, unknown, contaminated or degraded samples, and the detection of particular proteins of interest. The detection of banned substances in athletic competitions (“anti-doping”), led by the World Anti-Doping Agency (WADA), makes extensive use of mass spectrometric assays to detect performance-enhancing proteins and peptides. This effort takes place in a highly structured, highly regulated, and highconsequence environment and is therefore a good model for how proteomics might be deployed in criminal forensics. Another related area is the use of proteomics to characterize archaeological, paleontological, or cultural heritage samples. Proteins can be extracted from fossilized bones up to 10,000 years old, and proteomics analyses of protein sequence can shed light on phylogenetic relationships of extinct animals. In Chapter 8, Michael Buckley of the University of Manchester and coworkers describe forensic, archaeological and paleontological uses of bone proteomics. Proteomics can also be used to identify the species from which archaeological leather or skin objects are derived, as well as various residues in archaeological samples. Proteomics analysis has been used in a number of cases to determine the nature and origin of biological materials used in paints from various objects of art. Proteomics for Microbial Forensics A significant body of work exists on genomic and genetic characterization of pathogenic microbes, including those of national security concern. Less well known is the use of proteomics to characterize these pathogens. A number of studies have shown that proteomics can help provide answers to questions that have been intractable by genomics, such as the identification of the host in which a virion particle was grown (24), the growth phase at isolation of Clostridium botulinum cultures (25), the composition of the host media in which bacteria were cultured (26), and in distinguishing wild bacterial isolates from laboratory-adapted descendents (21, 22). In Chapter 9, I discuss these topics, along with the sizable body of work on using proteomics to identify microbial samples, and by extension, to characterize the taxonomy of unknown samples in general. Protein Toxins A number of toxins that are important in both forensics and national security are themselves proteins. The most famous of these is undoubtedly ricin, a toxin found in the seeds of the castor plant (27). Numerous cases of ricin being prepared for or used in crimes or attempted crimes have been documented (28). Ricin is specifically prohibited by the Chemical Weapons Convention, and was historically investigated as a chemical weapon by more than one nation. Other protein toxins include the related abrin (another plant toxin, found in the seeds of Abrus precatorius or jequirity pea), and bacterial toxins such as Clostridium botulinum neurotoxins and the staphylococcal enterotoxins. The application of proteomics to protein toxins is an obvious fit, because the dangerous (or illegal) agents 4

are themselves proteins, and detection of the corresponding nucleic acids may or may not indicate a crime or a threat to public health. Mass spectrometric analysis can identify and quantitate protein toxins with a high degree of specificity, providing the information needed by public health or law enforcement authorities. In Chapter 10, Alaine Garrett and Stephen Cendrowski of the National Bioforensics Analysis Center and coworkers describe the creation and validation of an untargeted proteomics method for ricin. In addition to illustrating the utility of mass spectrometry approaches for protein toxins, this work shows the value of untargeted approaches and provides an example of the type of rigorous validation that forensics assays, including proteomics assays, must undergo. Chapter 11 describes the combination of immunoaffinity purification of protein toxins with targeted mass spectrometry assays. In this chapter, Francois Becher of the French Atomic Energy Commission and Suzy Kalb of the Centers for Disease Control and Prevention summarize the challenges and successes of such methods. Although creating known authentic standards is rare in basic-biology proteomics studies, authentic standards are critical for forensics. Chapter 12 deals with the challenges associated with creating certified standard reference materials for regulated protein toxins in the context of the European Union’s EquaTox program (http://www.equatox.org/index.html). Dr. Brigitte Dorner of the Robert Koch Institute, the EquaTox coordinator, led this team of authors. Statistical Concerns Any book on scientific methods for forensics would be incomplete without a discussion of statistical issues. Since the Daubert ruling in 1993, scientific evidence has come under increasing scrutiny, as evidenced by a number of high-profile government reports on the state of forensic science (29, 30). In the final chapter (Chapter 13), statistician Kristin Jarman and I lay out the elements of a defensible forensics method. There are two important aspects to a legally defensible forensics science method, reliability and relevance. Reliability involves ensuring the scientific validity and appropriate application of scientific methods—the type of considerations that are traditionally associated with method development, verification, and validation. Relevance, often less appreciated, deals with the meaningfulness of the results compared in a particular context, such as background levels or frequencies. An analytical method for the detection of toxic industrial chemicals might give the same reading at an airport and at a chemical factory, but they could mean very different things: a terrorist attack or accidental release at the airport, and normal background levels at the factory. Forensic science methods development and validation need to consider both reliability and relevance. Applying this framework to the development of proteomics for forensics application will be critical for the future adoption of proteomics methods. Community-wide standards for the application of proteomics methods, especially the bioinformatics analysis of proteomics data, will also be crucial for the success of forensic proteomics as a discipline. This chapter also lays out our recommendations for analyzing and reporting targeted and untargeted proteomics data.

The Future In recent years, the applicability of proteomics to important forensic questions has been increasingly recognized. This is true both for the strict definition of forensics—applied to questions of crime and criminal justice—but also in areas that have scientifically related problems, such as national security and biodefense, paleontology, archaeology, and cultural history. In all these areas, proteomics can answer questions that DNA analysis cannot. As the utility of proteomics in forensics and related areas becomes increasingly apparent, it is critical for forensic scientists, funding agencies, 5

and policy makers to understand the role proteomics can play. This book illustrates the scope and current state of forensic applications of proteomics, highlights the ways in which proteomics technologies fill critical needs in the fields of forensic science and biosecurity, and calls attention to outstanding research questions in forensic proteomics. It is my hope that this volume will also serve to build a community of researchers within the disparate areas that make up the specialty of forensic proteomics. By including tutorial chapters on mass spectrometry and proteomics sample preparation, this book will also be useful for students of forensic science, both as an introduction to diverse areas of biological forensics and as an example of the issues that must be considered for any analytical method in forensic science. Proteomics is an inherently interdisciplinary science, bringing in aspects of physics and physical chemistry (peptide fragmentation behavior), electrical engineering (mass spectrometer design), analytical chemistry (measurement fundamentals and variability), and bioinformatics and statistics (data analysis). Last but not least, a thorough understanding of the cell biology, biochemistry, and physiology of the system under study is necessary to interpret and apply the results. In practice, no individual proteomics scientist masters all of those areas—collaborations are essential. The growing area of forensic proteomics will be no different. Forensic proteomics layers on top of all these areas the requirements for legal admissibility, operational feasibility, and regulatory controls. Collaborations between proteomics scientists and the forensic science community, especially operational forensics laboratories, will be crucial to the continued development of the field. I hope that this volume will also facilitate such collaborations.

Acknowledgments The author would like to thank Chris Ehrhardt for helpful discussions.

References 1. 2. 3.

4.

5.

6.

Butler, J. M. The Future of Forensic DNA Analysis. Philos. Trans. R. Soc. Lond. B Biol. Sci. 2015, 370, 20140252. van Oorschot, R. A. H.; Ballantyne, K.; MItchell, R. J. Forensic Trace DNA: A Review. Invest. Genet. 2010, 1, 1–17. Skowronski, E.; Lipkin, W. I. , Moleucular Microbial Surveillance and Discovery Bioforensics. In Microbial Forensics; Budowle, B., Schutzer, S. E., Breeze, R. G., Keim, P. S., Morse, S. A. , Eds.; Elsevier, 2011; pp 173−185. Sjödin, A.; Broman, T.; Melefors, Ö.; Andersson, G.; Rasmusson, B.; Knutsson, R.; Forsman, M. The Need for High-Quality Whole-Genome Sequence Databases in Microbial Forensics. Biosecurity and Bioterrorism: Biodefense Strategy, Practice, and Science 2013, 11, S78–S86. Keim, P.; Grunow, R.; Vipond, R.; Grass, G.; Hoffmaster, A.; Birdsell, D. N.; Klee, S. R.; Pullan, S.; Antwerpen, M.; Bayer, B. N.; Latham, J.; Wiggins, K.; Hepp, C.; Pearson, T.; Brooks, T.; Sahl, J.; Wagner, D. M. Whole Genome Analysis of Injectional Anthrax Identifies Two Disease Clusters Spanning More Than 13 Years. EBioMedicine 2015, 2, 1613–1618. Lippert, C.; Sabatini, R.; Maher, M. C.; Kang, E. Y.; Lee, S.; Arikan, O.; Harley, A.; Bernal, A.; Garst, P.; Lavrenko, V.; Yocum, K.; Wong, T.; Zhu, M.; Yang, W.-Y.; Chang, C.; Lu, T.; Lee, C. W. H.; Hicks, B.; Ramakrishnan, S.; Tang, H.; Xie, C.; Piper, J.; Brewerton, S.; Turpaz, Y.; Telenti, A.; Roby, R. K.; Och, F. J.; Venter, J. C. Identification of Individuals by

6

7.

8.

9. 10. 11.

12.

13.

14.

15.

16. 17.

18.

19.

Trait Prediction Using Whole-Genome Sequencing Data. Proc. Natl. Acad. Sci. U. S. A. 2017, 114, 10166–10171. Legg, K. M.; Powell, R.; Reisdorph, N.; Reisdorph, R.; Danielson, P. B. Discovery of Highly Specific Protein Markers for the Identification of Biological Stains. Electrophoresis 2014, 35, 3069–3078. Deininger, L.; Patel, E.; Clench, M. R.; Sears, V.; Sammon, C.; Francese, S. Proteomics Goes Forensic: Detection and Mapping of Blood Signatures in Fingermarks. Proteomics 2016, 16, 1707–1717. Duracova, M.; Klimentova, J.; Fucikova, A.; Dresler, J. Proteomic Methods of Detection and Quantification of Protein Toxins. Toxins (Basel) 2018, 10, 99. Buckley, M. Ancient Collagen Reveals Evolutionary History of the Endemic South American “Ungulates”. Proc. R. Soc. B-Biol. Sci. 2015, 282, 9. Brandt, L. Ø.; Schmidt, A. L.; Mannering, U.; Sarret, M.; Kelstrup, C. D.; Olsen, J. V.; Cappellini, E. Species Identification of Archaeological Skin Objects from Danish Bogs: Comparison between Mass Spectrometry-Based Peptide Sequencing and Microscopy-Based Methods. PLoS One 2014, 9, e106875. Yang, Y.; Shevchenko, A.; Knaust, A.; Abuduresule, I.; Li, W.; Hu, X.; Wang, C.; Shevchenko, A. Proteomics Evidence for Kefir Dairy in Early Bronze Age China. J. Archaeol. Sci. 2014, 45, 178–186. van den Broek, I.; Blokland, M.; Nessen, M. A.; Sterk, S. Current Trends in Mass Spectrometry of Peptides and Proteins: Application to Veterinary and Sports-Doping Control. Mass Spectrom. Rev. 2015, 34, 571–594. Ohana, D.; Dalebout, H.; Marissen, R. J.; Wulff, T.; Bergquist, J.; Deelder, A. M.; Palmblad, M. Identification of Meat Products by Shotgun Spectral Matching. Food Chem. 2016, 203, 28–34. Spinler, J. K.; Haidacher, S. J.; Hoch, K. M.; Luna, R. A.; Haag, A. M. Discerning StrainSpecific Β-Lactam Drug Resistance by Clonal Isolates of Multi-Drug Resistant Pseudomonas Aeruginosa Using Selected Reaction Monitoring. Int. J. Mass Spectrom. 2019, 438, 36–43. Kalb, S. R.; Barr, J. R. Mass Spectrometric Identification and Differentiation of Botulinum Neurotoxins through Toxin Proteomics. Rev. Anal. Chem. 2013, 32, 189–196. Parker, G. J.; Leppert, T.; Anex, D. S.; Hilmer, J. K.; Matsunami, N.; Baird, L.; Stevens, J.; Parsawar, K.; Durbin-Johnson, B. P.; Rocke, D. M.; Nelson, C.; Fairbanks, D. J.; Wilson, A. S.; Rice, R. H.; Woodward, S. R.; Bothner, B.; Hart, B. R.; Leppert, M. Demonstration of Protein-Based Human Identification Using the Hair Shaft Proteome. PLoS One 2016, 11, e0160653. Legg, K. M.; Powell, R.; Reisdorph, N.; Reisdorph, R.; Danielson, P. B. Verification of Protein Biomarker Specificity for the Identification of Biological Stains by Quadrupole Time-of-Flight Mass Spectrometry. Electrophoresis 2017, 38, 833–845. Thevis, M.; Loo, J. A.; Loo, R. R. O.; Schänzer, W. Recommended Criteria for the Mass Spectrometric Identification of Target Peptides and Proteins (150 kDa), very small (1 Million Years Old) Fossil Material: Pitfalls, Possibilities and Grand Challenges. Anal. Chem. 2014, 86 (14), 6731–6740. Hofman, L. F. Human Saliva as a Diagnostic Specimen. J. Nutr. 2001, 131 (5), 1621S–5S. Pieper-Bigelow, C.; Strocchi, A.; Levitt, M. D. Where Does Serum Amylase Come from and Where Does It Go? Gastroenterol. Clin. North Am. 1990, 19 (4), 793–810. Levitt, M. D.; Ellis, C. J.; Engel, R. R. Isoelectric Focusing Studies of Human Serum and Tissue Isoamylases. J. Lab. Clin. Med. 1977, 90, 141–152. Hochmeister, M. N.; Budowle, B.; Rudin, O.; Gehrig, C.; Borer, U.; Thali, M.; Dirnhofer, R. Evaluation of Prostate-Specific Antigen (PSA) Membrane Test Assays for the Forensic Identification of Seminal Fluid. J. Forensic Sci. 1999, 44 (5), 1057–1060. Pentyala, S.; Whyard, T.; Pentyala, S.; Muller, J.; Pfail, J.; Parmar, S.; Helguero, C. G.; Khan, S. Prostate Cancer Markers: An Update. Biomed. Rep. 2016, 4 (3), 263–268. Mashkoor, F. C.; Al-Asadi, J. N.; Al-Naama, L. M. Serum Level of Prostate-Specific Antigen (PSA) in Women with Breast Cancer. Cancer Epidemiol. 2013, 37 (5), 613–618. Lange, V.; Picotti, P.; Domon, B.; Aebersold, R. Selected Reaction Monitoring for Quantitative Proteomics: A Tutorial. Mol. Syst. Biol. 2008, 4, 222. Kuhn, E.; Carr, S. A. Multiplexed Immunoaffinity Enrichment of Peptides with Anti-Peptide Antibodies and Quantification by Stable Isotope Dilution Multiple Reaction Monitoring Mass Spectrometry. Methods Mol. Biol. 2016, 1410, 135–167. Newbrun, E. Observations on the Amylase Content and Flow Rate of Human Saliva Following Gustatory Stimulation. J. Dent. Res. 1962, 41 (2), 459–465. Yang, H.; Zhou, B.; Prinz, M.; Siegel, D. Proteomic Analysis of Menstrual Blood. Mol. Cell. Proteomics 2012, 11 (10), 1024–1035. Friedman, J. H. Greedy Function Approximation: A Gradient Boosting Machine. Ann. Stat. 2001, 29 (5), 1189–1232. Hastie, T.; Tibshirani, R.; Friedman, J. The Elements of Statistical Learning: Data Mining, Inference, and Prediction; Springer Science & Business Media, 2013. 89

23. Xu, R.; Wunsch, D. Clustering; John Wiley & Sons, 2008. 24. Jolliffe, I. T. Principal Component Analysis; Springer Science & Business Media, 2013. 25. Gordon, A. D.; Breiman, L.; Friedman, J. H.; Olshen, R. A.; Stone, C. J. Classification and Regression Trees. Biometrics 1984, 874. https://doi.org/10.2307/2530946 26. Guyon, I.; Weston, J.; Barnhill, S.; Vapnik, V. Gene Selection for Cancer Classification Using Support Vector Machines. Mach. Learn. 2002, 46 (1), 389–422. 27. Goodfellow, I.; Bengio, Y.; Courville, A. Deep Learning; MIT Press, 2016. 28. Braga-Neto, U. M.; Dougherty, E. R. Is Cross-Validation Valid for Small-Sample Microarray Classification? Bioinformatics 2004, 20 (3), 374–380. 29. Kushner, I. K.; Clair, G.; Purvine, S. O.; Lee, J.-Y.; Adkins, J. N.; Payne, S. H. Individual Variability of Protein Expression in Human Tissues. J. Proteome Res. 2018, 17 (11), 3914–3922. 30. Ma, S.; Ren, J.; Fenyö, D. Breast Cancer Prognostics Using Multi-Omics Data. AMIA Jt. Summits. Transl. Sci. Proc. 2016, 2016, 52–59. 31. Liu, W.; Payne, S. H.; Ma, S.; Fenyö, D. Extracting Pathway-Level Signatures from Proteogenomic Data in Breast Cancer Using Independent Component Analysis. Mol. Cell. Proteomics 2019, 18 (8) (suppl 1), S169–S182.

90

Chapter 6

Fingermarks as a New Proteomic Specimen: State of the Art and Perspective of In Situ Proteomics Simona Francese* and Cristina Russo Centre for Mass Spectrometry Imaging, Biomolecular Science Research Centre, Howard Street, S1 1WB Sheffield, United Kingdom *E-mail: [email protected].

For at least the first three decades since its advent, proteomics has exclusively largely belonged to a clinical, diagnostic, or fundamental biology context. However, the range and the significance of information that proteomes can disclose have led this discipline to be also applied to forensics, ranging from human identification from hair samples, identification of bodily fluids, and microbial forensics to doping investigations. Fingermarks are a relatively new specimen for proteomic studies with any form of proteomic investigation only appearing in 2012 with the analysis of intact peptides and small proteins in situ published by the research group at Sheffield Hallam University. It was not until 2015 that further developments allowed bottom-up proteomics to be also applied directly in situ. While in situ proteomics of fingermarks has many advantages, encompassing simplified sample preparation protocols, speed and the opportunity to perform molecular imaging analyses, this area remains under-investigated. This is probably due to the unique challenges of working with fingermark specimens. The relatively low protein content and the predominantly eccrine origin of fingermarks have been shown to severely impact protein detection at least when the “intact” protein approach is used both in full scan and using a top down approach. In this chapter, advantages, application, challenges and perspective of in situ fingermark proteomics are discussed and compared with classic approaches.

Introduction Historically, forensic analytical science has always been the playfield for small molecule analysis. The most commonly analyzed forensically relevant molecules, such as explosives, gunshot residues, poisons, medications/drugs of abuse and associated metabolites, make for a huge range of analytes of interest. The detection of the aforementioned analytes contributes to significant intelligence informing both investigations and judicial debates.

© 2019 American Chemical Society

However, forensic science has kept up with modern times and technologies; it has adopted more and more sophisticated analytical approaches to increase the quality and quantity of the forensic information, enabling much bigger molecules (in the order of kDa), such as DNA and proteins, to be investigated. While genomics and the completion of the human genome project have allowed DNA and mRNA typing for suspect identification, proteomics has enabled the detection and identification of proteins which has boosted biomarker discovery and promoted tailored medicine in clinical settings. Detection of proteins, encompassing knowledge on the four levels of structure organization as well as interaction with other molecules, has found application in a number of forensic biology areas covering the analysis of human samples, microbial forensics, evaluation of food species using libraries of unidentified spectra, archaeology/cultural heritage (1). With respect to the analysis of human samples, proteomics of tissues, bones, hair and biological fluids have contributed to methods informing on the of post-mortem interval of skeletal remains (2), human identification, determination of facts and possible responsibilities around an individual’s death (3), type of crime(s) being committed (4) (for example, presence of seminal fluid on a murder victim points to a contextual rape/sexual assault crime), in anti-doping tests (5, 6) and finally in an approach which holds the promise for the provision of an individual’s genetic information from genetically variant peptides (7). The vast majority of such proteomic approaches have involved the use of protein extraction, insolution proteolysis and LC-MS/MS based analysis. This is understandable as, to date, LC-MS/MS approaches still provide the highest number of protein identifications confirmed by the use of tandem mass spectrometry. Only a few approaches have been published reporting the use of in situ bottom-up proteomics, that is, enzymatic proteolysis undertaken directly on the specimen of interest with no prior sample homogenization/protein extraction. Not only does in situ proteomics, offer a simpler and much less time consuming sample preparation, but it also provides the possibility to apply mass spectrometric techniques that can operate in imaging mode to yield the spatial distribution of peptides and proteins of interest. The spatial information offers the link between protein presence and function as well as possibly another informative dimension to the type of forensic intelligence that can be gained, as it is discussed later in this chapter. One of the closest examples to in situ forensic proteomics, concerns the application to biofluid analysis. Martin et al (8) undertook dried blood spot analysis using Liquid Extraction Surface Analysis (LESA) coupled with automated sample preparation and LC-MS/MS. These authors of this chapter define this method as “hybrid” because, while the blood spot does not undergo homogenization and subsequent extraction in solution, proteins are “extracted in situ” through liquid microjunction (LESA) and transferred robotically into an Eppendorf for subsequent proteolysis prior to LC-MS/MS analysis using an autosampler. While the authors had in mind potential clinical applications in screening programs for LESA, refinement and transferability to a forensic proteomic context did not require a huge leap as demonstrated by the work of Bailey et al (9) discussed in this chapter. There is a huge amount of literature covering on tissue (in situ) proteomics, since 1997 when Caprioli et al (10) demonstrated the ability to both detect and visualize peptides and small proteins directly in human tissues (a small aggregate of human buccal mucosa (cheek) cells) by Matrix Assisted Laser Desorption Ionization Mass Spectrometry Imaging (MALDI MSI). The first on tissue bottom-up proteomic approach was published by Chaurand et al in 1999, who added trypsin as 1 µL droplets on tissue blots. This approach was followed by automatically “printing” trypsin on biological 92

tissues as published by Shimma et al in 2006 (11). To the best of our knowledge, the first group reporting spray coating of trypsin for in situ digestion was that led by Clench (12). This method, published in 2009, enabled in situ digestion as well as the opportunity to map peptides at a very high spatial resolution when compared to samples digested by spotting trypsin. Nonetheless, at large, it is technically since 1999 that in situ bottom-up proteomics methods have been developed to digest samples followed by MALDI MSI. The employment of ambient techniques such as Desorption Electrospray Ionization Mass Spectrometry (DESI MS) (13) and LESA (14, 15) for in situ bottom-up proteomics is dated instead to 2008 (although for LESA, proteolytic digestion is performed after the in situ micro-extraction). While MALDI requires sample preparation and largely operates in vacuum, DESI requires none of the above, thus speeding up analysis and avoiding sample degradation. Furthermore, while MALDI mainly generates mono-charged species, DESI, like Electrospray Ionization, yields multiply charged species, thus greatly facilitating fragmentation in MS/MS analyses aiming at protein identification. LESA exhibits advantages over MALDI too. In addition to the multiply charged ion formation, the decoupling of the surface sampling and ionization stages enables future opportunities for superior quantification. However, presently both techniques suffer lower spatial resolution for imaging applications (50-100 microns for DESI and 1000 microns for LESA compared to 10-100 for MALDI) as well as having a lower dynamic and upper m/z exploitable range. Although the vast majority of the tissue proteomic studies undertaken through MALDI, DESI or LESA, are intended for the research to be applied in a clinical diagnostic context, we have for a long time anticipated translation and implementation of these methods in a forensic context to detect and localize the distribution of story-teller protein biomarkers. A notable example is the work of Li et al (16) to determine post-mortem intervals from liver tissues. Nonetheless, among all human biological specimens, it appears that fingermarks have been one of the most explored in the context of forensic proteomics, although these methods are younger than a decade. Here protocols are still being reported in a rough 50/50 split between in situ and non in situ proteolysis. In the next section, the few in situ methods reported in the literature will be reviewed and their merits limits and challenges compared and contrasted with extraction and insolution proteolysis.

Fingermark Proteomics In the context of forensics, the interest in fingermarks as evidence has increased in the last decade (the term “fingermarks” is used here to refer to ridge impressions left on a surface unintentionally, as opposed to fingerprints intended as control prints of the type taken at police stations or airports). This is because the analytical community has experienced a paradigm shift with respect to what use can be made of a fingermark, by acting upon the notion that fingermarks are not just an ordered and unique pattern of lines linking directly to someone’s identity (17); in our opinion, they would be better described as “a molecular pattern of lines”. In other words, fingermarks are the “sexier” version of sweat which is transferred on a surface upon fingertip contact, thus forming a molecular impression of the ridges. Fingermarks are indeed the transfer of endogenous, semi-endogenous and exogenous substances, with the third category representing those molecular species that are external to the body and that have come in contact with fingertips through pollution or contact (contaminants) (18). Among the variety of biological molecules and electrolytes contained in sweat, proteins are present in the region of 15-25 mg/dL (19), and it their abundance is relatively low in a fingermark according 93

to recent estimates ranging between 0.2 to 51.0 μg (20). These species belong to “eccrine sweat”, which is the type of sweat secreted from eccrine glands through sweat pores. As such, proteins exist in an aqueous environment composed of 98% water and 2% of inorganic species such as electrolytes as well as organic molecules such as amino acids and urea. Depending on concentration and on type of mass spectrometric technique, electrolytes generally act as suppressor of the mass spectrometric signal. In other words, peptides and protein ion signals may be masked by the presence of electrolytes. This has been shown by Ferguson et al (21) and Francese (22) who showed that no MALDI MS peptide or protein signal was observed from an eccrine fingermark unless the mark was washed using a salt such as ammonium acetate. For the reader’s benefit, an eccrine mark is obtained through thoroughly washing hands with a 50/50 ethanol/deionized water solution, followed by drying time and placement of the hand in a plastic bag for 15 min to induce sweat. At the end of the 15 min the mark is produced by making contact with a deposition surface Polypeptides were also detected in fingermarks by Infrared Spectroscopy (23). However, despite the prior appearance of these data in the literature, the very nature of IR does not allow determination of the peptide sequence, only the inference that polypeptides are present through their characteristic bands (absorption bands associated with the chemical groups present). Other groups determined and identify the presence of small and higher molecular weight proteins in collected sweat through various techniques ranging from SDS PAGE/Western Blot, immunoistochemistry, Surface Enhanced Mass Spectrometry and LC-MS/MS (24–30). In 2012, Ferguson et al (21) were the first to report on direct detection of peptides and small proteins from eccrine, ungroomed and groomed fingermarks (the definition of which is reported in Ferguson et al (21)) without any prior washing procedure as shown in Figure 1. Some individuals have been observed to yield scarce ion population, possibly due to presence of ion suppression contaminants and/or to intrinsically being “poor donors”.

Figure 1. MALDI MS Profiling of an unwashed male natural mark showing (intact) peptides and proteins.

94

The work by Ferguson et al followed that of other groups putatively detecting, in many cases, the presence of the same species detected and identified in sweat using different analytical techniques (24–30). These species were employed to build a statistical model using Partial Least Squares Discriminant analysis (PLS-DA) enabling the determination of sex from a fingerprint with the 85% of accuracy of prediction. The opportunity to determine sex from a single mark helps to narrow down the range of suspects with minimal destructiveness (the marks are spotted with matrix in three discreet locations only). Since then, these studies have progressed to include a much larger cohort of donors (172) and investigation of natural marks (the only realistic type found at crime scenes, obtained with no prior washing of fingertips) and natural marks which are preliminarily enhanced with a common CSI powder (“indestructible white” which is TiO2-based) prior to MALDI MS analysis. Data processing is approaching completion and will inform evaluation of the robustness and feasibility of implementation in casework (Heaton et al, manuscript in preparation). Another example of direct detection of peptides in fingermarks was offered by Bailey et al in 2016 using LESA (9). Here the authors showed detection of multi-charged species belonging to peptides and proteins. Although from this work it is not clear how many species have been detected, the authors highlight the possibility of detecting species of as high molecular weight as 8600 Da through its +6 charge state ion at m/z 1434.68. This shows potential to identify small proteins by performing MS/MS experiments as well, as good dynamic detection range, although not as good as that offered by MALDI which was able to detect peptides and small proteins species from m/z 2000 to 14000 (21). Although the authors are correct that introducing a chromatographic step between LESA and MS analysis of fingerprints, following sample digestion, will yield a higher number of proteins to be detected and identified at a single fingerprint location, this method further distances itself from the simplicity of in situ protocols as well as making analyses more time consuming. In any case, approach selection will always come down to a trade between comprehensiveness of the information and speed of analysis; different methods will serve different purposes and a balanced approach would be that identifying what is a reasonable and acceptable compromise in the domain of forensic science and at which point additional information may become expendable. Considering the MALDI approach, while peptides and proteins could be detected in situ in fingermarks without any extraction as LESA entails, their visualization on the fingermark ridges has never been accomplished using this or any other mass spectrometry imaging technique. Francese’s group tested a variety of different sample preparation methods, acquisition conditions and instrumentation without success (22). It had become clear that none of the modifications to the relevant method variables could compensate for the low sensitivity of detection of these species (Figure 2). The only peptide/protein images obtained show a speckled distribution (Figure 2C), which reflects the secretion of these species from the sweat pores. Further attempts to rub fingertips against each other before deposition of the fingerprint in order to spread these species uniformly and improve the ridge pattern continuity within the molecular images was not successful; it is speculated that the low amounts of these species, furthered lowered by spreading them across the fingertip, as well as the notoriously lower sensitivity of the detection with the increase of the molecular weight, are the barrier to overcome to obtain acceptable fingerprint images in this mass range. Additionally, the generally lower signal intensity derived from spraying the matrix, as opposed to spotting it, is a known occurrence as Figure 2 A-B shows. It is possible that an alternative method or device for matrix deposition may perform better for imaging purposes, (this study used spray coating with the SunCollect pneumatic sprayer, Sunchrom, Germany) or may “extract” the analyte from the surface more efficiently and generate a higher 95

number of analyte-matrix co-crystals thus improving the resulting ion intensity. It is also possible that recent advancement in mass spectrometry instrumentation with reportedly higher sensitivity are the key to protein mapping in fingermarks. In 2015, Patel et al (31) addressed the issue of lower sensitivity related to the higher molecular weight detection range, by applying, in a pioneering approach, in situ bottom-up proteomics. Patel et al (31), tested a range of detergents (varying concentrations too) to be employed in the proteolysis solution to improve peptide yield (range and intensity) when performing in situ proteolysis. As a model, ungroomed fingermarks were still employed as they exhibit the lowest level of electrolytes and lipids (known ion suppressors in MALDI analyses of peptides), although it is also expected that the amount of peptides and proteins, by the very nature of ungroomed fingermarks, is lower.

Figure 2. MALDI MS Profiling and Imaging of peptides/small proteins in a latent fingermark. Half (Panel A) of a split mark was prepared using the dried-droplet method spotting 5 mg/mL CHCA in 25:25:50 acetonitrile/ethanol/0.5% TFA. Other fingermark half (Panel B) was prepared using the same matrix solution but sprayed using the SunCollect autospraying system (Sunchrom, Germany) showing a rather decreased peptide/protein ion signal intensity. Panel (C) shows the image of a latent mark prepared using the dry-wet method (32) by dusting with CHCA and spraying a 70:30 ACN/ 0.5% TFA solution using a SunCollect autosprayer and acquired on a more sensitive instrument. The speckled image of a peptide at m/ z 3369 is shown. Adapted and reproduced with permission from reference (22) Copyright (2016) Springer Nature. The different combinations of detergents/detergent concentrations were also tested in combination with spotting versus spraying the proteolytic enzyme (trypsin). All in situ proteolysis protocols successfully yielded peptides. It must be said that these kinds of studies require an objective 96

method of evaluation in order to establish which detergent(s)/concentration are most effective and under which method of enzyme deposition. Therefore, ion signals were evaluated only if their m/z fell within the fractional range 0.4-0.8, which is the range that the analytical community agrees the peptides m/z to fall within. Furthermore, in order to identify significant differences in the efficiency of the detergents investigated, StatsDirect statistical software (ver. 2.7.8) was used. Data were shown to be nonparametric via the Shapiro Wilk test; therefore, a Kruskall-Wallis with Conover-Inman post hoc analysis test was implemented (P≤0.05). According to the evaluation criteria selected (reproducibility, range of detectable m/z, ion intensity and number of yielded peptides), MEGA-8 in 2% concentration, appeared to be the best performing detergent. Despite the fact that many proteins were detected in both instances, statistically significant differences in ion intensity and the number of peptides (as well as different m/z of the detected peptides) were observed, depending on whether the enzyme was sprayed or spotted. Interestingly, ion intensity was higher in enzyme spraying than enzyme spotting experiments This is a different result than that typically observed for both: (1) what is conventionally observed for intact larger peptides and small proteins in fingermarks (see Figure 2 A-B), and (2) digested protein analysis in biological tissue specimens (when the same concentration of trypsin is used). With respect to instance (2), it was speculated that the lower abundance of proteins in fingermarks as opposed to tissue specimens makes the trypsin: protein ratio more favorable/efficient than that observed in tissues. This circumstance would suggest the potential to resolve the low intensity issue observed within the intact peptide/protein direct detection when the matrix is sprayed. Interestingly, spraying experiments yielded the presence of more peptides with smaller m/z and from lower abundance proteins than spotting experiments. This is an encouraging occurrence for both the possibility to perform more successful MS/MS experiments for protein identification and to have higher chances to detect disease biomarkers, which are typically present in low abundance and at the lower end of the large dynamic protein concentration range in protein-rich biofluids/tissues. The latter opportunity is now being explored in our laboratories in the context of non-invasive cancer detection in fingertip smears. As spraying the enzyme yields lower m/z peptides, belonging to least abundant proteins, the keratin abundance issue, obvious in fingermarks might be bypassed. Oonk et al report cytokeratins to make up for the 35% of a fingermark proteome (20), preventing the detection of other potentially more biologically interesting proteins. Figure 3 shows the extent to which keratins dominate the mass spectrum of a fingermark proteome digest, in an enzyme spotting experiment. Of course, one obvious way to resolve “the keratins issue” is by adopting non in situ proteomics methods. Oonk et al have recently published a method to swab extract, digest and analyse the fingermark proteome by LC-MS/MS (20) with the aim to find possible biomarkers of aging. As expected, a high number of proteins, fifty-three, was detected and identified. Among these, keratins, were of course also identified (types 1, 2, 5, 6, 9, 10, 14 and 17). However, the separation step prevented or minimized ion suppression phenomena from these highly abundant proteins, allowing detection of several proteins not otherwise seen without the chromatographic separation. While a very efficient system in terms of protein yield and identification, sample preparation and analytical complexity is much higher than in situ methods and five fingerprints per donor had to be pooled for a single donor analysis which appears to be unrealistic in a forensic scenario. Rapigest SF is another interesting detergent that is being used more and more frequently by the proteomic community; already Patel et al (31) showed that its use in 0.1% concentration, within in situ proteomics of fingermarks, produced a similar number of peptides as that generated by MEGA97

8 in spraying experiments, (albeit of lower intensity, except in the region 1000-1200 Da). In these experiments, Rapigest SF was tested at the concentration recommended by the manufacturer. Given its performance, there is scope in testing different concentrations to maximize both number of peptides and their ion signal intensities. Further proteolysis method development and improvement could also well further contribute to successful peptide mapping experiments which are in progress in our laboratories.

Figure 3. MALDI MS spectrum of in situ proteolysis of a fingermark. Trypsin was spotted on top the fingermark which was then incubated at 50°C for 2 hr in a K2SO4 saturated atmosphere. The spectrum shows a large abundance of putatively assigned keratin ion signals with high mass accuracy (only the most abundant have been labeled). While we believe that reconstructing a molecular image of a fingerprint through mapping proteolytic peptides is ultimately feasible, this would only be a technical demonstration. In our view, there is no real need for imaging peptides in latent marks, unless: (1) peptide species of different origin become forensically relevant to infer intelligence about a suspect’s prior movements, lifestyle etc., or (2) the demonstration of the presence of these species, specifically on the fingermark ridge detail, becomes important from an associative evidence point of view. On this particular point, of course, one should not forget consideration of activity levels, whereby the sole association of certain molecules with fingermark ridges may lead to erroneous conclusions, when gathering intelligence around offender’s actions prior to or during committing the crime. An interesting system which could potentially be developed further to enable an alternative method for in situ proteolysis is provided by the “lab on plate” approach. This approach employs the Vmh2 hydrophobin to preliminarily coat the MALDI target plate. Vmh2 belongs to the class I hydrophobins (amphiphatic self-assembling proteins) and it has been proven to strongly bind proteins, including trypsin, in their active form, (33). Initially reported by Longobardi et al (34) as an extremely quick desalting method for the rapid proteolysis of protein standards and in mixture, the lab on plate approach was subsequently developed to digest blood in as little as five minutes versus the most optimized in-solution proteolysis method needing 1 hour for efficient digestion (35). While the 98

conventional procedure is to spot a microliter of the solution to digest onto the Vmh2-immobilized trypsin wells, in our laboratory we are investigating a number of approaches to test the opportunity to contact-digest specimens such as fingermarks; on such initial example is provided by the spectrum obtained through lightly pressing the fingertip on the lab on plate. Figure 4 shows a MALDI MS peptide spectrum obtained with only 5 minute digestion (contact) time (Figure 4A) versus a blank run after leaving an ammonium bicarbonate solution for 5 minutes (Figure 4B) prior to washings. Within these experiments it was possible to observe the presence of peptides that were absent in the blank. While very far from a method that could be operationally deployable in a forensic scenario, successful demonstration of contact digestion would provide encouraging indicators for subsequent developments of the lab on plate approach having the remarkable advantage of very short proteolysis time and increased signal-to-noise ratio in MALDI MS.

Figure 4. MALDI MS spectra of a fingermark digest obtained by contacting the fingertip with Vmh2 immobilized trypsin for just 5 minutes (A) and of a blank (B) containing ammonium bicarbonate only. Ion signals only present in the fingermark digest are labeled. On one occasion, an ion signal that could be mistakenly thought to be present in both spectra has been labeled in both (A) and (B) to show the different m/z.

In Situ Proteomics and Peptide Imaging in Blood Marks The previous section has discussed the detection of endogenous fingermark proteins and peptides. However the most significant example of the importance of mapping peptides on the ridge pattern does not concern endogenous peptides. Since 2014, method development to detect blood protein signatures from stains and blood fingermarks has been pursued (35, 36, 37). The rationale behind this area of research stemmed from the observation that CSI and crime lab blood 99

enhancement techniques (BET) are only presumptive and may give rise to false positives. Depending on the technique used, horseradish, leather, plant extracts, egg yolk, bleach and even other biological fluids, such as saliva, sweat and semen may be mistaken for blood with these tests. Therefore while the current BET remain an important first screening test, a confirmatory test would be of fundamental importance to either avoid miscarriage of justice and/or better inform the investigations. Additionally, while DNA would greatly help to gather intelligence, even on the provenance of the biofluid (human or animal), DNA evidence is less stable than proteins, it may be sometimes present in hard-to-resolve mixtures, or not recoverable. Even when available, DNA alone still would not indicate the type of biofluid involved. mRNA could be the key to the latter issue but the technique is not yet consolidated.

Figure 5. MALDI MS Imaging analysis of a blood fingermark. The mark was split in quarters and spray coat using different trypsin concentrations prior to matrix application. a. blood mark; b. blood mark quarters after in situ proteolysis and matrix spray coat (1 quarter is missing as at the highest trypsin concentration, the high viscosity determined a capillary blockage); c-d. blood protein mapping through peptide derived peptides from Complement C3, α- Hemoglobin, hemopexin and serotransferrin. Adapted and reproduced with permission from (37) Copyright (2016) John Wiley and Sons. In the first ever publication of its kind, Bradshaw et al successfully detected and mapped both heme and intact hemoglobin (Hb) subunits directly on the ridge pattern of even pre-enhanced 100

blood marks (36). Further research lead to bottom-up proteomics method development enabling the recovery of additional blood protein signatures, initially, in extracted stains and (pre-enhanced) swabbed palm prints (35). This work showed improved reliability and robustness of the method as it enables the detection of blood in specimens as old as 9 years and thus opening up a new avenue for the investigation of cold cases. Additionally, Patel et al. demonstrated the possibility to differentiate blood from mixed provenance (equine and human) supporting the intact protein work carried out by Bradshaw et al (36). Although Patel et al. used MALDI MS and MS/MS in their work, thus saving some time by avoiding LC-MS/MS analysis, the method was based on in-solution proteolysis. It was not long before method refinement resulted in the work of Deininger et al. (37) in 2016 whereby blood protein signatures including those of hameoglobin, serotransferrin, complement C3 and hemopexin were visualized directly on the ridge pattern through in situ bottom-up proteomics (Figure 5). This was certainly a milestone holding the promise of a link between the biometric information (fingermark ridge pattern) and the circumstances of the crime (bloodshed), and especially useful in those cases where CSI BET may give rise to false positive or even false negatives if the sensitivity is insufficient. The latter circumstance has been recently reported by Francese (38) showing that the third fingermark of a serial depletion from a bloody fingertip was not enhanced by ninhydrin (which reacts with amine groups in the blood cells/proteins and amino acids in plasma), though blood was clearly there and present on the ridges as revealed by the application MALDI MSI. Following the work of Deininger et al (37), Kamanna et al (39) confirmed the possibility of mapping blood on the ridge detail through an in situ bottom-up proteomic approach using different MALDI MS instrumentation. In contrast to the results of Deininger et al (37), only Hb derived peptides were used to determine the presence of blood and on a few occasions their identity was determined through in situ MS/MS. In particular, Kamanna et al. exploited the detection of mainly Hb proteotypic peptides belonging to human blood and of those belonging to a number of marsupials typically inhabiting Australia, to reconstruct molecular images of the blood contaminated fingermark. Although Kamanna et al did not manage to differentiate blood provenance using intact hemoglobin analysis, (previously successfully shown by Bradshaw et al (36)), these authors showed that the in situ bottom-up approach, focused on detection and mapping of Hb peptides, may indicate a future for blood provenance determination directly in fingermarks, subjected to validation which is being undertaken in our laboratory. The same in situ bottom-up proteomic approach, followed by mass spectrometry imaging could in principle be used for fingermarks contaminated with other biofluids in order to provide the link between the biometric information and the circumstances of the crime as informed by the presence of the specific biofluid retrieved. Presently this work has only been shown by Oonk et al by extracting the contaminated marks (thus being totally destructive of the fingermark), followed by in-solution digestion and LC-MS/MS analysis to prove donor contact with urine, saliva and vaginal fluids (20). Differently from in situ proteomics of latent fingermarks for molecular reconstruction of the ridge pattern, the actual operational application of in situ methods for blood marks may be more problematic. Presumed blood marks are not usually lifted from surfaces but rather only photographed for evidence purposes. This immediately raises issues with respect to the surface of deposition which would need to be directly analyzed; in fact, if surfaces are curved, while the enzyme application could be performed with alternative devices other than the SunCollect used in our laboratory, the shape and size of the surface where the mark is present is still a limiting factor for 101

all the enzyme depositors on the market. In those cases where blood marks are/can be lifted, it is usually gelatin lifters that are employed. While some gelatin lifters work better than others (40), and it is debatable whether they can work for old and pre-enhanced blood marks, gelatin lifters present an issue for mass spectrometers operating in vacuum. For example, classic MALDI instrumentation, operating in vacuum and allowing visualization of the peptides generated from in situ proteolysis, will not actually permit analysis because evaporation of water from the highly humid surface of the gelatin lifters prevents the mass spectrometer from achieving the necessary high vacuum. Conventional tape lifters do not cause this issue. Additionally, it has not been tested yet if the spraying of the enzyme on a gelatin lifter can be performed avoiding ridge merging which would defeat the object of imaging analyses. Techniques like DESI could in the future overcome the “vacuum issue” experienced by MALDI with gelatin lifters. Presently, the MALDI imaging analysis following in situ digestion of the blood mark remains confined to paper, cardboard, thin plastics and generally to flat and thin surfaces of deposition no bigger (approximatively) than 10 cm2.

Conclusions The gold standard for proteomic analysis is, undoubtedly, still in-solution proteolysis and detection and identification via LC-MS/MS analysis. Proteins can be story-tellers of physiological as well as pathological states of an individual, as well as revealing contact with biofluids in a forensic scenario involving, for example, fingerprint evidence. However, while in-solution proteomics and LC-MS/MS analysis method provides a very high number of protein identifications, especially when 2D chromatography is used up front, there are some drawbacks that are important to consider especially in a forensic scenario. First and foremost is the level of destructivity of the method, which is here the highest that can be envisaged given that it requires sample homogenization/extraction prior proteolysis and analysis. This sample preparation step may not always be granted in a forensic workflow especially if the specimen bears biometric information as in the case of fingerprints. Additionally, the time consuming and labor-intensive aspects of this method should also need to be taken in account within an ever growing demand for fast turnaround results in an ongoing investigation. For the reasons above and given that fingerprints have proven to be the custodian of diverse molecular intelligence about their owner, it would be desirable for the proteomic analysis of this type of evidence to have a method enabling in situ proteomics, either for quick subsequent molecular survey by profiling mass spectrometry or to enable visualization of the peptides directly on the ridge pattern thus potentially providing the link between the biometric information and the circumstances of the crime. While studies have been published since 2015 showing the opportunity to perform in situ proteomics, the reports are very few. This occurrence may be explained by the inherent difficulties connected to performing in situ proteomics in fingermarks, especially for subsequent molecular mapping experiments, mainly due to sensitivity and limitation of the information available obtained by direct methods such as MALDI mass spectrometry. The latter can be ascribed to competitive ionization effects, the difficulty to select single ions and to fragment singly-charged ions. The introduction of additional gas-phase separation modes, such ion mobility separation in a mass spectrometer operating with a MALDI source has improved the success of MS/MS analyses (12) for protein identification and could prove very helpful for forensic proteomics. Though the overall performance of in situ approaches is still far inferior to that of LC-MS/ MS analyses, it is important to always bear in mind the aim of the analysis being performed. As an example, for discovery studies and in clinical settings, sample extraction, in-solution digestion and 102

LC-MS/MS analysis are crucially needed in order to have a comprehensive picture of the proteome and of its changes in pathological conditions. Nonetheless in a forensic context, what still needs to be determined is whether a handful of blood specific proteins, as detected through in situ proteomics approaches, are enough to substantiate a “blood presence” claim. This entirely reasonable hypothesis can only be verified through systematic blind validation studies as those in progress in our laboratories.

References 1.

Merkley, E. D.; Wunschel, D. S.; Wahl, K. L.; Jarman, K. H. Applications and Challenges of Forensic Proteomics. Forensic Sci. Int. 2019, 297, 350–363. 2. Procopio, N.; Williams, A.; Chamberlain, A. T.; Buckley, M. Forensic Proteomics for the Evaluation of the Post-Mortem Decay in Bones. J. Proteomics. 2018, 177, 21–30. 3. Pieri, M.; Lombardi, A.; Basilicata, P.; Mamone, G.; Picariello, G. Proteomics in Forensic Sciences: Identification of the Nature of the Last Meal at Autopsy. J. Proteome Res. 2018, 17, 2412–2420. 4. Illiano, A.; Arpino, V.; Pinto, G.; Berti, A.; Verdoliva, V.; Peluso, G.; Pucci, P.; Amoresano, A. Multiple Reaction Monitoring Tandem Mass Spectrometry Approach for the Identification of Biological Fluids at Crime Scene Investigations. Anal Chem. 2018, 90, 5627–5636. 5. Walpurgis, K.; Thomas, A.; Vogel, M.; Reichel, C.; Geyer, H.; Schänzer, W.; Thevis, M. Testing for the Erythropoiesis-Stimulating Agent Sotatercept/ACE-011 (ActRIIA-Fc) in Serum by means of Western Blotting and LC-HRMS. Drug Test. Anal. 2016, 8, 1152–1161. 6. Thevis, M.; Thomas, A.; Geyer, H.; Schänzer, W. Mass Spectrometric Characterization of a Biotechnologically Produced Full-Length Mechano Growth Factor (MGF) Relevant for Doping Controls, Growth Horm. IGF Res. 2014, 24, 276–280. 7. Borja, T.; Karim, N.; Goecker, Z.; Salemi, M.; Phinney, B.; Naeem, M.; Rice, R.; Parker, G. Proteomic Genotyping of Fingermark Donors with Genetically Variant Peptides. Forensic Science International: Genetics 2019, 42, 21. https://doi.org/10.1016/j.fsigen.2019.05.005 8. Martin, N. J.; Bunch, J.; Cooper; Helen, J. Sample Preparation and Mass Spectrometry Analysis, Dried Blood Spot Proteomics. J. Am. Soc. Mass Spectrom. 2013, 24, 1242–1249. 9. Bailey, M.; Randall, E. C.; Costa, C.; Salter, T. L.; Race, A. M.; de Puit, M.; Koeberg, M.; Baumert, M.; Bunch, J. Analysis of Urine, Oral fluid and Fingerprints by Liquid Extraction Surface Analysis Coupled to High Resolution MS and MS/MS -Opportunities for Forensic and Biomedical Science. Anal. Methods. 2016, 16, 3373–3382. 10. Caprioli, R. M.; Farmer, T. B.; Gile, J. Molecular Imaging of Biological Samples: Localization of Peptides and Proteins Using MALDI-TOF MS. Anal. Chem. 1997, 69, 4751–4760. 11. Shimma, S.; Furuta, M.; Ichimura, K.; Yoshida, Y.; Setou, M. A Novel Approach to In Situ Proteome Analysis Using Chemical Inkjet Printing Technology and MALDI-QIT-TOF Tandem Mass Spectrometer. J. Mass Spectrom. Soc. Jpn. 2006, 54, 133–140. 12. Djidja, M. J.; Francese, S.; Loadman, P. M.; Sutton, C. W.; Scriven, P.; Claude, E.; Snel, M. F.; Franck, J.; Salzet, M.; Clench, M. R. Detergent Addition to Tryptic Digests and Ion Mobility Separation prior to MS/MS Improves peptide Yield and Protein Identification for In Situ Proteomic Investigation of Frozen and Formalin-Fixed Paraffin-Embedded Adenocarcinoma Tissue Sections. Proteomics 2009, 9, 1–15. 103

13. Takats, Z.; Wiseman, J. M.; Ifa, D. R.; Cooks, G. In Situ Desorption Electrospray Ionization (DESI) Analysis of Tissue Sections. Cold Spring Harb. Protoc. 2008doi:10.1101/ pdb.prot4994. 14. Sarsby, J.; Martin, N. J.; Lalor, P. F.; Bunch, J.; Cooper, H. J. Top-Down and Bottom-up Identification of Proteins by Liquid Extraction Surface Analysis Mass Spectrometry of Healthy and Diseased Human Liver Tissue. J. Am. Soc. Mass Spectrom. 2014, 25, 1953–1961. 15. Wisztorski, M.; Quanico, J.; Franck, J.; Fatou, B.; Salzet, M.; Fournier, I. Droplet-Based Liquid Extraction for Spatially-Resolved Microproteomics Analysis of Tissue Sections. Methods Mol. Biol. 2017, 1618, 49–63. 16. Li, C.; Li, Z.; Tuo, Y.; Ma, D.; Shi, Y.; Zhang, Q.; Zhuo, X.; Deng, K.; Chen, Y.; Wang, Z.; Huang, P. MALDI-TOF MS as a Novel Tool for the Estimation of Postmortem Interval in Liver Tissue Samples. Sci. Rep. 2017, 7, 4887. 17. Knowles, A. M. Aspects of Physicochemical Methods for the Detection of Latent Fingerprints. J. Phys. E: Sci. Instrum. 1978, 11, 713–721. 18. Francese, S.; Bradshaw, R.; Ferguson, L. S.; Wolstenholme, R.; Clench, M. R.; Bleay, S. Beyond the Ridge Pattern: Multi-Informative Analysis of Latent Fingermarks by MALDI Mass Spectrometry. Analyst 2013, 138, 4215–4228. 19. Ramotowski, R. Composition of Latent Print Residue. In Advances in Fingerprint Technology, II ed.; Lee, H. C, Gaensslen, R. R., Eds.; CRC Press: Boca Raton, 2001; pp 63−104. 20. Oonk, S.; Schuurmans, T.; Pabst, M.; de Smet, L. C. P. M.; de Puit, M. Proteomics as a New Tool to Study Fingermark Ageing in Forensics. Sci. Rep. 2018, 8, 16425. 21. Ferguson, L. S.; Wulfert, F.; Wolstenholme, R.; Fonville, J. M.; Clench, M. R.; Carolan, V. A.; Francese, S. Direct Detection of Peptides and Small Proteins in Fingermarks and Determination of Sex by MALDI Mass Spectrometry Profiling. Analyst 2012, 137 (20), 4686–4692. 22. Francese, S. Techniques for Fingermark Analysis Using MALDI MS-a Practical Overview. In Advances in MALDI and Laser Induced Soft Ionisation Mass Spectrometry; Cramer, R., Ed.; Springer: New York, 2016; pp 93−128. 23. Ricci, C.; Phiriyavityopas, P.; Curum, N.; Chan, K. L.; Jickells, S.; Kazarian, S. G. Chemical Imaging of Latent Fingerprint Residues. Appl. Spectrosc. 2007, 61, 514–522. 24. Rieg, S.; Seeber, S.; Steffen, H.; Humeny, A.; Kalbacher, H.; Stevanovic, S.; Kimura, A.; Garbe, C.; Schittek, B. Generation of Multiple Stable Dermcidin-Derived Antimicrobial Peptides in Sweat of Different Body Sites. J. Invest. Dermatol. 2006, 126, 354–365. 25. Rieg, S.; Steffen, H.; Seeber, S.; Humeny, A.; Kalbacher, H.; Dietz, K.; Garbe, C.; Schittek, B. Deficiency of Dermcidin-Derived Antimicrobial Peptides in Sweat of Patients with Atopic Dermatitis Correlates with an Impaired Innate Defense of Human Skin In Vivo. J. Immunol. 2005, 174, 8003–8010. 26. Baechle, D.; Flad, T.; Cansier, A.; Steffen, H.; Schittek, B.; Tolson, J.; Herrmann, T.; Dihazi, H.; Beck, A.; Mueller, G. A.; Mueller, M.; Stevanovic, S.; Garbe, C.; Mueller, C. A; Kalbacher, H. Cathepsin D is Present in Human Eccrine Sweat and Involved in the Post-Secretory Processing of the Antimicrobial Peptide DCD-1L. J. Biol. Chem. 2006, 281, 5406–5415. 27. Lee, D. Y.; Yamasaki, K.; Rudsil, J.; Zouboulis, C. C.; Park, G. T.; Yang, J. M.; Gallo, R. L. Sebocytes Express Functional Cathelicidin Antimicrobial Peptides and Can Act to Kill Propionibacterium Acnes. J. Invest. Dermatol. 2008, 128, 1863–1866. 104

28. Schittek, B.; Paulmann, M.; Senyeurek, I.; Steffen, H. The Role of Antimicrobial Peptides in Human Skin and in Skin Infectious Diseases. Curr. Drug Targets 2008, 8, 135–143. 29. Drapel, V.; Becue, A.; Champod, C.; Margot, P. Identification of Promising Antigenic Component in Latent Fingermark Residues. Forensic Sci. Int. 2009, 184, 47–53. 30. Flad, T.; Bogumil, R.; Tolson, J.; Schittek, B.; Garbec, C.; Deega, M.; Muellera, C. A.; Kalbacher, H. Detection of Dermcidin-Derived Peptides in Sweat by ProteinChip Technology. J. Immunol. Methods 2002, 270, 53–62. 31. Patel, E.; Clench, M. R.; West, A.; Marshall, S.; Marshall, N.; Francese, S. Alternative Surfactants for Improved Efficiency of In Situ Tryptic Proteolysis of Fingermarks. J. Am. Soc. Mass Spectrom. 2015, 26, 862–872. 32. Ferguson, L. S.; Bradshaw, R.; Wolstenholme, R.; Clench, M. R.; Francese, S. Two-Step Matrix Application for the Enhancement and Imaging of Latent Fingermarks. Anal. Chem. 2011, 83, 5585–5591. 33. Longobardi, S.; Gravagnuolo, A.; Rea, I.; De Stefano, L.; Marino, G.; Giardina, P. Hydrophobin-Coated Plates as Matrix-Assisted Laser Desorption/Ionization Sample Support for Peptide/Protein Analysis. Anal. Biochem. 2014, 449, 9–16. 34. Longobardi, S.; Gravagnuolo, A.; Funari, R.; Della Ventura, B.; Pane, F.; Galano, E.; Amoresano, A.; Marino, G.; Giardina, P. A Simple MALDI Plate Functionalization by Vmh2 Hydrophobin for Serial Multi-Enzymatic Protein Digestions. Anal. Bioanal. Chem. 2015, 407, 487–496. 35. Patel, E.; Cicatiello, P.; Deininger, L.; Clench, M. R.; Marino, G.; Giardina, P.; Langenburg, G.; West, A.; Marshall, P.; Sears, V.; Francese, S. A Proteomic Approach for the Rapid, MultiInformative and Reliable Identification of Blood. Analyst 2016, 141, 191–198. 36. Bradshaw, R.; Bleay, S.; Clench, M. R.; Francese, S. Direct Detection of Blood in Fingermarks by MALDI MS Profiling and Imaging. Sci. Jus. 2014, 54, 110–117. 37. Deininger, L.; Patel, E.; Clench, M. R.; Sears, V.; Sammon, C.; Francese, S. Proteomics Goes Forensic: Detection and Mapping of Blood Signatures in Fingermarks. Proteomics 2016, 16, 1707–1717. 38. Francese, S. Criminal Profiling Through MALDI MS bBased Technologies – Breaking Barriers Towards Border-Free Forensic Science. Aust. J. Forensic Sci. 2019, 51, 623. DOI:10.1080/ 00450618.2018.1561949. 39. Kamanna, S.; Henry, J.; Voelcker, N.; Linacre, A.; Kirkbride, K. P. “Bottom-up” In Situ Proteomic Differentiation of Human and non-Human Haemoglobins for Forensic Purposes by Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Tandem Mass Spectrometry. Rapid Commun. Mass Spectrom. 2017, 31, 1927–1937. 40. Deininger, L.; Francese, S.; Clench, M. R.; Langenburg, G.; Sears, V.; Sammon, C. Investigation of Infinite Focus Microscopy for the Determination of the Association of Blood with Fingermarks. Sci. Jus. 2018, 58, 397–404.

105

Chapter 7

Human Identification Using Genetically Variant Peptides in Biological Forensic Evidence Fanny Chu,1,2 Katelyn E. Mason,1 Deon S. Anex,1 Phillip H. Paul,1 and Bradley R. Hart*,1 1Forensic Science Center, Lawrence Livermore National Laboratory,

Livermore, California 94550, United States 2Department of Chemistry, Michigan State University,

East Lansing, Michigan 48824, United States *E-mail: [email protected].

Proteins in biological evidence offer a pathway for human identification when DNA is absent or compromised and can augment existing intact DNA evidence, as collectives of single amino acid substitutions (SAPs) within protein sequences can serve as individual-specific markers. Peptides containing SAPs are known as genetically variant peptides (GVPs). Key to using GVPs in forensics is their link to associated single nucleotide polymorphisms (SNPs) in the corresponding proteincoding DNA. As such, SNP population frequencies can be used to calculate statistics, such as random match probability (RMP), derived from protein evidence, and rules of genetic inheritance can be applied. Proteomic analysis of forensic samples guided by predictions from DNA exomic analysis (i.e., of exons in the genome) can locate these GVPs. Protein-based identification was first demonstrated in 2016 using hair shafts for a cohort of over 60 individuals, producing RMPs up to 1 in 14,000 and ancestry determination. GVPs were shown to persist in archaeological hairs over 250 years old. Subsequent studies have extended GVP capabilities to bone and tooth tissues and shed skin cells. Improved sample preparation and bioinformatics have enabled greater numbers of identified SNPs; a 12,000-fold increase in maximum discriminative power has been achieved even with 100-fold reduction in sample size, from bulk quantities to a single inch of hair. Further, independence of GVP identification from body location-specific protein expression has been demonstrated. Continued development of this technology through common or rare GVP panels and concurrent GVP and mitochondrial DNA analysis provides powerful tools for individual identification and enhanced discriminative power.

© 2019 American Chemical Society

Introduction There are an alarming number of cases in which expert examination of hair evidence contributed to felony convictions against innocent individuals who have later been exonerated. Richard Danziger spent 12 years in prison before genetic profiling of DNA found on the victim’s body was used exonerate him. During his trial, a claim made by an analyst, who stated that hairs found at the crime scene were similar and consistent with a match when microscopically compared to Danziger’s, had significant influence on his conviction (1). In another example, William Barnhouse served 25 years in prison until DNA testing performed on biological material found on the victim collected during the investigation excluded Barnhouse as the source (2). A statement that hair found on the victim’s body matched that of Barnhouse given by a forensic analyst during trial had similar sway as in Danziger’s trial, leading to another false conviction. Although the analyst’s statement about consistency between the two sources of hair in both cases may have been correct, disclosure of the conclusive limitations of this technique is paramount. Failure to explain these limitations results in flawed expert testimony that, in some cases, facilitates false convictions. In the case of Santae Tribble, he spent 32 years in prison before mitochondrial DNA testing of hair evidence that belonged to the suspect revealed that Tribble was not the source of the hair; it belonged to a canine officer (3). During his trial, an analyst working the case for the FBI claimed there was “one chance in ten million” that hair evidence collected at the crime scene could have belonged to anyone other than Tribble. In this instance, the FBI analyst not only misstated the evidence in terms of a mathematical probability based upon population statistics that do not exist, the analyst committed the classical “prosecutor’s fallacy” by inferring that the probability that the two hairs matched was the same as the probability that the suspect was guilty (4–6). These noteworthy cases have several things in common: (1) Potentially incorrect interpretations of microscopic hair analysis was presented to the jury in their trials, (2) the men found guilty at trial were later found to be innocent, and (3) many years of life were lost in prison by each of these men. The serious consequences of flawed convictions in these examples highlight the crucial need for additional science-based forensic techniques. Specifically, outcomes of these cases involving hair, and other cases involving other biological materials such as bone and skin, could benefit from forensic tools that rely less on subjective expert opinion and more on analysis rooted in objective scientific analysis and based on a sound statistical footing (7–11).

Protein-Based Human Identification In contrast to subjective inspection and analysis of evidence, instrument-based approaches benefit from being more objective and quantifiable. Current conventional instrument-based analytical approaches to forensic identification of individual subjects using biological materials often rely on DNA. However, if biological evidence contains little or no available DNA or DNA has been damaged, then there are limited alternative analysis methods. As an example, consider hair shafts from shed (telogen) hair, which typically do not include the root. These hairs often do not have nuclear DNA that is suitable for DNA profiling. During hair growth, the transition of living keratinocytes into the keratinized form found in the hair shaft involves cleavage of nuclear DNA into short segments (12). This cleaved DNA is not amenable to the analysis of short tandem repeats (STR profiling) used in typical forensic DNA analysis (13, 14). Instead, hairs lacking suitable nuclear DNA for STR profiling may be analyzed under a microscope for appearance (color, morphology, etc.) and compared on this basis (15). This method relies on the expert opinion of the examiner and can be subject to error as illustrated in the cases cited above. Alternatively, mitochondrial DNA may 108

be extracted from hair shafts and analyzed (16). Analysis of the mitochondrial DNA, however, only shows connections along maternal lines and is far less discriminating than STR profiling of nuclear DNA. Over the past few years, the Forensic Science Center (FSC) at Lawrence Livermore National Laboratory has developed methods that exploit identifying information encoded in proteins. These identification methods rely on detecting single amino acid substitutions in proteins (single amino acid polymorphisms or SAPs) and linking them to the corresponding single nucleic acid substitutions in DNA (non-synonymous single nucleotide polymorphisms or nsSNPs). Peptides that contain informative SAPs are termed genetically variant peptides (GVPs). Initial research focused primarily on hair proteins, but more recent studies have extended the approach to proteins in tooth and bone samples and shed skin cells that are present in touch-type evidence. The connection between SNPs in DNA and SAPs in the proteins it encodes is shown schematically in Figure 1. Here, three codons (GGT/AGA/TGC) in the DNA sequence to the left are shown with the corresponding amino acids that they encode in the protein (glycine/arginine/ cysteine). In the DNA sequence on the right, an nsSNP (A→T) in the middle codon changes the corresponding amino acid from arginine to serine. Sequence analysis of a protein that reveals an amino acid substitution is evidence of the underlying SNP in the DNA that encoded it. These SAPs, when detected in GVPs, form the basis of protein-based identification. Connections of SAPs found in GVPs to the underlying SNPs provide the basis for statistical evaluation of discriminative power of a profile from a protein sample within a population, such as calculation of random match probability (RMP). In practice, a collection of GVPs are considered for analysis of a particular type of evidence (e.g., a panel of GVPs for hair analysis). The GVPs in the panel are chosen beforehand based on factors such as ease of detection via mass spectrometry, chemical stability, uniqueness, population frequency of the underlying allele, and statistical independence from other GVPs. Protein-containing evidence (such as a hair from a crime scene) can be analyzed for the presence of SAP-containing GVPs in the panel, which in turn correspond to SNPs in the DNA of the individual from whom the evidence originated. Both forms of the variant peptides, that is, the form containing the SAP and the form without the polymorphism, are included in the analysis. Using the population frequencies of the corresponding SNPs, statistical treatment of the match quality of the evidence to a suspect in the crime may be performed.

Figure 1. Illustration of an nsSNP in a DNA sequence and the corresponding SAP in the protein that is coded by that DNA sequence. The sequence on the left shows the reference sequence and three codons that encode for three amino acids. A SNP in the variant sequence on the right encodes for a different amino acid in the protein. Knowledge of the SNP profile of the individual may be obtained from a corresponding proteincontaining sample (e.g., a hair taken from the suspect to compare to a hair from the crime scene) or preferably by simply sequencing the subject’s DNA to determine the presence or absence of the SNPs corresponding to the GVP panel. Once the SNP profiles from the evidence and the suspect are 109

determined, then concepts that are familiar in STR profiling, such as RMP and likelihood ratio, can be applied to GVP analysis.

Advancements in Genetically Variant Peptide Discovery Capabilities Discovery of GVPs in a variety of matrices has not only enhanced forensic identification capabilities using an approach not typically considered in forensic analyses, but has also furthered our knowledge of protein chemistry in matrices that are not well-studied. Reliant upon a combination of techniques from different disciplines, including genetics, bioanalytical chemistry, and statistics, GVP identification is possible from hair shaft, bone, tooth, and skin cells, ranging from bulk quantities of material to forensically relevant sample sizes. GVP analysis not only enables identification of individuals, but also provides a means for distinction of individuals by ancestry (17, 18) and differentiation of monozygotic twin pairs from unrelated individuals (19). We review advancements in the GVP discovery process from work performed at the FSC, compare the proteomes among the different matrices and their effects on GVP identification, and briefly describe further considerations for translating this technology into routine operation in forensic investigations. Exome-Driven Approach to GVP Discovery Transition to an exome-driven process has substantially improved GVP discovery from protein digests of matrices of interest. Efficient and maximal GVP discovery for SNP inference is critically important for improving discriminative power, as measured via RMPs. Initial demonstration of human identification with GVPs utilized a custom in-house protein sequence database for matching to experimentally identified peptides (17). Deviating from conventional approaches in proteomics, identification of peptides with SAPs necessitates matching within databases that contain protein sequences with amino acid mutations as opposed to canonical protein sequences found in curated databases (e.g., UniProt KnowledgeBase SwissProt database (20)). To include SAPs relevant for GVP analysis in reference databases, protein sequences must be mutated to reflect SNPs. One method for generating relevant mutated protein sequences is to use known SNPs from databases such as the Single Nucleotide Polymorphism Database (dbSNP) (National Center for Biotechnology Information, U.S. National Library of Medicine) (21). However, roughly 12 million SNPs are annotated, and though a large fraction of them do not result in SAPs (i.e., synonymous SNPs do not effect amino acid mutations), an exhaustive search with all proteins affected by non-synonymous SNPs becomes a computationally expensive process. Furthermore, not all proteins may be expressed in the matrix of interest and not all SNPs may be expressed in each individual. Instead, a focused search strategy using an individualized mutated protein database from exome sequence information eliminates searching of SNPs not expressed within the individual while maximizing GVP identification to include SNPs that may not be curated in public databases. Additionally, other mutations to DNA sequences (e.g., insertions and deletions, proximal nsSNPs acting on the same codon) are considered when generating individualized protein databases using exome information for more accurate mutated protein sequence prediction. Figure 2 illustrates the exome-driven workflow for more efficient GVP discovery (18). From an individual-matched DNA sample, exome sequencing enables detection of all variants relevant to the individual, and these are then filtered for expression in genes of interest to the type of evidence, such as hair shaft, and ultimately annotated in protein sequences within an individualized mutated database along with their unmutated sequences, which includes isoforms (22). Applied to single one-inch scalp hairs, this process permitted identification of 20 SNPs from GVPs in 16 keratin-associated proteins (22), a class 110

of highly abundant hair structural proteins not included in Parker et al. as a source of GVPs (17). Consequently, inclusion of GVPs from this protein class and other cellular proteins yielded RMPs up to 1 in 167,000,000 with a single inch of hair (22), which represents approximately 12,000-fold improvement in discriminative power, from an RMP of 1 in 14,000 achieved in Parker et al. (17), with 100-fold less material.

Figure 2. Exome-driven approach for maximizing GVP discovery. Mass Spectrometry Considerations in GVP Discovery Inference of SNPs from GVPs in proteins adds an additional layer of complexity to GVP discovery, as proteins commonly undergo post-translational modifications. These chemical modifications to specific amino acid residues may not be distinguished from amino acid polymorphisms when detected in tandem mass spectra, as they share similar masses, and would need to be removed from consideration during GVP identification. Examples include oxidized methionine to phenylalanine, asparagine to aspartate, and cysteine to serine mutations. Tissue-specific chemical modifications also need to be accounted for; in collagen, a highly abundant protein class in bone, proline oxidation, whose product is similar in mass to isoleucine and leucine, is prevalent (23). As such, the proline to isoleucine and proline to leucine polymorphisms need to be excluded from the analysis (18), though the SAP may be less likely to occur than the chemical modification, given the differences between proline codons (e.g., CCT, CCA) and isoleucine codons (e.g., ATT, ATA). Additionally, isobaric amino acid polymorphisms such as isoleucine to leucine are excluded as they are indistinguishable via mass spectrometry. Table 1 displays a comprehensive list of SAPs and the chemical modifications that cannot be distinguished from the respective SAPs. For example, the SAP Met/Phe (methionine to phenylalanine mutation) shares a similar mass difference with methionine oxidation. The mass difference of the SAP, 16.0279 Da, cannot be distinguished from oxidation of methionine to methionine sulfoxide, which has a mass difference of 15.9949 Da; the mass difference between SAP and chemical modification is 0.0330 Da. Although many of these polymorphisms are indistinguishable from known post-translational modifications, the use of sufficiently high-resolution mass analyzers allows distinction of Met/Phe from methionine oxidation and Pro/Ile from proline oxidation, and differentiation of glutamine from lysine residues. Whereas protein digests were initially analyzed via a quadrupole-time-of-flight mass analyzer, increased mass resolution to 35,000 during tandem mass spectra acquisition with an Orbitrap™ (Thermo Scientific, Waltham, MA) and implementation of a 5-ppm mass error tolerance

111

for peptide identification enables distinction between SAP and chemical modification while minimizing false positive GVP identifications (22). The combination of high-resolution tandem mass spectrometry and exome sequencing enables GVP identification from bulk quantities to single hairs. Further, with improvements to hair sample preparation, greater numbers of SNPs are identified in single hairs (19 ± 5 (s.d.)) as compared to bulk hair quantities (12 ± 4; p = 0.0272) (22), demonstrating the success of protein-based human identification for application to forensically relevant sample sizes. The combination of ultrasonication at elevated temperatures and use of harsh detergent sodium dodecanoate, which is removed prior to mass spectrometry analysis, entirely degrades the hair matrix and provides greater exposure of hair proteins for enzymatic digestion, thus enabling a greater number of identified proteins and unique peptides, and consequently, SNPs inferred from GVPs (22). Table 1. List of SAPs Not Distinguished via Mass Spectrometry and the Corresponding Theoretical Mass Shifts SAP

ΔMass of SAP (Da)

ΔMass of Modification (Da)

Modification

ΔΔMass (Da)

Met/Phe

16.0279

15.9949

Methionine Oxidation

0.0330

Asn/Asp

0.9840

0.9840

Asparagine Deamidation

0

Gln/Glu

0.9840

0.9840

Glutamine Deamidation

0

Ser/Cys

15.9771

15.9949

Cysteine Oxidation to Sulfenic Acid

-0.0178

Pro/Ile or Pro/Leu

16.0313

15.9949

Proline Oxidation

0.0364

Gln/Lys

0.0364

--

--

--

Ile/Leu

0

--

--

--

Figure 3. Comparison of identified proteins between single one-inch and bulk (10 mg) quantities of scalp hair. Data are from reference (22).

112

Effects of Protein Chemistry in Hair, Bone, and Skin on GVP Identification Hair Proteome Hair shaft boasts enormous diversity in proteins, with at least 150 proteins identified per sample, ranging from highly abundant structural proteins (i.e., keratins and keratin-associated proteins) (17) to catalytic and binding proteins involved in metabolism and regulation (22), permitting discovery of a diverse set of GVPs. Similar protein populations can even be identified within a single hair (22), which represents 100-fold less material compared to 10 mg bulk hair quantities, demonstrating the abundance of total protein content within hair (Figure 3). Further, a handful of cellular proteins persisted in archaeological hair shaft dating from 250 years ago, though as expected, the majority of proteins that remain are structural proteins (Figure 4) (17).

Figure 4. Distribution of identified proteins in modern and archaeological hair shaft specimens by protein function. Reproduced with permission from reference (17). Creative Commons CC0. Table 2. SNPs Identified in Bulk Quantities and Single Hairs Also Identified in Archaeological Hair Samples

Gene

SNP Reference ID

SAP

S100A3

rs36022742

Arg/Lys

HEXB

rs10805890

Ile/Val

rs72830046

Arg/His

rs2071563

Thr/Met

rs2071561

Ser/Tyr

rs12451652

Cys/Tyr

X

rs743686

Ser/Pro

X

rs2071601

Pro/Thr

X

KRT33A

rs12937519

Ala/Val

SERPINB5

rs1455555

Ile/Val

GSTP1

rs1695

Ile/Val

KRT81

rs2071588

Gly/Arg

BLMH

rs1050565

Ile/Val

KRT83

rs2852464

Ile/Met

KRT32

KRT35

113

X

X

X

Despite the diversity of proteins identified from hair, many of the same GVPs are repeatedly identified in different cohorts, indicating the reliability of using these markers to differentiate within a population. For example, 14 SNPs in 10 proteins were inferred from GVPs identified from two separate cohorts, in both single inch and bulk hair quantities (Table 2), which represent 25% and 42% of the total number of SNPs identified in each cohort, respectively (17, 22). Six of these SNPs were further identified in archaeological hair shafts, making up the majority of the SNPs found in highly degraded specimens, all of which were from keratins (17). Table 2 lists the SNPs identified in bulk quantities and single hairs, the genes in which the SNPs are contained, the amino acid polymorphism corresponding to the SNP, and whether the SNP was also identified in archaeological hair samples. The prevalence and persistence of keratins in hair fibers naturally makes these proteins attractive targets for robust GVP identification. Many GVPs are also identified in keratin-associated proteins, as demonstrated in Mason et al. (22), though these shorter proteins do not contain as many enzymatic cleavage sites for digestion with trypsin, which may result in unreliable identification of non-tryptic GVPs (24) with the current approach. Comparison of GVPs identified in head, arm, and pubic hair also confirmed differential hair protein expression, in agreement with a previous study (25), and vast differences in protein abundance permit distinction of hair fibers by body location (24). Not surprisingly, localized proteins are mostly cellular proteins and keratin-associated proteins (Figure 5), which suggest that keratinassociated proteins may affect physicochemical properties of hair such as tensile strength and thickness (24). Despite protein expression differences in hair fibers from different body locations, SNPs are inferred from identified GVPs in an entirely different set of proteins, mostly in keratins that are not differentially expressed. The majority of SNPs are identified at all body locations (Figure 6), and those not reliably identified at the three body locations are not from differentially expressed proteins (24), demonstrating no effect of body location-specific protein expression on GVP identification. Figure 6 depicts overlapping SNPs that are inferred from both the major and minor forms of the GVPs among the three body locations. The major form of the GVP (i.e., the non-mutated form) is derived from the reference allele, usually with a higher population frequency, while the minor SAPcontaining form originates from the alternate allele. These findings lend greater confidence to the protein-based human identification approach for forensic purposes, as similar GVPs can be identified in hair regardless of body location origin, which is often not known during evidence recovery. Bone and Skin Proteome In comparison to the hair proteome, skin cells and bone tissue present different sets of proteins from which GVPs are identified. Skin cells are structurally more similar to hair, as both comprise differentiated keratinocytes that rely on coiled-coil keratin dimers for stability (26). Thus, skin cells are also highly enriched with keratins, though these are primarily soft keratins (e.g., K2) with lower cysteine content, as opposed to harder α-keratins localized in hair (e.g., K81) (27); a less rigid structure (28) may contribute to the ease with which proteins are extracted from skin cells. Other shared proteins between skin cells and hair include junction plakoglobin (JUP) and desmoplakin (DSP) (17, 26), which are peripheral structural proteins, and protein-glutamine gammaglutamyltransferase E (TGM3) (22), an enzyme that catalyzes isopeptide bonds for promotion of protein crosslinks (29). Proteins have been co-extracted with DNA from unfired brass bullet 114

cartridges, demonstrating the ability to obtain protein information from skin cells (26). Ongoing work is in progress to identify GVPs from proteins in recovered skin cells as contact traces.

Figure 5. Subset of differentially expressed proteins among scalp, arm, and pubic hairs. Adapted from reference (24) with permission. Creative Commons BY. While keratins dominate skin cells and hair, collagens are the most prevalent structural proteins in bone, four of which contain GVPs. Bone tissue also comprises a wide variety of noncollagenous proteins, such as the enzyme prothrombin and tissue-specific proteins osteocalcin and chonroadherin. Similar to the distribution of hair shaft proteins, catalytic and binding proteins make up the majority of the protein diversity (Figure 7), though these proteins are not the most abundant (18). The presence of a host of cellular proteins indicates minimal tissue degradation, as ancient bones dating from greater than 4,000 years old yield few noncollagenous proteins (30).

115

Figure 6. Distribution of the numbers of inferred SNPs from major and minor GVPs in hair by body location. Reproduced with permission from reference (24). Creative Commons BY.

Figure 7. Distribution of proteins in rib bone tissue by biological function. Data are from reference (18). GVP identification from bone tissue permits calculation of RMPs that span a wide range, from 1 in 6 to 1 in 42,472 (using European population frequencies annotated in 1000 Genomes (31)), with a true positive detection rate of 57% (18). Detection rates were calculated based on identification of the SNP through inference from presence of the GVP using mass spectrometry and directly using exome sequencing. The latter observation formed the basis for true and false positive and negative detections; a true positive detection indicates inference of the SNP from identification of the GVP where the SNP has been detected in an individual’s exome sequence. Comparison of RMPs calculated with European and African population frequencies as likelihood ratios showed greater statistical probability that the specimens were of European ancestry, with 50% of specimens over 10 times more likely to be of European than African ancestry. In comparison to hair shaft, bone tissues yield few inferred SNPs (19 SNPs from 15 proteins) relative to the number of identified proteins (18, 22), contributing to higher RMPs. Application of this protein-based approach to bone tissue likely finds greater utility in ancestry determination in archaeological analyses, as a method for exclusion in 116

human identification, or in mass casualty situations where subject DNA or DNA from close relatives is available to detect rare GVPs that would aid in the distinction of a particular individual within a restricted group of victims. Translating GVP Technology for Forensic Investigations Subsequent work following the initial characterization of GVPs in hair shaft has extended the capabilities of protein-based human identification to translate this technology for forensic investigations. Not only have we demonstrated the ability to identify GVPs in matrices beyond hair (18), but GVP identification is possible even from minute sample sizes (22), and equally probative in hair from different body locations (24). In propelling this technology forward for forensic applications, in a format similar to STR profiling, a few considerations need to be addressed, including selection, detection, and confirmation of GVPs in unknown samples, and statistical treatments for minimizing RMPs and maximizing differentiative potential. Although hair GVPs have demonstrated successful ancestry determination via likelihood ratios, they can be further exploited to differentiate within a population with accumulation and discovery of more GVPs. In this vein, differentiation of monozygotic twin pairs from unrelated individuals was achieved, where GVP profiles of true twin pairs exhibited the least amount of variation in contrast to pairwise comparisons of unrelated individuals (19), a measure of discriminative power. Where genotype frequency information is not available, allele frequencies have been used in conjunction with assumption of Hardy-Weinberg equilibrium (32) to estimate population frequency. However, this alternative introduces unnecessary error to RMP calculations. For example, the SNP referenced in dbSNP accession rs61814939, with a mutation C→T which produces the SAP G2054S in the protein hornerin, is observed in 32,040 alleles from a total of 90,178 alleles, with four instances of homozygosity for the SNP in a public database. Based on this data, the allele frequency is 0.36 and population frequency for the SNP (heterozygotes and homozygotes) is 0.71, but under assumption of Hardy-Weinberg equilibrium, the population frequency is calculated to be 0.58, which represents an error of 18% from the observed frequency. Observations of this SNP in the population do not follow Hardy-Weinberg equilibrium, and thus, population frequency from empirical observations should be used whenever possible. The expansion of databases containing genotype frequency information, such as 1000 Genomes (31) and the Genome Aggregation Database (gnomAD) (33) facilitates annotation of SNP population frequencies for RMP calculations, though the two databases mentioned are curated to different extents. The former contains approximately 1,100 samples while the latter contains an estimated 150,000 samples, with a combination of exome and genome sequences. To generate RMPs that represent a population, population genotype frequencies must also be obtained from a representative sample set, the obvious point being that larger sample sizes are generally indicative of a more representative population. Further, with a larger database such as gnomAD, more SNPs are annotated that can then be used in RMP calculations. In addition to population frequencies, the degree of statistical dependence of SNPs is of concern, as the product-rule calculation for RMPs assumes statistical independence between SNP loci. Mason et al. (22) applied a set of conservative rules that are consistent with the available analysis of SNP covariance (34, 35) and that aim to balance systematic error (in application of linkage measured in one population to a different population) with sampling error (linkage as measured on relatively small sample sizes). They assumed negligible covariance between SNPs in different chromosomes, between SNPs separated by more than 2 × 105 base pairs within a chromosome, and between those where one has an allele frequency less than 1 × 10-3. 117

Key to the GVP discovery approach is the use of data-dependent mass spectrometry, an untargeted method for maximizing protein sequence coverage and subsequently, GVP identification. In our current scheme, only the 10 most abundant ions in each MS1 survey scan are selected for fragmentation. However, a disadvantage of this approach is some irreproducibility in GVP identification. Owing to the complexity of the matrix not all ions undergo MS/MS; thus, from run-to-run differences in ion intensity or shifts in retention time, different ions may be selected for fragmentation, resulting in variations in SAPs identified in sample replicates. Consequently, differences in identified SAPs among sample replicates produce vastly different RMPs. For example, Mason et al. showed up to 1 in 167,000,000 RMP from a single inch of scalp hair, though its replicate yielded an RMP value that differed by approximately two orders of magnitude (Figure 8) (22), indicating large intraindividual variation among replicates. As we pivot to operational use of GVP identification, GVP detection methods must be modified to eliminate intraindividual variation introduced in the analytical process.

Figure 8. Random match probabilities of single and bulk hair samples as a function of the number of inferred SNPs from GVPs. Two single-hair sample replicates are labeled. Data are from reference (22). Targeted mass spectrometry methods such as selected ion monitoring or parallel reaction monitoring provide more reproducible and higher throughput means for detection of GVPs in a panel, parallel to STR profiling (22). Development of a panel of GVPs for profiling of individuals requires verification of the masses and retention times of candidate GVPs, typically confirmed with isotopically-labeled synthetic peptide standards. Orthogonal validation with Sanger sequencing may also be advantageous during development with known samples, in addition to the exome-driven approach, to obtain accuracy and detection rates (22). A focused search for select GVPs within an unknown sample achieves improved signal-to-noise, as only those GVPs are detected. Consequently, the combined presence or absence of panel GVPs creates a highly discriminating profile for the unknown sample with an associated statistical probability. There are two general approaches to creation of a GVP panel. For standard (or general GVP) identification, the panel is assembled with GVPs corresponding to SNPs with population frequencies in the range of 10% to 70%. The GVPs, both the non-mutated forms and the SAP-containing forms, are selected from proteins having a substantial chance of being represented in the sample, established in previous analyses during GVP discovery. Such a panel can be applied to any protein-containing evidence type (e.g., hair) without prior knowledge of the suspect. The population frequencies are selected to be high enough that a practical number of GVPs will be detected in the evidence and 118

low enough to obtain a reasonable level of discrimination. The GVPs are also selected to minimize covariances. For a panel with N components, each with population frequency p, the expected number expressed in a sample is np, hence the expected match probability is RMP = pnp (e.g., a 50component panel with p = 20% gives RMP ≈ 10-7). Specific GVP identification is an alternate approach that requires genetic information about the suspect in the form of DNA from the suspect or from a close family member (or members). The genetic information is analyzed to identify a modest number of low-probability GVPs in proteins that are expected to be present in the sample. The GVPs are also chosen to have suitably low levels of interference from other mutations and are again selected to minimize covariances. This approach sacrifices applicability to a general population, but has the advantage of generating extremely low RMPs using a small number of well-selected and targeted GVPs. For example, three independent GVPs each having a population frequency of 10-6 would yield an RMP of 10-18. A customized panel of rare GVPs based on an individual’s DNA is only useful for identification of that target individual. Rare GVPs are also advantageous in samples that may have multiple contributors such as shed skin cells in touch-type evidence. Finally, we aim to maximize differentiative potential by augmenting RMPs through a combination of protein- and DNA-based approaches. Though SNPs in mitochondrial DNA are inherited from the maternal line and thus possess limited statistical power relative to GVPs, the combination of both techniques strengthens confidence in profiling individuals, more powerful than either technique alone. Further demonstration of successful co-extraction of DNA and protein from even skin cells recovered as a contact trace (26) provides a pathway for concurrent identification of GVPs and mtDNA SNPs for enhanced discriminative power in forensic identification.

Conclusion With the development of protein-based identification methods, the set of tools available for the forensic examination of biological evidence is expanding. Robust proteins can comprise a long-lived component of forensic evidence and the link of amino acid substitutions in GVPs to SNPs in DNA provides a solid foundation for the use of GVPs for identification. Specifically, the ever-expanding collections of DNA SNP data provide not only information for the identification of candidate GVPs, but also provide frequency information regarding the occurrence of those markers in target populations. The connection to DNA also allows application of genetic rules of heredity to predict GVPs, assessment of familial relationships, and determination of ancestry or biogeographical origin. The earliest stages of research into GVPs focused on discovery of GVP markers that are suitable for reliable detection by LC/MS and provide consistent indication of the presence of an underlying SNP. Considerations of suitability include expression of the proteins that contain the candidate GVPs in the tissue of interest, the uniqueness of the peptide sequence of the GVP, the allele frequency of the corresponding SNP, chemical stability of the GVP, and the amenability of detection and characterization of the GVP, including both the non-mutated and the SAP-containing forms of the variant peptide, by tandem mass spectrometry. Early research also identified two general regimes of GVP strategies. In the first, termed the general GVP identification approach, a panel of common GVPs, with their associated non-mutated forms, are assembled and applied generally to evidence. In this case, the GVPs, that is, the SAPcontaining forms, must be common enough to be detected, but rare enough to provide discrimination. Here, the evidence is analyzed for the GVPs in the panel and a pattern of absence or presence of each marker, which enables inference of SNP genotype, is determined. This pattern is 119

then compared to other evidence to determine match quality. Alternatively, the theoretical pattern could be predicted from the DNA of a suspect and then compared to the evidence, analogous to STR profiling; this strategy is depicted in Figure 9. Figure 9 illustrates a workflow for the general GVP identification approach comparing evidence from a crime scene to a suspect’s DNA and the resultant output from a general GVP panel. In the second regime, referred to as the specific GVP identification approach, a panel can be constructed from a small set of rare GVPs. In this case, the specific set of rare markers would not be expected to be found in the general population, so by necessity, this would be a targeted approach. The subject DNA would be needed beforehand for analysis of a set of rare SNPs that would lead to detectable GVPs. Protein-containing evidence would be analyzed specifically for the set of rare GVPs in this individually targeted panel. The rare approach trades the high discrimination obtainable from the rare GVP panel for the restriction that requires a subject’s DNA in the analysis.

Figure 9. Workflow of a general GVP identification approach for comparison of forensic hair evidence to a suspect’s DNA. GVPs from forensic evidence are identified via targeted liquid chromatography-tandem mass spectrometry given a list of targets from the general GVP panel. Predicted GVPs from a suspect’s DNA are compared to GVPs identified from hair evidence. In this scenario, both the suspect and the individual to which the evidence belong are heterozygotes for the SNP rs1732263 in K82, as both the non-mutated and SAP-containing forms of the GVP are identified. There are several open issues that must be addressed before widespread use of GVP analysis for forensics will be possible. The statistics of evaluating matches is foremost. STR profiling has a long history of developing such a statistical treatment and concepts like RMP and likelihood ratio are well known. Analogous statistical treatment needs to be developed for GVP analysis. Elementary multiplication of allele or population frequencies to generate an RMP via the product rule ignores important considerations. Statistical independence of the GVPs is of primary concern. It is important to know if two markers arising from the same gene, for example, are linked. Other considerations include whether one should consider alleles or genotypes in the calculation and if pertinent SNP databases are large enough to produce accurate allele frequencies or assess linkage. In addition, if GVP analysis is to augment other methods such as STR profiling or analysis of mitochondrial DNA, then the statistical independence of these methods must be established. Finally, GVP analysis must be made robust and routine for use in forensic laboratories, presumably by the development of kits for sample extraction and analysis, and achieve acceptance by the legal community. 120

Future work in this area will focus on several areas. GVP analysis will be expanded more extensively into a variety of tissue types. Shed skin cells present in touch samples or fingerprints provides a particularly exciting area of future expansion. Challenges in this area include mixed samples, potentially small sample sizes, and the challenge of preserving other evidence present such as the fingerprint image for friction ridge analysis and DNA for STR profiling. Expansion of preliminary work in bone and tooth GVP analysis will also be a fruitful area of research. General expansion of SNP databases to include more samples to improve statistics and to include specific populations of interest will be important. Additionally, there are several questions related to practical application of GVP analysis. For example, it will be important to evaluate the robustness of the GVPs relative to environmental exposures such as sunlight and heat, and chemical exposure. Outside of the forensic arena, analysis of GVPs will have other applications. For example, SNPs are implicated as disease markers. What is the role of the proteins expressed from these mutated DNA and are there any benefits from detecting the associated SAPs in expressed proteins as a diagnostic or research tool? Information from GVPs will also find further application in archaeological studies where they would, for example, help elucidate population movements or understand familial groups from archaeological evidence. Finally, in a forensic setting, GVP analysis is not a substitute for DNA evidence. However, GVP analysis will be especially useful to augment DNA analysis if the sample DNA is missing or compromised. Once fully developed, GVP analysis is expected to provide tools to link individuals to crime scenes or objects, track yet unidentified individuals, identify unknown remains, and aid in mass casualty scenarios where there is a limited list of victims and family DNA is available.

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

8.

9.

Findley, K. A. Learning From Our Mistakes: A Criminal Justice Commission to Study Wrongful Convictions. Cal. WL Rev. 2001, 38, 333. Kouwenhoven, M. Focus On The Task At Hand: Contextual Bias in the Forensic Examination of Handwriting; University of Otago, 2019. Garrett, B. L. Constitutional Regulation of Forensic Evidence. Wash. & Lee L. Rev. 2016, 73, 1147. Norton, J.; Anderson, W. E.; Divine, G. Flawed Forensics: Statistical Failings of Microscopic Hair Analysis. Significance 2016, 13 (2), 26–29. De Macedo, C. Guilt by Statistical Association: Revisiting the Prosecutor’s Fallacy and the Interrogator’s Fallacy. The Journal of Philosophy 2008, 105 (6), 320–332. Leung, W.-C. The Prosecutor’s Fallacy—A pitfall in Interpreting Probabilities in Forensic Evidence. Medicine, Science and the Law 2002, 42 (1), 44–50. Ulery, B. T.; Hicklin, R. A.; Buscaglia, J.; Roberts, M. A. Accuracy and Reliability of Forensic Latent Fingerprint Decisions. Proceedings of the National Academy of Sciences 2011, 108 (19), 7733. Haber, L.; Haber, R. N. Error Rates for Human Latent Fingerprint Examiners. In Automatic Fingerprint Recognition Systems; Ratha, N., Bolle, R., Eds.; Springer: New York, 2004; pp 339−360. Spradley, M. K.; Jantz, R. L.; Robinson, A.; Peccerelli, F. Demographic Change and Forensic Identification: Problems in Metric Identification of Hispanic Skeletons*. Journal of Forensic Sciences 2008, 53 (1), 21–28. 121

10. Rogers, T.; Saunders, S. Accuracy of Sex Determination Using Morphological Traits of the Human Pelvis. Journal of Forensic Science 1994, 39 (4), 1047–1056. 11. Christensen, A. M.; Crowder, C. M. Evidentiary Standards for Forensic Anthropology*. Journal of Forensic Sciences 2009, 54 (6), 1211–1216. 12. Eckhart, L.; Lippens, S.; Tschachler, E.; Declercq, W. Cell Death by Cornification. Biochimica et Biophysica Acta (BBA)-Molecular Cell Research 2013, 1833 (12), 3471–3480. 13. Grisedale, K. S.; Murphy, G. M.; Brown, H.; Wilson, M. R.; Sinha, S. K. Successful Nuclear DNA Profiling of Rootless Hair Shafts: A Novel Approach. International Journal of Legal Medicine 2018, 132 (1), 107–115. 14. Brandhagen, D. M.; Loreille, O.; Irwin, A. J. Fragmented Nuclear DNA Is the Predominant Genetic Material in Human Hair Shafts. Genes 2018, 9 (12) 15. Microscopic Hair Comparison. In Forensic Science Reform; Koen, W. J., Bowers, C. M., Eds.; Academic Press: San Diego, 2017; Chapter 2, pp 25−55. 16. Just, R. S.; Irwin, J. A.; Parson, W. Mitochondrial DNA Heteroplasmy in the Emerging Field of Massively Parallel Sequencing. Forensic Science International: Genetics 2015, 18, 131–139. 17. Parker, G. J.; Leppert, T.; Anex, D. S.; Hilmer, J. K.; Matsunami, N.; Baird, L.; Stevens, J.; Parsawar, K.; Durbin-Johnson, B. P.; Rocke, D. M.; Nelson, C.; Fairbanks, D. J.; Wilson, A. S.; Rice, R. H.; Woodward, S. R.; Bothner, B.; Hart, B. R.; Leppert, M. Demonstration of Protein-Based Human Identification Using the Hair Shaft Proteome. PLoS One 2016, 11 (9), e0160653. 18. Mason, K. E.; Anex, D.; Grey, T.; Hart, B.; Parker, G. Protein-based Forensic Identification Using Genetically Variant Peptides in Human Bone. Forensic Sci. Int. 2018, 288, 89–96. 19. Wu, P.-W.; Mason Katelyn, E.; Durbin-Johnson Blythe, P.; Salemi, M.; Phinney Brett, S.; Rocke David, M.; Parker Glendon, J.; Rice Robert, H. Proteomic Analysis of Hair Shafts from Monozygotic Twins: Expression Profiles and Genetically Variant Peptides. PROTEOMICS 2017, 17 (13-14), 1600462. 20. Boutet, E.; Liberherr, D.; Tognolli, M.; Schneider, M.; Bairoch, A. UniProtKB/Swiss-Prot. Methods in Molecular Biology 2007, 406, 89–112. 21. Kitts, A.; Sherry, S. , The Single Nucleotide Polymorphism Database (dbSNP) of Nucleotide Sequence Variation. In The NCBI Handbook; McEntyre, J., Ostell, J., Eds.; National Center for Biotechnology Information (US): Bethesda, MD, 2002 [Updated 2011]. 22. Mason, K. E.; Paul, P. H.; Chu, F.; Anex, D. S.; Hart, B. R. Development of a Protein-based Human Identification Capability from a Single Hair. Journal of Forensic Sciences 2019, 64 (4), 1152. 23. Li, P.; Wu, G. Roles of Dietary Glycine, Proline, and Hydroxyproline in Collagen Synthesis and Animal Growth. Amino Acids 2018, 50 (1), 29–38. 24. Chu, F.; Mason, K. E.; Anex, D. S.; Jones, A. D.; Hart, B. R. Hair Proteome Variation at Different Body Locations on Genetically Variant Peptide Detection for Protein-Based Human Identification. Scientific Reports 2019, 9 (1), 7641. 25. Laatsch, C. N.; Durbin-Johnson, B. P.; Rocke, D. M.; Mukwana, S.; Newland, A. B.; Flagler, M. J.; Davis, M. G.; Eigenheer, R. A.; Phinney, B. S.; Rice, R. H. Human Hair Shaft Proteomic Profiling: Individual Differences, Site Specificity and Cuticle Analysis. PeerJ 2014, 2, e506.

122

26. Sterling, S.-A.; Mason, K. E.; Anex, D. S.; Parker, G. J.; Hart, B.; Prinz, M. Combined DNA Typing and Protein Identification from Unfired Brass Cartridges. Journal of Forensic Sciences 2019, 65 (5), 1475. 27. Yu, J.; Yu, D.-w.; Checkla, D. M.; Freedberg, I. M.; Bertolino, A. P. Human Hair Keratins. Journal of Investigative Dermatology 1993, 101 (1) (Supplement), S56–S59. 28. McKittrick, J.; Chen, P. Y.; Bodde, S. G.; Yang, W.; Novitskaya, E. E.; Meyers, M. A. The Structure, Functions, and Mechanical Properties of Keratin. JOM 2012, 64 (4), 449–468. 29. Rice, R. H.; Rocke, D. M.; Tsai, H.-S.; Silva, K. A.; Lee, Y. J.; Sundberg, J. P. Distinguishing Mouse Strains by Proteomic Analysis of Pelage Hair. Journal of Investigative Dermatology 2009, 129 (9), 2120–2125. 30. Wadsworth, C.; Buckley, M. Proteome Degradation in Fossils: Investigating the Longevity of Protein Survival in Ancient Bone. Rapid Communications in Mass Spectrometry 2014, 28 (6), 605–615. 31. The 1000 Genomes Project, C., A Global Reference for Human Genetic Variation. Nature 2015, 526, 68. 32. Wigginton, J. E.; Cutler, D. J.; Abecasis, G. R. A Note on Exact Tests of Hardy-Weinberg Equilibrium. The American Journal of Human Genetics 2005, 76 (5), 887–893. 33. Lek, M.; Karczewski, K. J.; Minikel, E. V.; Samocha, K. E.; Banks, E.; Fennell, T.; O’DonnellLuria, A. H.; Ware, J. S.; Hill, A. J.; Cummings, B. B.; Tukiainen, T.; Birnbaum, D. P.; Kosmicki, J. A.; Duncan, L. E.; Estrada, K.; Zhao, F.; Zou, J.; Pierce-Hoffman, E.; Berghout, J.; Cooper, D. N.; Deflaux, N.; DePristo, M.; Do, R.; Flannick, J.; Fromer, M.; Gauthier, L.; Goldstein, J.; Gupta, N.; Howrigan, D.; Kiezun, A.; Kurki, M. I.; Moonshine, A. L.; Natarajan, P.; Orozco, L.; Peloso, G. M.; Poplin, R.; Rivas, M. A.; Ruano-Rubio, V.; Rose, S. A.; Ruderfer, D. M.; Shakir, K.; Stenson, P. D.; Stevens, C.; Thomas, B. P.; Tiao, G.; TusieLuna, M. T.; Weisburd, B.; Won, H.-H.; Yu, D.; Altshuler, D. M.; Ardissino, D.; Boehnke, M.; Danesh, J.; Donnelly, S.; Elosua, R.; Florez, J. C.; Gabriel, S. B.; Getz, G.; Glatt, S. J.; Hultman, C. M.; Kathiresan, S.; Laakso, M.; McCarroll, S.; McCarthy, M. I.; McGovern, D.; McPherson, R.; Neale, B. M.; Palotie, A.; Purcell, S. M.; Saleheen, D.; Scharf, J. M.; Sklar, P.; Sullivan, P. F.; Tuomilehto, J.; Tsuang, M. T.; Watkins, H. C.; Wilson, J. G.; Daly, M. J.; MacArthur, D. G.Exome Aggregation Consortium Analysis of Protein-coding Genetic Variation in 60,706 Humans. Nature 2016, 536, 285. 34. Smith, A. V.; Thomas, D. J.; Munro, H. M.; Abecasis, G. R. Sequence Features in Regions of Weak and Strong Linkage Disequilibrium. Genome Research 2005, 15 (11), 1519–1534. 35. Pengelly, R. J.; Tapper, W.; Gibson, J.; Knut, M.; Tearle, R.; Collins, A.; Ennis, S. Whole Genome Sequences are Required to Fully Resolve the Linkage Disequilibrium Structure of Human Populations. BMC Genomics 2015, 16 (1), 666.

123

Chapter 8

Proteomics in the Analysis of Forensic, Archaeological, and Paleontological Bone Michael Buckley* Manchester Institute of Biotechnology, School of Earth and Environmental Sciences, University of Manchester, 131 Princess Street, Manchester M1 7DN, United Kingdom *E-mail: [email protected].

Proteomics is becoming ever more popular across a wide variety of disciplines, not only for medical and food-related science but also subject areas such as archaeology, paleontology, and more recently in forensics. As bone is the biological tissue that is the last to decompose, it is the most interesting in the study of proteome decomposition. For ancient bone, the application of proteomic techniques relates to the greater longevity of proteins over DNA, a phenomenon that has been utilized as a means of species identification of ancient bone fragments and the molecular phylogeny of enigmatic extinct species. However, one of the main advantages for forensic scenarios is the greater dynamics of the proteome over the genome. Although our earlier research investigated the decreasing complexity of the bone proteome through deep time (over hundreds of thousands of years), we have recently begun evaluating the changes that occur on more recent, forensic timescales. Published results clearly demonstrate changes in relative abundance of particular proteins that relate to biological age, yet remain informative even on archaeological timescales. However, it is the changes in posttranslational modifications that are often utilized to ensure endogeneity in ancient samples, which have recently been found of great use to forensics in their correlation with post-mortem interval estimation. This chapter discusses these and other advances in the application of proteomic methods in forensic science.

Introduction Proteomics, the study of collections of proteins within a tissue, is a discipline that emerged in the 1990s following the development of soft-ionization mass spectrometry (1) and has been applied to numerous scientific fields ranging from medical and food-related investigations (2, 3), to more applied subject areas such as studying ancient (4) or forensic remains (see current volume). As bone is the last tissue to decompose it has great value in studying long-term proteome decomposition, containing thousands of proteins, not only those specific to bone (e.g., serum proteins (5)). For © 2019 American Chemical Society

ancient tissues, one of the main interests in the application of proteomics has related to the greater longevity of proteins over DNA (6); the oldest uncontested ancient DNA sequences so far reported derives from permafrost bone ~0.7 Ma (7) whereas protein sequences have been matched from bone ~3.5 Ma using proteomic approaches (8, 9). The first application of proteomic methods to the study of ancient bone was by Ostrom et al. (4), reporting on the detection of osteocalcin (OC) from extinct bison (Bison priscus) via Matrix Assisted Laser Desorption Ionization Time of Flight Mass Spectrometry (MALDI-ToF-MS), which led on to other species such as horses (10) and the extinct camelid Camelops hesternus (11). The same team also were able to show the survival of the secondary structure of OC, even from specimens ~42,000 years old (10). These results were promising for the recovery of phylogenetic information from deep into the past, but a thorough study comparing the survival of OC with mtDNA was discouraging (6), finding greater success rates in the latter (a more informative molecule). Nevertheless, although intact OC (the intactness being important because the first component to degrade away was also the most variable in terms of sequence information (4)) could not compete with collagen, which was clearly present in all samples analyzed in the study by Buckley et al. (6), including those that failed DNA analysis; collagen was also known to survive long periods of time from various other stable isotope studies on Pleistocene remains (12). Subsequent analyses of collagen, first using bacterial collagenase to digest all but the telopeptides (ends of the chains) which managed to yield some but limited species information (13), then resorting to trypsin for studying peptides from the helical regions of collagen. Through using solid phase extraction, this latter approach was able to clearly show differences in collagen extracted from sheep (Ovis) and goat (Capra), two species with morphologically similar post-cranial remains that are particularly difficult for archaeologists to discriminate between, resulting in a viable method of species identification using a robust target molecule (14, 15). This has subsequently led on to the analysis of a wider range of domesticated and wild mammals likely used by past humans, beginning with terrestrial (16, 17), but also marine mammals (18, 19). More recently, and perhaps of greatest interest to forensic science has also been its use in identifying human remains (e.g., Figure 1; (20)) along with the remains of other locally extinct (21) or globally extinct species (22). More recently this area of research has developed to utilize machine learning, an approach which greatly reduces the subjectivity of biomarker identification (23). However, although this peptide mass fingerprint (PMF) approach is increasing in popularity, the standard proteomic workflow does offer more information (24) but at higher analysis time and costs with greater risks of false positive peptide matches. There are also many cases in wildlife forensics whereby this rapid means of species identification could be meaningfully implemented (e.g., (25)). Where forensic science is typically considered to be concerned with the use of analytical scientific methods to generate evidence for legal proceedings, this is usually in relation to human-based evidence. In wildlife forensics, the questions being asked may relate to the identification of the perpetrators, but they are more commonly considered to relate to the identification of the wildlife product in trade. Although there are many published methods for the species identification of animal tissues, the relatively fewer validated methods for wildlife forensics is thought to relate to the lack of a commercial market for the development and application of wildlife forensic tests (26). There is also the issue of the quality management of proposed wildlife forensic methods, similar to ISO17025 or GLP/GMP in traditional forensic science, which would be further complicated for samples that could potentially derive from mixtures (e.g., traditional Chinese medicine). Recently the Scientific Working Group on Wildlife Forensic Methods (SWGWILD) has worked with the Society for Wildlife Forensic Science to establish a set of standards and guidelines 126

which laboratories could choose to follow (rather than be required to do so). In terms of the techniques available, the most common approaches are either DNA profiling for species identification, which can also shed light on geographic origin to some extent (e.g., (27)), or radiocarbon dating for establishing the age (since death) of a specimen (e.g., (28)). Although proteomic information cannot tackle either of these questions with the same resolution, its main advantages are that it can be cheaper and more rapid than DNA-based approaches for species identification (e.g., identification of hundreds, potentially thousands of samples can be obtained within a few hours), but they can shed light on the particular tissue from which residues derive, and more specifically the approximate biological age of the specimens (discussed later in this chapter).

Figure 1. Example spectra of peptide mass fingerprints from an ancient bear (top) and ancient human (bottom) bone sample showing selected peptides annotated with their respective sequence differences (smal lettering used to highlight amino acid substitutions). In addition to species identification, proteomic techniques have recently taken advantage of the ever-increasing amount of genetic data becoming publicly available for attempting finer scale resolution, particularly in the study of genetically variant peptides. Although initially applied to hair (29), the study of these single amino acid polymorphisms in bone (30) has also widened the scope of this approach across forensic investigations. However, this is covered in detail elsewhere within this book (see Chapter 6) and so will not be discussed here. To date, research into proteomic analyses of ancient remains has been mainly used for species identification (14) with a few cases where, taking this one step further, molecular phylogenies have been reconstructed (31–33). However, one of the main advantages of proteomics for forensic applications is the greater dynamics of the proteome over the genome, which allows for the study of not only ontogenetic variations more readily but also degradation properties at a fine resolution, both of which will be discussed later in this chapter. In order to do so it must introduce the structure and development of bone.

127

Bone Structure Given that the bone proteome clearly changes throughout the lifetime of an individual, and through cycles of growth, fracture, and repair, proteomics offers a promising tool for forensic applications. These could particularly help age-at-death estimation (34) as well as post-mortem interval estimation (35), but potentially also the timing of related injuries if refined further. However, to consider its use for interpreting such information in a forensic scenario it is necessary to understand these developmental processes in bone growth and aging. Bone is one of the few biomineralized tissues in the human body, being a composite of ~70% inorganic phase and the remaining ~30% organic phase (36) – the former allowing the tissue to resist compression and the latter allowing it to resist torsion and tension. Its function in vertebrates is to lend protection to internal organs of the organism while also providing attachment sites for muscles and as levers for locomotion, but it also acts as a store for minerals, particularly calcium; in times of chronic calcium stress it is also a source of calcium (37). There are two main types of bone in which these roles are considered to differ, compact (also known as cortical) and spongy (also known as cancellous or trabecular) bone. Cortical bone is a dense form of bone made up of microscopic columns called osteons, forming concentric layers of osteoblasts and osteocytes that surround longitudinal vessels within (e.g., Haversian systems) the shafts of long bones; the osteons are connected to each other via Volkmann’s canals often running at obtuse angles to the Haversian canals and transmit blood from the periosteum into the back. Cancellous bone is more of an internal framework of struts that provide lightweight support at the ends of long bones and in the axial skeleton (i.e., the head and trunk of the body), being a more open and less dense type of material. Other tissues found in or associated with bone include the endosteum and periosteum (linings on the inside and outside of cortical bone respectively), as well as cartilage and bone marrow. The endosteum former is where bone resorption occurs, and the periosteum is where deposition of new bone takes place, but also contains the blood vessels, nerves and lymphatic vessels that nourish the bone). Marrow is a highly vascularized tissue containing hematopoietic stem cells (HSCs) and mesenchymal stem cells (MSCs) that create progenitor cells which differentiate into a variety of types. These not only include bone cells but also fat cells (adipocytes), muscle cells (myocytes), red blood cells (erythrocytes) and white blood cells (granulocytes). The main difference between the two types of marrow in bone (red and yellow) is the levels of their components, with red (also known as myeloid tissue) primarily producing HSCs that differentiate into cells generating red blood cells, white blood cells and platelets (the MSCs generate the other bone, fat, muscle and cartilage cell types), whereas some white blood cells develop in yellow marrow, but along with a much greater amount of fat cells with their carotenoids causing the yellow colouration. Of relevance to our ontogenetic variations in proteomic analyses of bone discussed later, at birth all bone cavities only contain red marrow, some of which (primarily in the medullary cavities of long bones) is converted to yellow marrow throughout the lifetime of an individual. At the molecular level, the inorganic phase, or mineral component, is predominantly dahllite (carbonated hydroxyapatite), which is largely composed of calcium phosphate but with detectable amounts of other elements (38). The organic phase is largely made up of proteins, particularly dominated by the structural protein collagen, which is also considered the most abundant protein in the animal kingdom (39). However, there are many other proteins known to be present that have roles in various activities, such as mineral co-ordination, the most abundant of which is osteocalcin (40).

128

Bone is formed via two processes of ossification, intramembranous ossification for the flat cranial bones and clavicles which form directly from sheets of mesenchymal cells, and endochondral ossification for the long bones which are developed upon cartilaginous matrices that serve as templates that become replaced. Longitudinal bone growth results from the activities of chondrocytes in the epiphyseal plate, the plate itself is divided into distinct zones that each have different phases of the chondrocyte lineage within them, resting, proliferating and hypertrophing chondrocytes, characterized by different sizes and shapes (41). After recruitment from the resting into the proliferative (dividing cells) zone, the cells subsequently hypertrophy and produce an abundant extracellular matrix based on a collagenous structure that becomes mineralized. Although there are more than 27 types of collagen known in vertebrates, in bone it is type 1 collagen, also known as collagen (I), that dominates the tissue; collagen (I) is composed of two identical and one genetically distinct alpha chains (each coiling in a left handed helix) that supercoils in a right-handed manner to form single tropocollagen molecules (42). These tropocollagen molecules line up headto-tail to form fibrils, which themselves combine to form repeating sets of fibers that add a level of flexibility to the structure of the tissue (43). The secreted osteoid matrix is made up of such collagen fibers and an amorphous ground substance made up of water and the several thousand other proteins loosely named ‘non-collagenous proteins’ (NCPs) present (5) that include glycosaminoglycans, proteoglycans and glycoproteins. In both the growth plate and osteoid, mineralization is associated with alkaline phosphatase activity (44) as well as many other proteins, such as chondrocalcin, proteoglycans, osteonectin, and osteocalcin (45). Bone growth and remodeling allows the tissue to maintain a mechanically competent skeleton (46) and remove areas of microdamage (47, 48). Earlier in life (i.e., during growth), bone formation exceeds resorption, whereas later in life this reverses, leading to a slow decline in bone mass with age (49). This is largely controlled by its three cell types, the osteoblasts, osteoclasts and osteocytes. Osteoblasts are a single layer of cells found on the surface of the bone, but can either be the matrixproducing cells or flattened lining cells (50), with parathyroid hormone (PTH) possibly acting as a switch between these types. Osteoblasts go on to become osteocytes, entrapped in the bone matrix during formation. They inhabit the lacunar-canalicular system and communicate with other osteocytes as well as the bone lining cells, with increasing evidence that implies they can act as mechano-receptors (51). The osteoclasts are large multinucleate cells that can resorb the equivalent amount of bone that is typically formed by 100 to 1,000 osteoblasts. They contact the bone via a ‘ruffled border’ and generate an acidic environment through the use of a proton pump (50). The regulation of osteoclast activity involves a variety of factors, but remains poorly understood. Interestingly, each of the three cell types produces different levels of proteins. Osteoblasts produce VEGFA promoting osteoblastogenesis, as well as a range of others (e.g., RANKL/OPG, M-CSF, SEMA3A) that regulate osteoclastogenesis. Osteocytes produce sclerostin which promotes osteoclast formation and inhibits osteoblast formation. Osteoclasts secrete factors that influence osteogenesis, including BMP6, CTHRC1, EFNB2, S1P among others. It is obvious that the measurements of each of these could in theory be useful to infer biological age in forensic cases, but much future work needs to be carried out in terms of validation, particularly with respect not only to differences between species (especially given that the majority of the discovery work is being carried out on animal models), but also in terms of reproducibility, potentially complicated by a range of disease states unknown to the forensic scientist carrying out the investigation (although it is possible that identification of such disease states from the proteome data (52) may assist with forensic analysis). 129

Methodologies for the Analysis of Bone Proteomes Although instrumentation for carrying out proteomic analyses has already been introduced elsewhere (53), including within this volume (Chapter 1), particularly those linked to electrospray ionization mass spectrometry, the methodology for extracting proteins from bone is typically less straightforward than for other non-mineralized tissues. This is due to the need to decalcify the tissue in order to better access the full range of proteins present. Optimum protocols for protein extraction from bone have been regularly re-evaluated with the advent of newer technologies. Comparisons between the uses of different acids have long been explored, particularly between the strong acids (e.g., hydrochloric acid; HCl) and the “softer” chelating agents, namely ethylenediaminetetraacetic acid (EDTA). The latter is preferred when attempting to yield greatest amounts of protein in order to avoid laboratory-induced hydrolysis (e.g., (54)) but for carbon isotope analyses the introduction of modern carbon should be avoided. It is for this reason that HCl has been the acid of choice for radiocarbon and stable isotope analyses (e.g., (55)). It is also currently the main approach to the ever-increasing applications of collagen fingerprinting for the species identification of archaeological bone (14, 18), given that the identification of a particular species (e.g., ancient hominin remains) may warrant radiocarbon dating of limited material (20). Rather than measuring the ratios of isotopes from particular elements for absolute geological age determination or for dietary inferences in which the state of the extracted proteins is not of primary concern, proteomic techniques including collagen fingerprinting rely of the extraction of proteins with minimal damage. This damage can be readily noticed through the observation of unwanted peptide hydrolysis and/or post-translational modification (PTM), such as the deamidation of asparagine (Asn) and glutamine (Gln) residues (15, 56–58), but the oxidation of methionine (Met) residues as well as a large range of others (carbonylation, nitration, racemization and glycation). However, it is the deamidation changes that have been proposed as a means to quantify decay, whereby they are thought to act as ‘molecular clocks’ because they occur physiologically in a time-dependent manner (59, 60). Deamidation is the non-enzymatic hydrolysis of the side-chain amide group in Asn and Gln that forms the carboxylic acids aspartic and glutamic acid respectively (Figure 2). Although the rate at which these spontaneously decay is known to be linked with primary sequence structure, other factors such laboratory and environmental conditions are also known to affect them (61, 62). As the measure of deamidation is potentially more useful in forensics than archaeology or paleontology, and reproducibility more important, method development is required in these early stages of forensic proteomics. Therefore, a recent re-evaluation of the use of a range of acids for decalcification (hydrochloric, formic, nitric, acetic and EDTA) found it important to use either EDTA or formic acid (54), along with optimization of other aspects of the protein extraction process (e.g., guanidine hydrochloride followed by enzymatic digestion in ammonium acetate). Although these refined approaches only retrieved 64% of the number of unique peptides observed under the most aggressive HCl-based approaches, with a 28% drop in unique proteins observed, the extent of deamidation was reduced by ~50% (54). However, it is important to note that even extraction protocols for other applications, such as for the isotope analysis methods described above, can be utilized for proteomic information of potential use to the forensic investigator (63), to the extent that even a surface rubbing has proven capable of yielding proteomic information (64). However, as mentioned already, the main difference between the large number of “palaeoproteomic” studies on ancient bone and the aspirations to carry out such proteomics in a forensic manner is greatly influenced by the issues with reproducibility. For example, Procopio 130

et al. (34) clear demonstrated that substantial variations between re-analyses of the same sample aliquots will be observed if analyzed in different instrumental runs. Nonetheless, the extent to which this hinders the forensic science with regards relative change of the desired signals (e.g., relative abundance of one protein to another within that dataset), and whether these variations can be readily mitigated needs much further evaluation.

Figure 2. Racemization of L-aspartic acid (L-Asp) and deamidation of L-asparagine (L-Asn) via a succinimide intermediate (Asu) to form D-aspartic acid (D-Asp).

Ancient and Forensic Proteins for Species Identification As the sequences of proteins are coded for by DNA, they retain a level of genetic information that can be specific to a particular evolutionary lineage. However, the taxonomic level at which this is can be heavily influenced by the functional constraints on them. In the case of bone, many of these functions relate to structural organization but as discussed in the last section, there are many proteins present that are more widely involved in transport processes, such as the circulation of blood serum, etc. Through the identification of 6,202 unique peptides, Jiang et al. (63) matched 2,479 unique proteins in their study of modern dog bone, including more than 40 which were specific to bone. By far the most abundant proteins were intracellular, but closely followed by extracellular proteins and membrane proteins with the most abundant functions relating to binding or catalytic activity. Those of structural importance were nearly as frequently observed, but collagen (I) dominates the tissue. Transport and signaling proteins are more typically observed at lower abundances, etc. However, the majority of these proteins have not been observed in ancient remains, or even those of forensic ages, although the methods of choice have not been as aggressive in terms of acid use, and the amount of starting material clearly has an impact on the results (8). The majority of studies report ~30-100 unique proteins, with collagens dominating the tissue as would be expected; although it is possible to use the proteome to inform the investigator of the tissue type, this will not be discussed here in relation to our evaluation of forensic bone proteomics. Although collagen possesses a highly repetitive amino acid sequence motif of a glycine (Gly) residue followed by two amino acids that are often the imino acids proline (Pro) and/or hydroxyproline (Hyp), due to its very 131

large size (~300 kDa) and apparent reduced level of conservation of the imino acid component of this motif in one of its three chains (the alpha 2 chain in most vertebrates), it has found particular value in retrieving genus- or species-level determinations of faunal remains as discussed at the start of this chapter. However, a range of NCPs are also usually observed in ancient samples, whether they are bone matrix proteins such as secreted phosphoprotein 24, vitamin K-dependent protein, and matrix Gla protein, small leucine-rich proteoglycans such as chondroadherin, biglycan or lumican, or plasm proteins such as albumin, fetuin, high temperature requirement serine protease A1, pigment epithelium-derived factor, vitamin D-binding protein, tetranectin, thrombospondin, or the various coagulation factors and complement component proteins (65). Yet the extent to which each of the proteins identified in ancient and decomposed tissues can be used for species determination (and phylogenetic investigations) can vary by orders of magnitude (65) and relates directly to their functional properties. However, in the case of forensic applications it might be more appropriate to focus on proteins that are more ubiquitous throughout an organism, such as albumin or fetuin. In forensic applications there will always be the need to understand the statistical rigor of the proposed identifications, which will relate to the analytical method used. The MALDI PMF approach discussed at the start offers what should be statistically a more robust form of identification, but more taxonomically limited by comparison to standard LC-MS/MS approaches (see (66) for comparison). Even so, the PMF approach is more restricted by the need to create a reference database of spectra that themselves could be influenced by various specimen processing differences (e.g., defatting, bleaching, etc.). The LC-MS/MS approach on the other hand is substantially much better at retrieving matches to the more variable, and therefore useful peptide sequences (e.g., Figure 3), but as it is using probability-based matching algorithms it becomes reliant on the presented sequences themselves. The consideration of this aspect of forensic proteomic sequence matching has been discussed elsewhere (53), but will undoubtedly progress over the coming years as more truly forensic applications are tested.

Figure 3. Typical protein identification via peptide spectra matching showing example of pig bone (from (34)) showing (A) good sequence coverage (>50%), (B) an example of the interpretation of a peptide spectrum match informing the score given, and (C) comparison of this particular peptide sequence with those of other animals. 132

Proteome Changes with Ontogeny The ability to observe distinct signals relating to biological age from a known part of the skeleton is not surprising given the expected changes during skeletal growth and aging in adulthood (i.e., continuous remodeling that changes throughout life), but whether or not these signals are detectable in the forensic and archaeological records (i.e., remain after degradation) is of interest to many scholars. Although morphological analyses are often used for the estimation of biological age, with skeletal remains this becomes substantially more difficult with advancing biological age. This is why other microstructural properties have been widely studied over the past few decades, with recent interest in the application of proteomics. Even within the same tissue remarkable differences in the rates of bone turnover is known to occur, with adult cortical bone having a rate of ~2-3% per year compared with ~28% per year for trabecular bone (67). Upon studying proteomes (with five replicates at each site) between the middle and ends of various pig long bones, Procopio et al. (34) found a much greater level of variation within each skeletal element than between the elements (i.e., the tibiae midshaft samples were more similar to the mid-metatarsus and scapula samples than to the ends of the tibiae; the same also true of the femur samples studied). Yet there were some noticeable differences between elements, such as a higher relative abundance of some bone-specific proteins including lumican, PSG2 and DERM in the tibiae as opposed to the femora (34). Of greatest interest to the forensic scientist is perhaps the observation of trends in the relative abundances of particular serum proteins with biological age of the individual (Figure 4); the proteins alpha-1 antitrypsin (A1AT) and chromogranin A (CMGA) were seen to increase with age, whereas the protein fetuin-A (FETUA; also known as alpha-HS-2-glycoprotein in humans) was seen to decrease with biological age (34), the latter of which was also shown to remain in human burials several hundred years old (68) and even much older archaeological cattle several thousand years old (69) albeit all with very small sample sizes.

Figure 4. Normalized protein abundances for the three proteins of interest that increase (A1AT & CMGA) or decrease (FETUA) with biological age, with their 3D structures. The relationship of the abundance of two of these proteins with biological age remains unclear, as A1AT is a type of enzyme inhibitor that protect tissues from the enzymes of inflammatory cells, and chromogranin A is the precursor of peptides that modulate neuroendocrine function. However, the decreasing abundance of fetuin-A (A2HSG) does make biological sense because it takes its name from being notably more abundant in fetal blood (70), likely becoming entrapped within remodeled bone through its known interaction with calcium (71). Its function has long been thought to relate to the regulation of endochondral ossification through the inhibition of mineralization (72), therefore affecting long bone growth. However, the relative abundances of this protein after long bone growth is complete need further investigation with a much wider range of samples. 133

The potential for biological age determination via proteomic analysis in this manner has the potential to complement other lines of forensic evidence, both for the analysis of human remains and in wildlife forensics. However, the most common approaches that use macroscopic changes in bone and teeth are known to have much reduced accuracy in older age with different skeletal elements better at different ages (73). Other methods include the racemization of aspartic acid, lead accumulation, the build-up of collagen cross-links and DNA methods such as telomere shortening and epigenetic modifications and error rates from ~3-15 years. Therefore, for the rather complex proteomic approach of relative abundance measures would need to deliver on age estimates that are within the lower end of the error range already provided by other biochemical methods.

Proteome Changes with Length of Burial Time Although there have been a substantial number of ‘palaeoproteomes’ (i.e., ancient proteomes) now reported, there have been few that consider the process of proteome degradation over time. This is perhaps not unexpected, or at least not as readily feasible as might be initially presumed, given the high level of differences between analytical runs, even with repeats of the same peptide digest aliquots run on the same instrument months apart (34). Yet of those that have attempted this (74), it is clear that it is the bulk of the proteins not associated with either mineral or collagenous matrix that are leached out of the tissue first, followed by the apparent survival of some serum proteins that are likely trapped within the bone mineral (e.g., fetuin and albumin; although this may relate to initial relative abundance) and collagen-associated proteins, and finally the structural and highly-insoluble collagen itself. Prieto-Bonete et al. (75) investigated the presence or absence of a large number of proteins in human remains from two post-mortem interval (PMI) categories, less than 12 years PMI and 12-20 years PMI, and made conflicting observations in that alpha-HS-2-glycoprotein was observed in the former but not in the latter grouping. This is particularly surprising given the much greater relative survival of this protein deep into the archaeological record (i.e., nearly one million years (74)) by comparison to many of the other proteins observed. It should also be noted that although CMGA was not observed, A1AT (coded for by the SERPIN A1 gene in humans) was, despite a more complex proteome being reported. Although the sample sizes in this latter study are much greater than in each of the earlier studies, with the additional limitations on working with animal remains in some countries, such as in the UK, there is notably a need for much further research in this area. Similar to models of DNA degradation, collagen decay was initially modeled on the premise of random chain scission, resulting in fragment lengths that would decrease with burial time (76). However, more recent studies (e.g., (77)) triggered by earlier observations of the apparent survival of relatively intact collagen from Pleistocene sub-fossils (78) indicate that it should not be considered in this way. It appears that although some collagen will undoubtedly hydrolyze with time, and likely be lost to the environment, the collagen molecules that remain within ancient tissues remain relatively intact, although note that their ends, or telopeptides, which do not have the typical collagenous sequence motif, do rapidly degrade away (13), at least until present below a critical mass – which is apparently ~1% dry weight yield (77). It may be that glycation-based cross-links aid the preservation of partial collagen fragments deep into the fossil record, and some proteomic studies are only just beginning to look into this (79). However, reports of collagen from the fossil record, such as those from dinosaur remains (80, 81) remain contested, and have been proposed to be crosscontamination (82, 83), with other supporting evidence such as morphology being linked with biofilm and our limited understanding of diagenesis (84). The predominantly globular NCPs present 134

in bone are more likely to degrade in a scission-like manner more similar to DNA degradation, with the exception that there would be particular parts more likely to preserve than others such as hydrophobic core regions, even more so those that interact with the mineral phase of bone as in OC (Figure 5).

Figure 5. 3D structure of osteocalcin showing (A) its alpha helical structure and (B) the mineral-binding region and tightly packed hydrophobic core (with green spheres representing calcium ions), produced using 1Q8H pdb file. (Created at https://www.rcsb.org/structure/1Q8H). However, it is the PTMs that are often utilized to ensure endogeneity in ancient samples, which have recently been found useful for forensics in their correlation with PMI. Although the concept of using PTMs as ‘molecular timers’ has existed for decades (60), it has only been within the last decade that they have been studied in ancient materials (note that (4) in the first use of proteomic techniques for ancient samples evaluated the integrity of tertiary structure by carrying out laboratory reduction and alkylation of the cysteine residues). Notably, Cleland et al. (85) do make a thorough evaluation of PTMs in the remains of the extinct moa (palaeognath bird) that does include several additional modifications such as fucosylation, and go as far as suggesting the effects of diagenesis even on the biological PTMs, specifically proline hydroxylation. If this were validated it could also add another means to evaluating PMI in forensics, but would require much further investigation. The promise of a new technique for PTM estimation is perhaps the simplest of the forensic challenges that proteomics can tackle, that is after species identification (which arguably utilize proteomic methods but is not reliant on the analysis of whole proteomes, e.g., peptide mass fingerprinting can be carried out following the isolation of a single protein such as collagen).

Future Directions of Forensic Bone Proteomics and Concluding Remarks Clearly there is great potential for the use of proteomic techniques in the study of forensic bone, whether this is for the more simplistic analysis of peptide mass fingerprints for the species identification of animal tissues, including discrimination from human remains but also the potential to span a wide range of other vertebrate animals, or the more sophisticated inferences made from in-depth proteome analyses, such as biological and ‘geological’ (PMI) age estimation or potentially even individual-level identification through genetically variant peptides. However, the latter is fraught with particular issues that relate to column carry over (8). This could have much more serious 135

effects on forensic analyses than in archaeology or paleontology, if not heeded early on. The other major issue so far, at least with in-depth proteomics, is that of limited sample number. This is linked to method development, as the lack of reproducibility even with repeated analyses of the exact same aliquots several months later (34) makes such proteomics a costly endeavor for those that do not manage their own instrumentation. In fact, the poor reproducibility of sample reanalysis is perhaps the greatest weakness to forensic proteomics at the moment, which will need to be addressed through greater use of calibration. Nevertheless, the use of proteomics will need to achieve a standard of admissibility in court for it to become a widely used forensic method, with similar governmental bodies around the world requiring proofs of validation of the technique (in the UK methods to be validated are guided by the Forensic Science Regulator’s Codes of Practice and Conduct, but accreditation to the international standard (ISO 17025) is required where external scrutiny is provided by the United Kingdom Accreditation Service). This is the main difference between the archaeological and forensic sciences, with the former having a much greater range of potential method variations, etc. With this in mind it should be noted that there have only been very few forensic studies on human bone, alongside method development studies carried out on pig bone. Although the use of proteomics more generally, regardless of tissue type, could be considered validated as a tool (53), variations in methods used in treating different tissue types could significantly alter the observations. let alone the issues with same sample reproducibility mentioned above. Therefore, at some point in the near future, the forensic proteomic community will undoubtedly need to establish validation criteria. More fundamentally this would need to carefully consider the variability between labs, clearly highly problematic given that some of the proteins that our lab and others routinely see in bone (e.g., fetuin) and which we propose as a biological age marker, apparently is not seen by others. Understanding the reasons behind such discrepancies is vital before establishing whether or not they can still serve as biomarkers for those labs/methods that retrieve them.

Acknowledgments The author declares no conflicts of interest and acknowledges support from the Royal Society in funding a University Research Fellowship.

References 1. 2. 3.

4.

5. 6.

Anderson, N. L.; Anderson, N. G. Proteome and Proteomics, New Technologies, New Concepts, and New Words. Electrophoresis 1998, 19, 1853–1861. Hochstrasser, D. F. Clinical and Biomedical Applications of Proteomics. Proteome Research, New Frontiers in Functional Genomics; Springer, 1997; pp 187−219. Wheat-Grain Proteomics, The Full Compliment of Proteins in Developing and Mature Grain. Wheat Gluten Proceedings of the 7th International Workshop Gluten 2000, Bristol, UK, 2-6 April 2000;Rathmell, W., Skylas, D., Bekes, F., Wrigley, C., Eds.; Royal Society of Chemistry, 2000. Ostrom, P. H.; Schall, M.; Gandhi, H.; Shen, T. L.; Hauschka, P. V.; Strahler, J. R.; Gage, D. A. New Strategies for Characterizing Ancient Proteins Using Matrix-Assisted Laser Desorption Ionization Mass Spectrometry. Geochim. Cosmochim. Acta 2000, 64, 1043–1050. Jiang, X.; Ye, M.; Jiang, X.; Liu, G.; Feng, S.; Cui, L.; Zou, H. Method Development of Efficient Protein Extraction in Bone Tissue for Proteome Analysis. J. Proteom. Res. 2007, 6, 2287–2294. Buckley, M.; Anderung, C.; Penkman, K.; Raney, B. J.; Gotherstrom, A.; Thomas-Oates, J.; Collins, M. J. Comparing the Survival of Osteocalcin and mtDNA in Archaeological Bone From Four European Sites. J. Archaeol. Sci. 2008, 35, 1756–1764.

136

7.

8.

9.

10.

11.

12.

13. 14.

15.

16. 17.

18. 19.

20.

21. 22. 23.

Librado, P.; Gamba, C.; Gaunitz, C.; Der Sarkissian, C.; Pruvost, M.; Albrechtsen, A..; Fages, A.; Khan, N.; Schubert, M.; Jagannathan, V.; Serres-Armero, A. Ancient Genomic Changes Associated With Domestication of the Horse. Science 2017, 356, 442–445. Buckley, M.; Lawless, C.; Rybczynski, N. Collagen Sequence Analysis of Fossil Camels, Camelops and cf Paracamelus, from the Arctic and Sub-Arctic of Plio-Pleistocene North America. J. Proteom. 2019, 194, 218–225. Rybczynski, N.; Gosse, J. C.; Harington, C. R.; Wogelius, R. A.; Hidy, A. J.; Buckley, M. Mid-Pliocene Warm-period Deposits in the High Arctic Yield Insight Into Camel Evolution. Nat. Commun. 2013, 4, 1550. Ostrom, P. H.; Gandhi, H.; Strahler, J. R.; Walker, A. K.; Andrews, P. C.; Leykam, J.; Stafford, T. W.; Kelly, R. L.; Walker, D. N.; Buckley, M.; Humpula, J. Unraveling the Sequence and Structure of the Protein Osteocalcin from a 42 ka Fossil Horse. Geochim. Cosmochim. Acta 2006, 70, 2034–2044. Humpula, J. F.; Ostrom, P. H.; Gandhi, H.; Strahler, J. R.; Walker, A. K.; Stafford, T. W., Jr; Smith, J. J.; Voorhies, M. R.; Corner, R. G.; Andrews, P. C. Investigation of the Protein Osteocalcin of Camelops hesternus, Sequence, Structure and Phylogenetic Implications. Geochim. Cosmochim. Acta 2007, 71, 5956–5967. Palmqvist, P.; Gröcke, D. R.; Arribas, A.; Fariña, R. A. Paleoecological Reconstruction of a Lower Pleistocene Large Mammal Community Using Biogeochemical (δ13C, δ15N, δ18O, Sr, Zn) and Ecomorphological Approaches. Paleobiology 2003, 29, 205–229. Buckley, M.; Collins, M.; Thomas-Oates, J. A Method of Isolating the Collagen (I) Alpha 2 Chain Carboxytelopeptide for Species Identification in Bone Fragments. Anal. Biochem. 2008, 374, 325–334. Buckley, M.; Collins, M.; Thomas-Oates, J.; Wilson, J. C. Species Identification by Analysis of Bone Collagen Using Matrix-Assisted Laser Desorption/Ionisation Time-Of-Flight Mass Spectrometry. Rapid Commun. Mass Spectrom. 2009, 23, 3843–3854. Buckley, M.; Kansa, S. W.; Howard, S.; Campbell, S.; Thomas-Oates, J.; Collins, M. Distinguishing Between Archaeological Sheep and Goat Bones Using a Single Collagen Peptide. J. Archaeol. Sci. 2010, 37, 13–20. Buckley, M.; Kansa, S. W. Collagen Fingerprinting of Archaeological Bone and Teeth Remains from Domuztepe, South Eastern Turkey. Archaeol. Anthropol. Sci. 2011, 3, 271–280. von Holstein, I. C.; Ashby, S. P.; van Doorn, N. L.; Sachs, S. M.; Buckley, M.; Meiri, M.; Barnes, I.; Brundle, A.; Collins, M. J. Searching for Scandinavians in Pre-Viking Scotland, Molecular Fingerprinting of Early Medieval Combs. J. Archaeol. Sci. 2014, 41, 1–6. Buckley, M.; Fraser, S.; Herman, J.; Melton, N.; Mulville, J.; Pálsdóttir, A. H. Species Identification of Archaeological Marine Mammals Using Collagen Fingerprinting. J. Archaeol. Sci. 2014, 41, 631–641. Evans, S.; Godino, I. B.; Álvarez, M.; Rowsell, K.; Collier, P.; de Goodall, R. N. P.; Mulville, J.; Lacrouts, A.; Collins, M. J.; Speller, C. Using Combined Biomolecular Methods to Explore Whale Exploitation and Social Aggregation in Hunter–Gatherer–Fisher Society in Tierra del Fuego. J. Archaeol. Sci. Rep. 2016, 6, 757–767. Brown, S.; Higham, T.; Slon, V.; Pääbo, S.; Meyer, M.; Douka, K.; Brock, F.; Comeskey, D.; Procopio, N.; Shunkov, M.; Derevianko, A. Identification of a New Hominin Bone from Denisova Cave, Siberia Using Collagen Fingerprinting and Mitochondrial DNA Analysis. Sci. Rep. 2016, 6, 23559. Hufthammer, A. K.; Arntsen, L.; Kitchener, A. C.; Buckley, M. Grey Whale (Eschrichtius robustus) in Norwegian Waters 2000 Years Ago. Palaeogeog. Palaeoclim. Palaeoecol. 2018, 495, 42–47. Buckley, M.; Cosgrove, R.; Garvey, J.; Prideaux, G. J. Identifying Remains of Extinct Kangaroos in Late Pleistocene Deposits Using Collagen Fingerprinting. J. Quatern. Sci. 2017, 32, 653–660. Gu, M.; Buckley, M. Semi-Supervised Machine Learning for Automated Species Identification by Collagen Peptide Mass Fingerprinting. BMC Bioinform. 2018, 19, 241.

137

24. Buckley, M.; Gu, M.; Shameer, S.; Patel, S.; Chamberlain, A. High-Throughput Collagen Fingerprinting of Intact Microfaunal Remains, a Low-Cost Method For Distinguishing Between Murine Rodent Bones. Rapid Commun. Mass Spectrom. 2016, 30, 1–8. 25. Linacre, A.; Tobe, S. S. An Overview to the Investigative Approach To Species Testing in Wildlife Forensic Science. Invest. Genet. 2011, 2, 2. 26. A Review of Wildlife Forensic Science and Laboratory Capacity To Support the Implementation and Enforcement of CITES; Ogden, R., Mailley, J., Eds.; Review commissioned by the Secretariat of the Convention on International Trade in Endangered Species of Wild Fauna and Flora and review undertaken by the United Nations Office on Drugs and Crime, 2016. 27. Ogden, R.; Linacre, A. Wildlife Forensic Science, a Review of Genetic Geographic Origin Assignment. Forensic Science International. Genetics 2015, 18, 152–159. 28. Cerling, T. E.; Barnette, J. E.; Chesson, L. A.; Douglas-Hamilton, I.; Gobush, K. S.; Uno, K. T.; Wasser, S. K.; Xu, X. Radiocarbon Dating of Seized Ivory Confirms Rapid Decline in African Elephant Populations and Provides Insight Into Illegal Trade. Proc. Nat. Acad. Sci. 2016, 113, 13330–13335. 29. Parker, G. J.; Leppert, T.; Anex, D. S.; Hilmer, J. K.; Matsunami, N.; Baird, L.; Stevens, J.; Parsawar, K.; Durbin-Johnson, B. P.; Rocke, D. M.; Nelson, C. Demonstration of Protein-Based Human Identification Using the Hair Shaft Proteome. PloS ONE 2016, 11, e0160653. 30. Mason, K. E.; Anex, D.; Grey, T.; Hart, B.; Parker, G. Protein-Based Forensic Identification Using Genetically Variant Peptides in Human Bone. Forensic Science International 2018, 288, 89–96. 31. Buckley, M. A Molecular Phylogeny of Plesiorycteropus Reassigns the Extinct Mammalian Order ‘Bibymalagasia’. PloS ONE 2013, 8, e59614. 32. Buckley, M. Ancient Collagen Reveals Evolutionary History of the Endemic South American ‘Ungulates’. Proc. R Soc B 2015, 282, 1806. 33. Welker, F.; Collins, M. J.; Thomas, J. A.; Wadsley, M.; Brace, S.; Cappellini, E.; Turvey, S. T.; Reguero, M.; Gelfo, J. N.; Kramarz, A.; Burger, J. Ancient Proteins Resolve the Evolutionary History of Darwin’s South American Ungulates. Nature 2015, 522 (7554), 81–84. 34. Procopio, N.; Chamberlain, A. T.; Buckley, M. Intra-and Interskeletal Proteome Variations in Fresh and Buried Bones. J. Proteom. Res. 2017, 16, 2016–2029. 35. Procopio, N.; Williams, A.; Chamberlain, A. T.; Buckley, M. Forensic Proteomics for the Evaluation of the Post-Mortem Decay in Bones. J. Proteom. 2018, 177, 21–30. 36. Biltz, R. M.; Pellegrino, E. D. The Chemical Anatomy of Bone, I. A Comparative Study of Bone Composition in Sixteen Vertebrates. J. Bone Joint Surg. 1969, 51, 456–466. 37. Quarles, L. D. Endocrine Functions of Bone in Mineral Metabolism Regulation. J. Clin. Inv. 2008, 118, 3820–3828. 38. Rey, C.; Combes, C.; Drouet, C.; Glimcher, M. J. Bone Mineral, Update on Chemical Composition and Structure. Osteoporosis Int. 2009, 20, 1013–1021. 39. Shoulders, M. D.; Raines, R. T. Collagen Structure and Stability. Ann. Rev. Biochem. 2009, 78, 929. 40. Ducy, P.; Desbois, C.; Boyce, B.; Pinero, G.; Story, B.; Dunstan, C.; Smith, E.; Bonadio, J.; Goldstein, S.; Gundberg, C.; Bradley, A. Increased Bone Formation in Osteocalcin-Deficient Mice. Nature 1996, 382, 448. 41. Sims, N. A.; Gooi, J. H. Bone Remodeling, Multiple Cellular Interactions Required for Coupling of Bone Formation and Resorption. Sem. Cell Develop. Biol. 2008, 19, 444–451. 42. Wood, E.; Smith, C. Biological Molecules; Chapman and Hall: London, 1991. 43. Currey, J. D. Bones, Structure and Mechanics; Princeton University Press: Princeton, NJ, 2002; Vol. xii, p 436. 44. Siffert, R. S. The Role of Alkaline Phosphatase in Osteogenesis. J. Exp. Med. 1951, 93, 415–426.

138

45. Termine, J. D.; Wientroub, S.; Fisher, L. W. Methods for the Isolation and Testing of Skeletal Tissue Matrix Proteins. In Methods of Calcified Tissue Preparation; Dickson, G. R., Ed.; Elsevier, 1984; pp 547−563. 46. Wolff, J. The Law of Bone Transformation; Hirschwald: Berlin, 1892. 47. Burr, D. B.; Forwood, M. R.; Fyhrie, D. P.; Martin, R. B.; Schaffler, M. B.; Turner, C. H. Bone Microdamage and Skeletal Fragility in Osteoporotic and Stress Fractures. J. Bone Min. Res. 1997, 12 (1), 6–15. 48. Burr, D. B.; Martin, R. B.; Schaffler, M. B.; Radin, E. L. Bone Remodeling in Response to in vivo Fatigue Microdamage. J. Biomech. 1985, 18, 189–200. 49. Dequeker, J. Bone and Ageing. Ann. Rheumatic Dis. 1975, 34 (1), 100. 50. Baron, R. Molecular Mechanisms of Bone Resorption an Update. Acta Orthopaedica Scandinavica 1995, 66, 66–70. 51. Spadaro, J. A. Mechanical and Electrical Interactions in Bone Remodeling. Bioelectromagnetics 1997, 18, 193–202. 52. Hanash, S. Disease Proteomics. Nature 2003, 422, 226. 53. Merkley, E. D.; Wunschel, D. S.; Wahl, K. L.; Jarman, K. H. Applications and Challenges of Forensic Proteomics. Forensic Sci. Int. 2019. 54. Procopio, N.; Buckley, M. Minimizing Laboratory-Induced Decay in Bone Proteomics. J. Proteom. Res. 2016, 16, 447–458. 55. Schoeninger, M. J.; Moore, K. Bone Stable Isotope Studies in Archaeology. J. World Prehistory 1992, 6, 247–296. 56. Wilson, J.; van Doorn, N. L.; Collins, M. J. Assessing the Extent of Bone Degradation Using Glutamine Deamidation in Collagen. Anal. Chem. 2012, 84, 9041–9048. 57. Doorn, N. L.; Wilson, J.; Hollund, H.; Soressi, M.; Collins, M. J. Site‐Specific Deamidation of Glutamine, a New Marker of Bone Collagen Deterioration. Rapid Commun. Mass Spectrom. 2012, 26, 2319–2327. 58. Simpson, J. P.; Penkman, K. E. H.; Demarchi, B.; Koon, H.; Collins, M. J.; Thomas-Oates, J.; Shapiro, B.; Stark, M.; Wilson, J. The Effects of Demineralisation and Sampling Point Variability on the Measurement of Glutamine Deamidation in Type I Collagen Extracted from Bone. J. Archaeol. Sci. 2016, 69, 29–38. 59. Robinson, A. B.; Rudd, C. J. Deamidation of Glutaminyl and Asparaginyl Residues in Peptides and Proteins. In Current Topics In Cellular Recognition; Einsele, G., Ricken, W., Seilacher, A., Eds.; Academic Press: New York, 1974; Vol. 8, pp 247−294. 60. Robinson, N. E.; Robinson, A. B. Molecular Clocks. Proc. Nat. Acad. Sci. USA 2001, 98, 944–9. 61. Robinson, A. B.; McKerrow, J. H.; Cary, P. Controlled Deamidation of Peptides and Proteins, an Experimental Hazard and a Possible Biological Timer. Proc. Nat. Acad. Sci. USA 1970, 66, 753–757. 62. Robinson, N. E.; Robinson, A. B. Prediction of Protein Deamidation Rates from Primary and ThreeDimensional Structure. Proc. Nat. Acad. Sci. USA 2001, 98, 4367–4372. 63. Wadsworth, C.; Buckley, M. Characterization of Proteomes Extracted Through Collagen-Based Stable Isotope and Radiocarbon Dating Methods. J. Proteom. Res. 2017, 17, 429–439. 64. Cicatiello, P.; Ntasi, G.; Rossi, M.; Marino, G.; Giardina, P.; Birolo, L. Minimally Invasive and Portable Method for the Identification of Proteins in Ancient Paintings. Anal. Chem. 2018, 90, 10128–10133. 65. Buckley, M.; Wadsworth, C. Proteome Degradation in Ancient Bone, Diagenesis and Phylogenetic Potential. Palaeogeog. Palaeoclim. Palaeoecol. 2014, 416, 69–79. 66. Buckley, M. Species Identification of Bovine, Ovine and Porcine Type 1 Collagen, Comparing Peptide Mass Fingerprinting and LC-Based Proteomics Methods. Int. J. Mol. Sci. 2016, 17, E445.

139

67. Hedges, R. E.; Clement, J. G.; Thomas, C. D. L.; O’Connell, T. C. Collagen Turnover in the Adult Femoral Mid‐Shaft, Modeled from Anthropogenic Radiocarbon Tracer Measurements. Am. J. Phys. Anthropol. 2007, 133, 808–816. 68. Sawafuji, R.; Cappellini, E.; Nagaoka, T.; Fotakis, A. K.; Jersie-Christensen, R. R.; Olsen, J. V.; Hirata, K.; Ueda, S. Proteomic Profiling of Archaeological Human Bone. Open Sci. 2017, 4, 161004. 69. Procopio, N.; Chamberlain, A. T.; Buckley, M. Exploring Biological and Geological Age-Related Changes Through Variations in Intra-and Intertooth Proteomes of Ancient Dentine. J. Proteom. Res. 2018, 17, 1000–1013. 70. Dziegielewska, K.; Matthews, N.; Saunders, N.; Wilkinson, G. α2HS–Glycoprotein is Expressed at High Concentration in Human Fetal Plasma and Cerebrospinal Fluid. Fetal Diag. Ther. 1993, 8, 22–27. 71. Suzuki, M.; Shimokawa, H.; Takagi, Y.; Sasaki, S. Calcium‐Binding Properties of Fetuin in Fetal Bovine Serum. J. Exp. Zool. 1994, 270, 501–507. 72. Brylka, L.; Jahnen-Dechent, W. The Role of Fetuin-A in Physiological and Pathological Mineralization. Calc. Tissue Int. 2013, 93, 355–364. 73. Adserias-Garriga, J.; Thomas, C.; Ubelaker, D. H.; Zapico, S. C. When Forensic Odontology Met Biochemistry, Multidisciplinary Approach in Forensic Human Identification. Archiv. Oral Biol. 2018, 87, 7–14. 74. Wadsworth, C.; Buckley, M. Proteome Degradation in Fossils, Investigating the Longevity of Protein Survival in Ancient Bone. Rapid Commun. Mass Spectrom. 2014, 28, 605–615. 75. Prieto-Bonete, G.; Pérez-Cárceles, M. D.; Maurandi-López, A.; Pérez-Martínez, C.; Luna, A. Association Between Protein Profile and Postmortem Interval in Human Bone Remains. J. Proteom. 2019, 192, 54–63. 76. Collins, M. J.; Riley, M. S.; Child, A. M.; Turner-Walker, G. A Basic Mathematical Simulation of the Chemical Degradation of Ancient Collagen. J. Archaeol. Sci. 1995, 22, 175–83. 77. Dobberstein, R. C.; Collins, M. J.; Craig, O. E.; Taylor, G.; Penkman, K. E. H.; Ritz-Timme, S. Archaeological Collagen, Why Worry About Collagen Diagenesis? Archaeol. Anthropol. Sci. 2009, 1, 31–42. 78. Tuross, N.; Eyre, D. R.; Holtrop, M. E.; Glimcher, M. J.; Hare, P. E. Collagen in Fossil Bones. In Biogeochemistry of Amino Acids; Hare P. E., Hoering, T. C.; King, K., Jr. , Eds.; Wiley: New York, 1980; pp 193−201. 79. Hill, R. C.; Wither, M. J.; Nemkov, T.; Barrett, A.; D’Alessandro, A.; Dzieciatkowska, M.; Hansen, K. C. Preserved Proteins from Extinct Bison Latifrons Identified by Tandem Mass Spectrometry, Hydroxylysine Glycosides are a Common Feature of Ancient Collagen. Mol. Cell. Proteom. 2015, 14, 1946–1958. 80. Asara, J. M.; Schweitzer, M. H.; Freimark, L. M.; Phillips, M.; Cantley, L. C. Protein Sequences from Mastodon and Tyrannosaurus rex Revealed by Mass Spectrometry. Science 2007, 316, 280–285. 81. Schweitzer, M. H.; Zheng, W.; Organ, C. L.; Avci, R.; Suo, Z.; Freimark, L. M.; Lebleu, V. S.; Duncan, M. B.; Vander Heiden, M. G.; Neveu, J. M.; Lane, W. S. Biomolecular Characterization and Protein Sequences of the Campanian Hadrosaur B. canadensis. Science 2009, 324, 626. 82. Buckley, M.; Walker, A.; Ho, S. Y.; Yang, Y.; Smith, C.; Ashton, P.; Thomas-Oates, J.; Cappellini, E.; Koon, H.; Penkman, K.; Elsworth, B.; Ashford, D.; Solazzo, C.; Andrews, P.; Strahler, J.; Shapiro, B.; Ostrom, P.; Gandhi, H.; Miller, W.; Raney, B.; Zylber, M. I.; Gilbert, M. T.; Prigodich, R. V.; Ryan, M.; Rijsdijk, K. F.; Janoo, A.; Collins, M. J. Comment on “Protein Sequences from Mastodon and Tyrannosaurus rex Revealed by Mass Spectrometry”. Science 2008, 4, 33c. 83. Buckley, M.; Warwood, S.; van Dongen, B.; Kitchener, A. C.; Manning, P. L. A Fossil Protein Chimera, Difficulties in Discriminating Dinosaur Peptide Sequences from Modern Cross-Contamination. Proc. R. Soc. B 2017, 284, 20170544.

140

84. Kaye, T. G.; Gaugler, G.; Sawlowicz, Z. Dinosaurian Soft Tissues Interpreted as Bacterial Biofilms. PLoS ONE 2008, 3, e2808. 85. Cleland, T. P.; Schroeter, E. R.; Schweitzer, M. H. Biologically and Diagenetically Derived Peptide Modifications in Moa Collagens. Proc. R. Soc. B 2015, 282, 20150015.

141

Chapter 9

Proteomics for Microbial Forensics Eric D. Merkley* Chemical and Biological Signature Sciences Division, Pacific Northwest National Laboratory, Richland, Washington 99352, United States *E-mail: [email protected].

Mass spectrometry-based proteomics is a powerful tool for the detection and characterization of microbes of forensic and national security concern. Both targeted and untargeted proteomics methods have been developed for the taxonomic classification of unknown microbial samples. Targeted proteomics assays can be designed for specific microbes of security concern. Untargeted, library-based matrix-assisted laser desorption-ionization time of flight (MALDITOF) mass spectrometry is now extensively used in the medical field for microbial identification. Several research groups have developed data analysis pipelines for organism identification/taxonomic classification from untargeted liquid chromatography-tandem mass spectrometry (LC-MS/MS) data, an application which has some overlap with the field of metaproteomics. This chapter reviews these organism identification techniques. Whole genome sequencing/ metagenomics is becoming the standard for organism identification/classification. Using examples from published literature, this chapter highlights several examples of how proteomics approaches can provide information that cannot be acquired from DNA sequencing alone, such as distinguishing laboratory-adapted bacteria from closely related wild isolates, and characterizing the growth medium of bacteria and the host cells of virus particles.

Introduction Whole genome sequencing is an extremely powerful method for identifying and classifying microorganisms, even to the strain level. The question naturally arises as to why proteomics methods for essentially the same task are needed. But in truth, proteomics methods for microbial forensics should not be viewed as in competition with genomics. Rather, proteomics can confirm, extend, and complement nucleic-acid based methods. Genomics reveals genotype—the molecular potential of an organism—but proteomics provides a window into phenotype—the actual molecular functions being carried out by a cell, tissue, or microbial community. Genomic characterization of an isolate is straightforward, but taxonomic characterization of an environmental sample—a metagenome—is © 2019 American Chemical Society

still challenging. False positives are common in metagenomics (1), and different bioinformatics tools can report different results for the same sequencing data (2). Thus, proteomics methods can serve as an important independent validation. Analyzable proteins may be present in samples that lack DNA or whose DNA is degraded. Due to cell size or biases in extraction or amplification, metaproteomics can give a different view into the composition of a complex microbial community than metagenomics. In all these ways, proteomics provides data that is related to genomics data but complementary. Efforts to identify and distinguish biological organisms by mass spectrometry have a long history, with the use of proteins for this effort dating back to the 1990s. By 2001, enough work had been done for Fenselau and Demirev to publish a review (3) of methods attempted to that date, which emphasized instrumental and sample preparation methods for MALDI-TOF measurements on whole bacterial cells. Indeed, the work of Catherine Fenselau and coworkers was important in establishing the field: exploring the application of intact protein masses from MALDI-TOF-MS (4), peptide mass fingerprinting (5, 6), MS/MS based bottom-up proteomics (5), and even top-down LC-MS/MS proteomics and proteogenomics concepts (7) to microbial forensics. In this chapter, we will first review three mass spectrometric approaches to identification or typing of microbes: targeted LC-MS/MS methods based on biomarkers, intact protein patternmatching of MALDI-TOF-MS spectra, and identification by shotgun LC-MS/MS proteomics and database searching. Next, we will consider several published examples of the value added by proteomics in characterizing sources and methods of production, including the differentiation of laboratory-adapted populations and wild isolates of Yersinia pestis, identification of the culture conditions of poxviruses, and the composition and sourcing of bacterial cell culture media. In general, these methods are still in development, and not in common use in criminal-justice forensics. Indeed, much of the research is motivated by biosecurity and public health considerations, but the examples below are suggestive that these methods could be useful in traditional forensic settings.

Microbial Identification Microbial Identification by Targeted Proteomics A brief explanation of the differences between targeted and untargeted proteomics is found in Chapter 1 of this book. Specific peptides (targets) are designated as biomarkers of the organism of interest, and, using a triple quadrupole mass spectrometer in the multiple reaction monitoring (MRM), are the only analytes measured. If untargeted proteomics is loosely analogous to whole genome sequencing, then targeted proteomics is loosely analogous to real-time polymerase chain reaction (RT-PCR). Targeted proteomics, as the name implies, detects and measures only those peptides designated as targets. As with RT-PCR, targets must be selected carefully and subjected to rigorous validation. However, unlike RT-PCR there is no amplification or increase in the number of analyte molecules present. Therefore, while RT-PCR analysis is possible for even a few copies of a DNA molecule under favorable conditions, proteomics requires much more material. (Affinitybased targeted assays can sometimes detect target proteins at low ng/mL concentrations (8); this is highly dependent on application and methodological details.) Targeted proteomics is also more readily multiplexed than PCR. Due to the chromatographic separation of digested peptides, as many as 200 targets can be analyzed in a single LC-MS/MS experiment. Identification of these marker peptides usually involves a combination of bioinformatics analysis and untargeted proteomics measurements in order to assure the specificity (uniqueness to the target 144

organism), sensitivity, and detectability of the targeted biomarker peptides. The bioinformatics analysis is used to determine whether candidate protein or peptide markers appear in other known genome sequences. A key challenge here is often to identify peptides whose sequences differ between high-risk pathogens and closely related species that are not traditionally regarded as biothreats, such as Bacillus anthracis and B. cereus, or Yersinia pestis and Y. pseudotuberculosis. However, identical peptide sequences can occur in unrelated organisms and proteins by chance, so these comparisons should be made across large sequence databases wherever possible. In addition, the selected peptides must be expressed under the conditions targeted by the assay. For instance, B. anthracis exists in the environment as a spore, so a peptide from a protein expressed only in vegetative cells may not be useful for environmental detection. Finally, peptide markers must be amenable to mass spectrometry in terms of size, charge, hydrophobicity, and absence of amino acid residues that are prone to oxidation or deamidation during sample preparation. Some of these properties can be predicted a priori from sequence (9, 10), but the most straightforward way to address these issues is empirically, with an untargeted proteomics measurement. This is why the development of a targeted proteomics assay often begins with an untargeted proteomics study. Gekenidis et al. carried out such an untargeted study to discover markers distinguishing subspecies of Salmonella enterica (11). For more security-relevant pathogens, targets have been identified and targeted proteomics assays created for B. anthracis spore-specific proteins (12), B. anthracis small acid-soluble proteins (13), and for Yersinia pestis (14), These last two studies also included an antibody capture step prior to protein extraction and digestion, which improves sensitivity by separating out the targeted cells from the background matrix (which could be, e.g., soil, food, or blood serum or plasma). Antibody capture combined with MRM targeted mass spectrometry can lead to limits of detection of a few thousand colony-forming units, only a couple of orders of magnitudes less sensitive than RT-PCR assays. Targeted proteomics methods have also been used extensively for protein toxins (15–17) because of their low limit of detection. As with organism identification, some toxins, such as ricin, have near neighbors (proteins from the same organism with high sequence identity) that must be accounted for when selecting marker peptides (8, 18, 19). Targeted assays are the most sensitive, but they are expensive and time-consuming to develop and validate. More importantly, the ever-growing nature of public sequence databases means that the uniqueness of peptide markers must constantly be reevaluated. The addition of a new sequence to a database due to the discovery and genomic sequencing of a new species or strain could mean that a peptide once thought to be unique is now shared between the target strain and the new strain. This phenomenon is referred to as signature erosion. Precisely this scenario was reported by Pfrunder et al. (20) in the course of their work to identify species-specific peptide markers for members of the Bacillus cereus group. These researchers were forced to repeat their bioinformatics analysis because of additions to the database during the course of only a few months. Microbial Identification with MALDI-TOF MS Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry is a technique in which the sample is applied as a dried spot to a target along with a matrix, an organic molecule that absorbs light at the wavelength of the laser. Laser irradiation causes analyte molecules in the sample to form gas-phase ions that can then be measured in a time-of-flight mass analyzer (see Chapter 1). MALDI-TOF-MS can be used on peptides or intact proteins, as well as other molecules. Due to the speed of the measurement and the ability to spot many samples on one target, it can be higher throughput than liquid chromatography electrospray ionization methods. But since it 145

cannot be coupled on-line to chromatography, it is most often used with samples that do not require chromatographic separations. MALDI-TOF-MS is now routinely and extensively used for clinical identification of bacteria and fungi (21–23). Within the clinical realm, the technique is rapid, simple to implement, and reliable for clinical samples (with certain caveats). A clinical isolate is cultured according to a standard protocol, and material from a microbial colony is directly spotted on the MALDI target. (Alternatively, a simple extraction/lysis procedures can be applied.) Because samples are not digested with trypsin, this can be considered a “top-down” method (see Chapter 1). The observed spectrum consists of ribosomal proteins and other relatively small (below 20,000 molecular weight), common proteins. A “fingerprint” derived from this spectrum (24, 25) is compared to a library of such fingerprints and a similarity score is calculated. A sufficiently high score is taken as a species identification, which aids clinicians in prescribing appropriate antibiotics. MALDI-TOF-MS is faster than traditional culturebased methods because results can be obtained with shorter culture times, and cheaper than DNAbased methods because fewer specialized reagents are required. In routine clinical application, MALDI-TOF-MS microbe identification is accurate, but the need for complete libraries is a key limitation of the technique. Wiesner et al. (21) reviewed studies investigating the accuracy of species identification, and found that most studies report greater than 90% accuracy for many clinically relevant species, given that the subject species’ fingerprint is included in the library. These authors assert that some earlier studies reported lower accuracy because some species were missing from the library, and that this defect has been largely remedied for many clinically relevant species. Typically, these libraries are proprietary and sold by the instrument manufacturers. MALDI-TOF-MS microbe identification is also relevant for biosecurity-related bacteria. Several independent studies have investigated the use of MALDI-TOF-MS for biothreat agent bacteria (26–29). All of these studies agree on two main findings. First, a complete library is essential for accurate identifications. Given a complete enough library, accurate identification is often possible. One prominent vendor markets a security-relevant database which must be purchased separately. Use of this library provides substantial improvement for the represented organisms. Still greater improvement was provided by the use of expanded in-house libraries (26). Unfortunately, relatively few laboratories have access to the regulated and highly dangerous pathogens that are traditionally considered as bioterrorism threats, such as those on the Select Agents and Toxins List in the United States. Given the use of the commercial security-relevant database and/or customized expanded databases, most bioterrorism threat bacteria can be identified, and lower success rates for certain organisms were again attributed to deficiencies in the libraries (26). The second major finding is that for certain groups of closely related biothreat bacteria, such as Yersinia, Bacillus, and Brucella, reliable species-level identification not possible. For instance, Yersinia pestis samples yielded close matches for both Y. pestis and Y. pseudotuberculosis (26–28). In one case study, a clinical isolate of Y. pseudotuberculosis was misidentified as Y. pestis by MALDI-TOF-MS, but other culture-based and DNA-based methods made the correct identification (30) Members of phylogenetically tight genera like Bacillus and Brucella (26–29) exhibited high similarity scores for more than one species. In some cases, such as Brucella, for which all the species are dangerous, a genus level identification is sufficient to initiate a public health response (29). However, the appropriate response for Bacillus anthracis, the causative agent of anthrax, is very different than the response to B. thuringensis, a generally harmless soil organism.

146

These studies suggest that MALDI-TOF-MS identification of high-risk pathogens or potential bioterror agents could be a useful forensic and biosecurity tool, but that the results need to be interpreted in the light of knowledge of the contents of the library and of bacterial phylogeny. Although the technique has been evaluated for potential bioweapons pathogens, most validation studies have focused on clinical use cases that include a relatively small number of such pathogens. Further work will be necessary to evaluate the use of this technique in a criminal-justice forensics context which includes the possibility of completely unknown samples. Microbial Identification with LC-MS/MS Proteomics An organism’s DNA is a record of its phylogeny, (i.e., its evolutionary history and relatedness to other organisms) and since the genome codes for the proteome, phylogenetic information is also contained in the sequences of digested tryptic peptides. This logic suggests that untargeted proteomics can be used to identify organisms. Since protein-coding genes are under evolutionary pressure to conserve their function, their sequences change more slowly than those of non-coding regions, and diversity is lower. Since many DNA-based methods rely on non-coding sequences that evolve more rapidly, one expects a priori that DNA analysis will be more successful at distinguishing closely related strains, and indeed this is the case. Still, taxonomic characterization by bottom-up proteomics is possible. Several research groups have developed and characterized software tools for this task, such as ABOid (31–34), TCUP (35), MiCid (36, 37), and an unnamed organism identification algorithm by Jarman and coworkers (38). The basic approach is the same in all these studies. Bottom-up proteomics data is first analyzed by database search against a database consisting of a large list of microbial proteomes (either sequential or simultaneously), confident peptide-spectrum matches (PSMs) to each species in the database are tallied, and the peptide evidence for organisms in the database is assessed. All of these tools have been reported to successfully identify unknown bacterial samples. For instance, the TCUP tool had identification rates of 90% and higher on samples of Escherichia coli, Pseudomonas aeruginosa, Staphylococcus aureus, and Streptococcus pneumoniae (35). MiCid (36) was tested on multiple archived LC-MS/MS datasets for E. coli, Mycobacterium tuberculosis, Salmonella enterica, Yersinia pestis, Y. pseudotuberculosis, and Shewanella oneidensis; and new datasets for E. coli, P. aeruginosa, and S. enterica. At the genus and species levels, the correct organism was always the most frequently identified. Jarman et al. (38) analyzed 844 datasets from 41 different organisms, with an overall correct identification rate of 99.1% at the genus level and 83.4% at the species level. However, the various approaches differ in the choice of database used in database search, in which peptides are considered, in the way in which peptide evidence is used to support the identification of a particular taxon, and in how the results are statistically assessed (Table 1). As we will discuss below, these considerations are critical for the correct interpretation of such results in forensic applications. Database Selection The interpretation of the results of a shotgun proteomics organism identification procedure depends the composition of the sequence database used, and on the phylogenetic relationships between the organisms represented. For true unknowns, the most comprehensive database possible is desirable in order to avoid false negatives: a microbe that is not present in the database cannot be identified. However, the pitfalls of searching too large a database are well known in the proteomics community (see Chapter 1): long computational times, and an increased chance of false PSMs that requires more stringent score thresholds to preserve the same false discovery rate, and thus lower 147

sensitivity. A more subtle problem occurs when a taxon that has many members is represented in the database by only one. If the correct organism is missing, the closest relative present in the database will likely be the best match. This can lead to the erroneous conclusion that a strain- or subspecieslevel identification has been made, when in truth multiple strains may match equally well, indicating that classification is possible only to the species or genus level (39). This is demonstrated by the frequency of matches to multiple strains in the strain-level data presented for MiCid, for instance (see Table 5 in Alves et al. (36)). Alves et al. (36) put it succinctly: “If we are certain that the correct microorganism is present in the database, we should interpret the results as microbial identification. On the other hand, if we are sure that the correct microorganism is absent from the database, we may interpret the results as microbial classification.” In forensic cases, when the presence of the correct organism in the database is unknown, a conservative approach would seem to dictate that the results be regarded as classification, not identification. Table 1. Comparison of LC-MS/MS Shotgun Proteomics Approaches to Microbial Identification/Classification* ABOid What database was used?

TCUP

• Various; typically • Nonredundant all available proteins from bacterial genomes GenBank

MiCid

Jarman et al.

• 460 complete • Nonredundant genomes (190 peptides from NCBI bacterial genera)

• Top-ranked Which peptides peptide-spectrum are used to assign matches (PSMs) • Peptides unique taxonomy? to a taxon

• Top-ranked PSMs • Least common ancestor (LCA) peptides

• All high-ranking PSMs (not just topranked)

• Strong (statistically infrequent) peptides with respect to a taxon

• Exact match to sequence in database How are peptides • Hits to topmatched to taxa? ranked taxon removed from other taxa iteratively (winner-takes-all)

• All peptides matched to database by BLAST; 90% sequence identity required • Peptides assigned to (LCA) • Adjustment for expected number of LCA peptides

• Contribution of spectrum to organism score weighted by PSM score • Clustering by shared peptides at each taxonomic level

• Tree-based algorithm that alternates between taxonomy and clustering by shared peptides

• Expectation value calculated from perspectrum E-values • E-values can be calculated at any taxonomic level

• P-value from goodness-of-fit test • P-values can be calculated at any taxonomic level

What are the decision criteria for determining the presence of a taxon?

• Organisms with more unique peptides than • Unspecified expected number of false PSMs

* See text for discussion and references.

Choice of Peptides to Use in Identification Algorithms One obvious choice is to use all confidently identified peptides as input to an organism identification algorithm, as is done by ABOid. To account for the uncertainty present in even 148

confident PSMs, other tools used more sophisticated approaches. Rather than direct sequence matching, TCUP uses BLAST to align peptides to the search database and counts all matches with >90% sequence identity. MiCid, which is more closely integrated with the database search, uses multiple high-ranking PSMs per spectrum, rather than only the top-ranking PSM. Both of these approaches increase the number of peptides considered and the number of organisms to which a peptide is assigned. Similar to increasing the size of the database, increasing the number of peptides considered broadens the search space. This is important for forensic applications because the broader the search space the more unbiased the approach can be considered. Mapping Peptides to Taxa—“Organism Inference” Once the presence of a given peptide sequence has been established, it must be determined how it bears on the question of taxonomy. This concept is related to the protein inference problem (see Chapter 1): many peptide sequences are found in more than one protein, and in more than one organism. A simple approach is to count all peptides in all organisms in which they occur, such that closely related species receive a similar score. A slightly more sophisticated approach, which takes more advantage of the information contained in peptide sequences, is to use only unique peptides (i.e., those that come from only a single organism or protein) for scoring. However, the use of unique peptides poses fundamental problems. As more and more sequences are generated and sequence databases become larger and larger, the fewer unique peptides per organism remain. The concept of uniqueness presumes that the set of sequences used for analysis is complete. This is not a valid assumption, since new organisms and strains are constantly being sequenced and added to the database—the process of signature erosion defined above. Many algorithms for protein inference in general proteomics are based on parsimony: i.e., finding the smallest list of proteins that explains the observed peptide evidence. The analogous approach for what might be called organism inference is the least common ancestor (LCA) algorithm. Instead of identifying a single organism, the LCA approach reports the lowest (least basal) taxon which contains all known examples of that sequence. TCUP uses a version of this algorithm. However, the LCA algorithm implicitly assumes that the taxonomic tree as presently understood is phylogenetically accurate—an assumption that is known to be untrue in many cases. Further, the problems of uniqueness and signature erosion also apply to the LCA approach. To counteract the problem of signature erosion, and to take better advantage of the information contained in the many peptide sequences that occur mostly, but not exclusively, in a single taxon, Jarman et al. (38) introduced a measure of the statistical frequency of a peptide sequence (within the set of proteomes/organisms considered) called the peptide strength. Peptide strength is a function of the number of groups in the database (a group could be a taxon at any level or a protein family of related toxins) and the number of groups containing the given peptide. They showed that strong peptides are resistant to signal erosion, and that the number of strong peptides per group is a robust measure of the presence of that group in the dataset. Decision Criteria Decision criteria—a standardized set of rules for determining the meaning and relevance of the observed results of a test—are a key part of establishing a defensible forensic method (see the chapter by Jarman and Merkley in this book). In the context of shotgun proteomics organism classification, this means not only determining what metric will be used to evaluate the presence of an organism in a sample, as discussed above, but also determining the critical values of that metric. Both MiCid 149

and Jarman et al. use statistical parameters assigned at the taxon level. MiCid uses a taxon-level expectation value (or E-value) that is calculated from the E-values of the individual PSMs assigned to that taxon. Jarman et al. tally the strong peptides per taxon and apply a statistical hypothesis test called a goodness-of-fit test. Under the null hypothesis test that taxa in the database are not present in the sample, incorrect peptides will be distributed uniformly across taxa. If a given taxon is present (alternative hypothesis), it will have a number of hits that is above this background. The goodnessof-fit test produces a p-value statistic for each taxon at the desired taxonomic level. To ensure that the most specific identification possible is made, the test is applied iteratively to the set of taxa with low p-values in the previous level. For example, if only two genera are found, only species from those genera are tested in the next level. To account for the presence of multiple closely related taxa, the traditional taxonomic groupings are alternated with “supergenus” and “superspecies” groups that cluster organisms by shared tryptic peptides. The presence of clear decision criteria for these tools greatly expands their potential for use in forensic cases, because it addresses the Daubert requirement (40) for controlled processes and allows the empirical measurement of error rates. Future Directions: Increasingly Complex Mixtures The body of work referenced above clearly demonstrated that LC-MS/MS proteomics can be used for taxonomic characterization of pure bacterial samples. The evidence also suggests that simple mixtures are also amenable to this type of analysis. For example, for mixed samples of up to four organisms, TCUP’s estimates of the relative abundance of proteins were also approximately correct for ratios in the range of 1:1 to 1:4, and MiCid was recently expanded to account for mixtures (36). It is not yet clear how well such methods will work for complex environmental samples, that is, for true metaproteomes. The impact of the LC-MS/MS approach would be greatly increased if accurate identifications of pathogens of security concern could be made in samples with a complex background, and this is an important research direction. A step in this direction was recently taken by Potgeiter et al. (41), who used de novo peptide identification (42, 43) to produce a preliminary peptide list. Although far less accurate than database search, enough correct PSMs were made to give an indication of sample taxonomy and allow a smaller database to be constructed. Searching this smaller database yielded similar numbers of peptide hits as more traditional metaproteomic approaches, such as searching a matched metagenome. Applying similar strategies to forensic proteomics is an active area of research in our laboratory. The ability to identify and taxonomically characterize unknown samples of arbitrary complexity with statistical confidence would be a major milestone for forensic proteomics. Literature to date suggests that all the pieces needed to accomplish this feat are actively being developed. Palmblad and coworkers have taken a more empirical approach (44–47). They used unannotated (unidentified) spectra from standard reference samples, and directly compared spectra from the reference datasets to spectra from unknowns. This is similar to spectral library searching except that the peptide that give rise to the library spectra remain unknown—no attempt to identify the peptides is made. Instead, the important information is the empirical association of the spectrum with a known sample. Spectral similarity scores are calculated between all spectra in the library and all spectra in the unknown sample. The distribution of spectral similarity scores is used to determine the quality of the match between the sample and the unknown. This method has been used for several animal species (mostly in food quality inspection context), but not yet for microbes (44–47). A great advantage of this method is that it is as applicable to unsequenced organisms as it is to sequenced ones, provided an initial isolate can be obtained to create a reference dataset. An evaluation of how useful this method is for microbial identification will have to wait for further data. 150

Characterizing Methods of Production Distinguishing Laboratory-Adapted or Laboratory-Grown Pathogenic Bacteria from Wild Isolates: Serial Passaging of Yersinia pestis Wild Isolates and Classification by Machine Learning One of the challenges in microbial forensics is differentiating intentionally introduced agents that constitute evidence of a crime or attack from the naturally occurring microbiome. Every pathogen comes from somewhere, and all of the agents traditionally considered biological threats are naturally present in some environments. Anthrax exists as a disease of livestock and wildlife in many areas. Yersinia pestis is endemic in many regions globally, including the American Southwest, existing in a sylvatic life cycle that includes fleas and rodent hosts such as prairie dogs. A highly publicized example was the 2002 case of two tourists who were hospitalized in New York City and diagnosed with bubonic plague (48). As a plague epidemic in a city the size of New York would be a major public health emergency, this event naturally generated great concern, and even raised the possibility that a biological attack had occurred. However, careful case history and a series of genomic analyses from near the tourists’ home in New Mexico, where plague is endemic, established that the patients had been exposed near their home and traveled to New York City before becoming symptomatic. In this case it was possible to reach this conclusion because comparison samples were available. Without these, genomics might not have been helpful. In fact, whole genome sequencing efforts have historically not shed much light on the problem of natural versus anthropogenic pathogens. A few proteomics studies, on the other hand, are beginning to suggest that patterns of protein expression changes can distinguish wild isolates from laboratory-adapted pathogens. Bacteria change their expression patterns as a regulated response to changing environmental conditions, and as an adaptive (evolutionary) response to changing selective pressures. Both mechanisms are potentially at play when an organism is moved from its natural environment and cultured in a laboratory. Leiser et al. (49) compared the protein expression patterns of two wild isolates of Yersinia pestis with their laboratory-adapted descendants. Wild isolates were obtained from plague-infected fleas collected at field sites in Texas and Arizona. Fleas were homogenized and the homogenate plated on two rounds of selective media to allow growth of Y. pestis. Colonies were analyzed by PCR to confirm their identity. The resulting isolates were used to inoculate twelve replicate liquid cultures. This biomass was considered passage zero (P0). It actually represents three laboratory cultivation steps, but this is the minimum number that allows confident identification of the pathogen. Each of the twelve replicates were then serially passaged by periodically inoculating fresh medium with the previous culture. Sixty serial passages were performed, amounting to roughly 750 generations. Whole genome sequencing and LC-MS/MS proteomics were performed on samples from P0 and from P60. No universal genomics changes were found, although loss of virulence determinants such as the pCD1 plasmid and the pgm (pigmentation) locus were observed in some lineages. (Loss of virulence in general and of the pgm locus in particular is known to be common when passaging Y. pestis). There were mutational hotspots, genes that exhibited mutations in multiple evolved lineages, but none were universal. In addition, measurements of metabolites revealed that the levels of the polyamine putrescine were elevated in the laboratory-adapted populations. Many protein expression differences between P0 cultures and P60 cultures were observed. A key point here is that these differences cannot be attributed solely to a regulated physiological response by the bacterium to being grown in laboratory medium, because the experimental conditions were 151

the same at both time points. Rather, the differences must represent some form of adaptation that is not obvious from analysis of mutations in protein-coding genes, possibly epigenetic changes or mutations in regulatory elements. Key categories of proteins with strong expression changes included amino nitrogen metabolism (glutamine synthetase, glutamate dehydrogenase, and carbamoyl phosphate synthetase), virulence factors (the attachment invasion locus protein Ail), membrane/cell surface proteins, and others. Many of these changes can be rationalized as adaptations to growth on nutrient-rich (and especially nitrogen-rich) laboratory medium. Having established that protein expression changes do occur upon laboratory adaptation, this research team next sought to establish signatures of laboratory adaptation that could be used to classify Y. pestis samples as wild isolates or laboratory-adapted strains. In addition to the serially passaged samples, they collected proteomics data from several additional wild isolates grown in several different media. Previously acquired proteomics data from a number of published Y. pestis proteomics studies was also retrieved from a proteomics data archive (50). In total, data from 137 different cultures was assembled, representing eight wild isolates and thirteen laboratory-adapted populations, two different laboratory facilities and at least five different growth media. These data were analyzed using the MaxQuant database search/relative abundance calculation software (51, 52), and protein abundances were either normalized by range or converted to binary/presence absence data. To better classify samples, the authors turned to machine learning. The problem can be stated as: given a sufficient number of classified example entities, learn (i.e., discover) rules for correctly classifying additional unknown entities based on their properties. In this case, the entities are proteomics data sets that are known to be from either laboratory-adapted or wild isolates, and the properties are the measures of protein abundance derived from the proteomics measurements. The particular method used in this study is called a logistic regression classifier (53, 54). It is well suited for applications where there are a large number of input variables (called features, in this case, the abundance values of around 2000 proteins), and a classifier based on a small number of nonredundant (that is, uncorrelated) features is sought. The classifier is essentially an equation that uses the abundance values of selected proteins to calculate the probability that a new data set came from a laboratory-adapted sample. The accuracy of the classifier was estimated by an internal cross-validation method. This method involves randomly dividing the samples into ten equal-sized sets. Ninety percent of the data is used to calculate (or train) the classifier, and the remaining ten percent is used to test it. This process is repeated for each of the ten sets. The whole exercise can then be repeated with as many new sets of ten random partitions as desired. Two classifiers were created, one using protein abundance values and one using the binary presence/absence data. The abundance-based classifier had an overall accuracy of 99.5%. The presence/absence based classifier had an overall abundance of 98.8%. Interestingly, several of the proteins selected as discriminating features were also found to be differentially expressed in the earlier serial passaging study. Because the data came from archived sources, some experimental variables like laboratory facilities used, culture media, sample preparation procedures, and instrumentation were correlated with the wild/laboratory status of the samples. The authors investigated whether this fact influenced the classifier by retraining the classifier to predict analysis facility, and culture medium. Facility was predicted with greater than 98 % accuracy and culture medium was predicted with similarly high accuracy. Taken together, these results suggest that the overall approach is able to correctly classify wild and laboratory-adapted Y. pestis. This is a promising approach for answering a number of forensic 152

questions and should be pursued further. However, it is not without its limitations. The validation that was done in work published to date was internal validation. A more stringent approach is to use external validation, in which the classifier is tested on independent samples that were not used in creating the classifier. Obtaining sufficient wild and laboratory-adapted strains to carry out this kind of validation is non-trivial. In addition, except for Y. pestis strain KIM D27 samples, all of the samples in this study belong to what is called the 1.ORI molecular group (55), which roughly coincides with biovar Orientalis. Thus, while the published data strongly suggest that the classifier is accurate for the strains used in the training set, it is impossible to predict how well this classifier would work for more distantly related strains such as those from other geographical areas or phylogenetic groups. Lastly, in the future it will be critical to further elucidate the biological processes that underlie the observed changes. A well-understood biological mechanism for observed changes in protein abundance would provide a different kind of validation, and possibly yield improve methods of discrimination. The fact that the selected features were related to observed proteomic changes is a step in this direction, but much fundamental science work remains to provide a satisfactory mechanistic explanation. Determining Growth Conditions of Bacteria from Bacterial Protein Expression Data In the basic biological sciences, proteomics is often used in a comparative mode, to elucidate the organism’s response to changing environmental conditions. A recent paper by Deatherage Kaiser et al. (56) turns this paradigm on its head: if an organism’s proteomic responses are sufficiently well-characterized, protein abundance changes can be used to infer growth conditions. Deatherage Kaiser et al. studied four different strains of Clostridium botulinum in two different growth media over five days. Botulinum neurotoxin levels were higher at later time points, during the stationary phase of culture. Other proteins were found with similar expression patterns. For instance, at earlier time points (exponential growth phase), C. botulinum expresses proteins for a metabolic pathway that converts pyruvate to acetate, whereas at later time points (stationary phase) the enzymes for a different pathway leading to butyrate production are expressed. Thus, given a biomass sample, relative abundance of these and other identified proteins can be used to determine the age of the culture at the time it was harvested. Culture age may not seem like a parameter of critical forensic relevance, but it is relevant, since the botulinum neurotoxin is produced only during stationary phase. The remaining challenge is how to determine relative protein abundance for a single sample, in the absence of the control or comparison sample that proteomics usually relies on. Deatherage Kaiser et al. went on to show that this problem can be overcome by calculating ratios of selected protein abundances. These ratios separated stationary phase from exponential phase cultures, thus overcoming the problem of the lack of a comparison sample in forensic proteomics. Deatherage Kaiser et al.’s paper provides an example of how protein abundances could be utilized to characterize aspects of the growth environment of a microbial sample. Other possibilities include characterization of the growth medium’s carbon sources via expression of different metabolic enzymes, or characterization of the medium’s iron content via expression of iron uptake and homeostasis proteins. Most of biological forensics has traditionally been focused on agent identification. Forensic proteomics opens the possibility to additionally exploit signatures of production processes. Characterizing Microbiological Growth Medium Components Microbiological growth media contain macronutrients derived from sources such as milk protein (casein), soybean extract, yeast, and meat extract, among others, usually in the form of a 153

hydrolysate, or a mixture of peptides. Specific medium formulations use nutrients from different sources and are used for particular organisms. For example, the common medium component tryptone is a pancreatic digest of casein. Clowers et al. (57) demonstrated that tryptic peptides of casein adhere to the surface of Y. pestis cells cultured in tryptone-containing media. These peptides can be recovered by washing the cell pellet under conditions that minimize cell lysis, analyzed by LC-MS/MS measurements, and identified by database search. Thus information about the methods used to grow a culture can be derived from a sample of biomass. Database search hits for bovine casein might suggest various medium formulations containing tryptone; additional database search hits for soybean might suggest tryptic soy broth, and so forth. The patterns of peptides from common proteins were reproducible between analyses and even between samples that were inactivated with bleach or autoclaving. Beyond identifying the biological source of the medium peptides, the patterns of peptides found for particular proteins were reproducibly different for material from different suppliers (58), suggesting that analysis of medium peptides could illuminate the sources and methods of pathogen production at a more detailed level. These analyses were done with a database search based on a list of known organisms that are used in common growth media. The LC-MS/MS based species identification methods described above would be well suited to a broader and unbiased analysis of this type of data. Determining the Host Cell Species of a Virus Preparation Unlike bacteria or other microbes, which are grown on simple nutrient media, virus must be produced in a host cell. Preparations of virions (infectious virus particles) isolated from host cells inevitably contain host cell proteins. These proteins may be nonspecifically associated with the outside of the virion, part of a host-derived membrane (59) or simply co-purified (60). They may also be nonspecifically incorporated into the structure of the virus itself while it is maturing in the host cell (61–63). It may be possible to use shotgun proteomics analysis to identify these host cell proteins, and in favorable cases, to identify the host species. Wunschel et al. (64) demonstrated this approach with vaccinia virus. They compared virion preparations from three different cell lines (two human and one African green monkey) by LC-MS/ MS proteomics. The virion preparations used a range of centrifugation procedures and thus retained different amounts of host proteins. Database searching with the known host organism plus virus database demonstrated increasing proportion of viral proteins as the expected level of purification increased. The types of host proteins associated with the virions also changed, with a wide variety of proteins at low purification to mostly cytoskeletal and protein folding chaperone proteins at the highest level of purification. This finding suggests that proteomics could illuminate the degree of purification of a virion preparation. Next, they investigated how specifically the host cell organism could be determined from the available data. They analyzed the taxonomic distribution of host proteins identified in a broad database search. Several peptides were identified that could potentially limit host taxonomy, but in general, the sequence similarity of the human and monkey peptides was so high that the overall analysis points only to an undetermined primate host. This analysis was carried out by an ad hoc analysis of the taxonomic distribution of identified peptides, one at a time. The application of newer tools for metaproteomic analysis such as UniPept (65, 66), which analyzes lists of tryptic peptides using a LCA approach, would be informative. Likewise, applying LC-MS/MS based species identification methods described above (38) would better probe the limits of the information contained in the data. 154

Conclusions In this chapter we have attempted to show how a number of proteomics methods, especially untargeted LC-MS/MS methods, can provide not only taxonomic information, but also information about methods of production that is complementary to genomics and useful in many cases. For microbes, most of the identification methods are currently focused on public health and/or biosecurity applications, rather than criminal justice applications. However, these methods could prove critical to a criminal case if a biological attack were carried out and the perpetrator brought to trial. Thus, it is important for forensic scientists to be engaged with research in these areas and to shape it to promote standards that lead to legal admissibility.

List of Abbreviations ABOid BLAST DNA LC LCA MALDI MiCid MRM MS MS/MS NCBI pCD1 pgm PSM RT-PCR TCUP TOF

Agents of Biological Origin Identification (Software) Basic local alignment search tool (Software) Deoxyribonucleic acid Liquid chromatography Least common ancestor Matrix-assisted laser desorption/ionization Microbe Identification (Software) Multiple reaction monitoring mass spectrometry Mass spectrometry Tandem mass spectrometry National Center for Biotechnology Information A virulence plasmid in Yersinia pestis A genomic locus in Yersinia pestis that codes for several genes including virulence factors Proteomics (Software) Peptide-spectrum match Real-time polymerase chain reaction Typing and Characterization of Unknown Pathogens Using Time of flight mass analyzer

Acknowledgments The author would like to thank Janine Hutchison, Karen Wahl, David Wunschel, Helen Kreuzer, and Kristin Jarman for helpful discussions.

References 1. 2.

Gonzalez, A.; Vázquez-Baeza, Y.; Pettengill, J. B.; Ottesen, A.; McDonald, D.; Knight, R. Avoiding Pandemic Fears in the Subway and Conquering the Platypus. mSystems 2016, 1. McIntyre, A. B. R.; Ounit, R.; Afshinnekoo, E.; Prill, R. J.; Hénaff, E.; Alexander, N.; Minot, S. S.; Danko, D.; Foox, J.; Ahsanuddin, S.; Tighe, S.; Hasan, N. A.; Subramanian, P.; Moffat, K.; Levy, S.; Lonardi, S.; Greenfield, N.; Colwell, R. R.; Rosen, G. L.; Mason, C. E. Comprehensive Benchmarking and Ensemble Approaches for Metagenomic Classifiers. Genome Biol. 2017, 18, 182. 155

3. 4.

5.

6. 7.

8.

9. 10.

11.

12.

13.

14.

15. 16.

Fenselau, C.; Demirev, P. A. Characteriztion of Intact Microorganisms by Maldi Mass Spectrometry. Mass Spectrom. Rev. 2001, 20, 157–171. Pineda, F. J.; Lin, J. S.; Fenselau, C.; Demirev, P. A. Testing the Significance of Microorganism Identification by Mass Spectrometry and Proteome Database Search. Analytical Chemistrry 2000, 72, 3739–3744. Warscheid, B.; Fenselau, C. A Targeted Proteomics Approach to the Rapid Identification of Bacterial Cell Mixtrues by Matrix-Assisted Laser Desorption/Ionization Mass Spectrometry. Proteomics 2004, 2004, 2877–2892. Fenselau, C.; Russell, S.; Swatkoski, S.; Edwards, N. Proteomic Strategies for Rapid Characterization of Micro-Organisms. Eur. J. Mass Spectrom. 2007, 13, 35–39. Wynne, C.; Edwards, N. J.; Fenselau, C. Phyloproteomic Classification of Unsequenced Organisms by Top-Down Identification of Bacterial Proteins Using CapLC-MS/MS on an Orbitrap. Proteomics 2010, 10, 3631–3643. Dupré, M.; Gilquin, B.; Fenaille, F.; Feraudet-Tarisse, C.; Dano, J.; Ferro, M.; Simon, S.; Junot, C.; Brun, V.; Becher, F. Multiplex Quantification of Protein Toxins in Human Biofluids and Food Matrices Using Immunoextraction and High-Resolution Targeted Mass Spectrometry. Anal. Chem. 2015, 87, 8473–8480. Fusaro, V. A.; Mani, D. R.; Mesirov, J. P.; Carr, S. A. Prediction of High-Responding Peptides for Targeted Protein Assays by Mass Spectrometry. Nat. Biotechnol. 2009, 27, 190. Mallick, P.; Schirle, M.; Chen, S. S.; Flory, M. R.; Lee, H.; Martin, D.; Ranish, J.; Raught, B.; Schmitt, R.; Werner, T.; Kuster, B.; Aebersold, R. Computational Prediction of Proteotypic Peptides for Quantitative Proteomics. Nat. Biotechnol. 2006, 25, 125. Gekenidis, M.-T.; Studer, P.; Wüthrich, S.; Brunisholz, R.; Drissner, D. Beyond the MatrixAssisted Laser Desorption Ionization (MALDI) Biotyping Workflow: In Search of Microorganism-Specific Tryptic Peptides Enabling Discrimination of Subspecies. Appl. Environ. Microbiol. 2014, 80, 4234–4241. Chenau, J.; Fenaille, F.; Caro, V.; Haustant, M.; Diancourt, L.; Klee, S. R.; Junot, C.; Ezan, E.; Goossens, P. L.; Becher, F. Identification and Validation of Specific Markers of Bacillus Anthracis Spores by Proteomics and Genomics Approaches. Mol. Cell. Proteomics 2014, 13, 716–732. Chenau, J.; Fenaille, F.; Ezan, E.; Morel, N.; Lamourette, P.; Goossens, P. L.; Becher, F. Sensitive Detection of Bacillus Anthracis Spores by Immunocapture and Liquid Chromatography–Tandem Mass Spectrometry. Anal. Chem. 2011, 83, 8675–8682. Chenau, J.; Fenaille, F.; Simon, S.; Filali, S.; Volland, H.; Junot, C.; Carniel, E.; Becher, F. Detection of Yersinia Pestis in Environmental and Food Samples by Intact Cell Immunocapture and Liquid Chromatography–Tandem Mass Spectrometry. Anal. Chem. 2014, 86, 6144–6152. Kalb, S. R.; Barr, J. R. Mass Spectrometric Detection of Ricin and Its Activity in Food and Clinical Samples. Anal. Chem. 2009, 81, 2037–2042. Kalb, S. R.; Goodnough, M. C.; Malizio, C. J.; Pirkle, J. L.; Barr, J. R. Detection of Botulinum Neurotoxin a in a Spiked Milk Sample with Subtype Identification through Toxin Proteomics. Anal. Chem. 2005, 77, 6140–6146.

156

17. Gilquin, B.; Jaquinod, M.; Louwagie, M.; Kieffer-Jaquinod, S.; Kraut, A.; Ferro, M.; Becher, F.; Brun, V. A Proteomics Assay to Detect Eight Cbrn-Relevant Toxins in Food. Proteomics 2017, 17, 1600357. 18. Merkley, E. D.; Jenson, S. C.; Arce, J. S.; Melville, A. M.; Leiser, O. P.; Wunschel, D. S.; Wahl, K. L. Ricin-Like Proteins from the Castor Plant Do Not Influence Liquid ChromatographyMass Spectrometry Detection of Ricin in Forensically Relevant Samples. Toxicon 2017, 140, 18–31. 19. Fredriksson, S.-Å.; Hulst, A. G.; Artursson, E.; de Jong, A. L.; Nilsson, C.; van Baar, B. L. M. Forensic Identification of Neat Ricin and of Ricin from Crude Castor Bean Extracts by Mass Spectrometry. Anal. Chem. 2005, 77, 1545–1555. 20. Pfrunder, S.; Grossmann, J.; Hunziker, P.; Brunisholz, R.; Gekenidis, M.-T.; Drissner, D. Bacillus cereus Group-Type Strain-Specific Diagnostic Peptides. J. Proteome Res. 2016, 15, 3098–3107. 21. Wieser, A.; Schneider, L.; Jung, J.; Schubert, S. Maldi-Tof Ms in Microbiological Diagnostics—Identification of Microorganisms and Beyond (Mini Review). Appl. Microbiol. Biotechnol. 2012, 93, 965–974. 22. Rodríguez-Sánchez, B.; Cercenado, E.; Coste, A. T.; Greub, G. Review of the Impact of MALDI-TOF MS in Public Health and Hospital Hygiene, 2018. Eurosurveillance 2019, 24, 1800193. 23. Duriez, E.; Armengaud, J.; Fenaille, F.; Ezan, E. Mass Spectrometry for the Detection of Bioterrorism Agents: From Environmental to Clinical Applications. J. Mass Spectrom. 2016, 51, 183–199. 24. Jarman, K. H.; Daly, D. S.; Petersen, C. E.; Saenz, A. J.; Valentine, N. B.; Wahl, K. L. Extracting and Visualizing Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectral Fingerprints. Rapid Commun. Mass Spectrom. 1999, 13, 1586–1594. 25. Jarman, K. H.; Cebula, S. T.; Saenz, A. J.; Petersen, C. E.; Valentine, N. B.; Kingsley, M. T.; Wahl, K. L. An Algorithm for Automated Bacterial Identification Using Matrix-Assisted Laser Desorption/Ionization Mass Spectrometry. Anal. Chem. 2000, 72, 1217–1223. 26. Lasch, P.; Wahab, T.; Weil, S.; Pályi, B.; Tomaso, H.; Zange, S.; Kiland Granerud, B.; Drevinek, M.; Kokotovic, B.; Wittwer, M.; Pflüger, V.; Di Caro, A.; Stämmler, M.; Grunow, R.; Jacob, D. Identification of Highly Pathogenic Microorganisms by Matrix-Assisted Laser Desorption Ionization–Time of Flight Mass Spectrometry: Results of an Interlaboratory Ring Trial. J. Clin. Microbiol. 2015, 53, 2632–2640. 27. Tracz, D. M.; Antonation, K. S.; Corbett, C. R. Verification of a Matrix-Assisted Laser Desorption Ionization–Time of Flight Mass Spectrometry Method for Diagnostic Identification of High-Consequence Bacterial Pathogens. J. Clin. Microbiol. 2016, 54, 764–767. 28. Rudrik, J. T.; Soehnlen, M. K.; Perry, M. J.; Sullivan, M. M.; Reiter-Kintz, W.; Lee, P. A.; Pettit, D.; Tran, A.; Swaney, E. Safety and Accuracy of Matrix-Assisted Laser Desorption Ionization–Time of Flight Mass Spectrometry for Identification of Highly Pathogenic Organisms. J. Clin. Microbiol. 2017, 55, 3513–3529. 29. Cunningham, S. A.; Patel, R. Importance of Using Bruker’s Security-Relevant Library for Biotyper Identification of Burkholderia pseudomallei, Brucella Species, and Francisella tularensis. J. Clin. Microbiol. 2013, 51, 1639–1640. 157

30. Gérôme, P.; Le Flèche, P.; Blouin, Y.; Scholz, H. C.; Thibault, F. M.; Raynaud, F.; Vergnaud, G.; Pourcel, C. Yersinia pseudotuberculosis St42 (O:1) Strain Misidentified as Yersinia pestis by Mass Spectrometry Analysis. Genome Announcements 2014, 2, e00435–00414. 31. Dworzanski, J. P.; Snyder, A. P.; Chen, R.; Zhang, H. Y.; Wishart, D.; Li, L. Identification of Bacteria Using Tandem Mass Spectrometry Combined with a Proteome Database and Statistical Scoring. Anal. Chem. 2004, 76, 2355–2366. 32. Dworzanski, J. P.; Deshpande, S. V.; Chen, R.; Jabbour, R. E.; Snyder, A. P.; Wick, C. H.; Li, L. Mass Spectrometry-Based Proteomics Combined with Bioinformatic Tools for Bacterial Classification. J. Proteome Res. 2006, 5, 76–87. 33. Deshpande, S. V.; Jabbour, R. E.; Snyder, P. A.; Stanford, M. F.; Wick, C. H.; Zulich, A. W. Aboid: A Software for Automated Identification and Phyloproteomics Classification of Tandem Mass Spectrometric Data. Journal of Chromatography and Separation Techniques 2011, 5, 001–006. 34. Jabbour, R. E.; Deshpande, S. V.; Wade, M. M.; Stanford, M. F.; Wick, C. H.; Zulich, A. W.; Skowronski, E. W.; Snyder, A. P. Double-Blind Characterization of Non-Genome-Sequenced Bacteria by Mass Spectrometry-Based Proteomics. Appl. Environ. Microbiol. 2010, 76, 3637–3644. 35. Boulund, F.; Karlsson, R.; Gonzales-Siles, L.; Johnning, A.; Karami, N.; AL-Bayati, O.; Ahren, C.; Moore, E. R. B.; Kristiansson, E. Tcup: Typing and Characterization of Bacteria Using Bottom-up Tandem Mass Spectrometry Proteomics. Mol. Cell. Proteomics 2017, 16, 1052–1063. 36. Alves, G.; Wang, G.; Ogurtsov, A. Y.; Drake, S. K.; Gucek, M.; Suffredini, A. F.; Sacks, D. B.; Yu, Y.-K. Identification of Microorganisms by High Resolution Tandem Mass Spectrometry with Accurate Statistical Significance. J. Am. Soc. Mass Spectrom. 2016, 27, 194–210. 37. Alves, G.; Wang, G.; Ogurtsov, A. Y.; Drake, S. K.; Gucek, M.; Sacks, D. B.; Yu, Y.-K. Rapid Classification and Identification of Multiple Microorganisms with Accurate Statistical Significance Via High-Resolution Tandem Mass Spectrometry. J. Am. Soc. Mass Spectrom. 2018, 29, 1721–1737. 38. Jarman, K. H.; Heller, N. C.; Jenson, S. C.; Hutchison, J. R.; Kaiser, B. L. D.; Payne, S. H.; Wunschel, D. S.; Merkley, E. D. Proteomics Goes to Court: A Statistical Foundation for Forensic Toxin/Organism Identification Using Bottom-up Proteomics. J. Proteome Res. 2018, 17, 3075–3085. 39. Padliya, N. D.; Garrett, W. M.; Campbell, K. B.; Tabb, D. L.; Cooper, B. Tandem Mass Spectrometry for the Detection of Plant Pathogenic Fungi and the Effects of Database Composition on Protein Inferences. Proteomics 2007, 7, 3932–3942. 40. Daubert V. Merrell Dow Pharmaceuticals, Inc. In United States Reports; 1993; Vol. 509, p 579. 41. Potgieter, T.; Nel, A. J.; Tabb, D. L.; Fortuin, S.; Garnett, S.; Blackburn, J.; Mulder, N. Metanovo: A Probabilistic Approach to Peptide and Polymorphism Discovery in Complex Mass Spectrometry Datasets. bioRxiv 2019, 605550. 42. Ma, B.; Johnson, R. De Novo Sequencing and Homology Searching. Mol. Cell. Proteomics 2012, 11. 43. Hoopmann, M. R.; Moritz, R. L. Current Algorithmic Solutions for Peptide-Based Proteomics Data Generation and Identification. Curr. Opin. Biotechnol. 2013, 24, 31–38. 158

44. Palmblad, M.; Deelder, M. A. Molecular Phylogenetics by Direct Comparison of Tandem Mass Spectra. Rapid Commun. Mass Spectrom. 2012, 26, 728–732. 45. Ohana, D.; Dalebout, H.; Marissen, R. J.; Wulff, T.; Bergquist, J.; Deelder, A. M.; Palmblad, M. Identification of Meat Products by Shotgun Spectral Matching. Food Chem. 2016, 203, 28–34. 46. Palmblad, M.; Deelder, A. M. Molecular Phylogenetics by Direct Comparison of Tandem Mass Spectra. Rapid Commun. Mass Spectrom. 2012, 26, 728–732. 47. Wulff, T.; Nielsen, M. E.; Deelder, A. M.; Jessen, F.; Palmblad, M. Authentication of Fish Products by Large-Scale Comparison of Tandem Mass Spectra. J. Proteome Res. 2013, 12, 5253–5259. 48. Centers for Disease Control and Prevention. Imported Plague—New York City, 2002. https://www.cdc.gov/mmwr/preview/mmwrhtml/mm5231a1.htm (accessed April 2, 2018). 49. Leiser, O. P.; Merkley, E. D.; Clowers, B. H.; Deatherage Kaiser, B. L.; Lin, A.; Hutchison, J. R.; Melville, A. M.; Wagner, D. M.; Keim, P. S.; Foster, J. T.; Kreuzer, H. W. Investigation of Yersinia Pestis Laboratory Adaptation through a Combined Genomics and Proteomics Approach. PLoS One 2015, 10, e0142997. 50. Kiebel, G. R.; Auberry, K. J.; Jaitly, N.; Clark, D. A.; Monroe, M. E.; Peterson, E. S.; Tolić, N.; Anderson, G. A.; Smith, R. D. Prism: A Data Management System for High-Throughput Proteomics. Proteomics 2006, 6, 1783–1790. 51. Cox, J.; Hein, M. Y.; Luber, C. A.; Paron, I.; Nagaraj, N.; Mann, M. MaxLFQ Allows Accurate Proteome-Wide Label-Free Quantification by Delayed Normalization and Maximal Peptide Ratio Extraction. Mol. Cell. Proteom. 2014, M113.031591. 52. Cox, J.; Mann, M. Maxquant Enables High Peptide Identification Rates, Individualized P.P.B.Range Mass Accuracies and Proteome-Wide Protein Quantification. Nat. Biotech. 2008, 26, 1367–1372. 53. Ritchie, M. D.; Holzinger, E. R.; Li, R.; Pendergrass, S. A.; Kim, D. Methods of Integrating Data to Uncover Genotype–Phenotype Interactions. Nature Reviews Genetics 2015, 16, 85. 54. Tibshirani, R. Regression Shrinkage and Selection Via the Lasso. Journal of the Royal Statistical Society: Series B (Methodological) 1996, 58, 267–288. 55. Achtman, M.; Morelli, G.; Zhu, P.; Wirth, T.; Diehl, I.; Kusecek, B.; Vogler, A. J.; Wagner, D. M.; Allender, C. J.; Easterday, W. R.; Chenal-Francisque, V.; Worsham, P.; Thomson, N. R.; Parkhill, J.; Lindler, L. E.; Carniel, E.; Keim, P. Microevolution and History of the Plague Bacillus, Yersinia pestis. Proceedings of the National Academy of Sciences of the United States of America 2004, 101, 17837–17842. 56. Deatherage Kaiser, B. L.; Hill, K. K.; Smith, T. J.; Williamson, C. H. D.; Keim, P.; Sahl, J. W.; Wahl, K. L. Proteomic Analysis of Four Clostridium Botulinum Strains Identifies Proteins That Link Biological Responses to Proteomic Signatures. PLoS One 2018, 13, e0205586. 57. Clowers, B. H.; Wunschel, D. S.; Kreuzer, H. W.; Engelmann, H. E.; Valentine, N.; Wahl, K. L. Characterization of Residual Medium Peptides from Yersinia Pestis Cultures. Anal. Chem. 2013, 85, 3933–3939. 58. Merkley, E. D.; Kaiser, B. L. D.; Wunschel, D.; Wahl, K. L. Proteomics for Bioforensics. In Microbial Forensics, 3rd ed.; Budowle, B., Schutzer, S. E., Breeze, R. G., Keim, P. S., Morse, S. A., Eds.; Elsevier Academic Press, 2019; in press. 159

59. Krauss, O.; Hollinshead, R.; Hollinshead, M.; Smith, G. L. An Investigation of Incorporation of Cellular Antigens into Vaccinia Virus Particles. J. Gen. Virol. 2002, 83, 2347–2359. 60. Manes, N. P.; Estep, R. D.; Mottaz, H. M.; Moore, R. J.; Clauss, T. R. W.; Monroe, M. E.; Du, X.; Adkins, J. N.; Wong, S. W.; Smith, R. D. Comparative Proteomics of Human Monkeypox and Vaccinia Intracellular Mature and Extracellular Enveloped Virions. J. Proteome Res. 2008, 7, 960–968. 61. Vanderplasschen, A.; Mathew, E.; Hollinshead, M.; Sim, R. B.; Smith, G. L. Extracellular Enveloped Vaccinia Virus Is Resistant to Complement Because of Incorporation of Host Complement Control Proteins into Its Envelope. Proc. Nat. Acad. Sci. U.S.A. 1998, 95, 7544–7549. 62. Varnum, S. M.; Streblow, D. N.; Monroe, M. E.; Smith, P.; Auberry, K. J.; Pasa-Tolic, L.; Wang, D.; Camp, D. G., 2nd; Rodland, K.; Wiley, S.; Britt, W.; Shenk, T.; Smith, R. D.; Nelson, J. A. Identification of Proteins in Human Cytomegalovirus (Hcmv) Particles: The Hcmv Proteome. J. Virol. 2004, 78, 10960–10966. 63. Castro, A. P. V.; Carvalho, T. M. U.; Moussatché, N.; Damaso, C. R. A. Redistribution of Cyclophilin a to Viral Factories During Vaccinia Virus Infection and Its Incorporation into Mature Particles. J. Virol. 2003, 77, 9052–9068. 64. Wunschel, D.; Tulman, E.; Engelmann, H.; Clowers, B. H.; Geary, S.; Robinson, A.; Liao, X. F. Forensic Proteomics of Poxvirus Production. Analyst 2013, 138, 6385–6397. 65. Mesuere, B.; Debyser, G.; Aerts, M.; Devreese, B.; Vandamme, P.; Dawyndt, P. The Unipept Metaproteomics Analysis Pipeline. Proteomics 2015, 15, 1437–1442. 66. Mesuere, B.; Devreese, B.; Debyser, G.; Aerts, M.; Vandamme, P.; Dawyndt, P. Unipept: Tryptic Peptide-Based Biodiversity Analysis of Metaproteome Samples. J. Proteome Res. 2012, 11, 5773–5780.

160

Chapter 10

ISO 17025 Accreditation of Method-Based Mass Spectrometry for Bioforensic Analyses Stephen R. Cendrowski* and Alaine M. Garrett National Bioforensic Analysis Center, National Biodefense Analysis and Countermeasures Center, Frederick, Maryland 21702, United States *E-mail: [email protected].

Ricin, a ribosomal-inactivating protein toxin produced by the castor plant (Ricinus communis) is considered a biological threat due to its accessibility and low human lethal dose. Method-based mass spectrometry (MS) techniques for the identification of protein toxins such as ricin are needed to enhance analytical processes for data collection on bioforensic samples. MS-based identification of ricin involves comparison of MS measurements between test samples and reference standards in addition to identification of ricin-specific peptides by searching proteome databases. The National Bioforensic Analysis Center (NBFAC) validated two MS techniques for identification of ricin in bioforensic samples following the ISO 17025 Standard. These methods included: 1) one-dimensional sodium dodecylsulfate polyacrylamide gel electrophoresis (SDS-PAGE), for separation and identification by molecular weight, followed by in gel trypsin digestion and MALDI TOF/TOF MS analysis and 2) in-solution trypsin digestion and nanoLC-HRMS/MS analysis. Robust method acceptance criteria were established for this ISO 17025 accreditation process to provide confidence in results. Ricin identification is made after all MS data are searched using commercially available proteomics software, compared to data obtained from reference materials analyzed in parallel, and manually reviewed to confirm peptide identifications. The final ricin identification is made when at least one peptide unique to this protein toxin is identified. This validation performed under ISO 17025 standards was completed by multiple technicians over a year-long process, which provided established limits of detection, sensitivity, false positive/negative rates and variability assessments of the methods. ISO 17025 accreditation described in this chapter for these MS techniques for identification provides confidence in the robustness of the processes used during characterization of bioforensic samples containing ricin.

© 2019 American Chemical Society

This work was funded under Agreement No. HSHQDC-15-C-00064 awarded to Battelle National Biodefense Institute by the Department of Homeland Security Science and Technology Directorate (DHS S&T) for the management and operation of the National Biodefense Analysis and Countermeasures Center a Federally Funded Research and Development Center. The views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the U.S. Department of Homeland Security or the U.S. Government. The Department of Homeland Security does not endorse any products or commercial services mentioned in this presentation. In no event shall the DHS, BNBI or NBACC have any responsibility or liability for any use, misuse, inability to use, or reliance upon the information contained herein. In addition, no warranty of fitness for a particular purpose, merchantability, accuracy or adequacy is provided regarding the contents of this document

Introduction Biological agents and toxins have been used throughout the ages as weapons. In more modern times, use of biological agents has continued (1), such as the anthrax spores (from Bacillus anthracis) sent through the mail during the Amerithrax letter attacks that killed 5 and infected 17 other individuals (2). This attack on the heels of 9/11 and other evidence demonstrated a need that led to the Presidential directive to establish a National Bioforensic Analysis Center (NBFAC) to support biocrime investigations as part of U.S. biodefense preparedness (3). Ricin is one example of a biological toxin which had been developed for use as a weapon and whose production has been described in scientific articles as well as in more nefarious publications (4). Ricin detection by the NBFAC involves the use of sensitive immunoassays (5, 6) to identify the toxin and its activity, which have been ISO 17025 accredited and in use for well over a decade. The use of these ricin immunoassays are complemented by analytical chemistry capabilities used to confirm the presence of protein sequences specific to ricin. As part of ongoing activities to expand capabilities at NBFAC, analytical chemistry methods including mass spectrometry (MS) have been established and validated. Methods recently validated at NBFAC include two ISO 17025 accredited MS processes to identify and characterize ricin. The contents of this chapter provide an overview of ricin and the steps used at NBFAC toward method validations using MALDI TOF/TOF MS and nanoLC-HRMS/MS instruments as presented at a recent ACS meeting (7). Ricinus communis Toxin In addition to anthrax, biological toxins have also been used during attacks on the U.S. government, most recently in 2013 letters were sent to the President and other government officials containing the plant toxin ricin (8). Beans/seeds from the castor plant (Ricinus communis) are the source for the protein toxin commonly referred to as ricin. This aqueous-soluble protein toxin is composed of two subunits: an A chain that is the active toxin component, and a B chain, which is the cell-binding component. Ricin is classified as a type II RIP (ribosome inactivating protein), and the A chain acts as an rRNA N-glycosidase, which irreversibly hydrolyzes an rRNA adenine residue (9). 162

The ricin B chain binds to cell surface receptors, allowing for A chain entry into cells and targeting of the protein translation machinery, leading to cell death (10). ISO 17025 Accreditation Process Validation is a requirement for labs performing bioforensics as results can impact decisions on criminal proceedings, national security, and potentially international relations (11). Lab testing validation is complex, and the major steps during any validation process are illustrated in Figure 1. At the outset is the initial method development stage (Figure 1A), which entails identification of methods to be used, and development of the method for use within a bioforensic setting and samples. Additionally, collaboration and “buy-in” from external subject matter experts has proven useful to stakeholders given that results from evidentiary testing may be scrutinized by independent expert witnesses called on by the defense during criminal proceedings. Once developed, validation planning and testing is performed (Figure 1B) under the watchful eye of independent Quality Assurance personnel (QA) as part of an overarching Quality Management System (QMS). The validation planning is used to describe the scope and purpose of the method, identifies critical reagents, instruments and how the performance parameters that will be evaluated during the validation will be judged using acceptance criteria determined prior to performance. Once validation of the method is accepted, proficiency testing is completed to ensure the laboratory is capable of performing the method at the expected levels of accuracy. Once reviewed by QA, all of these results are submitted to an independent accreditation agency (e.g., A2LA or PJLA). Producing test evaluations under a structured and rigorous validation plan that is reviewed and approved by independent QA personnel under an overarching QMS provides robust assurances that methods provide accurate results, and are continuously monitored (Figure 1C) so that results can be considered reliable for use in prosecution of biocrime or bioterror investigations. Ricin Characterization Methods Workflow Immunoassay characterization: Initial NBFAC testing of unknown samples is performed using enzyme-linked immunosorbent assay (ELISA) and if ricin is detected, these samples are carried forward into an in vitro activity assay termed cell-free translation (CFT) that provides evidence ricin is intact (5, 6). These assays have previously been validated and time-tested to provide evidence that ricin is present, and can also be used for ricin concentration estimation. However, given that immunoassays also detect the highly similar ricin agglutinin protein typically coincident in ricin samples or ricin may be present but denatured/inactivated, MS methods also allow for a structurefunction agnostic confirmation of ricin protein sequences in these samples. SDS-PAGE: Analyses of unknowns using gel electrophoresis provided an initial visual assessment of proteins present in samples, estimation of protein molecular weight (MW) calculated using reference standards, refinement of downstream MS analytics by identifying protein bands similar to a ricin reference standard that was run in parallel (i.e., ricin), and minimized potential matrix interference by overabundant proteins or unknown sample components with downstream MS analyses (i.e., MALDI TOF/TOF). The gel system and Coomassie-staining technique selected for validation (Criterion™ XT BisTris and Bio-Safe™, Bio-Rad) were compared with similar SDS-PAGE systems that were suggested by collaborators, and provided several advantages over other systems, including: reproducible protein separation, proven compatibility with downstream MS analyses, and relatively long shelf-life for the reagents. The general qualification process that this selection and development falls under 163

is outlined below (Figure 1A). The precision and accuracy of ricin SDS-PAGE MW estimation, and between-day, technical staff, varying concentration, and detection limit reproducibilities were assessed during validation testing (Figure 1B). MALDI TOF/TOF MS The ricin MALDI TOF/TOF MS method development was tied to sample preparation using SDS-PAGE, band excision and in gel tryptic digestion of products running within the expected ricin MW ranges for non-reduced and reduced forms (Figures 1A and 2). This combined process allowed for effective identification of peptide mass spectra acquisitions to characterize protein sequences associated with the SDS-PAGE gel bands. Given that MALDI TOF/TOF MS analyses benefits from using low complexity samples, SDS-PAGE sample processing provided a means to reduce interference by overabundant proteins and other signal suppressors such as salts and detergents that negatively impact ionization. Parameters assessed during validation of the ricin MALDI TOF/ TOF MS method were used to determine detection limit, as well as ruggedness (between-day and technical staff variability) and reliability (false negative and positive results). Validation planning used preliminary qualification data from the established method to define criteria used during testing (Figure 1B). LC-HRMS/MS Analysis of ricin by LC-HRMS/MS involves separation of in-solution trypsin-digested peptides based on hydrophobicity followed by mass spectral measurements used to identify peptide masses and a second set of mass spectral determinations, which are used to identify the amino acid composition of selected peptides. As in the in gel digestion process mentioned above, trypsin is used for its highly specific cleavage of the carboxyl-terminal side of lysine and arginine amino acids which allows for the creation of theoretical digests of known protein sequence for purposes of in silico identification. MS Data Analysis The mass spectra from MALDI TOF/TOF MS were analyzed using an algorithm termed peptide-spectral match (PSM) analysis (Mascot Server v2.5 or higher, Matrix Science). Identification of ricin in both non-reduced and reduced in gel tryptic digested samples was based on the detection of peptides with at least one PSM per peptide sequence obtained from the raw mass spectra data. The LC-HRMS/MS sample PSM analysis was performed using the SEQUEST HT algorithm available through Thermo Scientific (Proteome Discoverer, 2.0 or higher). For this process the identification of ricin from in-solution trypsin-digested samples was based on the detection of peptides with at least one PSM per peptide sequence obtained from the raw mass spectra data. MS Methods Maintenance ISO 17025 method maintenance is a dynamic process that includes continual monitoring of assay and instrument performance by tracking and trending control results, performance and recording of routine maintenance and calibrations, competency training and evaluation, and proficiency testing (Figure 1C). These maintenance processes provide mechanisms to ensure that personnel, instruments, and lab environments are capable of producing accurate results and data that are useful for troubleshooting both analytical and training issues. 164

Figure 1. ISO 17025 Accreditation Trajectory for Bioforensic Lab Testing. From development through final validation and accreditation; requires (A) collaboration with subject matter experts; (B) validation planning, testing, reporting through a QMS, and accreditation through an independent entity; and (C) establishment of an assay maintenance program to ensure the methods continue to meet ISO 17025 standards.

Validation of MS Methods for Ricin Sample Preparation Sample types that may be received for analysis using these validated methods include environmental swabs, solid substances (e.g., castor beans, bean mashes, crude extracts, purified powders, etc.), and aqueous or organic solutions. Regardless of sample type, all are resuspended in PBS, pH7.4 + 0.01% Tween20 (minimum sample volume: 56 µL), and heat inactivated for safety purposes at 100 °C for 30 minutes using a beadbath (SDS-PAGE samples were heat inactivated in XT gel loading buffer). Protein separations by SDS-PAGE were achieved using 12% Criterion XT BisTris Gels, XT MOPS buffer, and XT gel loading buffer with and without XT sample reducing agent (tris(2-carboxyethyl)phosphine, TCEP) according to recommended procedure (Bio-Rad). These sample preparation steps relied on buffers proven to provide ricin stability (PBS+0.01% Tween-20) and reliable, commercially available SDS-PAGE reagents from an ISO accredited manufacturer (BioRad). 165

Fixation and staining were respectively accomplished using a methanol:water:acetic acid (5:4:1) solution followed by Coomassie Blue (Bio-Safe™, Bio-Rad) for 1 hour, and destaining steps (3x10 minutes) and overnight (10 - 18 hours) at ambient temperature in ultrapure water. The gels were then imaged and analyzed using a Gel Doc™ EZ and Image Lab™ Software (Bio-Rad). The controls established for SDS-PAGE include a MW reference standard used to calculate MW estimations of ricin positive control and protein bands in unknown samples within a 10 kDa – 250 kDa range (Precision Plus, Bio-Rad). Additionally, ricin reference material has been established at 50 µg/mL for a positive control (RCA60; Vector Labs, Burlingame CA) and negative controls consisted of sample buffer alone. These controls have been processed in parallel with unknown samples. The established criteria for controls include consistent detection of ricin bands in the positive control non-reduced and reduced samples, and no detectable protein bands in negative controls. Note: Positive control ricin data are not presented here, however, SDS-PAGE results for ricin at 37.5 µg/ mL are similar (Figure 2, Lane 2). SDS-PAGE Results

Figure 2. SDS-PAGE of ricin. Ricin non-reduced (A) and reduced (B) samples electrophoresed using a BisTris/MOPS gel system (Bio-Rad). Precision Plus MW standards are run in lane 1 of both gels, and ricin sample concentrations are represented below each lane number. No sample (NS) was loaded in lane 10. Brackets indicate regions in each lane where gel bands were assessed for volume, MW, and MALDI TOF/ TOF MS after excision and in gel trypsin digestion. (B) The doublet seen in the upper bracketed band is likely due to varying levels of subunit glycosylation (12). The accuracy of the MW estimates for the protein toxins’ bands using Image Lab™ software (v 2.5 Security Edition) was determined based on comparison with the calculated theoretical MW available from the UniProt/Swissprot website for ricin (accession B9T8T0; note: this sequence is identical to the reviewed P02979 database entry). As determined from the UniProt website and ExPASy Peptide Mass tool, ricin MWs, 64.091 kDa, 29.91 kDa, and 30.32 kDa were used as the theoretical masses for the respective holotoxin, and A and B subunits, respectively. Calculation of the MW estimates were performed by Image Lab™ using a log-linear regression method based on the Rf of MW reference standards that were run in parallel with the ricin samples (Lane 1, Figure 2) and a percent error calculation was performed separately (Equation 1).

MW Estimation by SDS-PAGE During the validation process for SDS-PAGE, a pre-set MW correlation coefficient (≥ 0.95; this cutoff was estimated from data acquired during pre-qualification gel runs) was met for all gel 166

MW regression lines established using the MW standards (Precision Plus, Bio-Rad) along with precision acceptance criteria for the MW estimation variability (Table 1). For variability assessment, the acceptable %Relative Standard Deviation x 2 (k values) were ≤ 30% for non-reduced and ≤ 50% for reduced sample bands at concentrations in the detectable range for ricin samples (data not shown).

Figure 3. Four Parameter Logistic Regression Curves for Volume Measurements. Mean volume (relative units based on gel band pixel density) measurements plotted against ricin concentrations from 6 SDS-PAGE replicates. Volume measurements obtained from NRl bands for ricin, imaged using Gel Doc™ EZ and Image Lab™ software (Bio-Rad). Data fit to a 4 parameter logistic curve using displayed equations. Error bars represent ± SD. Limit of detection (LOD) estimated for ricin (1.2 µg/mL) represented by dotted line. Table 1. Accuracy of MW Estimates SDS-PAGE Band ID

Overall Error %

Tech 1 Day 1 Error %

Tech 1 Day 2 Error %

Tech 2 Day 1 Error %

Tech 2 Day 2 Error %

Non-reduced Band

10.5

10

11.3

10.6

10.2

Reduced Band 1

14.6

14.5

15.4

14.5

16.8

Reduced Band 2

3.1

3.3

4

3

5

Table 2. SDS-PAGE Detection of Ricin Bands Ricin (µg/mL)

Non-reduced Band

Reduced Bands

High

35.7

11/11 Detected

11/11 Detected

Intermediate

3.57

19/19 Detected

19/19 Detected

Low

2.69

11/11 Detected

11/11 Detected

Blank

0

0/19 Detected

0/19 Detected

Level Tested

For the MW accuracy pre-set acceptance criteria, the acceptable % Error for MW estimation was achieved (≤ 30%) for non-reduced bands and (≤ 50%) for reduced bands in the detectable range (Table 1). These limits were based on initial pre-qualifying gel runs, which demonstrated higher variability than what was experienced during validation runs. For limit of detection (LOD) pre-set acceptance criteria, a sensitivity of ≥ 80% for detection of ricin by Coomassie staining at the low level concentration (2.69 µg/mL) was achieved (Table 2). Note: automated detection using the Image Lab 167

software was possible at lower concentrations (Figure 3), however, visual confirmation of gel bands could only be achieved at this low level concentration. Volume measurement variability was too high to provide accurate concentration estimates; Figure 3 is presented here for information purposes. Additionally, the false positive and false negative rates were 0% for all samples tested.

In Gel Trypsin Digestion and MALDI TOF/TOF MS Sample Preparation Once bands were identified by SDS-PAGE in non-reduced and/or reduced samples corresponding to the expected MW for ricin, gel slices were excised and macerated using a disposable scalpel, destained with 25mM ammonium bicarbonate/50% acetonitrile, and proteolytically digested based on a standard in gel trypsin digestion method (Kit No. 89871X, Thermo Fisher). After peptide extraction and further processing by C18-tip clean-up, the samples were mixed with matrix (α-cyano-4-hydroxycinnamic, α-CHCA, Alfa Aesar), spotted onto stainless steel target plates, and analyzed using an AB Sciex 5800 (Applied Biosystems, Framingham, MA, USA) MALDI TOF/TOF instrument. Mass spectra of each spot were obtained by scanning from m/z 800 to 3100 in MS-positive ion reflector mode with a focus mass of 1800, with MS/MS data collected on the top 40 most abundant peaks with 1000 laser shots acquired and averaged for each spot using and NdYAG laser with wavelength 355 nm. Parent and fragment ion spectra were collected using Explorer software (AB Sciex), and analyzed using Mascot (Matrix Science). The spectral data were searched against the ricin genome obtained from UniProt, updated quarterly. Mascot search parameters for MS peak filtering include a peptide charge of +1, peptide tolerance of ± 100 ppm and a maximum of one missed cleavage. The MS/MS Mascot search parameter included the MS/MS tolerance of ± 0.2 Da. Results The validation of MALDI TOF/TOF MS was based on qualitative parameters used to determine the presence or absence of the protein toxin following pre-established criteria for identification of ricin in non-reduced and reduced samples. This method is used to identify proteins based on peptides obtained from tryptic digestion of ricin first separated from complex mixtures by SDS-PAGE, which is described above. Analysis of peptides by MALDI TOF/TOF MS first results in mass spectra used to obtain peptide masses and a second mass spectra, which represents the individual amino acid composition for selected peptides. Trypsin is used for its highly specific cleavage of the carboxyl-terminal side of lysine and arginine residues, which allows for the creation of ricin theoretical trypsin digestions for use during analyses of the MALDI TOF/TOF MS data. The mass spectra of the MALDI TOF/TOF MS analyzed samples are compared to a theoretical trypsin digest of toxin using proteomics database searching with Mascot (Mascot Server version 2.5 or higher). Identification of protein toxin in both non-reduced and reduced is based on the detection of peptides with at least one PSM per peptide sequence obtained from the raw mass spectral data. Pre-established criteria were then used to manually determine quality of peptide spectral data, and whether a minimal protein sequence coverage is achieved in which at least four peptides representing both the A and B chains from the ricin holotoxin, and coverage of at least 10% of the holotoxin sequence length. The specificity of the peptides detected was also evaluated with regard to sequence identity shared with peptides from other organisms. By doing so, identification of unique protein toxin 168

peptides resulted in a positive for ricin determination, while peptide identities shared with other proteins found in the toxin-producing plant species (R. communis), but not associated with other proteins outside of the plant resulted in a positive for R. communis but inconclusive for ricin determination. This latter determination was scored as a positive result since the appearance of plant-specific proteins still provides support to evidence obtained from other methods regarding the presence of ricin. The LOD for the identification of ricin using specific peptides in non-reduced samples was found to be 35.7 µg/mL with a sensitivity of 73% and a false negative rate (FNR) of 27% which did not meet the set criteria of sensitivity of 80% and FNR of 20%. The method will be expected to result in positive identification of ricin using a higher concentration of ricin standard. The highest ricin concentration reduced sample was 35.7 µg/mL, yielding a sensitivity of 55% and FNR of 45% which did not meet the LOD criteria. As such, the positive control used for the method was increased to 50 µg/mL and verified during proficiency testing (not shown). The LOD for non-reduced samples at the identification of peptides at the R. communis level was determined at 3.57 µg/mL with a sensitivity of 79% and FNR of 21%, which did not meet the set criteria but was just barely out and is not expected to have a large impact on the results of the assay (Table 3). Table 3. LOD Estimates for Ricin by MALDI TOF/TOF MS Ricin (µg/mL)

Band (Sensitivity/False Negative Rate)

35.7

Positive for ricin in non-reduced samples (73/27)

17.9

Positive for R. communis in reduced samples (100/0)

3.57

Positive for R. communis in non-reduced samples (79/21)

The sensitivity and FNR for reduced samples at the same concentration was 47% and 53%, respectively, which did not meet the pre-set criteria. Reduced samples were not expected to meet the set criteria at the same frequency as the non-reduced samples given that the effective protein concentration of the reduced sample is half that of the non-reduced samples. The LOD for reduced samples R. communis identification is ≤ 17.9 µg/mL. No false positives were identified for blank samples run in parallel with the ricin samples, which met the criteria of ≤ 5% false positivity. Specificity testing was also performed for SDS-PAGE and MALDI TOF/TOF MS using a panel of abrin (a similar plant RIP), R, communis agglutinin (RCA120) and other plant lectins at 50 µg/ mL concentrations in parallel with ricin: abrin, RCA120, peanut agglutinin, jack bean agglutinin, soybean agglutinin, lentil agglutinin, red kidney bean agglutinin, and Bandeiraea simplicifolia lectin. No false positives were detected by either method (data not shown). The method ruggedness was assessed by comparison of two analyses performed by two technicians on different days with the set criteria of 100% concurrency in results. The method was found to meet the set criteria for the identification of ricin at the R. communis level but did not meet the set criteria for the identification at the ricin-specific level. However, since R. communis identification results in a sample score of positive, this degree of ruggedness was acceptable.

169

In-Solution Trypsin Digestion and LC-HRMS/MS Sample Preparation As for the MALDI TOF/TOF MS method validation described above, data acquisition requires the generation of tryptic peptides that are subsequently used for the analysis by the LC-HRMS/ MS. The in-solution tryptic digestion method was performed based on the manufacturer’s protocol (Trypsin Gold: V5280, Promega). However, given the resolving power of a high-resolution tandem mass spectrometric (HRMS/MS) instrument such as the Thermo Scientific Q Exactive Plus Orbitrap coupled to a Waters nanoflow liquid chromatograph (nanoLC), samples can be proteolytically digested directly in-solution without the need for SDS-PAGE. The LC-HRMS/MS process uses a nanoflow rate liquid chromatography (LC) system to first separate peptides based on hydrophobicity. As peptides are eluted from the LC system they are introduced to the mass spectrometer by electrospray ionization (ESI) where multiple charges are picked up by the peptides. The ionized peptides are then passed through a series of ion optics and the Orbitrap mass analyzer which allow for the measurement of the peptide mass-to-charge ratio (m/z) and assigned charge, from which peptide MWs can then be determined. After the m/z values of the intact peptides have been determined, selection for fragmentation is performed, which allows for the assignment of amino acid sequence through peptide-spectrum matching. As done for MALDI TOF/TOF MS data, the mass spectra of the LC-HRMS/MS analyzed samples are compared to a theoretical trypsin digest using an algorithm termed PSM analysis (Proteome Discoverer 2.0 or higher). Identification of protein toxin will be based on the detection of peptides with at least one PSM per peptide sequence obtained from the raw mass spectra data. Again as for MALDI TOF/TOF MS, each sample is reviewed following a two-tiered approach for the nanoLC-HRMS/MS data. Results The final estimate for the method LOD (mLOD) was obtained after testing 6 dilutions of ricin ranging from 0.1 ng/mL - 10 µg/mL. The method LOD was determined at the concentration where the protein identification sensitivity (Equation 2) was ≥ 50% at the R. communis level, and FNRs (Equation 3) ≤ 50%. The resulting LOD estimate for ricin identification is 1 µg/mL.

The ruggedness, which is the ability of the method to remain unaffected by variations in method parameters associated with differences in analyst performance, as well as between-day and betweeninstrument variations. The variations produced by different analysts and analysis on different days were assessed in the determination of the ruggedness of the identification of protein toxins by insolution trypsin digestion and LC-HRMS/MS for ricin reference samples. The method ruggedness was assessed by both intraday and between-day sensitivity for ricin digested at a high concentration (20 µg/mL) and negative control digested in a minimum of three replicates by two technicians. The high concentration for ricin was aliquoted, dried and reconstituted in 0.1% formic acid before analysis. Ruggedness was established given that all the high concentrations of toxin digested standard sensitivity measures were ≥ 80% and FNRs were ≤ 20%. Sensitivity was calculated using Equation 1 and FNR was calculated using Equation 2.

170

Within-day sensitivity was determined for ricin using replicate in-solution trypsin digestion of high concentration (20 µg/mL) samples and negative controls. Ricin within-day sensitivity was performed for a total of three days by two technicians using high concentration samples, which were aliquoted to a final concentration of 5 µg/mL before analysis by nanoLC-HRMS/MS to mitigate carryover. The results of the sensitivity and FNR analysis are reported below (Table 4). Table 4. Sensitivity and False Negative Rates for Ricin by nanoLC-HRMS/MS Ricin

R. communis

Test Conc. µg/mL

Sensitivity %

False Negative Rate %

Sensitivity %

False Negative Rate %

20

100

0

100

0

10

91

9

100

0

2

73

27

100

0

1

83

17

100

0

0.2

36

64

73

27

It is expected that sample matrix effects may have an impact on the ability of the in-solution trypsin digestion and LC-HRMS/MS to identify ricin because samples may contain additional unknown proteins. To test the effect of complex matrices on the identification of ricin, the results from 11 samples ranging in ricin concentration 0.0035 – 1.43% w/w per total dry mass of an unknown dry matrix were analyzed. The accuracy of ricin identification for the complex samples were assessed for accuracy based on the presence of ricin determined by orthogonal immunoassay methods (data not shown). The effect of the complex matrices on downstream analysis was also analyzed for effects of carry over and results provided below (Table 5). The matrices of these samples included unknown environmental contaminants and other proteins present from castor bean extractions. The reportable range, defined as the highest and lowest concentrations at which the test accuracy is maintained, has been evaluated for the LC-HRMS/MS instrument using digested ricin standard in which 100 ng of ricin was loaded on the instrument by injecting 2 µL from a high concentration sample. At 100 ng, detection of peptides originating from the digested standard were observed in sequential blank analysis and thus the upper limit for toxin load on the instrument has been established. Higher concentrations of ricin digest were not tested, as higher concentration of ricin are not expected to effect the digestion nor interfere with the accuracy of the analysis. Higher toxin load on the column may result in carryover which may interfere with the accuracy of sequential sample analyses. The method has been limited to no more than 10 ng of ricin to be loaded for analysis. Previously acquired data has demonstrated that a sample concentration of 20 µg/mL provides a sensitivity ≥ 95% (Equation 2) and carryover was not observed in the column after autosampler needle and column washing, and subsequent column blank analysis. The lower reportable range has been determined by the method LOD estimate (Table 4). No acceptance criteria were established for this range determination, and results are for information purposes only to guide analytical practices. The reliability measures of the false positive and false negative rates has been determined as a percentage from the positive or negative detection results from the identification of ricin tested at a high, intermediate, and zero (blank) level concentrations obtained during validation of the in171

solution trypsin digestion and LC-HRMS/MS procedures. The false positive and negative rates were calculated using the listed equations (Equations 2 and 3).

Table 5. Matrix Effect of Ricin Samples on MS Analyses

Sample ID

%Ricin* (w/w)

Sample 1

MALDI TOF/TOF MS (Non-reduced Band Results)

nanoLCHRMS/MS

ELISA

CFT

Carryover

0.93

Ricin Detected

Ricin Detected

Positive for Ricin

Positive for Ricin

–

Sample 2

0.94

Ricin Detected

Ricin Detected

Positive for Ricin

Positive for Ricin

–

Sample 3

1.21

Ricin Detected

Ricin Detected

Positive for Ricin

Positive for Ricin

–

Sample 4

0.67

Ricin Detected

Ricin Detected

Positive for Ricin

Positive for Ricin

–

Sample 5

0.72

Ricin Detected

Ricin Detected

Positive for Ricin

Positive for Ricin

–

Sample 6

0.55

Ricin Detected

Ricin Detected

Sample 7

1.29

Ricin Detected

Ricin Detected

Sample 8

0.82

Ricin Detected

Ricin Detected

Sample 9

1.43

Ricin Detected

Ricin Detected

Sample 10

1.06

Ricin Detected

Sample 11

0.0035

Ricin Detected

Positive for R. communis Positive for but inconclusive for Ricin Ricin

Yes

Positive for Ricin

Yes

Positive for Ricin

Positive for R. communis Positive for but inconclusive for Ricin Ricin

–

Positive for Ricin

Positive for Ricin

–

Ricin Detected

Positive for Ricin

Positive for Ricin

Yes

Ricin Detected

Analysis Not Performed**

Negative for Ricin

–

* Ricin

content of samples estimated from ELISA results per dry weight of original sample. ** Sample 11 estimated ricin content was below the expected limit of detection for SDS-PAGE and MALDI TOF/TOF MS analysis and analysis was not performed. – No carryover detected in the subsequent column blank processed immediately after the ricin sample.

A minimum of 10 replicates was used to test the reliability of the method, and represented as total sensitivity and calculated using Equation 1. A total six of these replicates were completed during method pre-qualification testing and used as part of the validation analyses along with an additional four replicates.

172

No false positive identification were made for either ricin, or at the R. communis detection level. The sensitivity and FNR criteria for high concentration samples is ≥ 80% and ≤ 20%, respectively. No set criteria was set for intermediate or low concentration samples.

Conclusions For analysis of ricin, SDS-PAGE followed by in gel trypsin digestion and MALDI TOF/TOF MS analysis provides a very specific method for the qualitative identification and characterization of ricin, however, detection limits of these combined analytical methods reduce the value of their use during routine confirmatory testing. The lower ricin concentrations detectable during in-solution digestion followed by LC-HRMS/MS indicates that this analytical process offers a more robust and sensitive methodology for confirmatory testing. Higher concentrations can be analyzed with the expectation that samples with an estimated concentration greater than 2 µg/mL will be treated after in-solution digestion in such a way that the amount of estimated ricin injected on the column be ≤ 10 ng for LCHRMS/MS analysis. The method was determined to be rugged as both the intraday and the between-day variation met the set criteria of FNR of ≤ 20% up to the 1 µg/mL concentration (Table 4). The sample matrix impact on the accuracy of the method was determined to be negligible based on the analysis of 11 samples with ricin concentrations 0.0035 – 1.43% w/w per total dry mass of an unknown dry matrix analyzed and compared to the results from immunoassay analyses; all but one samples was concurrent with the results of the immunoassays and concurrent or more specific for all samples tested compared to MALDI TOF/TOF MS analysis (Table 5). The one sample that did not match results with the immunoassays had an estimated concentration approximately 30-fold lower than the LC-HRMS/MS method LOD for ricin. The method limit of detection (mLOD) for the identification of ricin using ricin toxin specific peptides at 1.0 µg/mL with a sensitivity of 83% and a FNR of 17%. The mLOD for identification of peptides at the R. communis level but inconclusive for ricin identification was determined at 0.2 µg/ mL with a sensitivity of 73% and FNR of 27% (Table 3). No false positives were identified for either ricin blank samples. Criteria for false positive rate (FPR) of blank samples: FPR < 5%; was met. The FNR for ricin tested at the high sample concentration of (20 µg/mL) met the set criteria for FNR of sensitivity ≥ 80%, FNR ≤ 20%; for ricin at the specific and R. communis levels (Table 3). Overall, both mass spectrometry platforms provide specific confirmatory testing methods critical for the confirmation of ricin, however, in-solution digestion followed by LC-HRMS/MS analysis is more sensitive and provides for a relatively faster methodology. Additional approaches to enhance these methods could include upstream affinity chromatographic methods to concentrate ricin and remove interfering proteins and/or matrix. Subsequently, the method developed and validated for ricin has been successfully employed for five additional biological toxins, and more will be added, which gives credence to the robust nature of these method-based techniques in bioforensic confirmatory testing.

References 1. 2.

Maurer, S. M. Technologies of Evil: Chemical, Biological, Radiological, and Nuclear Weapons. In WMD Terrorism, Science and Policy Issues; MIT Press: Cambridge, MA, 2009; pp 47−110. U.S. Department of Justice. Amerithrax Investigative Summary; 2010. https://www.hsdl. org/?abstract&did=28996 (accessed April 2019)

173

HSPD-10 (Homeland Security Presidential Directive). Biodefense for the 21st Century; 2004. https://www.hsdl.org/?abstract&did=784400 (accessed April 2019) 4. Preisler, S. Ricin: Kitchen Improvised Devastation. In Uncle Fester: Silent Death, Revised and Expanded, 2nd ed.; Festering Publications: Green Bay, WI, 1989; pp 107−118. 5. Frawley, D. A.; Samaan, M. N.; Bull, R. L.; Robertson, J. M.; Mateczun, A. J.; Turnbull, P. C. B. Recovery Efficiencies of Anthrax Spores and Ricin from Nonporous or Nonabsorbent and Porous or Absorbent Surfaces by a Variety of Sampling Methods. J. Forensic Sci. 2008, 53, 1102–1107. 6. Lindsey, C. Y.; Richardson, J. D.; Brown, J. E.; Hale, M. L. Intralaboratory Validation of Cell-Free Translation Assay for Detecting Ricin Biological Activity. J. AOAC Int. 2007, 90, 1316–1325. 7. Garrett, A.; Vereecke, K.; Brown, N.; Hanlon, A.; Cardamone, A.; Lehman, R.; Merkley, E.; Jarman, K.; Wunschel, D.; Cendrowski, S.; Wahl, K. L.; Burans, J. ISO 17025 Validation of MethodBased Mass Spectrometry Techniques for the Identification of Ricin in Bioforensic Samples. Presented at the ACS National Meeting, August 2018 Program Book p 96, ANYL 485. 8. Bozza, W. P.; Tolleson, W. H.; Rivera Rosado, L. A.; Zhang, B. Ricin Detection: Tracking Active Toxin. Biotech. Adv. 2015, 33, 117–123. 9. Endo, Y.; Tsurugi, K. RNA N-Glycosidase Activity of Ricin A-chain. Mechanism of Action of the Toxic Lectin Ricin on Eukaryotic Ribosomes. J. Biol. Chem. 1987, 262, 8128–8130. 10. Walsh, M. J.; Dodd, J. E.; Hautbergue, G. M. Ribosome-Inactivating Proteins, Potent Poisons and Molecular Tools. Virulence 2013, 4, 774–784. 11. Budowle, B.; Schutzer, S. E.; Morse, S. A.; Martinez, K. F.; Chakraborty, R.; Marrone, B. L.; Messenger, S. L.; Murch, R. S.; Jackson, P. J.; Williamson, P.; Harmon, R.; Velsko, S. P. Criteria for Validation of Methods in Microbial Forensics. App. Env. Microbiol. 2008, 74, 5599–5607. 12. Worbs, S.; Skiba, M.; Söderström, M.; Rapinoja, M.; Zeleny, R.; Russmann, H.; Schimmel, H.; Vanninen, P.; Fredriksson, S.; Dorner, B. G. Characterization of Ricin and R. communis Agglutinin Reference Materials. Toxins 2015, 7, 4906–4934. 3.

174

Chapter 11

Unambiguous Identification of Ricin and Abrin with Advanced Mass Spectrometric Assays Suzanne R. Kalb*,1 and François Becher*,2 1National Center for Environmental Health, Centers for Disease Control and Prevention,

4770 Buford Highway, NE, Atlanta, Georgia 30341, United States 2Service de Pharmacologie et Immunoanalyse, Laboratoire d’Etude du Métabolisme des

Médicaments, Commissariat à l’Énergie Atomique et aux Énergies Alternatives, Institut National de la Recherche Agronomique, Université Paris Saclay, 91191 Gif-sur-Yvette, France *E-mails: [email protected]; [email protected].

Due to the potential misuse as biothreat agents, ricin and abrin are the focus of surveillance by international and national authorities. Ricin is a protein toxin produced by the castor bean plant (Ricinus communis), whereas abrin is contained in the seeds of Abrus precatorius found in tropical regions. Ricin and abrin belong to the ribosomal inactivating protein class II (RIP II) which possess A- and B-chains. The A-chain mediates the cessation of protein synthesis due to depurination of 28S ribosomal RNA, and the B-chain binds to glycoconjugates of target cells. Fast, sensitive and reliable methods are of great importance for early identification. Advanced mass spectrometric (MS) assays were reported to unambiguously identify ricin or abrin and to detect RIP activity in samples with complex matrices. These assays enable detection and differentiation of ricin or abrin through specific determination of the amino acid sequence of the protein, and active toxins can be monitored from the depurination of a nucleic acid substrate. Because of the inherent susceptibility of MS to matrix effects in the wide range of complex environmental or biological samples investigated in biodefense scenarios, prior purification and enrichment of toxins is most critical for efficient detection. In this review, we describe recent advances in MS-based methods for ricin and abrin detection, with a particular emphasis on affinity extraction protocols.

Introduction Ricin and abrin are protein toxins contained in the seed of the plants Ricinus communis and Abrus precatorius, respectively. Ricinus communis is widespread throughout subtropical and temperate regions, whereas Abrus precatorius is found in tropical regions. Both toxins are similarly composed © 2019 American Chemical Society

of two polypeptide chains, known as the A and B-chains, which are each approximately 30-32 kDa, making intact ricin and abrin approximately 60-64 kDa with glycosylation (1). Ricin and abrin belong to the family of type 2 ribosome-inactivating protein toxins (RIP-II toxins). The A-chain has a specific enzymatic activity, as it depurinates a single adenosine that is part of a GAGA tetraloop of 28S ribosomal RNA (2, 3). This prevents the binding of elongation factor 2 (EF-2) to the ribosome, leading to the cessation of protein synthesis (4). The termination of protein synthesis is responsible for cellular apoptosis. The B-chain has lectin activity and brings the enzymatically active A-chain to its target through cell receptor binding and endocytosis (5). Both chains are therefore needed for in vivo toxicity (6). Ricin and abrin are highly dangerous with similar minimum lethal doses in mice at 2.7 and 0.7 µg/kg body weight after i.v. injection, respectively (7). In addition to the N-glycosylation sites on the A or B-chain, sequence heterogeneity comes from different isoforms. Ricin isoforms D and E differ in their B-chain sequence (8), with the Bchain of ricin E sharing its N-terminus with ricin D and the C-terminus with R. communis agglutinin (RCA120), a protein which is significantly less toxic than ricin. Regarding abrin, four different isoforms with 78% protein similarity were reported (9, 10), abrin-a, -b, c and –d respectively. In contrast to the D and E isoforms of ricin, amino-acid variations in abrin isoforms are distributed all along the sequence (11). Reliable identification of the toxins is further challenged by co-isolated protein from the seeds. Abrus and Ricinus agglutinins have high sequence homology (up to 89% for ricin and agglutinin), making the toxins and agglutinins hard to distinguish by antibodies due to cross-reactivity (12). Mass spectrometry techniques involving either electrospray ionization (ESI) or matrix-assisted laser desorption/ionization (MALDI) have been widely used for identification and/or quantification of protein toxins in complex environments (13). Mass spectrometry provides precise mass measurement of protein sequences for accurate identification, differentiation of protein with sequence homology, and robust quantification. A wide variety of protocols involving MS and MS/ MS measurements were published in the last fifteen years for ricin (14). Methods mostly relied on ricin digestion into peptides by trypsin, and MS detection of peptides unique to the protein, i.e. signature or proteotypic peptides. Proteomic methods were based on data-dependent acquisitions (DDA), where the 3 to 10 most intense peptide’s parent ions in the MS survey scan are selected for MS/MS analysis. Targeted detection of ricin was subsequently developed for higher selectivity and sensitivity (15). The best unique ricin peptides with favorable physico-chemical properties regarding LC-MS/MS analysis were pre-selected, and detected by tandem mass spectrometry. The selected reaction monitoring mode (SRM) is based on two stages of ion selection, the parent ion and fragment ions generated by collision induced dissociation (CID), in a triple quadrupole instrument (16). Immuno-enrichment was combined with targeted mass spectrometry to extract and purify ricin from complex samples before MS (17–22). Such a combination of separation techniques constitutes an efficient methodology to detect trace amounts in the diversity of matrices from the security sector. To gain further information on sample toxicity, mass spectrometry protocols based on the RNA Nglycosidase activity were developed for ricin (17, 20, 23, 24). In the last five years, improvements to the mass spectrometry protocols for unambiguous identification of ricin or abrin in a wide range of matrices were reported. Recent developments in targeted analyses, proteomics, activity-based methods and affinity enrichment are reviewed in this paper.

176

Targeted MS/MS Monitoring of specific SRM transitions from selected peptides allows an analysis of ricin peptides with high selectivity and generally achieves detection of a peptide at a lower limit of detection than other types of mass spectrometers, as seen in several reports in which this approach was used to identify ricin spiked into complex matrices (20, 22, 25). Determination of the amount of toxin in the unknown sample is also of prime importance for public health considerations. Absolute toxin quantification was achieved by isotope dilution mass spectrometry (22, 26). Synthetic versions of each signature peptide containing an amino acid labeled with a stable isotope (for example, 15N or 13C) were spiked after digestion by trypsin and used as internal standards. The labeled internal standards behave identically to their native counterparts during LC and tandem MS analysis. The approach was applied to beverages (milk, tap-water and juices (26), and more recently to the accurate quantification of ricin and agglutinin (RCA120) in extracts of castor beans (22). Quantification of ricin in the seeds of 18 representative cultivars of R. communis showed concentrations ranging from 5.7 to 20.5 mg/g among the cultivars with an average ricin content of 9.3 mg/g seed. The methods also identified similar amounts of ricin and RCA120. Application of the assay in the security sector is expected for accurate analysis of “white powder”, supporting public health and law enforcement. SRM assays can be multiplexed for the simultaneous analysis of several proteins in a single assay, as recently illustrated with eight Chemical, Biological, Radiological and Nuclear defense (CBRN)relevant toxins, including ricin, in a food matrix consisting of soup (25). Absolute quantification was reached by an alternative approach to the peptide-based tactic presented above that included the use of isotopically labelled proteins, analogues of the proteins to be assayed. Protein Standards for Absolute Quantification (PSAQ) standards (27) of each of the eight toxins were spiked early in the samples, thus correcting for protein losses during sample preparation and incomplete toxin digestion. Good quantitative performances were demonstrated in the soup samples (25). In addition, the PSAQ strategy would avoid the preparation of external protein calibration standard curves for quantification in unknown samples. High resolution mass spectrometry can further increase sensitivity and specificity performance in the field of toxin analysis. The Q-Exactive instrument (28) is composed of the orbitrap analyser combined with a quadrupole mass filter as front-end. The PRM mode or parallel reaction monitoring mode performed on the high resolution instrument demonstrated higher selectivity in comparison with SRM in complex matrices where the proteomic background has similar composition to the protein of interest (29, 30). The simultaneous high resolution/accurate mass detection in the Orbitrap of all potential fragment ions from targeted signature peptide precursors contribute to the higher identification confidence (18). The PRM mode was applied to the multiplex analyses of three CBRN toxins: ricin, SEB and epsilon toxin in food samples (18). In this work, mass spectrometric sensitivity was further increased by summing the signal of several non-interfered fragment ions from a common peptide precursor to provide one extracted ion chromatogram (XIC) for each targeted peptide. This process allowed up to a 10-fold final increase in the XIC area. Combined with PSAQ internal standards, PRM demonstrated accurate quantification of the three toxins in human serum and milk with LLOQ at or below 1 ng/mL. The PRM mode was recently extended to the plant toxin abrin, mentioned on the U.S. Select Agents and Toxins list (31). Abrin belongs, as well as ricin, to the ribosomal inactivating protein class II (RIP II) with similar estimated human lethal dose (7). A reliable method was therefore of great importance for identification and quantification of abrin in forensic samples. In view of the 177

protein sequence heterogeneity and sequence overlap with Abrus agglutinin, the multiplex ability of PRM was used for the specific monitoring of 14 abrin peptides. Peptides common or specific to isoforms were targeted allowing rapid differentiation, and simultaneous quantification of abrin and its four isoforms in milk, ham, river water, soil and human serum (Figure 1). In addition, the method verified the main occurrence of abrin-a and minor, but similar, amounts of abrin-b and abrin-c/-d in purified extract material of Abrus precatorius seeds.

Figure 1. Workflow for the rapid absolute quantification of abrin and its isoforms. Abrin spiked in different complex matrices interacts with antibodies immobilized on magnetic beads. After several washing steps to remove unspecific bindings, an on-bead tryptic digest is performed with the assistance of a sonicator and a surfactant. The labeled peptides are added for absolute quantification purposes, and the surfactant is removed. The resulting peptides were analyzed with the LC-MS/MS (PRM) method. Reproduced with permission from reference (31). Copyright 2017 American Chemical Society.

Proteomic MS/MS Scanning tandem mass spectrometers such as quadrupole-time-of-flight (QTOF) produce a full spectrum of fragment ions from a peptide precursor ion selected for MS/MS analysis. If used in combination with LC, the retention time, mass of the intact peptide and the masses of fragment ions can be collectively used to identify a specific peptide based on its amino acid sequence. Provided that the sequence of the peptide is unique, this can be used to unambiguously identify a protein. Such a process was first used in 2005 to identify ricin, yielding 14 peptides from the A-chain of ricin and 16 peptides from the B-chain (32). Similar methods have been reported to identify ricin spiked into complex matrices (20) as well as abrin (33). Also of note are improvements in the digestion process, either through the use of solvent-assisted tryptic digestion (34), or microwave-assisted hot acid digestion (35) both of which shorten the time needed for production of tryptic peptides. Further improvement in digestion time could be considered based on recent works. For instance, online digestion using immobilized trypsin was demonstrated to enhance assay precision and maximize throughput for other proteins (36, 37). The use of PCR thermocycler was also reported to significantly shorten digestion time, resulting in only 5-minutes cycle (38). Ricin and abrin could benefit from those innovative strategies. One interesting recent use of proteomic MS/MS as it relates to ricin involves the study of tryptic peptides emanating from ricin and how they relate to other proteins (39). In this work, the authors 178

compare the amino acid sequence of ricin to other proteins produced by R. communis which have a similar amino acid sequence, such as RCA120 which is 89% identical to ricin as well as ricin-like proteins, whose sequence identity varies from 34% to 94% compared to ricin. The accurate amino acid identification of large portions of proteins provided by proteomic MS/MS study allows for differentiation between these closely related proteins in many cases. Tryptic peptides were identified and classified as being either unique to ricin, existing in ricin and one (or more) other R. communis proteins, or belonging to ricin and a protein from another species. Through this and by learning that ricin-like proteins appear to be 1-2 orders of magnitude less abundant than ricin or RCA120, (as estimated by the low-precision spectral counting method) it is possible to use proteomics MS/MS to unambiguously identify the presence of ricin.

Measuring Ricin’s Activity with Mass Spectrometry As stated earlier, ricin and abrin arrest protein synthesis and cause cell death by depurinating an adenosine in the GAGA tetraloop of 28S ribosomal RNA. The enzymatic activity of the A-chain of ricin upon a synthetic substrate which mimics the depurination site of 28S ribosomal RNA can be monitored by mass spectrometry. When ricin depurinates the adenosine in the GAGA tetraloop, this results in the formation of free adenine, which can be measured by mass spectrometry. This was first reported in 2004 using selected ion monitoring in which the m/z of adenine (136) was recorded following incubation of ricin with a synthetic RNA substrate (23). While this approach successfully identified the presence of enzymatically active ricin, there was little specificity, both in terms of biological specificity and analyte specificity. Several years later, a similar method improved upon this idea by monitoring the depurination of a synthetic RNA substrate with MS/MS, using selected reaction monitoring (SRM) to study the fragmentation of adenine from m/z 136 to 119, yielding analyte specificity (17). Biological specificity was also added through an additional technique discussed shortly. The specificity of SRM was used a few years later to study the enzymatic activity of ricin in 18 different R. communis cultivars (22), in addition to the quantification of the ricin in these cultivars as discussed earlier. The formation of free adenine following the depurination of a synthetic DNA substrate was monitored through the fragmentation of three adenine transitions; m/z 136 to 119, 136 to 92, and 136 to 65. This method was used to quantify the overall N-glycosidase activity in 18 cultivars. Because the R. communis cultivars also contain RCA120, it was not possible to determine the enzymatic activity of ricin alone. One serious limitation of detection of ricin via its enzymatic activity is the lack of biological specificity as several RIP-II toxins have the same enzymatic activity as ricin. Specificity can be addressed through the addition of another detection method. Nonetheless, detection of the enzymatic activity of ricin and abrin is important as it is through the enzymatic activity that ricin and abrin become health risks; therefore, the ability to differentiate between active and inactive toxin is essential to determine the health threat posed by ricin (40). For this reason, public health agencies in particular have utilized SRM methods to detect enzymatically active ricin (22) and have developed (20) and optimized (24) MALDI-TOF mass spectrometry methods which measure depurination of a synthetic DNA or RNA substrate as a measure of toxin activity (Figure 2).

179

Figure 2. Detection of the cleavage product from RNA14 by various concentrations of ricin after incubation for 0.5 h (A) and 4 h (B). The toxin was spiked in 0.5 mL of PBST buffer (●) or 2% milk (○) followed by enrichment with specific antibodies immobilized on magnetic beads. (C). Some typical mass spectra obtained from the aliquots of the ricin activity reaction (55 °C, 4 h). Reproduced with permission from reference (24). Copyright 2016 American Chemical Society.

180

Affinity Extraction A purification and enrichment step prior to MS or MS/MS analysis of ricin or abrin aids in decreasing background interferences and lowers the limit of detection. This step can also add biological specificity to the mass spectrometric analysis provided that the technique used for enrichment has high and selective affinity for that analyte. One popular method used is immunoaffinity, or the use of an antibody to select for ricin. This technique was first combined in 2007 with LC-MS/MS analysis of the adenine released by depurination of a synthetic RNA substrate by ricin (17). It is important to note that this technique employed a monoclonal antibody against the B-chain of ricin, eliminating the potential for inhibition of the enzymatically-active A-chain. Immunoaffinity has been used in combination with mass spectrometry for several ricin detection methods, including MS analysis of ricin’s enzymatic activity (20) and MALDI-TOF analysis of a tryptic digest of ricin (19). One recent interesting use of immunoaffinity coupled with mass spectrometry for ricin analysis involves a multiplex analysis for multiple toxins. The first approach is able to detect four separate protein toxins using immunoaffinity followed by a tryptic digest of the toxin and analysis of the tryptic fragments by MALDI-TOF (21). A second method already mentioned earlier is applicable to three protein toxins (ricin, SEB, and epsilon toxin) in which immunoaffinity supplemented the PRM analysis of the toxins spiked into foods (18). The increase in analyte specificity was achieved through the PRM analysis, but immunoaffinity assisted in increasing biological specificity. Both methods report the ability to detect and differentiate multiple protein toxins, including ricin. Another recent interesting combination of immunoaffinity and mass spectrometry involves differentiation of four abrin isoforms as mentioned earlier (31). This approach is similar to a multiplex assay as it used multiple antibodies to capture the isoforms of abrin from complex matrices (Figure 1) with differentiation through examination of the amino acid sequence of the isoforms. In these cases, immunoaffinity was critical to the success of these methods, especially when the toxin(s) were present in a complex matrix. An alternative to antibodies for affinity extraction involves the use of carbohydrate ligands. Galactose affinity in particular has been reported for large-scale purification of RIP-II toxins (41, 42). Carbohydrate affinity combined with mass spectrometric analysis of ricin first involved the use of lactose immobilized to silica for ricin extraction followed by tryptic digestion and LC-MS identification of the tryptic fragments (43). A more recent report used galactose coupled to chromatographic resin to extract multiple RIP-II toxins (including ricin and abrin) followed by LCMS/MS to identify the toxins after tryptic digestion (33). Carbohydrate affinity is an attractive alternative to immunoaffinity for ricin and abrin extraction due to its decreased cost; however, it has less specificity than immunoaffinity. This lack in specificity can often be addressed through the accuracy afforded by the mass spectrometer.

Conclusions Mass spectrometric measurements of ricin and abrin are important for accurate identification of these toxic proteins, particularly in the presence of complex matrices. Some mass spectrometric measurements, such as enzymatic activity measurements, act as auxiliary methods which, when used in concert with other methods, yield information to assist in unambiguous ricin and abrin determination and differentiation. Other mass spectrometric measurements, such as targeted and proteomic MS/MS measurements, work alone as a definitive identification of ricin or abrin through determining the unique amino acid sequence of ricin or abrin. In many cases, the use of an affinity 181

technique prior to mass spectrometric analysis is a necessity for unambiguous toxin detection, especially in the presence of complex matrices. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.

References 1. 2. 3.

4.

5.

6.

7. 8.

9. 10. 11.

12.

13.

14.

15.

Bradberry, S. M.; Dickers, K. J.; Rice, P.; Griffiths, G. D.; Vale, J. A. Ricin Poisoning. Toxicol. Rev. 2003, 22, 65–70. Amukele, T. K.; Roday, S.; Schramm, V. L. Ricin A-Chain Activity on Stem-Loop and Unstructured DNA Substrates. Biochemistry 2005, 44, 4416–4425. Endo, Y.; Mitsui, K.; Motizuki, M.; Tsurugi, K. The Mechanism of Action of Ricin and Related Toxic Lectins on Eukaryotic Ribosomes. The Site and the Characteristics of the Modification in 28 S Ribosomal RNA Caused by the Toxins. J. Biol. Chem. 1987, 262, 5908–5912. Montanaro, L.; Sperti, S.; Mattioli, A.; Testoni, G.; Stirpe, F. Inhibition by Ricin of Protein Synthesis in Vitro. Inhibition of the Binding of Elongation Factor 2 and of Adenosine DiphosphateRibosylated Elongation Factor 2 to Ribosomes. Biochem. J. 1975, 146, 127–131. Simmons, B. M.; Stahl, P. D.; Russell, J. H. Mannose Receptor-Mediated Uptake of Ricin Toxin and Ricin A Chain by Macrophages. Multiple Intracellular Pathways for a Chain Translocation. J. Biol. Chem. 1986, 261, 7912–7920. Foxwell, B. M.; Blakey, D. C.; Brown, A. N.; Donovan, T. A.; Thorpe, P. E. The Preparation of Deglycosylated Ricin by Recombination of Glycosidase-Treated A- and B-Chains: Effects of Deglycosylation on Toxicity and in Vivo Distribution. Biochim. Biophys. Acta 1987, 923, 59–65. Gill, D. M. Bacterial Toxins: A Table of Lethal Amounts. Microbiol. Rev. 1982, 46, 86–94. Araki, T.; Funatsu, G. The Complete Amino Acid Sequence of the B-Chain of Ricin E Isolated from Small-Grain Castor Bean Seeds. Ricin E Is a Gene Recombination Product of Ricin D and Ricinus Communis Agglutinin. Biochim. Biophys. Acta. 1987, 911, 191–200. Lin, J. Y.; Lee, T. C.; Hu, S. T.; Tung, T. C. Isolation of Four Isotoxic Proteins and One Agglutinin from Jequiriti Bean (Abrus Precatorius). Toxicon. 1981, 19, 41–51. Tahirov, T. H.; Lu, T. H.; Liaw, Y. C.; Chen, Y. L.; Lin, J. Y. Crystal Structure of Abrin-A at 2.14 A. J. Mol. Biol. 1995, 250, 354–367. Hung, C. H.; Lee, M. C.; Lee, T. C.; Lin, J. Y. Primary Structure of Three Distinct Isoabrins Determined by CDNA Sequencing. Conservation and Significance. J. Mol. Biol. 1993, 229, 263–267. Simon, S.; Worbs, S.; Avondet, M. A.; Tracz, D. M.; Dano, J.; Schmidt, L.; Volland, H.; Dorner, B. G.; Corbett, C. R. Recommended Immunological Assays to Screen for Ricin-Containing Samples. Toxins (Basel) 2015, 7, 4967–4986. Duriez, E.; Armengaud, J.; Fenaille, F.; Ezan, E. Mass Spectrometry for the Detection of Bioterrorism Agents: From Environmental to Clinical Applications. J. Mass Spectrom. 2016, 51, 183–199. Kalb, S. R.; Schieltz, D. M.; Becher, F.; Astot, C.; Fredriksson, S. A.; Barr, J. R. Recommended Mass Spectrometry-Based Strategies to Identify Ricin-Containing Samples. Toxins (Basel) 2015, 7, 4881–4894. Gillette, M. A.; Carr, S. A. Quantitative Analysis of Peptides and Proteins in Biomedicine by Targeted Mass Spectrometry. Nat. Methods 2013, 10, 28–34.

182

16. McReynolds, J. H.; Anbar, M. Isotopic Assay of Nanomole Amounts of Nitrogen-15 Labeled Amino Acids by Collision-Induced Dissociation Mass Spectrometry. Anal. Chem. 1977, 49, 1832–1836. 17. Becher, F.; Duriez, E.; Volland, H.; Tabet, J. C.; Ezan, E. Detection of Functional Ricin by Immunoaffinity and Liquid Chromatography-Tandem Mass Spectrometry. Anal. Chem. 2007, 79, 659–665. 18. Dupre, M.; Gilquin, B.; Fenaille, F.; Feraudet-Tarisse, C.; Dano, J.; Ferro, M.; Simon, S.; Junot, C.; Brun, V.; Becher, F. Multiplex Quantification of Protein Toxins in Human Biofluids and Food Matrices Using Immunoextraction and High-Resolution Targeted Mass Spectrometry. Anal. Chem. 2015, 87, 8473–8480. 19. Duriez, E.; Fenaille, F.; Tabet, J. C.; Lamourette, P.; Hilaire, D.; Becher, F.; Ezan, E. Detection of Ricin in Complex Samples by Immunocapture and Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry. J. Proteome Res. 2008, 7, 4154–4163. 20. Kalb, S. R.; Barr, J. R. Mass Spectrometric Detection of Ricin and Its Activity in Food and Clinical Samples. Anal. Chem. 2009, 81, 2037–2042. 21. Kull, S.; Pauly, D.; Stormann, B.; Kirchner, S.; Stammler, M.; Dorner, M. B.; Lasch, P.; Naumann, D.; Dorner, B. G. Multiplex Detection of Microbial and Plant Toxins by Immunoaffinity Enrichment and Matrix-Assisted Laser Desorption/Ionization Mass Spectrometry. Anal. Chem. 2010, 82, 2916–2924. 22. Schieltz, D. M.; McWilliams, L. G.; Kuklenyik, Z.; Prezioso, S. M.; Carter, A. J.; Williamson, Y. M.; McGrath, S. C.; Morse, S. A.; Barr, J. R. Quantification of Ricin, RCA and Comparison of Enzymatic Activity in 18 Ricinus Communis Cultivars by Isotope Dilution Mass Spectrometry. Toxicon. 2015, 95, 72–83. 23. Hines, H. B.; Brueggemann, E. E.; Hale, M. L. High-Performance Liquid Chromatography-Mass Selective Detection Assay for Adenine Released from a Synthetic RNA Substrate by Ricin A Chain. Anal. Biochem. 2004, 330, 119–122. 24. Wang, D.; Baudys, J.; Barr, J. R.; Kalb, S. R. Improved Sensitivity for the Qualitative and Quantitative Analysis of Active Ricin by MALDI-ToF Mass Spectrometry. Anal. Chem. 2016, 88, 6867–6872. 25. Gilquin, B.; Jaquinod, M.; Louwagie, M.; Kieffer-Jaquinod, S.; Kraut, A.; Ferro, M.; Becher, F.; Brun, V. A Proteomics Assay to Detect Eight CBRN-Relevant Toxins in Food. Proteomics 2017, 17. 26. McGrath, S. C.; Schieltz, D. M.; McWilliams, L. G.; Pirkle, J. L.; Barr, J. R. Detection and Quantification of Ricin in Beverages Using Isotope Dilution Tandem Mass Spectrometry. Anal. Chem. 2011, 83, 2897–2905. 27. Dupuis, A.; Hennekinne, J. A.; Garin, J.; Brun, V. Protein Standard Absolute Quantification (PSAQ) for Improved Investigation of Staphylococcal Food Poisoning Outbreaks. Proteomics 2008, 8, 4633–4636. 28. Michalski, A.; Damoc, E.; Hauschild, J. P.; Lange, O.; Wieghaus, A.; Makarov, A.; Nagaraj, N.; Cox, J.; Mann, M.; Horning, S. Mass Spectrometry-Based Proteomics Using Q Exactive, a HighPerformance Benchtop Quadrupole Orbitrap Mass Spectrometer. Mol. Cell. Proteomics 2011, 10, M111.011015. 29. Gallien, S.; Duriez, E.; Crone, C.; Kellmann, M.; Moehring, T.; Domon, B. Targeted Proteomic Quantification on Quadrupole-Orbitrap Mass Spectrometer. Mol. Cell. Proteomics 2012, 11, 1709–1723.

183

30. Peterson, A. C.; Russell, J. D.; Bailey, D. J.; Westphall, M. S.; Coon, J. J. Parallel Reaction Monitoring for High Resolution and High Mass Accuracy Quantitative, Targeted Proteomics. Mol. Cell. Proteomics 2012, 11, 1475–1488. 31. Hansbauer, E. M.; Worbs, S.; Volland, H.; Simon, S.; Junot, C.; Fenaille, F.; Dorner, B. G.; Becher, F. Rapid Detection of Abrin Toxin and Its Isoforms in Complex Matrices by Immuno-Extraction and Quantitative High Resolution Targeted Mass Spectrometry. Anal. Chem. 2017, 89, 11719–11727. 32. Fredriksson, S. A.; Hulst, A. G.; Artursson, E.; de Jong, A. L.; Nilsson, C.; van Baar, B. L. Forensic Identification of Neat Ricin and of Ricin from Crude Castor Bean Extracts by Mass Spectrometry. Anal. Chem. 2005, 77, 1545–1555. 33. Fredriksson, S. A.; Artursson, E.; Bergstrom, T.; Ostin, A.; Nilsson, C.; Astot, C. Identification of RIP-II Toxins by Affinity Enrichment, Enzymatic Digestion and Lc-Ms. Anal. Chem. 2015, 87, 967–974. 34. Ostin, A.; Bergstrom, T.; Fredriksson, S. A.; Nilsson, C. Solvent-Assisted Trypsin Digestion of Ricin for Forensic Identification by LC-ESI MS/MS. Anal. Chem. 2007, 79, 6271–6278. 35. Chen, D.; Bryden, W. A.; Fenselau, C. Rapid Analysis of Ricin Using Hot Acid Digestion and MALDI-ToF Mass Spectrometry. J. Mass Spectrom. 2018, 53, 1013–1017. 36. Kuklenyik, Z.; Jones, J. I.; Toth, C. A.; Gardner, M. S.; Pirkle, J. L.; Barr, J. R. Optimization of the Linear Quantification Range of an Online Trypsin Digestion Coupled Liquid Chromatography–Tandem Mass Spectrometry (LC–MS/MS) Platform. Instrumentation Science & Technology 2018, 46, 102–114. 37. Toth, C. A.; Kuklenyik, Z.; Barr, J. R. Nuts and Bolts of Protein Quantification by Online Trypsin Digestion Coupled LC-MS/MS Analysis. Methods Mol. Biol. 2019, 1871, 295–311. 38. Turapov, O. A.; Mukamolova, G. V.; Bottrill, A. R.; Pangburn, M. K. Digestion of Native Proteins for Proteomics Using a Thermocycler. Anal. Chem. 2008, 80, 6093–6099. 39. Merkley, E. D.; Jenson, S. C.; Arce, J. S.; Melville, A. M.; Leiser, O. P.; Wunschel, D. S.; Wahl, K. L. Ricin-Like Proteins from the Castor Plant Do Not Influence Liquid Chromatography-Mass Spectrometry Detection of Ricin in Forensically Relevant Samples. Toxicon. 2017, 140, 18–31. 40. Bozza, W. P.; Tolleson, W. H.; Rivera Rosado, L. A.; Zhang, B. Ricin Detection: Tracking Active Toxin. Biotechnol. Adv. 2015, 33, 117–123. 41. Olsnes, S.; Pihl, A. Different Biological Properties of the Two Constituent Peptide Chains of Ricin, a Toxic Protein Inhibiting Protein Synthesis. Biochemistry 1973, 12, 3121–3126. 42. Lin, T. T.; Li, S. L. Purification and Physicochemical Properties of Ricins and Agglutinins from Ricinus Communis. Eur. J. Biochem. 1980, 105, 453–459. 43. Kanamori-Kataoka, M.; Kato, H.; Uzawa, H.; Ohta, S.; Takei, Y.; Furuno, M.; Seto, Y. Determination of Ricin by Nano Liquid Chromatography/Mass Spectrometry after Extraction Using Lactose-Immobilized Monolithic Silica Spin Column. J. Mass Spectrom. 2011, 46, 821–829.

184

Chapter 12

Challenges in the Development of Reference Materials for Protein Toxins R. Zeleny,1 A. Rummel,2 D. Jansson,3 and B. G. Dorner*,4 1European Commission, Joint Research Centre (JRC), 2440 Geel, Belgium 2toxogen GmbH, 30625 Hannover, Germany 3Swedish Defence Research Agency, 90182 Umeå, Sweden 4Robert Koch Institute, Biological Toxins (ZBS3), 13353 Berlin, Germany *E-mail: [email protected].

High molecular weight protein toxins produced by bacteria, e.g. staphylococcal enterotoxins and botulinum neurotoxins, as well as plant toxins such as ricin and abrin, are relevant analytes in different application areas: food safety, public health, civil security and defense sector, and – in case of botulinum neurotoxins – also in pharmaceutics. For their reliable and accurate detection, identification and quantification, reference materials (RMs), in particular certified reference material (CRM), are required. The present article focuses on challenges in the development (production and certification) of such RMs. Firstly, it highlights the role of RMs and CRMs, what they can be used for, the nature of certified properties, metrological traceability, and uncertainty of certified values, as well as commutability of RMs. Secondly, the molecule-specific technical challenges are highlighted using the example of the mentioned toxins. This includes for instance the choice of a suitable purification strategy (recombinant expression and purification versus the purification of toxin from natural sources), the in-depth characterization of the obtained preparations by a comprehensive set of methods including immunochemical assays, mass spectrometry, and functional assays to verify their identity and establish their purity and activity, and finally, suitable approaches for determining reference values of important toxin properties (protein mass concentration in solution, biological activity). The article summarizes ongoing activities in a new European initiative called EuroBioTox, which aims at the production and certification of RMs for selected protein toxins and the establishment of validated procedures for the detection and identification of biological toxins.

© 2019 American Chemical Society

Biological Toxins Biological toxins are a large group of hazardous substances produced by living organisms such as bacteria, plants, or animals, which exert a detrimental effect on other organisms upon uptake. Among those, high molecular weight protein toxins deserve special attention since they are relevant in different fields: Bacterial toxins such as staphylococcal enterotoxins (SE) or botulinum neurotoxins (BoNT) are known as causative agents of food poisoning outbreaks and are therefore monitored by national and international food and health agencies (1–3). On the other hand, the same toxins have been linked to military research programs in the past where staphylococcal enterotoxin B (SEB) has been explored as incapacitating agent and BoNT has been weaponized under the code name “X” (4–6). Along the same line, the plant toxins ricin or abrin have long been known to induce natural intoxications in humans and animals (7, 8). Ricin, similar to BoNT, has a history of military research (code name “W”) and is the only protein toxin listed in Schedule 1 of the Chemical Weapons Convention (9). Based on the potential threat level in public health incidents, BoNT as well as ricin and SE have been classified as bioterrorism agents of the highest (BoNT) or second highest (ricin, SE) category (10, 11). Recent incidents in Europe and worldwide have threatened civil society by the attempted use of different biological toxins. Exemplarily, in June 2018 a biological terror attack was thwarted in Cologne, Germany, where the suspect was accused of having manufactured ricin and acquired bomb-making materials for a serious act of violence against the state (12). Therefore, increased vigilance and adequate preparation is of importance in a world facing growing risks of man-made disasters. From an analytical perspective, protein toxins have several commonalities even though they are produced by different organisms and have quite different structures and functions (13): 1) They are toxic in the absence of the producing organism and its genetic information. Therefore, detection has to focus on protein-based methods, DNA-based methods are not sufficient in most instances. 2) Their very high toxicity demands for highly sensitive methods, optimally with detection limits in the pg/mL range when different sample types are analyzed. 3) They are rapidly metabolized after uptake in the human body, limiting the time window for detection in clinical specimens. 4) Finally and most challenging, they are often produced in multiple variants or isoforms which might differ in their toxicokinetics and toxicodynamics. Detailed information on the individual toxins, their biological structure and function has been reviewed elsewhere (for an overview please see (13–16)). Briefly, for the detection of biological toxins, a variety of methods has been established based on immunological, spectrometric, and functional assays or combinations thereof – all of those have advantages and limitations. Among those, immunoassays such as sandwich enzyme-linked immunosorbent assays (ELISA) display the presence of the analyte and highlight the native folding of the molecule depending on the antibodies used (13, 17–19). Among all technologies available, ELISA-based methods still provide the highest sensitivity with detection limits in the ng/mL to fg/mL range provided that high-affinity antibodies are used (18, 19). However, for ELISA-based methods, the recognition of individual toxin variants has to be comprehensively tested, and the discrimination of closely related toxin variants is not always feasible (20–22). In this context, mass spectrometry methods (e.g., matrix-assisted laser desorption 186

ionization—time of flight mass spectrometry (MALDI-TOF-MS) and liquid chromatographyelectrospray ionization-tandem mass spectrometry (LC-ESI MS)) are able to deliver unambiguous sequence information and therefore provide reliable identification of toxins. With modern instrumentation, sensitivities are still somewhat limited and can reach down to a few ng/mL of toxin, especially when combinations of immunoaffinity-based enrichment, tryptic digestion plus MS-based detection and identification of specific peptides is applied (18, 19, 23, 24). Finally, functional methods display the biological activity or potency of protein toxins. Here, many different approaches have been described for the individual toxins ranging from in vivo assays (e.g., mouse bioassay for BoNT), ex vivo assays using animal tissues (e.g., mouse phrenic nerve hemidiaphragm assay for BoNT) or in vitro assays which renounce te use of animal tissues (25–27). Depending on the protein toxin, functional in vitro assays display receptor binding, internalization and enzymatic activity or only parts thereof (e.g., cytotoxicity assay versus adenine release assay for ricin and abrin; endopeptidase assay for BoNT (28–30)) or they display the interaction of toxins with their physiological targets (e.g., TCR / MHC-binding in a mixed lymphocyte reaction for SE (31)). In the past, an objective comparison of the different methods has not been possible since no suitable RM and no opportunity for proficiency testing has been available. In light of the relevance of biological toxins in the food, health and security sectors and in order to establish the status quo of detection capabilities within the European Union (EU) and beyond, a project called EQuATox was funded by the EU’s 7th framework program from 2012 to 2014 (EQuATox, “Establishment of Quality Assurances for the Detection of Biological Toxins of Potential Bioterrorism Risk”) (1, 32, 33). Here, the evaluation of technical capabilities in a series of proficiency tests (PTs) showed that mostly satisfactory results were obtained in international expert laboratories when dealing with basic analytical tasks; still the equivalence of analytical results would clearly profit from further technical improvement (1, 13, 18, 19, 34). Among other findings, it became clear that expert laboratories use indeed a broad panel of different tools and technical approaches for detection, identification and quantification of biological toxins resulting in data of variable quality. Though validation studies were published for individual methods, there are currently hardly any agreed-upon reference methods available, nor are there RMs available for the toxins in focus (ricin, abrin, SE, BoNT). So expert laboratories currently use either own in-house purified materials or commercial toxins of varying quality as quality-control (QC) samples or for calibration, which makes any comparison of different methods applied in different laboratories questionable (13). As an outcome of the project, a roadmap for harmonization of detection methods for biological toxins was drafted (13) which is now implemented in the ongoing EuroBioTox project funded by the EU’s Horizon 2020 program from 2017 to 2022 (“European Programme for the Establishment of Validated Procedures for the Detection and Identification of Biological Toxins”) (35, 36). In this project, 13 core members join their forces together with 48 network partners from 23 countries to work on a comprehensive package of quality assurance measures, including the production and certified reference materials (CRMs), the refinement of analytical procedures, the availability of tools in a European repository, state-of-the-art training on good analytical strategies, and the establishment of a comprehensive proficiency testing scheme. A major focus of the project is the production of CRMs for prioritized biological toxins such as ricin, BoNT and SEB.

On the Importance of Quality-Control Tools In case of a natural (i.e., food poisoning outbreak) or intentional release (i.e., bioterrorism threat scenario) of biological toxins, decision makers rely on correct and reliable laboratory data to make 187

appropriate and timely decisions, to manage the threat and to alleviate the outcomes on society. In this context, a high level of analytical capability is required to understand the scale of the incident and to take qualified decisions on countermeasures (13). Generally, several components are needed in order to perform reliable measurements: • Good analytical strategies • Validated methods for screening, identification, quantification and measurement of biological activity; the latter allows for the determination of the biological threat level • RMs, preferably certified RMs (ideally both pure toxin solutions (“calibrants”) as well as toxin in matrix materials) • Standard operating procedures • Regular training of personnel • External evaluation of measurement capabilities by proficiency testing • Continuous refinement of methods and development of innovative/superior methods The use of QC tools properly characterized with respect to identity and purity is an important requirement to achieve accurate and reliable results in the measurement of biological toxins (37). By definition, ISO Guide 30 defines an RM as a material which is sufficiently homogeneous and stable with respect to one or more specified properties, which has been established to be fit for its intended use in a measurement process (38). The property is the entity for which a reference value is established. This property can be qualitative (e.g., identity of species: genomic DNA of Listeria monocytogenes, strain 4B, NCTC 11994), but in most cases it is quantitative (e.g., mass concentration of a given protein in a matrix: amyloid β1-42 peptide in human cerebrospinal fluid, 0.72 ± 0.11 µg/L [examples from JRC RM catalog (39), CRMs IRMM-449 and ERM-DA481/IFCC, respectively]). A special category are RMs used for presence/absence testing, where the property is typically expressed as probability of detection (e.g., Staphylococcus aureus enterotoxin A (SEA) in cheese, CRM IRMM-359) (40). A certified reference material (CRM) is an RM characterized by a metrologically valid procedure for one or more specified properties, accompanied by an RM certificate that provides the value of the specified property (the certified value), its associated uncertainty, and a statement of metrological traceability (38). The associated uncertainty is expressed as so-called expanded uncertainty, which has the meaning of a confidence interval in which the true value lies with a certain probability, typically 95%. The qualification requirements for CRM producers are laid down in ISO 17034 (“General requirements for the competence of reference material producers”) (41), a standard which recently has been published after its conversion from the former ISO Guide 34 (42). The Reference Materials Unit within JRC Geel (formerly Institute for Reference Materials and Measurements, IRMM) obtained accreditation to ISO Guide 34 in 2004, and is meanwhile accredited according to ISO 17034. Whenever possible, RM production (processing and certification) projects are carried out under this accreditation. In addition, all finalized projects undergo review by subject-specific external expert panels before respective materials are released for sales. Metrological traceability is a key concept of metrology and is strongly connected with uncertainty. Metrological traceability is defined as “property of a measurement result whereby the result can be related to a reference through a documented unbroken chain of calibrations, each contributing to the measurement uncertainty” (43, 44). The practical meaning is that if a valid metrological traceability chain is established, the measurement result in a laboratory using its in188

house routine method and analyzing a routine sample can be linked to a metrological reference as anchor point, which can be the practical realization of an SI unit (e.g., sample mass is 5.02 kg), a measurement procedure (e.g., amyloid β1-42 peptide as obtained by solid phase extraction and subsequent quantification by liquid chromatography with mass spectrometry detection, according to the reference methods (45, 46)), or an artifact (e.g., 1 international unit (IU) is equivalent to 0.0347 mg of human insulin (47)). Metrological traceability is essential to make results comparable over time and space, with profound consequences, e.g., in pharmaceutical products and in international trade (import and export of goods). Metrological traceability is undoubtedly important for measurement results, but equally important for certified values of a CRM. Establishment of metrological traceability is accomplished by using existing CRMs of qualified National Metrology Institutes (NMIs) or other institutions of equivalently demonstrated competence. In this specific case, no such CRMs are available, which rules out this possibility. Another means, however, is a demonstration of the RM producer’s competence, e.g., through participation in key comparisons under the umbrella of the CIPM-MRA (Mutual Recognition Arrangement under the Comité International des Poids et Mesures/International Committee for Weights and Measures). JRC Geel (formerly IRMM) has participated in various studies organized in dedicated CCQM (Consultative Committee for Amount of Substance: Metrology in Chemisty and Biology) working groups (e.g., Bioanalysis Working Group, recently further split down into Protein Analysis (48), DNA Analysis, and Cell Analysis Working Groups; Organic Analysis Working Group, Inorganic Analysis Working Group). The intercomparisons among NMIs serve as benchmark and aim at demonstrating the degree of equivalence of results obtained by the participants. Another important metrological concept shall be briefly mentioned: commutability of a CRM. It shall be understood as the degree of equivalence in the analytical behavior of real samples and a CRM with respect to various measurement procedures (methods). The term commutability originated from the clinical chemistry field (49). It describes the ability of an RM to have inter-assay properties comparable to the properties demonstrated by authentic clinical samples when measured by more than one analytical method (50). Preparation of the CRM (formulation procedures as lyophilization, additives for preservation) can potentially lead to non-commutability of the material, for instance due to alteration of the analyte and/or changes in the matrix which impairs extraction efficiency of the analyte. Investigation of suitable commutability is especially important for matrix CRMs. Again, various examples can be listed from the clinical RM sector where dedicated commutability studies aim to demonstrate that the RM investigated behaves in the same or comparable way to patient samples (51). The certificate of a CRM describes the material, indicates the property value (certified value), its corresponding uncertainty, in case of method-dependent measurands the method(s) that was (were) used to obtain the data contributing to establish the certified value and its uncertainty, the reference to which the certified value is traceable, and the name and function of the RM producer’s approving officer (52). Moreover, other information typically present on CRM certificates include safety information, instructions for storage, how and what to use the material for, reconstitution protocol, if applicable, and a legal disclaimer. The typical uses of CRMs comprise the following applications: • Calibration of a method. Typically, calibrants are pure substances or solutions of a pure substance (a small molecule, but also biomolecules such as a protein). There are some 189

exceptions, especially in the clinical field, where matrix CRMs – a calibrant spiked in defined amount into a given matrix – are used for calibration: If an isolated pure protein would behave differently in a measurement process compared to a sample where that protein is in its natural matrix (e.g., human serum), matrix CRMs are recommended for method calibration. • Validation of methods. For assessing the trueness of a method, a matrix CRM is indispensable since all extraction steps are included in the analytical workflow. • Method performance verification. In this context, the CRM is used as QC sample. The aim is to demonstrate that when applying the method, the certified value is found, thus the method performs correctly. It is important to note that for each CRM it has to be defined case-by-case for which purpose and application it can be used. For instance, a calibrant solution can be used in method validation studies, but not for the parameter trueness (matrix is absent). As mentioned earlier, a matrix CRM on the other hand is usually not used for calibration of a method.

Development of RMs and CRMs The development of RMs and especially CRMs is a complex process. Exemplarily, Figure 1 displays the general outline of a CRM production at JRC Geel.

Figure 1. General procedure for CRM production applied at JRC Geel (53). Material selection can sometimes be difficult, especially for some matrix RMs (e.g., decision on exact matrix or accessibility of suitable raw materials). Often orienting feasibility studies are required, for instance to investigate how processing of a suitable RM can be performed. Processing usually is a multi-step process, i.e. to convert a raw material such as plant seeds containing a biological toxin into an RM preparation, here a purified toxin isoform. Processing typically contains several steps. It can comprise but is not limited to extraction, purification by means of preparative chromatography (different modes), formulation, aliquoting into containers, and lyophilization. Once the so-called candidate RM batch is produced, it is aliquoted into the final storage containers (e.g., plastic vials), ensuring that no filling trends occur over the large number of containers filled.

190

Homogeneity is a key requirement for any RM aliquoted into units. To demonstrate suitable homogeneity, a study is carried out to verify equivalence between the units produced and to quantify the between-unit variation (54). Stability testing is necessary to establish the conditions for storage (long-term stability) as well as the conditions for dispatch of the materials to the customers (shortterm stability). Again, dedicated studies are executed to demonstrate at which temperatures the RM (CRM) can be safely stored (e.g., -80 °C) and how shipments have to be executed (e.g., dry ice shipment to customer). Whenever possible, so-called isochronous stability studies are the primary choice, as measurements will be carried out in a large series, thereby adhering to repeatability conditions and thus minimizing the variability that stems from measurements (55). One central aspect as concerns measurements within an RM (CRM) project is the availability and accessibility of high-performing methods, i.e. methods to be used for homogeneity and stability measurements (key parameters precision and intermediate precision) and essentially for characterization (key parameters trueness, precision and intermediate precision). Characterization of RMs is defined as determination of the property values or attributes of an RM, as part of the production process. These property values (e.g., protein mass concentration of toxin X as established by method Y is 1.52 mg/mL) once established are then called certified values when the RM project is completed and certificates have been established (value, assigned to a property of an RM that is accompanied by an uncertainty statement and a statement of metrological traceability, identified as such in the RM certificate (38)). Strictly speaking, validation of methods – especially the parameter trueness –require CRMs, and, vice versa, the production of CRMs (establishment of certified values for a given property) require validated methods, whereby trueness and precision (“accuracy”) is key. This vicious circle or chicken-egg dilemma can only be overcome by a step-wise approach. To this end, it is crucial to understand the methods (measurement processes) to the best possible extent. This comprises but is not limited to comprehensive validation (e.g., assessment of specificity, extraction recoveries and matrix effects), use of existing suitable CRMs whenever possible, and regular participation in PT studies to demonstrate the competent application of a method in a laboratory. RMs from JRC Geel come with comprehensive documentation: the RM certificate and a certification report, which outlines the background, production of the material, homogeneity, stability and characterization studies and results, value assignment, uncertainty calculation, metrological traceability statement, list of references, and annexes in most cases containing graphics and charts.

Challenges in the Development of CRMs for Protein Toxins As mentioned above, the current EU-project EuroBioTox addresses the production and characterization of CRMs for different plant and bacterial protein toxins, among them ricin, BoNT and SEB. Specifically for those high molecular weight toxins, there are a number of practical challenges to deal with. 1) The purification or production strategy has to be decided on – shall the CRMs be produced by purification of toxins from their natural sources or shall they be produced by recombinant technologies? Actually, the answer has to be given on a case-by-case basis and is closely linked to the molecular characteristics of the toxin CRM to be produced. As an example, ricin is a plant toxin with a complex and variable glycosylation pattern containing four glycosylation sites, two on the A and B chains each (8, 56). Importantly, the type and level of glycosylation affects the functional activity of ricin (20, 57, 58): of differently glycosylated ricin isoforms tested, the highest glycosylated form, containing 191

more hybrid/complex-type glycans, was most toxic in different biological assays tested. Conversely, chemically deglycosylated ricin A turned out to be approximately 1000-fold less toxic than glycosylated ricin A (59). N-glycosylation was shown to promote the toxicity of the ricin A chain by promoting its transport out of the endoplasmic reticulum (60). In light of the available information on ricin’s functional activity, the authentic glycosylation pattern seems to be crucial for its enzymatic activity. Therefore, production from natural sources is – for the glycoprotein ricin – a superior approach compared to recombinant expression, which most likely would not result in correctly glycosylated toxin. Also, considering the cellular toxicity of full-length ricin, recombinant production would be difficult to pursue in eukaryotic expression systems (which in principle are able to deliver glycosylation). As an alternative production strategy and to circumvent ricin’s cytotoxicity, advanced cell-free expression systems could be used which are able to add post-translational modifications; however, the authenticity of the glycosylation pattern compared to the natural plant-derived toxin would have to be proven. The method of choice for non-glycosylated bacterial toxins such as SE and BoNT is the recombinant production, e.g., in E. coli, which ensures reliable expression, high yields and purity following established protocols. Especially the anaerobic cultivation of Clostridium botulinum is empiric and error prone. In addition, BoNT is encoded together with up to five neurotoxin-associated proteins (NAPs) such as the non-toxic non-hemagglutin of 140 kDa and three hemagglutinins which form large toxin complexes up to 760 kDa. Isolating the 150 kDa BoNT from these toxin complexes is challenging and requires sophisticated procedures. The drug substances of all currently approved BoNT-based pharmaceuticals are isolated as toxin complex from C. botulinum culture supernatant, but only two drug products (inco- and daxibotulinumtoxin A) comprise just the pure 150 kDa BoNT/A illustrating the challenge to separate BoNT from the NAPs. Recently, the first recombinantly produced BoNT drug product (rBoNT-E) was successfully evaluated in a clinical phase I trial (EudraCT 2016-002609-20) and further recombinantly produced BoNTs are in development as pharmaceuticals. In case of BoNT and SE with their >40 and >26 toxin variants, respectively, it is extremely valuable being able to use codon-optimized synthetic genes. This approach allows the expression of any toxin whose sequence is known or deposited in a public database without the need to obtain the natural producing strain or its genomic DNA. One important prerequisite, however, is that it can be demonstrated in a pre-study that native and recombinant protein behave the same way or at least as similar as possible in different methods. For instance, if the toxin was equipped with an affinity peptide tag to increase the isolation efficiency, the tag removal after purification is important so that the interaction in an immunochemical assay (e.g., ELISA) or in a functional assay is not impaired. In contrast to SEs, BoNTs are classical AB-toxins and require proteolytic activation into a 50 kDa light chain (LC) and a 100 kDa heavy chain (HC) which remain covalently linked by a disulfide bridge for biological activity. Since recombinant expression of BoNTs in E. coli yields a single polypeptide chain which is biologically inactive, subsequent specific and quantitative hydrolysis (>95% di-chain BoNT) at a defined loop region between LC and HC is essential to obtain biological activity. This process is best controlled in vitro with specific proteases and, if necessary, with an engineered loop sequence. Of course, the used protease, any peptides cut out from the loop as well as affinity tags cleaved off the BoNT need to be quantitatively removed at a rather late stage of the purification process to ensure high purity of the RM. Some BoNT 192

variants like all BoNT/E subtypes are released by group II non-proteolytic C. botulinum and hence always occur as single polypeptide chain with low biological activity until being hydrolyzed in the patients’ gastrointestinal tract. In contrast, a recombinantly produced BoNT/E including a proteolytic activation step already yields the biological highly active toxin which is suitable as RM for sensitive detection methods. Whereas the biological consequences of the disulfide bond formation in e.g. SEB is yet unknown, BoNTs absolutely depend on the intact disulfide bond connecting LC and HC to exert their neurotoxicity. Due to the aerobic conditions during recombinant expression and isolation of the BoNT its formation has been proven for all established BoNT serotypes (61). Altogether, the ‘recombinant way’ is highly suitable for the production of SE and BoNT RM. 2) A second important decision in the planning phase of a toxin CRM project is the decision on which molecular toxin variant to use. Is it more appropriate to use one isolated toxin variant – and if yes, which one, or would a defined mixture of naturally occurring variants better reflect the analytical task? The large batch size of CRMs has to be taken into account: can the selected toxin variant be produced in sufficient amount and purity? Generally, a high purity of the analyte (≥ 95%) is desirable in a CRM production process. Lower purity preparations are acceptable if the uncertainty of the certified value is adjusted accordingly with properly estimated contributions from respective impurities. Residual impurities have to be identified and quantified individually whenever possible. This includes not only protein/peptide impurities, but also impurities of other nature, e.g. inorganic impurities (62) Considering the occurrence of plant toxins in multi-gene families, the differentiation of the analyte toxin in the CRM from co-purified residual impurities can be challenging when purifying from natural sources (20). As an example, ricin occurs in two major isolectins, ricin D and ricin E, which are present in most R. communis cultivars and are 97% identical on the protein level (63–65). Ricin is composed of a cell-binding B subunit and an enzymatically active A subunit, both linked by a single disulfide bond forming a 63 kDa protein (66). Additionally, the plant expresses a highly related molecule called Ricinus communis agglutinin (RCA120) which is a tetrameric protein of two ricin A-like and two ricin B-like subunits. The identity on the protein level between the A and B chains of ricin D to those of the related RCA120 is 94% and 84%, and 94% and 89% for ricin E, respectively (67). Starting from a cultivar expressing all three molecules, the presence of ricin E concomitant to RCA120 cannot be clearly differentiated and quantified from ricin D by LC-MS/MS on the protein level. Therefore, a more straight-forward strategy for the production of a ricin CRM is to start with the cultivar R. communis zansibarienis which has been shown to produce ricin D only (68). Here, ricin D and RCA120 can be separated in sufficient amount and purity by preparative chromatography. Due to gene synthesis, recombinant expression of a toxin offers maximal freedom with respect to selection of the variant. One criterium would be the epidemiological occurrence of the variant, but also biosecurity aspects need to be considered for selecting the toxin variant. As an example, BoNTs pathogenic to humans occur in more than 40 different variants called subtypes, which vary up to 36% on the amino acid level (69). Subsequently, molecule characteristics like high specific toxicity to humans (to stress the sensitivity of methods to be validated), high water solubility, sufficient stability with respect to protein 193

mass concentration as well as specific toxicity (in commonly usable formulations) and acceptable yield by the production process guide the selection process. 3) A third critical point to consider is the molecular integrity of the toxin CRM at the end of the production process. It has to be made sure that the production process (including purification/expression, filling, storage) does not compromise the identity and function of the analyte. To this end, the comparability of the toxin CRM to its natural, authentic analog should be demonstrated by a panel of different methods (see below). Generally, the formulation of the toxin CRM, e.g., the buffer composition in case of a CRM solution, has to be carefully selected so that a comprehensive characterization of the CRM is technically feasible. It has to be considered that additives might stabilize the toxin CRM, but might interfere with certain measurement procedures (e.g., addition of protein stabilizers or salt might interfere with MS-based methods, detergents impair analysis in cell-based methods); therefore, the use of additives should be minimized. 4) Finally, the whole production chain for toxin CRMs requires appropriate safety and security measures in place including a concept for safe storage and distribution. This goes beyond conventional safety and security measures since the large-scale production of toxin CRMs imposes additional challenges, namely upscaling issues. Generally, working with biological toxins requires physical security measures such as working in highly secured laboratories equipped with biological safety cabinets for delicate sample handling and having decontamination agents available in case of spill. In addition to these biosafety measures, biosecurity measures have to be implemented, such as facility security plans (e.g., data and IT security, emergency response plans, procedures for receipt, transfer and shipment of select agents), personnel access control, personnel registration and security vetting, operational control and regular staff training. Additionally, risk assessments for all procedures need to be documented prior to the practical work. This includes information on adequate personal protective equipment, decontamination procedures and operational safety measures. There are useful references, which describe safety and health considerations for conducting work with biological toxins (70–72).

Characterization of Toxin CRMs As mentioned above, availability of and accessibility to high-performance methods are key for the characterization of toxin CRMs. Apart from measurements for homogeneity, stability and characterization, this also accounts for the purity and identity assessment of the produced toxins. A variety of methods has to be applied to demonstrate identity and suitable purity of the preparations (Figure 2). These comprise but are not limited to liquid chromatography – (tandem) mass spectrometry (peptide fingerprinting and protein sequencing for identification; LC-MS of intact protein for exact mass determination), MALDI-TOF MS and other MS techniques to detect protein or peptide impurities, immunochemical methods such as ELISA and Western Blot as well as SDSPAGE and/or capillary gel electrophoresis for a purity profile. In addition, for glycosylated protein toxins a set of methods is required dedicated to comprehensive glycosylation analysis (identification of glycans and glycoforms). Typically, LC-MS methods are applied to identify the glycans, their structures and microheterogeneity, and to identify which N-glycosylation sites are occupied. The mass concentration of the protein toxin in solution will best be quantified using protein impurity corrected amino acid analysis. It involves quantification of constituent amino acids following complete hydrolysis of the material and correction for amino acids originating from 194

inherent structurally-related protein impurities. Individual amino acids are separated by liquid chromatography or gas chromatography (GC) and quantified, typically using isotope dilution mass spectrometry (62, 73–75). The protein mass concentration is calculated taking into account the amino acid sequence of the protein. This approach requires a highly purified protein preparation; therefore the above-mentioned analyses are performed first to confirm a high level of purity. The speed and completeness of acidic hydrolysis of the protein to amino acids, both indispensable requirements for obtaining correct results, is sequence-depending, thus hydrolysis might need optimization (76).

Figure 2. Characterization of toxin CRM requires a comprehensive and complementary panel of different methods. Finally, different functional assays are required to assess the toxicity and/or biological activity of the toxin preparation. This is challenging since it is not obvious to link the precise protein amount to its biological activity – the outcome of activity determination heavily depends on the method used. For example, for ricin the measurement of cytotoxicity in vitro displays the activity of the A and the B chain (cell binding plus enzymatic activity), while adenine release assay indicates enzymatic activity only (19, 20). Likewise, for BoNTs it has been shown that the potency of different serotypes relative to each other is different by its quantitative factor when either in vivo (mouse bioassay) or ex vivo (mouse phrenic nerve assay) methods are used (61). This can be explained by the fact that the mouse bioassay describes the pharmacodynamics plus the pharmacokinetics of BoNT including absorption, distribution, metabolism and elimination from circulation of the individual BoNTs, while factors such as distribution, metabolism or elimination are not displayed in the mouse phrenic nerve assay. This discrepancy was extended by comparing the biological activity of the six native BoNT/A-F in in vitro cell-based assays (neurotransmitter release vs. substrate [SNARE protein] cleavage), ex vivo MPN assay and in vivo sublethal digital abduction score assay (77). With respect to value assignment in a toxin CRM project it is therefore imperative to describe the functional method used in great detail so that it can be reproduced by the CRM customer.

Special Consideration of the Use of Toxin CRMs in Mass Spectrometry From the different methods in use for detection of biological toxins, MS-based methods clearly display the highest specificity and allow for unambiguous identification of the analyte (23, 24, 78, 79). Here, the availability of toxin CRMs is highly useful to develop the technology forward in terms of toxin identification, quantification and sample preparation. Toxin identification is typically based on peptide mass fingerprinting of digested protein matched against the theoretical peptide masses derived in silico, nowadays often realized in MS/ 195

MS-based approaches. The availability of pure CRMs makes it possible to generate high-quality peptide spectral libraries for forensic identification of closely related protein toxins. Based on an in silico digest, it is important to determine which of the theoretically possible prototypic peptides can be measured experimentally with sufficient intensity and/or sensitivity depending on the instrumentation used. This information is a prerequisite to select diagnostic peptides for an overall workflow to be applied in the course of a forensic investigation (80–82). Here, a regular verification of the selected diagnostic peptides against peptides derived from newly identified proteins is recommended to maintain specificity and hence unambiguous identification of the toxin (80). Many state-of-the-art MS-based approaches are quite sensitive (low limits of detection and quantification). This can trigger matrix effect issues if a highly pure toxin is present at very low mass concentration in a buffer/matrix (unfavorable mass ratio of matrix to analyte). Highly pure toxin CRMs are important for the improvement of analytical procedures with respect to developing sample preparation methods, assessing recovery and in documenting the overall efficiency and reproducibility of the workflow. Toxin CRMs will furthermore support the determination of the enzymatic digestion efficiency in order to obtain high sequence coverage.

Conclusions and Outlook CRMs, thoroughly validated methods and PT studies are cornerstones of applied quality assurance. They are important QC tools for laboratories to validate and safeguard analytical methods. One of the main deliverables in the EuroBioTox project is the development and production of CRMs for plant and bacterial protein toxins. As in other fields, ideally pure substances for calibration (calibrants, either solutions or solid materials) as well as matrix RMs containing the analyte in a defined amount in a complex matrix representing a typical sample material (e.g., toxin spiked into serum, food matrix, water) for method validation would be available. Producers of high-quality CRMs face several challenges regarding the purification or production process and the choice of the toxin variant to be produced. Safeguarding the molecular integrityof the analyte during the production process as well as high purity and biological activity of the CRM are key elements of the process. Above all, appropriate safety and security measures have to be in place to deal with the task. The provision of such materials shall improve the preparedness and competence of laboratories to reliably analyse samples for biological toxins. Application fields are not limited to the food safety and public health sectors, but are equally important in security, military, and verification sectors. Overall, harmonized and standardized laboratory detection will enhance preparedness and response planning and will help to maintain a high level of vigilance, thus increasing resilience of the civil society in the capacity to prepare and respond to an incident involving biological toxins.

Abbreviations BoNT CCQM CRM ELISA EQuATox EU

Botulinum neurotoxin Consultative Committee for Amount of Substance: Metrology in Chemisty and Biology Certified reference material Enzyme-linked immunosorbent assay EU-project, acronym: Establishment of Quality Assurances for the Detection of Biological Toxins of Potential Bioterrorism Risk European Union 196

EuroBioTox GC HC LC LC-ESI MS MALDI-TOF MHC MS NAP NMIs PT QC RM SDS-PAGE SE TCR

EU-project, acronym: European Programme for the Establishment of Validated Procedures for the Detection and Identification of Biological Toxins Gas chromatography Heavy chain of BoNT Light chain of BoNT Liquid chromatography-electrospray ionization-tandem mass spectrometry Matrix-Assisted Laser Desorption Ionization—Time of Flight Major histocompatibility complex Mass spectrometry Neurotoxin-associated proteins National Metrology Institutes Proficiency test quality-control Reference material Sodium dodecyl sulfate polyacrylamide gel electrophoresis Staphylococcal enterotoxin T cell receptor

Acknowledgments The authors acknowledge the funding of the EuroBioTox project under the European Union’s Horizon 2020 research and innovation program under Grant Agreement No 740189.

References 1. 2.

3.

4.

5.

6. 7. 8.

Dorner, B. G.; Rummel, A. Preface Biological Toxins—Ancient Molecules Posing a Current Threat. Toxins (Basel) 2015, 7, 5320–5321. European Centre for Disease Prevention and Control. Annual Epidemiological Report 2014 – Food- and Waterborne Diseases and Zoonoses; European Centre for Disease Prevention and Control: Stockholm, Sweden, Nov 2014, 2014. European Food Safety Authority; European Centre for Disease Prevention and Control. The European Union Summary Report on Trends and Sources of Zoonoses, Zoonotic Agents and Food-borne Outbreaks in 2014. EFSA Journal 2015, 13, 4329. Ahanotu, E.; Alvelo-Ceron, D.; Ravita, T.; Gaunt, E. Staphylococcal Enterotoxin B as a Biological Weapon: Recognition, Management, and Surveillance of Staphylococcal Enterotoxin. Appl. Biosaf. 2006, 11 (3), 120–126. Middlebrook, J.; Franz, D. Botulinum toxins. In Medical Aspects of Chemical and Biological Warfare; Sidell, F., Takafuji, E., Franz, D. , Eds.; Office of The Surgeon General, Department of the Army, United States of America: Washington, DC, 1997; pp 643−654. Manual of Security Sensitive Microbes and Toxins; CRC Press: Boca Raton, FL, 2014. Robb, C. S. The Analysis of Abrin in Food and Beverages. Trends Analyt. Chem. 2015, 67, 100–106. Worbs, S.; Köhler, K.; Pauly, D.; Avondet, M. A.; Schaer, M.; Dorner, M. B.; Dorner, B. G. Ricinus communis Intoxications in Human and Veterinary Medicine — A Summary of Real Cases. Toxins (Basel) 2011, 3, 1332–1372. 197

9.

10. 11. 12.

13.

14. 15. 16. 17. 18.

19.

20.

21.

22.

23. 24.

25.

Organisation for the Prohibition of Chemical Weapons (OPCW). Convention on the Prohibition of the Development, Production, Stockpiling and Use of Chemical Weapons and on Their Destruction (Chemical Weapons Convention); Technical Secretariat of the Organisation for the Prohibition of Chemical Weapons: The Hague, The Netherlands, 2005; p 181. Darling, R. G.; Catlett, C. L.; Huebner, K. D.; Jarrett, D. G. Threats in Bioterrorism. I: CDC Category A Agents. Emerg. Med. Clin. North Am. 2002, 20 (2), 273–309. Moran, G. J. Threats in Bioterrorism. II: CDC Category B and C Agents. Emerg. Med. Clin. North Am. 2002, 20, 311–330. Der Generalbundesanwalt beim Bundesgerichtshof [Public Prosecutor General of the Federal Court of Justice in Germany]. Press release 11/2019: Anklage Wegen des Vorwurfs der Vorbereitung einer Schweren Staatsgefährdenden Gewalttat u.a. Erhoben; Karlsruhe, 07.03.2019; https://www. generalbundesanwalt.de/prnt/showpress.php?newsid=822 (accessed 30.04.2019). Dorner, B. G.; Zeleny, R.; Harju, K.; Hennekinne, J.-A.; Vanninen, P.; Schimmel, H.; Rummel, A. Biological Toxins of Potential Bioterrorism Risk: Current Status of Detection and Identification Technology. Trends Analyt. Chem. 2016, 85, 89–102 Part B. Rummel, A. The Long Journey of Botulinum Neurotoxins into the Synapse. Toxicon 2015, 107, 9–24 Part A. Spooner, R. A.; Lord, J. Ricin Trafficking in Cells. Toxins (Basel) 2015, 7, 49–65. Krakauer, T.; Stiles, B. G. The Staphylococcal Enterotoxin (SE) Family. Virulence 2013, 4, 759–773. Duracova, M.; Klimentova, J.; Fucikova, A.; Dresler, J. Proteomic Methods of Detection and Quantification of Protein Toxins. Toxins (Basel) 2018, 10, e99. Worbs, S.; Fiebig, U.; Zeleny, R.; Schimmel, H.; Rummel, A.; Luginbühl, W.; Dorner, B. G. Qualitative and Quantitative Detection of Botulinum Neurotoxins from Complex Matrices: Results of the First International Proficiency Test. Toxins (Basel) 2015, 7, 4935–4966. Worbs, S.; Skiba, M.; Bender, J.; Zeleny, R.; Schimmel, H.; Luginbühl, W.; Dorner, B. G. An International Proficiency Test to Detect, Identify and Quantify Ricin in Complex Matrices. Toxins (Basel) 2015, 7, 4987–5010. Worbs, S.; Skiba, M.; Söderström, M.; Rapinoja, M.-L.; Zeleny, R.; Russmann, H.; Schimmel, H.; Vanninen, P.; Fredriksson, S.-Å.; Dorner, B. Characterization of Ricin and R. communis Agglutinin Reference Materials. Toxins (Basel) 2015, 7, 4906–4934. Simon, S.; Fiebig, U.; Liu, Y.; Tierney, R.; Dano, J.; Worbs, S.; Endermann, T.; Nevers, M.-C.; Volland, H.; Sesardic, D.; Dorner, M. B. Recommended Immunological Strategies to Screen for Botulinum Neurotoxin-Containing Samples. Toxins (Basel) 2015, 7, 5011–5034. Simon, S.; Worbs, S.; Avondet, M.-A.; Tracz, D.; Dano, J.; Schmidt, L.; Volland, H.; Dorner, B.; Corbett, C. Recommended Immunological Assays to Screen for Ricin-Containing Samples. Toxins (Basel) 2015, 7, 4967–4986. Kalb, S.; Baudys, J.; Wang, D.; Barr, J. Recommended Mass Spectrometry-based Strategies to Identify Botulinum Neurotoxin-Containing Samples. Toxins (Basel) 2015, 7, 1765–78. Kalb, S.; Schieltz, D. M.; Becher, F.; Åstot, C.; Fredriksson, S.-Å.; Barr, J. Recommended Mass Spectrometry-based Strategies to Identify Ricin-Containing Samples. Toxins (Basel) 2015, 7, 4881–4894. Dorner, M. B.; Schulz, K. M.; Kull, S.; Dorner, B. G. Complexity of Botulinum Neurotoxins: Challenges for Detection Technology. Curr. Top. Microbiol. Immunol. 2013, 364, 219–255. 198

26. Bigalke, H.; Rummel, A. Botulinum Neurotoxins: Qualitative and Quantitative Analysis Using the Mouse Phrenic Nerve Hemidiaphragm Assay (MPN). Toxins (Basel) 2015, 7, 4895–4905. 27. Stern, D.; von Berg, L.; Skiba, M.; Dorner, M. B.; Dorner, B. G. Replacing the Mouse Bioassay for Diagnostics and Potency Testing of Botulinum Neurotoxins – Progress and Challenges. Berl. Munch. Tierarztl. Wochenschr. 2018, 131, 375–394. 28. Kalb, S.; Boyer, A.; Barr, J. Mass Spectrometric Detection of Bacterial Protein Toxins and their Enzymatic Activity. Toxins (Basel) 2015, 7, 3497–3511. 29. Becher, F.; Duriez, E.; Volland, H.; Tabet, J. C.; Ezan, E. Detection of Functional Ricin by Immunoaffinity and Liquid Chromatography-Tandem Mass Spectrometry. Anal. Chem. 2007, 79, 659–665. 30. Pauly, D.; Worbs, S.; Kirchner, S.; Shatohina, O.; Dorner, M. B.; Dorner, B. G. Real-Time Cytotoxicity Assay for Rapid and Sensitive Detection of Ricin from Complex Matrices. PLoS One 2012, 7, e35360. 31. Rasooly, R.; Do, P. M. In Vitro Cell-based Assay for Activity Analysis of Staphylococcal Enterotoxin A in Food. FEMS Immunol. Med. Microbiol. 2009, 56, 172–178. 32. The EQuATox Consortium. EQuATox: Establishment of Quality Assurance for the Detection of Biological Toxins of Potential Bioterrorism Risk. http://www.equatox.eu (accessed 29.04.2019). 33. Community Research and Development Information Service. CORDIS EQuATox. Establishment of Quality Assurances for the Detection of Biological Toxins of Potential Bioterrorism Risk. https://cordis.europa.eu/project/rcn/103025/factsheet/en (accessed 29.04.2019). 34. Nia, Y.; Rodriguez, M.; Zeleny, R.; Herbin, S.; Auvray, F.; Fiebig, U.; Avondet, M. A.; Munoz, A.; Hennekinne, J. A. Organization and ELISA-based Results of the First Proficiency Testing to Evaluate the Ability of European Union Laboratories to Detect Staphylococcal Enterotoxin Type B (SEB) in Buffer and Milk. Toxins (Basel) 2016, 8, pii: E268. 35. Community Research and Development Information Service. CORDIS EuroBioTox. European Programme for the Establishment of Validated Procedures for the Detection and Identification of Biological Toxins. https://cordis.europa.eu/project/rcn/209945/factsheet/en (accessed 29.04.2019). 36. The EuroBioTox Consortium. European Programme for the Establishment of Validated Procedures for the Detection and Identification of Biological Toxins. www.eurobiotox.eu (accessed 29.0.2019). 37. Zeleny, R.; Schimmel, H. Influence of the Approach to Calibration on the Accuracy and the Traceability of Certified Values in Certified Reference Materials. Trends Analyt. Chem. 2012, 33, 107–116. 38. International Organization for Standardization. ISO Guide 30:2015 Reference Materials – Selected Terms and Definitions; Geneva, Switzerland, 2015. 39. Welcome to the Certified Reference Materials Catalogue of the JRC. https://crm.jrc.ec.europa.eu/ (accessed 10.04.2019). 40. Zeleny, R.; Nia, Y.; Schimmel, H.; Mutel, I.; Hennekinne, J. A.; Emteborg, H.; CharoudGot, J.; Auvray, F. Certified Reference Materials for Testing of the Presence/Absence of Staphylococcus aureus Enterotoxin A (SEA) in Cheese. Anal. Bioanal. Chem. 2016, 408, 5457–5465. 41. International Organization for Standardization. ISO 17034:2016, General Requirements for the Competence of Reference Material Producers; ISO: Geneva, Switzerland, 2016. 199

42. International Organization for Standardization. ISO Guide 34:2009, General Requirements for the Competence of Reference Material Producers; ISO: Geneva, Switzerland, 2009. 43. International Organization for Standardization. ISO/IEC Guide 99:2007 International Vocabulary of Metrology – Basic and General Concepts and Associated Terms (VIM); ISO: Geneva, Switzerland, 2007. 44. Koeber, R.; Linsinger, T.; Emons, H. An Approach for More Precise Statements of Metrological Traceability on Reference Material Certificates. Accredit. Qual. Assur. 2010, 15, 255–262. 45. Leinenbach, A.; Pannee, J.; Dulffer, T.; Huber, A.; Bittner, T.; Andreasson, U.; Gobom, J.; Zetterberg, H.; Kobold, U.; Portelius, E.; Blennow, K. Mass Spectrometry-based Candidate Reference Measurement Procedure for Quantification of Amyloid-beta in Cerebrospinal Fluid. Clin. Chem. 2014, 60, 987–994. 46. Korecka, M.; Waligorska, T.; Figurski, M.; Toledo, J. B.; Arnold, S. E.; Grossman, M.; Trojanowski, J. Q.; Shaw, L. M. Qualification of a Surrogate Matrix-based Absolute Quantification Method for Amyloid-beta(4)(2) in Human Cerebrospinal Fluid using 2D UPLC-Tandem Mass Spectrometry. J. Alzheimers Dis. 2014, 41, 441–451. 47. Burns, C.; Morris, T.; Jones, B.; Koch, W.; Borer, M.; Riber, U.; Bristow, A. Proposal to Initiate a Project to Evaluate a Candidate International Standard for Human Recombinant Insulin; WHO: Geneva, Switzerland, 2010. 48. Josephs, R. D.; Stoppacher, N.; Westwood, S.; Wielgosz, R. I.; Li, M.; Quaglia, M.; Melanson, J.; Martos, G.; Prevoo, D.; Wu, L.; Scapin, S.; Senal, M. Ö.; Wong, L.; Jeong, J.-S.; Chan, K. W. Y.; Arsene, C. G.; Park, S.-R. Concept Paper on SI Value Assignment of Purity – Model for the Classification of Peptide/Protein Purity Determinations. J. Chem. Metrol. 2017, 11, 1–8. 49. Franzini, C. Commutability of Reference Materials in Clinical Chemistry. J. Int. Fed. Clin. Chem. 1993, 5, 169–173. 50. Zeleny, R.; Emteborg, H.; Schimmel, H. Assessment of Commutability for Candidate Certified Reference Material ERM-BB130 “Chloramphenicol in Pork”. Anal. Bioanal. Chem. 2010, 398, 1457–1465. 51. Andreasson, U.; Kuhlmann, J.; Pannee, J.; Umek, R. M.; Stoops, E.; Vanderstichele, H.; Matzen, A.; Vandijck, M.; Dauwe, M.; Leinenbach, A.; Rutz, S.; Portelius, E.; Zegers, I.; Zetterberg, H.; Blennow, K. Commutability of the Certified Reference Materials for the Standardization of beta-Amyloid 1-42 Assay in Human Cerebrospinal Fluid: Lessons for Tau and beta-Amyloid 1-40 Measurements. Clin. Chem. Lab. Med. 2018, 56, 2058–2066. 52. International Organization for Standardization. ISO Guide 31:2015 Reference Materials – Contents of Certificates, Labels and Accompanying Documentation; ISO: Geneva, Switzerland, 2015. 53. Emons, H.; Linsinger, T.; Gawlik, B. Reference Materials: Terminology and Use. Can’t One See the Forest for the Trees? Trends Analyt. Chem. 2004, 23, 442–449. 54. Linsinger, T. P. J.; Pauwels, J.; van der Veen, A. M. H.; Schimmel, H.; Lamberty, A. Homogeneity and Stability of Reference Materials. Accredit. Qual. Assur. 2001, 6, 20–25. 55. Lamberty, A.; Schimmel, H.; Pauwels, J. The Study of the Stability of Reference Materials by Isochronous Measurements. Fresenius J. Anal. Chem. 1998, 360, 359–361.

200

56. Despeyroux, D.; Walker, N.; Pearce, M.; Fisher, M.; McDonnell, M.; Bailey, S. C.; Griffiths, G. D.; Watts, P. Characterization of Ricin Heterogeneity by Electrospray Mass Spectrometry, Capillary Electrophoresis, and Resonant Mirror. Anal. Biochem. 2000, 279, 23–36. 57. Sehgal, P.; Khan, M.; Kumar, O.; Vijayaraghavan, R. Purification, Characterization and Toxicity Profile of Ricin Isoforms from Castor Beans. Food Chem. Toxicol. 2010, 48, 3171–3176. 58. Sehgal, P.; Kumar, O.; Kameswararao, M.; Ravindran, J.; Khan, M.; Sharma, S.; Vijayaraghavan, R.; Prasad, G. B. K. S. Differential Toxicity Profile of Ricin Isoforms Correlates with their Glycosylation Levels. Toxicology 2011, 282, 56–67. 59. Soler-Rodríguez, A.-M.; Uhr, J. W.; Richardson, J.; Vitetta, E. S. The Toxicity of Chemically Deglycosylated Ricin A-Chain in Mice. Int. J. Immunopharmacol. 1992, 14, 281–291. 60. Yan, Q.; Li, X. P.; Tumer, N. E. N-Glycosylation Does Not Affect the Catalytic Activity of Ricin A Chain but Stimulates Cytotoxicity by Promoting its Transport Out of the Endoplasmic Reticulum. Traffic 2012, 13, 1508–1521. 61. Weisemann, J.; Krez, N.; Fiebig, U.; Worbs, S.; Skiba, M.; Endermann, T.; Dorner, M. B.; Bergström, T.; Muñoz, A.; Zegers, I.; Müller, C.; Jenkinson, S.; Avondet, M.-A.; Delbrassinne, L.; Denayer, S.; Zeleny, R.; Schimmel, H.; Åstot, C.; Dorner, B. G.; Rummel, A. Generation and Characterization of Six Recombinant Botulinum Neurotoxins as Reference Material to Serve in an International Proficiency Test. Toxins (Basel) 2015, 7, 5035–5054. 62. Josephs, R. D.; Martos, G.; Li, M.; Wu, L.; Melanson, J. E.; Quaglia, M.; Beltrão, P. J.; Prevoo-Franzsen, D.; Boeuf, A.; Delatour, V.; Öztug, M.; Henrion, A.; Jeong, J.-S.; Park, S.R. Establishment of Measurement Traceability for Peptide and Protein Quantification Through Rigorous Purity Assessment—A Review. Metrologia 2019, 56, 044006. 63. Araki, T.; Funatsu, G. The Complete Amino Acid Sequence of the B-Chain of Ricin E Isolated from Small-Grain Castor Bean Seeds. Ricin E Is a Gene Recombination Product of Ricin D and Ricinus communis Agglutinin. Biochim. Biophys. Acta 1987, 911, 191–200. 64. Ladin, B. F.; Murray, E. E.; Halling, A. C.; Halling, K. C.; Tilakaratne, N.; Long, G. L.; Houston, L. L.; Weaver, R. F. Characterization of a cDNA Encoding Ricin E, a Hybrid RicinRicinus communis Agglutinin Gene from the Castor Plant Ricinus communis. Plant Mol. Biol. 1987, 9, 287–295. 65. Mise, T.; Funatsu, G.; Ishiguro, M.; Funatsu, M. Isolation and Characterization of Ricin E from Castor Beans. Agric. Biol. Chem. 1977, 41, 2041–2046. 66. Mantis, N. J. Ricin Toxin. In Manual of Security Sensitive Microbes and Toxins; Liu, D., Ed.; CRC Press: Boca Raton, FL, 2014; pp 499-510. 67. Roberts, L. M.; Lamb, F. I.; Pappin, D. J.; Lord, J. M. The Primary Sequence of Ricinus communis Agglutinin. Comparison with Ricin. J. Biol. Chem. 1985, 260, 15682–15686. 68. Fredriksson, S.-Å.; Hulst, A. G.; Artursson, E.; de Jong, A. L.; Nilsson, C.; van Baar, B. L. Forensic Identification of Neat Ricin and of Ricin from Crude Castor Bean Extracts by Mass Spectrometry. Anal. Chem. 2005, 77, 1545–1555. 69. Peck, M. W.; Smith, T. J.; Anniballi, F.; Austin, J. W.; Bano, L.; Bradshaw, M.; Cuervo, P.; Cheng, L. W.; Derman, Y.; Dorner, B. G.; Fisher, A.; Hill, K. K.; Kalb, S. R.; Korkeala, H.; Lindstrom, M.; Lista, F.; Luquez, C.; Mazuet, C.; Pirazzini, M.; Popoff, M. R.; Rossetto, O.;

201

70. 71. 72. 73. 74.

75.

76. 77.

78.

79.

80.

81.

82.

Rummel, A.; Sesardic, D.; Singh, B. R.; Stringer, S. C. Historical Perspectives and Guidelines for Botulinum Neurotoxin Subtype Nomenclature. Toxins (Basel) 2017, 9, E38. Johnson, B.; Mastnjak, R.; Resnick, I. G. Safety and Health Considerations for Conducting Work with Biological Toxins. Appl. Biosaf. 2001, 6, 117–135. Kozlovac, J. P.; Hawley, R. J. , Biological Toxins: Safety and Science. In Biological Safety; American Society of Microbiology, 2006. Biosafety in Microbiological and Biomedical Laboratories, 5th ed.; HHS Publication No. (CDC) 21-1112; CDC: Atlanta, GA, 2009. Muñoz, A.; Kral, R.; Schimmel, H. Quantification of Protein Calibrants by Amino Acid Analysis Using Isotope Dilution Mass Spectrometry. Anal. Biochem. 2011, 408, 124–131. Pritchard, C.; Torma, F. A.; Hopley, C.; Quaglia, M.; O’Connor, G. Investigating Microwave Hydrolysis for the Traceable Quantification of Peptide Standards Using Gas ChromatographyMass Spectrometry. Anal. Biochem. 2011, 412, 40–46. Stoppacher, N.; Josephs, R. D.; Daireaux, A.; Choteau, T.; Westwood, S. W.; Wielgosz, R. I. Impurity Identification and Determination for the Peptide Hormone Angiotensin I by Liquid Chromatography-High-Resolution Tandem Mass Spectrometry and the Metrological Impact on Value Assignments by Amino Acid Analysis. Anal. Bioanal. Chem. 2013, 405, 8039–8051. Fountoulakis, M.; Lahm, H. W. Hydrolysis and Amino Acid Composition of Proteins. J. Chromatogr. A 1998, 826, 109–134. Donald, S.; Elliott, M.; Gray, B.; Hornby, F.; Lewandowska, A.; Marlin, S.; Favre-Guilmard, C.; Perier, C.; Cornet, S.; Kalinichev, M.; Krupp, J.; Fonfria, E. A Comparison of Biological Activity of Commercially Available Purified Native Botulinum Neurotoxin Serotypes A1 to F1 In Vitro, Ex Vivo, and In Vivo. Pharmacol. Res. Perspect. 2018, 6, e00446. Söderström, M.; Bossée, A.; Dorner, B.; Worbs, S.; Guo, L. Analysis of ricin: Analysis Strategy. In Recommended Operating Procedures for Analysis in the Verification of Chemical Disarmament; Vanninen, P., Ed.; University of Helsinki: Helsinki, Finland, 2017; pp 547−580. Skiba, M.; Dorner, B.; Guo, L.; Brinkworth, C.; Ginter, J.; Shefcheck, K. Analysis of Ricin: MALDI-MS. In Recommended Operating Procedures for Analysis in the Verification of Chemical Disarmament; Vanninen, P., Ed.; University of Helsinki: Helsinki, Finland, 2017. Merkley, E. D.; Jenson, S. C.; Arce, J. S.; Melville, A. M.; Leiser, O. P.; Wunschel, D. S.; Wahl, K. L. Ricin-like Proteins from the Castor Plant Do Not Influence Liquid ChromatographyMass Spectrometry Detection of Ricin in Forensically Relevant Samples. Toxicon 2017, 140, 18–31. Schieltz, D. M.; McGrath, S. C.; McWilliams, L. G.; Rees, J.; Bowen, M. D.; Kools, J. J.; Dauphin, L. A.; Gomez-Saladin, E.; Newton, B. N.; Stang, H. L.; Vick, M. J.; Thomas, J.; Pirkle, J. L.; Barr, J. R. Analysis of Active Ricin and Castor Bean Proteins in a Ricin Preparation, Castor Bean Extract, and Surface Swabs from a Public Health Investigation. Forensic Sci. Int. 2011, 209, 70–79. Schieltz, D. M.; McWilliams, L. G.; Kuklenyik, Z.; Prezioso, S. M.; Carter, A. J.; Williamson, Y. M.; McGrath, S. C.; Morse, S. A.; Barr, J. R. Quantification of Ricin, RCA and Comparison of Enzymatic Activity in 18 Ricinus communis Cultivars by Isotope Dilution Mass Spectrometry. Toxicon 2015, 95, 72–83.

202

Chapter 13

The Statistical Defensibility of Forensic Proteomics Kristin H. Jarman*,1 and Eric D. Merkley2 1Applied Statistics & Computational Modeling Group, Pacific Northwest National

Laboratory, Richland, Washington 99352, United States

2Chemical & Biological Signatures Group, Pacific Northwest National Laboratory,

Richland, Washington 99352, United States

*E-mail: [email protected]. Phone: (509) 375-4539.

The U.S. Federal court system maintains very high standards for admissibility of scientific evidence. In particular, to allow such evidence into court, a judge must be satisfied it is both reliable, in other words, based on a robust, reproducible, and accurate method, and relevant, meaning it adds weight to the prosecution or defense claims in a case. In 1993, the U.S. Supreme Court issued landmark guidelines for judges considering whether or not to admit scientific testimony. The resulting Daubert criteria changed the forensic sciences in a profound way, leading investigators, researchers, and legal experts to question the scientific validity of established forensic methods that had been widely accepted for years. The Daubert criteria also caused a paradigm shift in the development of new forensic methods, forcing researchers to better prepare new techniques for use in an increasingly adversarial environment. In this chapter, we discuss the current state of forensic proteomics in the context of modern forensics. We present the Daubert criteria, discuss seven elements of a defensible method, and provide guidelines for building statistical defensibility of this emerging discipline.

The Challenge of Modern Forensics The forensic science community has been under a large and very critical microscope for the past 20 years. This intense scrutiny dates back to the 1990s, when a series of U.S. Supreme Court rulings redefined the requirements for admissibility of expert testimony in court. Before that time, courtrooms commonly applied the Frye standard, which requires expert opinion to be based on methods generally accepted as reliable among the relevant scientific community. However, a 1993 product liability case, Daubert v. Merrell Dow Pharmaceuticals, Inc., brought to light the limitations of relying solely on general acceptance for admissibility, causing the Court to lay out more specific and systematic rules for allowing testimony to reach a jury. The Supreme Court made further refinements through two additional cases during the 1990s, and by 2000, these guidelines had been codified into © 2019 American Chemical Society

what is now called the Daubert standard (1–3). The Daubert standard consists of three important requirements. First, the trial judge must serve as the gatekeeper for admitting scientific testimony into court. Second, expert testimony must rest on a “reliable foundation”, and must also be “relevant” to the investigative question at hand. Third, the evidence must be a product of sound scientific methodology. Since judges, who may have a limited scientific background, are required to determine the scientific validity of forensic evidence, the Daubert standard goes on to suggest criteria for evaluating a method. These criteria can be summarized as follows: (1) whether the method has been tested, (2) if it has been peer reviewed, (3) if it has known error rates, (4) if it is widely accepted among the relevant scientific community, and (5) if there are standards controlling the method’s implementation. The Daubert rulings opened a floodgate of debate, criticism, and controversy among the forensic sciences (4–7). Since 2000, the Daubert criteria have been used to question the validity of new and established forensic methods alike. The pattern sciences, handwriting examination and tool marks for example, have received the most criticism, but the more scientific disciplines have also taken fire. This intense scrutiny is particularly evident at the National Academy of Sciences (NAS), which has published a number of expert panel studies criticizing specific methods (8–11) as well as the broader forensic science community (12, 13). These panels have had a major impact on the forensic sciences over the years. For example, in 2004, a review panel evaluated an established technique known as compositional analysis of bullet lead (CABL). The FBI had been using inductively coupled plasma-optical emission spectroscopy (ICP-OES) to compare the elemental composition of bullets in order to determine if a questioned bullet might have originated from a particular source. The NAS panel found some inconsistencies in how the analyses were conducted, along with inadequacies in the statistical methods used to arrive at conclusions. However, most significantly, it uncovered fundamental problems with the scientific underpinnings of the method. Specifically, after studying bullet lead manufacturing processes, the panel concluded that two bullets from a single box can easily contain significantly different elemental compositions, and conversely, bullets from different boxes can easily have indistinguishable elemental compositions. In other words, the conclusion that two bullets with similar composition come from the same source is faulty, as is the conclusion that two bullets with different composition come from different sources. This bombshell report dealt a blow to the forensic community, and the resulting controversy eventually contributed to the Federal Bureau of Investigation abandoning CABL analysis altogether. In 2009, another review panel evaluated the practice of the forensic sciences in the United States. That panel’s final report, “Strengthening the Forensic Sciences in the United States: A Path Forward” (12), was highly critical of the scientific validity of many forensic methods, and has become infamous among the forensic science community. Such is the playing field on which the discipline of forensic proteomics emerges. In some ways, it is ready for the adversarial climate of the U.S. court system. Proteomics has been around for more than 20 years, and there are now established methods available in application areas such as toxicology, clinical diagnostics, and pharmaceutical development (14–16). These twenty-plus years of research, development, and application demonstrate the validity of proteomics in general, but the majority of this work has been directed toward scientific discovery, leaving a number of open questions related to the forensic application of proteomics. In this chapter, we discuss proteomics in the context of modern forensics. We present requirements for establishing a defensible forensic method, review the current state of forensic proteomics, and make suggestions for future research.

204

Reliability and Relevance: The Foundation for a Defensible Method The Daubert standard dictates that admissible expert testimony must rest “on a reliable foundation” and be “relevant to the task at hand (1).” Reliability, a measure of what can be called technical error (17) or measurement uncertainty (12), can be thought of as its reproducibility and robustness. A reliable forensic proteomics method will consistently detect key peptides or proteins when present in a sample. It will rely on consistent and objective bioinformatics and data interpretation criteria. Its performance, including operational limitations, will be characterized for real-world forensic samples. Reliability tends to be the major focus of scientific method development, and standards organizations such as the Association of Official Analytical Chemists (AOAC), and the International Standards Organization (ISO) provide guidelines for establishing reliability through good laboratory practice. As a result, there’s a great deal of readily available information on establishing the reliability of laboratory methods (18–25, 26, 27–29, 30). In addition, there is a vast body of literature supporting the reliability of both targeted and untargeted proteomics (31–37). The second requirement, relevance to the task at hand, goes beyond experimental reliability. Often measured via error rates (12, 17, 38, 39), relevance refers to the strength of the evidence produced by a method, in other words, the “scientific connection to the pertinent inquiry (1)” necessary for admissibility in court. In the world of forensic proteomics, a relevant method confidently identifies target proteins or protein sources in a questioned sample to the exclusion of all other possible sources. Confident identification of a target protein or protein source is more complicated than confident identification of a peptide. Confident identification to the exclusion of all other possible protein sources (i.e. relevance) is unique to forensics. To illustrate, consider a shotgun proteomics assay to detect the ricin toxin in a questioned sample. If the method can reproducibly find the toxin peptides in a questioned sample, then the method is reliable. However, for the method to produce relevant evidence, it must also be able to confidently differentiate ricin from other protein sources that may be in the sample. This means demonstrating that the identified peptides produce strong evidence for ricin, where their presence increases the likelihood it is ricin and not some other protein source in the sample of interest. Relevance is often overlooked in the development and practice of forensic methods, making it one of the favorite points of attack by critics of the community. Indeed, the most damning criticism in the 2004 NAS panel report on compositional bullet lead analysis had to do with the relevance of the evidence produced by the method. The assertion that two matching bullets must, with a high degree of confidence, come from the same box was untested, and ultimately proved to be unfounded. In another high profile case, the investigation into the 2001 anthrax letters, the Government Accountability Office (GAO) questioned the relevance of several emerging microbial forensics methods used to identify the source of the Bacillus anthracis strain dispersed in the attack (39). In the end, the GAO reviewers speculated that certain genetic markers linking the crime scene samples to a specific stock at a known location could be the result of parallel evolution, not the fact that the unknown sample originated from the feedstock in question. Because this question hadn’t been tested or even asked, there was no way to address this criticism, and had this case gone to trial, the entire assay could have been called into question. Forensic proteomics – the use of proteomics to identify specific targets (i.e. organisms, tissues, and toxins) in biological samples of unknown origin—has a growing body of literature to support the relevance of both targeted and untargeted methods. For example, proteomics has been used to characterize items of historical and archaeological significance, including ancient paints (40, 41), tuberculosis-ridden skeletal remains (42), and South American ungulates from the Ice Age (43). 205

A number of law enforcement applications have been described, including microbial identification (44–49), characterization of microbiological growth media (50), human identification from hair samples (51, 52), identification of bodily fluids (53–55), species identification using bones (56–58) and other tissues (55), as well as detection of protein toxins such ricin (59–64), the botulinum neurotoxin (65, 66), and other protein toxins (64, 67, 68). This body of research lays a foundation for relevance, but more work is needed. The remainder of this chapter discusses how the forensic proteomics community is establishing a foundation for reliability and relevance of proteomics evidence, and discusses important areas for future research.

Questions an Investigator Might Ask Forensics is concerned not only with detection of target protein sources, but also the attribution, or association, of a questioned sample with a potential origin or production process. As part of a forensic proteomic analysis, there are several questions an investigator might like to answer: • Is protein source W present in the sample? • How much of protein source W is in the sample? • Do sample A and sample B originate from the same source? These questions determine how reliability and relevance must be demonstrated when developing a new forensic method. For example, the process of finding protein source W in a sample is very different from determining how much of it is present. To find protein source W, the method only needs to find sufficient numbers of peptide markers for W. On the other hand, to determine the concentration of W, the method needs to produce, for example, an extracted ion chromatogram peak whose area also correlates highly with the concentration of W. To correlate mass spectral data to concentration, researchers typically develop protocols using internal standards and calibration controls to account for the many potential sources of variation found in forensic samples. Such protocols may not be necessary if the goal is simply to identify the presence of a target protein in a sample. Is Protein Source W Present in the Sample? An Identification Problem. Identification is the process of determining which, if any, protein sources of concern are present in a sample. Most of the forensic proteomics literature to date is focused on identification of target proteins or protein sources. In peptide-based proteomics methods, protein identification is based on the amino acid sequences of detected peptides, which are inferred from peptide fragmentation mass spectra. Methods for identifying peptides based on their fragmentation spectra and methods for evaluating the reliability of those identifications (discussed in the chapter “A Proteomics Tutorial” in this volume) are at the core of proteomics research. How Much of Protein Source W Is Present in the Sample? A Quantitation Problem. Quantitation involves estimating the amount of one or more target proteins in a sample. In mass spectrometry, quantitation usually relies on peak areas measured relative to one another or to some internal standard (such as stable-isotope labeled standards). Quantitative methods as a means of protein toxin detection have been published (63, 64, 67). However, methods that rely on relative protein abundance, including many basics-science proteomics studies, are less common in forensics. One example can be found in Schieltz et. al . (63), where a quantitative ricin assay is presented, and Merkley et. al. (69), where the relative abundances of key proteins are used to determine if Y. pestis cultures come from a wild or lab-adapted strain. 206

Do Sample A and B Have the Same Source? A Sample Comparison Problem. In some cases, an investigator might like to compare two samples to determine if they have a similar composition of protein sources. Often it is a questioned sample, a sample of unknown origin, being compared to an exemplar or reference sample, whose origin is known. However, an investigator might also like to compare two unknown samples to one another. Sample comparison might involve identification, and/or quantitation, and in order to accurately interpret the results, a comprehensive understanding of the statistical variability of same-source and different-source samples is needed. In proteomics, sample comparison is common among samples collected from controlled environments, as with laboratory protein expression studies and clinical studies. However, variation in an uncontrolled environment, such as a crime scene, adds difficulty to the general problem of comparing forensic samples. Proteomic methods for comparing body fluids (53, 64) and tissues (55) have been proposed, however, to our knowledge, nobody has yet proposed a general statistical method for proteomic comparison of forensic samples, leaving this to be an open area of research. The Forensic Proteomics Method Development Process Demonstrating reliability and relevance should be a primary goal in the development of a new forensic proteomics technique. The method development process is illustrated in Figure 1. Three distinct phases are shown: proof-of-concept, developmental verification and validation (V&V), and routine implementation. The first stage, proof-of-concept, includes the initial studies establishing the scientific underpinnings of a method: protocol development, identification of limitations, determination of sensitivity and specificity, development of bioinformatics and data interpretation methods, and pilot demonstrations. At the end of the proof-of-concept stage, a standard operating protocol (SOP) is constructed and used in developmental V&V. Developmental V&V transitions a method from the research laboratory to routine operation. In this stage, the formal V&V experiments are performed, population studies are conducted, and a reference database, if needed, is constructed. A final SOP is determined, performance of the method is validated for forensic samples, operational limitations are identified, and calibration and quality control practices are put into place. All of these elements are used to develop reporting guidelines, obtain ISO or other accreditation, and transition the method into routine operation.

Figure 1. The forensic method development process. Researchers often use the term validation to refer to any experiments designed to demonstrate a method’s performance. On the other hand, standards organizations apply these terms much more precisely (18), defining V&V as a series of carefully planned studies designed to establish and verify specific statistical performance characteristics of an experimental method. In this chapter, we try to avoid confusion with the following terminology. Specifically, we use method development to refer to the overall beginning-to-end process of demonstrating the reliability and relevance of a forensic 207

method. Developmental validation and verification refers to one step of the method development process, the transition from proof-of-concept to implementation, where performance of the method is formally established and confirmed. Finally, validation (verification) study refers to a specific study within the developmental V&V process, where the study is carefully designed to establish (confirm) statistical performance characteristics of the method.

Seven Elements of a Defensible Method The Amerithrax case, the investigation into the anthrax mailings of 2001, was one of the first cases to call on the emerging field of microbial forensics (70). Investigators had identified the specific strain used in the attacks, and they were looking for clues about the source of the biomass. Though they had ample reference material for comparison, there were no established sample comparison methods available. Method development and implementation had to be conducted in parallel, through collaboration between the FBI and the contractors developing the genetic tests that would ultimately be used in the investigation. The FBI gave the contractors the freedom to develop the methods, with the stipulation that they must undergo validation testing prior to use on the case. In 2014, the U.S. Government Accountability Office (GAO) reviewed the methods developed and used during the Amerithrax case (39). Reviewers found several issues: • One contractor performing the genetic tests didn’t state a limit of detection prior to validation testing, an operational limitation which, if stated, might have changed the results of the testing. • One contractor did not have a method for reconciling contradictory results between its two assays, causing ambiguous results and forcing researchers to throw out validation samples after the fact. • Some of the error rates (false positives and false negatives) were higher in post-validation testing than validation testing, causing the auditors to question the robustness of the methods. Table 1. Seven Elements of a Defensible Method 1. A standard operating protocol, complete with calibration standards, controls, and proficiency requirements. 2. A standard data analysis and interpretation protocol, complete with decision criteria for all possible outcomes of the method. 3. Verified and validated statistical performance metrics with their associated uncertainties. 4. An interpretation framework for evaluating the strength of the evidence. 5. Clearly stated operational limitations. 6. A formal quality assurance program. 7. Peer-reviewed publication.

The GAO concluded that, while most contractors did perform some form of validation testing, this testing wasn’t sufficient, and the entire development effort would have benefited from a more formalized method development process. For example, the first issue found by the GAO, that of forgetting to state a limit of detection, can be alleviated by formal V&V testing. The second issue, that of ambiguous results caused by contradictory outcomes, can be addressed by constructing a standard data interpretation protocol with decision criteria for all possible outcomes of a method. The final issue, the unexpectedly high post-validation error rates, can be minimized through statistical experimental design focused on establishing key performance characteristics along with their margins 208

of error. In this section, we describe seven key outcomes of the forensic method development process. Summarized in Table 1, these outcomes serve as a checklist for formalizing a method, addressing V&V requirements, and, ultimately, satisfying the Daubert criteria. A Standard Operating Protocol (SOP) Proof-of-concept studies are typically designed to help researchers optimize and demonstrate the potential of a proposed method. Experimental protocols in this stage of research tend to be flexible. However, as the method transitions from proof-of-concept to developmental V&V, it is important to establish a well-documented standard operating protocol that states all relevant procedures, from sample collection to controls and calibration. According to Huber (71), a leading resource for FDA method validation, a well-written protocol should describe the analytical procedure “…in sufficient detail to allow a competent analyst to reproduce the necessary conditions and obtain results within the proposed acceptance criteria”, meaning consistent with the performance characteristics established during V&V. In Budowle (38), the authors provide further detail, stating that an SOP should contain “(i) all reagents critical to the procedure that should be tested before analyzing unknown samples; (ii) critical equipment, calibration, and certification requirements; and (iii) known positive‑, negative‑, and/or internalcontrol samples used with the analysis”. A rigid SOP can be a barrier in situations where a method must adapt to new investigative situations such as different sample types, new analytes, and so on. For this reason, flexibility can be written into the SOP, covering modifications not expected to impact the outcome of the method. In addition, quality assurance programs allow for minor modifications to SOPs, as long as those modifications can be planned and approved prior to analyzing an evidentiary sample, and this allows for further flexibility. For samples requiring significant deviations to an SOP, an on-the-fly validation protocol can be developed, where the modified protocol is evaluated through a limited V&V study and accepted if the key performance characteristics meet predetermined criteria. Standard methodologies and protocols have been developed for both targeted and untargeted proteomics, and numerous excellent reviews have been published (31–37). A typical proteomics experiment proceeds in three steps: (1) sample preparation, (2) data acquisition, and (3) peptide identification. Sample preparation includes extraction and proteolytic digestion of the proteins followed by sample cleanup. Data acquisition is typically done with an electrospray ionization (ESI) mass spectrometer coupled to a high-performance liquid chromatography (HPLC) for upfront separation of peptides, although offline fractionation followed by matrix-assisted laser desorption ionization (MALDI) mass spectrometry analysis is also used. Peptide identification uses modern statistical and bioinformatics methods to infer peptide sequence from the mass spectral data. A number of excellent, widely accepted peptide identification tools are available, and these will be discussed in the following section. A Standard Data Analysis and Interpretation Protocol Untargeted proteomics requires bioinformatics tools to extract identified peptides from the raw LC-MS/MS spectra. A number of such tools are available, relying on database search (34, 35, 72), de novo sequencing approaches (35, 73–75), and, to a lesser extent, spectral library search searching (76, 77). The choice of bioinformatics tools and how those tools are used can impact the results of a proteomics experiment. In order to make robust and generalizable statements about the performance of a given assay, it is therefore necessary to have an objective, standard data analysis and interpretation 209

protocol. This protocol should be set early in the development process, meaning parameters and bioinformatics tools are fixed, the data analysis process is clearly outlined, and decision criteria are established. This helps prevent inconsistencies in the interpretation of the results such as those encountered in the GAO review of the Amerithrax sample analysis (39). The data analysis process has two major components, peptide identification and interpretation of the evidence, both of which require a formal protocol. In addition, in the most common scenario where a search database is used, a protocol for database maintenance is also necessary. Peptide Identification. Peptide identification depends on the data acquisition strategy, whether untargeted or targeted. With traditional untargeted proteomics, the mass spectrometer collects data as peptides are eluted off the HPLC column, producing thousands of tandem (precursor and fragment) mass spectra. Identification of these peptides is most commonly done through a technique known as database searching (34, 35, 72), where all of these spectra are compared to a database containing theoretical tandem mass spectra for all peptide sequences that might be present in the sample. Though less common, spectral library search and de novo peptide identification are also accepted among the proteomics community. Spectral library search (76, 77) relies on a database of experimental rather than theoretical spectra, and therefore, can only be used for target protein sources with available libraries (e.g. common laboratory organisms such as Escherichia coli, yeast, human, and mouse). De novo peptide identification infers the peptide sequence directly from the fragment mass spectrum, building the sequence from the observed mass differences between peaks (35, 73, 74). Though not universally quantified, error rates for de novo sequencing methods are widely acknowledged to be higher than for the more popular database search (75, 78, 79). Nonetheless, de novo sequencing finds steady application among the metaproteomics community, where the uncharacterized nature of complex or environmental samples makes construction of a valid search database difficult (80). Some forensic proteomics applications face similar challenges, making de novo peptide sequencing an attractive alternative to database search tools (see below for further discussion). Peptide identification for targeted proteomics employs a more traditional analytical chemistry approach. Targeted methods rely on a data acquisition strategy tuned to a small set of pre-defined peptides. A triple quadrupole mass spectrometer is often used (37, 81) in single reaction monitoring (SRM) or multiple reaction monitoring mode (MRM) mode, which is also used for detection of small molecules. With this data acquisition strategy, three quadrupoles are used in series to isolate peptides of the target mass, fragment those isolated peptides, and isolate specific fragment ions, respectively. As a result, only those peak areas corresponding to specific fragment masses are measured. Stable-isotope labeled standards can be used for quantitation. Targeted proteomics assays have better sensitivity than untargeted proteomics assays, but the range of targets they can detect is limited to those proteins for which the instrument has been tuned. Data Interpretation. Obtaining a list of identified peptides is only the first step in the analysis of proteomics evidence. In the second step, this list of peptides is used to make a determination regarding the presence of target protein or protein source. In the basic sciences, a “2-peptide” rule is standard, where two identified peptides lead to a positive identification for a protein of interest, however, the question is more complicated for forensic samples where any number of unknown protein sources may be present. Though a variety of forensic proteomics assays have been proposed, very few have suggested statistical criteria for declaring a target protein source present in an unknown sample (44, 82). In order to generalize and formalize robust forensics proteomics methods, more work in this area is needed. 210

A Database Development, Update, and Maintenance Plan. A valid, comprehensive database is essential to the reliability and relevance of any forensic proteomics method (83). If peptides are to be identified by database searching, a database must be selected. The choice of a database critically affects the number and identity of peptides that can be detected. But how can a valid database be selected if the protein sources in the sample are not known in advance? This situation is also faced by the metaproteomics community, where researchers would often like to characterize proteins in an environmental sample containing an unknown community of organisms (84). In metaproteomics, metagenomics is often used to identify organisms to include in a protein sequence database for searching (85, 86), and a similar approach may be applicable in some forensic applications. Alternatively, de novo peptide sequencing may be used to identify peptides in an unknown sample (80). In any case, taxonomic or protein assignment of the identified peptides requires the use of a sequence database. As illustrated in Jarman et al. (44), selection of the database has a huge impact on the sensitivity and specificity of a forensic proteomics method. No general guidelines for selecting a proteomics search and/or sequence database have yet been created, and more work is this area is needed. Verified and Validated Statistical Performance Criteria Several statistical performance metrics are associated with developmental V&V. Table 2 lists those performance metrics applicable to detection, identification, and quantitation methods. (Other criteria apply to sample collection, extraction, and storage methods (38)). Because these statistics are calculated using experimental data, each one has uncertainty associated with it as well. Characterizing this uncertainty during V&V testing is important. To illustrate, in their review of the genetic testing methods developed during the Amerithrax investigation, one of the criticisms put forth by the GAO auditors was that the error rates in post-validation testing were larger than those calculated during validation. This can happen when the uncertainties inherent in the calculation of the error rates are not fully understood. For example, during validation testing, one of the assays produced no false negatives out of six known positive samples, for a false negative rate of 0%. In post-validation testing, the false negative rate for this assay was found to be a surprising 43% (13 out of the 30 samples). Several explanations are possible; however, had the researchers calculated a margin of error on the false negative rate during validation testing, they would have found it to be 0-64%, meaning that in repeated testing under similar conditions, 95% of the tests would be expected to produce a false negative rate somewhere between 0% and 64%. In other words, the limited number of true positives available in validation testing left so much uncertainty in the estimated false negative rate, a postvalidation value of 43% should have been expected, not surprising. With the widely popular untargeted proteomics approach to peptide identification, the proteomics community has had to deal with the statistical issues associated with big data for 20 years. As a result, there are many robust algorithms and tools for estimating error rates for peptide identification. The widely used target-decoy search is a strategy for estimating false discovery rate (FDR), or fraction of falsely identified peptides in a database search (87–89). Other methods are available for calculating reliability or error probability for a single peptide-spectrum match. Statistical learning tools such as PeptideProphet (90), Percolator (91, 92) and QVALITY (93) report a peptide error probability (PEP) value. Dynamic programming-based algorithms such as MS-GF+ (94, 95) and Crux (96) also report a spectrum-level reliability score. Beyond peptide identification, several authors have proposed statistical significance estimates for protein inference. Popular software tools

211

available include ProteinProphet (97), MSBayesPro (98), and ProteinLP (99). The reader is referred to Audain (100) for a review of current methods. Table 2. Statistical Performance Characteristics for Method Development Method Type

Performance Characteristic

Calculation

Error rates: FP=false positives FN=false negatives POD=probability of detection

FP=fraction of negative samples incorrectly identified as positives FN=fraction of positive samples incorrectly identified as negatives POD=1-FN

Detection/ identification Limit of detection(LOD) Sensitivity Specificity

Related to error rates. Sensitivity=100-FN% Specificity=100-FP%

Accuracy and precision

Accuracy = Mean error (bias) Precision = Standard deviation of error, or RMSEP

Quantitation Limit of quantitation (LOQ)

All

Lowest concentration for which POD exceeds some predetermined value, say 99%

Often 3x standard deviation of blank/background signal

Linearity

Visual inspection and regression R2

Range

The range of concentrations for which the method performs according to specified performance characteristics

Robustness

The reproducibility of a method under challenged (slightly abnormal) operating conditions

Ruggedness

The reproducibility of a method under normal laboratory operating conditions

Existing tools attach reliability or statistical significance values to individual peptide or protein identifications, however, they don’t estimate the overall error rates associated with a complete assay, where peptide and/or protein identification is just one of several important steps in the data interpretation process. In forensic proteomics assays, for example, the selection of search database, complexity of sample, and rules for declaring a positive result can all significantly impact the performance criteria in Table 2. A few statistical studies establishing methods for evaluating error rates for forensic proteomics methods have been published (44, 82, 101), however, most of the research is in the proof-of-concept stage. More work in this important area is needed. An Interpretation Framework for Evaluating the Strength of Evidence Any forensic method has the potential to make two types of error (17): (1) an error in the analysis of a sample, and (2) an error in the interpretation of the results. The analysis error results from an accumulation of all the uncertainties associated with sample collection, sample preparation, data acquisition, and bioinformatic analysis of the resulting mass spectral data. Traditional V&V testing characterizes the reliability, or experimental uncertainty, of a method, in other words, this

212

first type of error. The performance characteristics in Table 2 address some aspect of the experimental error, and the process for establishing them is, if not simple, at least relatively straightforward. Unfortunately, the same cannot be said for the second type of error, interpretation error. Interpretation error relates to the relevance of a method. It involves overstating or understating the significance—the strength of the evidence—in the context of an investigation. Many of the Daubert challenges to existing forensic methods over the years have focused on interpretation error (7, 21, 83, 102–124). Therefore, it is critical for a defensible forensic proteomics method to have a framework for evaluating the strength of the evidence produced by the assay. To illustrate its importance, consider a proteomics-based assay for detecting the ricin toxin in an unknown sample. Suppose just a single ricin peptide – LTTGADVR -- is detected in the sample in question. This peptide connects the unknown sample to ricin, however, it also happens to be present in a protein from the Central American freshwater snail Pomacea flagellata. How strong is this evidence for ricin? It depends on where the sample was collected, the likelihood the sample came into contact with a snail, potential for contamination, etc. A researcher developing a forensic proteomics method might be tempted to circumvent this problem by ignoring “non-unique” peptides, in other words, allowing only those peptides found in ricin and no other protein to be used as markers for identification. However, this reliance on unique peptides can create more problems than it solves. First, uniqueness of a forensic marker or signature is impossible to prove, making claims of uniqueness difficult to defend. Second, the uniqueness requirement creates a phenomenon known in whole genome sequencing as signature erosion, where the collection of markers available for identification shrinks over time as more organisms are sequenced (125, 126). Since the performance of this ricin assay depends on the number of detectable ricin markers, this can ultimately result in a method whose sensitivity decreases over time. Statistical methods enable the community to develop methods that are immune to signature erosion, at the same time providing ways to evaluate the strength of proteomics evidence. The development of such methods is an open area of research. Jarman et. al. have begun to address the issue of peptide marker selection, taking cues from forensic DNA analysis by selecting peptide markers based on probabilistic criteria (44). The selected peptides, called strong peptides, produce excellent error rates for microorganism identification to the species level. However, the approach has only been demonstrated in a single published study within the context of a limited database, and more work in this area is needed. Clearly Stated Operational Limitations One of the keys to developing a defensible forensic method is to understand the operational boundaries, in terms of both V&V testing and error rates. The following list includes experimental factors that typically must be set during V&V testing, and these factors often define the operational boundaries of a method: • • • • • • •

Sample collection method Sample size/volume/concentration Sample type Sample matrix/background Sample age Sample storage Age/quality of reagents and controls 213

• Instrument configuration (e.g. column age, column type, instrument resolution, etc) • Operator proficiency and training One of the roles of proof-of-concept research is to uncover the limitations of a method. For example, different sample volumes, sample types, reagents, and instrument settings are tested during method optimization, and through this process, limitations are identified. These limitations should be documented as the method transitions into developmental V&V. During V&V testing, they should either be verified or used to specify the boundaries of the studies. If verified, then the impact of a given limitation on the method should be documented. If used to specify the scope of V&V testing, then it should be documented that the method has not been evaluated outside the chosen boundaries. A Quality Assurance Program A great deal of attention has been paid to quality assurance of analytical laboratories, and there are many good references available (127–130). Several accreditation programs are available, most notably ISO 17025 for testing and calibration laboratories (https://www.iso.org/home/standards/ popular-standards/isoiec-17025-testing-and-calibra.html). Such programs include requirements for all elements of good laboratory practice, including the following: • • • • • •

training and proficiency testing, administrative review, standard operating protocols, quality control and calibration procedures, reporting guidelines, and procedures for on-the-fly validation (as needed).

Many challenges to forensic methods in recent years have argued poor sample quality, potential for contamination, and failure to follow laboratory QA/QC practices. It is impossible to completely eliminate the potential for such errors in all cases, however, in order to minimize risk, an effective laboratory management strategy should also include procedures for mitigation of errors through continual process improvement, corrective action procedures, and root cause analysis of any incidents that may occur. Management programs should be proactive rather than punitive, encouraging employees to report problems early, so that issues can be resolved quickly and efficiently. Peer-Reviewed Publication When it comes to strengthening the defensibility of a laboratory method, there is no substitute for scientific peer review. This type of review, done prior to publication in most reputable scientific journals, places the work under scrutiny by someone who is objective, critical, and versed in the science being developed. Peer review helps researchers identify and fix issues that have been overlooked. It establishes a level of acceptance by experts in the field. And most importantly for forensic applications, it is specifically stated as one of the Daubert criteria. Publication at every stage of the method development process, including proof-of-concept studies, V&V studies, and a demonstration of the relevance of the method can significantly bolster the defensibility of an emerging forensic proteomics method.

214

Additional Guidelines for Forensic Proteomics Over the years, the research community has developed standard practices for conducting and reporting the results of proteomics experiments. Several organizations have sought to formalize these practices. The so-called “Paris guidelines” (131), list experiment-specific details that should be reported in published articles. The Paris guidelines have been adopted as editorial policy by leading proteomics journals (such as Molecular and Cellular Proteomics and Journal of Proteome Research). They include the steps in a proteomic workflow needed to reproduce an experiment, in other words, the steps that should be specified in a standard operating protocol, monitored as part of the quality assurance program, and reported with the results of an analysis. The Paris guidelines include details, including the peak extraction method, database search software used, sequence database and/or spectral library used, full sequence of identified peptides, the precursor ion mass-to-charge ratio, any modifications observed, the peptide identification score(s), the protein accession number and database where it can be found, the number of distinct peptides for each protein, the percent sequence coverage for each protein, threshold peptide spectrum match (PSM) score for peptide identification, and the FDR along with a description of how it was calculated. Another set of standards is the “Human Proteome Project Mass Spectrometry Data Interpretation Guidelines 2.1” (132). The Human Proteome Project is an interlaboratory consortium, organized by the Human Proteome Organization (HUPO), whose goal is to find mass spectrometric evidence for every protein encoded by the human genome. These guidelines build on the Paris guidelines, but reflect the maturation of the field in the intervening decade, providing additional clarity on best practices for PSM‑, peptide- and protein-level FDR estimation, threshold values, and interpretation. These guidelines also introduce the concept of “extraordinary claims” peptides—peptides from neverbefore-detected proteins that a researcher is claiming to have detected for the first time. As such, they require a more stringent level of evidence, including annotated, high signal-to-noise, high-massaccuracy MS/MS spectra, comparison of such spectra with similar data collected for a synthetic peptide standard, and consideration of alternate spectral interpretations, including single amino acid variants, isobaric sequence or mass variants, and alternate peptide-to-protein mapping. Not all guidelines put forth by the Human Proteome Project are appropriate or practical for a forensic laboratory performing routine analysis. However, the requirements regarding extraordinary claims peptides highlight important issues relevant to untargeted forensic methods in two important ways. First, the consideration of alternate spectral interpretations directly relates to the relevance of forensic proteomics evidence; spectra with many possible alternate interpretations due to, for example, near-neighbor peptides with similar sequences (133), provide weaker evidence for the presence of a target protein than spectra with no other possible interpretations. Second, for a virus or protein toxin assay, where the ultimate decision depends on the detection of very few peptides, correct identification of those peptides is critical. Therefore, some of the extraordinary claim requirements may be appropriate for forensics, particularly in samples with a complex background. The Paris and HUPO guidelines focus on untargeted proteomics experiments. Similar standards for targeted experiments have been proposed in Isir et. al. (134). Three tiers of targeted studies, each for a different purpose, are listed. Tier 1 assays are absolute quantification measurements related to clinical, pharmaceutical or similarly regulated applications. Tier 2 assays are also absolute quantitative measurements, with a different focus, namely changes in protein levels in a non-clinical setting. Both Tiers 1 and 2 use stable-isotope labeled peptides as internal standards. Tier 3 experiments focus on relative protein quantitation in fundamental life sciences research, and as such, do not use internal standards. In quantitative forensics applications, the high regulatory and validation requirements for forensic proteomics resembles tiers 1 and 2. Agencies such as the U.S Food and Drug Administration 215

(FDA), the Clinical Laboratory Improvement Amendments (CLIA), and the World Anti-Doping Agency (WADA) have issued guidelines for validation of Tier 1 assays. With a few additions to accommodate the unknown background of forensic samples, these guidelines could serve as a foundation for developing and reporting targeted proteomics methods for forensic laboratories. WADA oversees the detection of banned performance-enhancing substances in international athletic competitions. Many of these banned substances are peptide or protein hormones, such as insulin variants or growth factors, and many can be detected by LC-MS methods (135–137). WADA has issued guidelines for qualitative LC-MS assays, many of which are relevant to data evaluation standards in forensics (138), particularly to targeted assays. For example, their guidelines for mass spectrometry of compounds with molecular weight greater than 800 require verification of the specificity of the targeted amino acid sequences by comparison with a sequence database (e.g., by BLAST), comparison of the relative abundances of the fragment ions in MS/MS to a standard (which could be isotopically labeled), and ensuring that 10% of the amino acid sequence of the targeted protein is covered by detected peptides. The Organization for the Prevention of Chemical Weapons (OPCW) is charged with verifying compliance with the Chemical Weapons Convention. The protein toxin ricin is one of the agents prohibited under that convention, and OPCW’s Scientific Advisory Board has considered standards for ricin detection, including by mass spectrometry. Included in this statement are the recommendations that (1) detection of at least four ricin peptides (presumably two each from the A and B chains) be required to declare a sample positive for ricin, (2) the detected peptides cover at least 10% of the sequence of the combined A and B chains, and (3) the uniqueness (i.e., species distribution) of the detected peptides be noted. These requirements serve as a reasonable model for other proteins. The requirement for number of peptides, if considered as two peptides per polypeptide chain, is in line with the widely used 2-peptide rule in proteomics (131). Requiring sequence coverage of 10% of the protein is more stringent than generally required by the proteomics community, and in this sense, is conservative. With regards to peptide uniqueness, the typical proteomics approach requires at least one, and usually two peptides total, unique (specific) peptides to determine the presence of a protein. However, as pointed out earlier in this section and elsewhere (44, 104), the concept of uniqueness in the forensic sciences is problematic, and more statistically sound approaches are needed. The documents put forth by the Paris guidelines, HUPO, WADA, and OPCW provide guidance specific to the mass spectrometry data and bioinformatics tools used in the proteomics community. These can be folded into the seven elements of a defensible method listed in the previous section to help steer the development of defensible forensic proteomics assays. The full details of the development process are assay-specific and, therefore, beyond the scope of this chapter. Here, we list important considerations to help researchers as they transition a method from proof-ofconcept to operations. Guidelines Relevant to All Forensic Proteomics Assays 1. The purpose of the assay, whether identification, detection, or quantitation, should be clearly stated, and the relevant performance metrics be calculated through V&V testing. This guideline draws from Table 2 and the recommendations of Carr et al. (139). For instance, if the purpose of the assay is identification, then establishing the limit of detection and the false positive and false negative rates becomes more important than establishing the limit of quantitation. An example of an early V&V study for untargeted assays can be found in Jarman et. al (44). 2. The SOP, established before V&V testing begins, should include experimental protocols, specification of the usage of all bioinformatics tools, and well-defined decision 216

criteria for declaring a positive, inconclusive, or not detected result. Although essential in the forensic world, and in many applications of targeted mass spectrometry, the concept of formal method validation is somewhat foreign to many proteomics scientists. We include this guideline to emphasize that not only experimental procedures, but also data analysis procedures (including software tools), and especially data evaluation criteria, are formal parts of the method and should be included in method validation as such. 3. The specificity of the peptides should be characterized. This guideline applies both to bioinformatic specificity (the number of proteins or species in which the chosen peptide occurs) and to analytical specificity (the presence or absence of interfering signals in the LC-MS/MS data). The bioinformatic specificity refers to the discriminating power of the peptide markers used in an assay. Quantifying bioinformatic specificity addresses the requirement for population studies in Figure 1 and helps to demonstrate the relevance of a forensic proteomics method. Such an assessment includes sequence analysis and examination of the peptide and fragment ions to determine (1) how frequently a peptide or protein occurs throughout nature, and (2) for targeted assays, if any other peptides or proteins would be expected to produce interfering signals. The database used to calculate bioinformatic specificity should be documented, and a plan for re-evaluating this value over time should be created. For targeted assays, analytical specificity is also important. The transition (or combination of transitions and their ratios) constitutes the signature in targeted proteomics. The relevance of this signature—that is, how rare it is and how strongly it indicates the intended peptide or protein—can be established by a combination of empirical and theoretical means. “Empirical means” consist of looking for the signature or signals that interfere with it in as comprehensive a series of relevant background and related samples as resources permit. These specificity studies have been amply described in relevant literature (140). 4. Names, version numbers, and parameters/settings of any data analysis software used (file format conversion, peak picking, peptide identification, etc.) should be reported. As above, the details of tools used likewise allow the analysis to be repeated, and the suitability of the tools used to be evaluated. Furthermore, the tools used must have been validated as an integral part of method validation. Additional Guidelines for Untargeted Assays An untargeted proteomics experiment relies on multiple bioinformatics and data interpretation tools in order to identify peptides and, ultimately, proteins of interest. The following list provides quality assurance and reporting guidelines specific to this type of assay. 5. The protein sequence databases used for database searching should be specified, including links to publicly available databases and dates of accession. Specifying the database allows others to repeat the analysis and provides context to the results. In particular, proteins not in the database will not be found by the database search, so the absence of a protein from the final list cannot be evaluated without full knowledge of the database used in a particular assay. A link (website, computer location, or file) to the specific database and bioinformatics tools used for each evidentiary sample should be stored securely so that the analyses associated with casework are traceable and easily reproduced. 6. A database maintenance and update plan should be incorporated into the quality assurance program. The number of genomes publicly available for proteomic analysis is large and growing. For assays involving environmental samples, it may be necessary to monitor the specificity 217

of target proteins/organisms against an increasing set of background matrices. A formalized database update and maintenance plan facilitates good record keeping and consistent use of proteomic databases over time. 7. The FDR at the peptide-spectrum match, peptide, and protein levels, and the expected number of true and false positives at each level, should be reported where relevant. A description of how these values were calculated should also be reported. When the entire list of proteins is to be presented, the protein-level FDR becomes important. It is important to remember that not all proteins on the list are “confidently identified,” and the protein-level FDR provides an estimate of how many are not. There are various ways of framing the calculation of FDR (141), and it is important to note that the simplest target-decoy approach will provide a protein-level FDR, defined as a protein inferred from incorrectly assigned peptides, not as an absent but identified protein. In line with the Human Proteome Project Guidelines, peptides with a threshold of no more than 1% FDR should be presented. 8. The approach used for protein inference, including any software tools used and associated parameters, should be reported. If protein groups are included, explain the basis for assigning proteins to groups. Many peptides map back to multiple proteins, and protein inference is the process of assembling identified peptides into a list of proteins. Decisions about the protein inference process influence the final protein list. 9. A minimum of two non-nested, peptide markers greater than nine amino acids long (if they exist) should be required to declare the protein of interest present. The requirement for two peptides per protein has been conventional in proteomics for a long time. The recent HUPO guidelines for extraordinary claims peptides expand this to two, unique, non-nested peptides of more than nine residues in length (132). 10. For assays that rely on a small number of peptides (such as an assay for a single protein toxin), confidence metrics, specified by the SOP, should be reported for peptide markers used in a determination. Metrics could include search engine scores, precursor mass errors, aggregate measures of fragment mass errors, p- or q-values, PEP values, number of qualifying spectral matches (spectral counts), and the results of any human expert review of the peptide-spectrum matches. At least one spectrum-level statistical score should be included. Aggregate values (median, minimum and maximum) over all spectra for a given peptide may be appropriate. 11. For assays that rely on a small number of peptides (such as an assay for a single protein toxin), critical peptide-spectrum matches should be manually verified by inspection and peak assignment of their high-resolution MS and MS/MS spectra, and according to rules specified in the SOP and validated as part of the method development process. By “critical PSM,” we mean any PSM that will be used or reported to meet data analysis criteria supporting the presence of the protein of interest. This guideline has its origins in the Human Proteome Project Mass Spectrometry Data Interpretation Guidelines 2.1, guideline 10 (132), and the purpose is not the identification of the peptides by a human expert, but rather a confirmation of the computational result. Search engines are not perfect, and experienced humans can consider information that a search engine does not, such as fragment intensities and unexpected modifications. Alternate sequences explaining the spectrum should also be considered at this stage, for instance, lowerranking database search hits, deamidated forms, etc. The procedure for manual inspection should be a formal part of standard operating procedures. A general protocol for this process can be found in Tabb et. al. (142)

218

References 1. 2. 3. 4.

5. 6. 7. 8. 9. 10. 11. 12.

13.

14. 15.

16.

17. 18. 19. 20.

Daubert v. Merrell Dow Pharmaceuticals, Inc. U.S., 1993. General Electric Co. v. Joiner. U.S., 1995; Vol. 522. Kumho Tire Co. v. Carmichael. U.S., 1999; Vol. 526. Imwinkelried, E. J. The Meaning of “Appropriate Validation” in Daubert v. Merrell Dow Pharmaceuticals, Inc., Interpretes in Light of the Broader Rationalist Tradition, Not the Narrow Scientific Tradition. 30 Fla. St U.L. 2003, 30 (4), 5. Imwinkelried, E. J. Peer Dialogue: The How and What of Appropriate Validation under Daubert: Reconsidering the Treatment of Einstein and Freud. 68 Mo. L. Rev. 2003, 68 (1), 7. Fradella, H. F.; O’Neill, L.; Fogarty, A. The Impact of Daubert on Forensic Science. Pepperdine Law Review 2004, 31, 323–362. Saks, M. J.; Faigman, D. L. Expert Evidence after Daubert. Annual Review of Law and Social Science 2005, 1, 105–130. National Research Council on Forensic Analysis. Weighing Bullet Lead Evidence; The National Academies Press: Washington, DC, 2004. National Research Council. The Polygraph and Lie Detection; The National Academies Press: Washington, DC, 2003; p 416. National Research Council. Ballistic Imaging; The National Academies Press: Washington, DC, 2008; p 344. National Research Council. Review of the Scientific Approaches Used During the FBI’s Investigation of the 2001 Anthrax Letters; The National Academies Press: Washington, DC, 2011; p 232. National Research Council, Committee on Identifying the Needs of the Forensic Sciences. Strengthening Forensic Science in the United States: A Path Forward; The National Academies Press: Washington, DC, 2009. National Academies of Sciences, Engineering Medicine. Support for Forensic Science Research: Improving the Scientific Role of the National Institute of Justice; The National Academies Press: Washington, DC, 2015; p 116. Aslam, B.; Basit, M.; Nisar, M. A.; Khurshid, M.; Rasool, M. H. Proteomics: Technologies and Their Applications. Journal of Chromatographic Science 2017, 55, 182–196. Bustamante, I. T.; Mata, F. S.; Gonzalez, N. H.; Gazapo, R. G.; Palau, J.; Ferreira, M. M. C. Application of Chemometric Tools for Automatic Classification and Profile Extraction of DNA Samples in Forensic Tasks. Analytica Chimica Acta 2007, 595, 43–50. Hochstrasser, D. F. Clinical and Biomedical Applications of Proteomics. In Proteome Research: New Frontiers in Functional Genomics; Wilkins, M. R., Williams, K. L., Appel, R. D., Hochstrasser, D. F., Eds.; Springer: Berlin, Heidelberg, 1997; pp 187−219. Velsko, S. P. Validation Strategies for Microbial Forensic Analysis; Lawrence Livermore National Laboratory, 2012. Project Management Institute. Pmbok Guide and Standards; 2016. U.S. Department of Health and Human Services, FDA. Guidance for Industry: Q2b Validation of Analytical Procedures: Methodology; 1996. Wood, R. How to Validate Analytical Methods. TrAC Trends in Analytical Chemistry 1999, 18, 624–632.

219

21. Kafadar, K. Critical Role of Statistics in Development and Validation of Forensic Methods. In AAAS, 2013; amstat.org/asa/files/pdfs/POL-aaas13.pdf. 22. van Zoonen, P.; Hoogerbrugge, R.; Gort, S. M.; van de Wiel, H. J.; van’t Klooster, H. A. Some Practical Examples of Method Validation in the Analytical Laboratory. TrAC Trends in Analytical Chemistry 1999, 18, 584–593. 23. Gonzalez, C.; Spinelli, S.; Gille, J.; Touraud, E.; Prichard, E. Validation Procedure for Existing and Emerging Screening Methods. TrAC Trends in Analytical Chemistry 2007, 26, 315–322. 24. Trullols, E.; Ruis Sanchez, I.; Rius, F. X. Validation of Qualitative Analytical Methods. TrAC Trends in Analytical Chemistry 2004, 23, 137–145. 25. Peters, F. T.; Drummer, O. H.; Musshoff, F. Validation of New Methods. Forensic Science International 2007, 165, 216–224. 26. Vessman, J. Selectivity or Specificity? Validation of Analytical Methods from the Perspective of an Analytical Chemist in the Pharmaceutical Industry. Journal of Pharmaceutical and Biomedical Analysis 1996, 14, 867–869. 27. Rozet, E.; Dew, W.; Ziemons, E.; Bouklouze, A.; Boulanger, B.; Hubert, P. Methodologies for the Transfer of Analytical Methods: A Review. Journal of Chromatography B 2009, 877, 2214–2223. 28. Bouabidi, A.; Rozet, E.; Fillet, M.; Ziemons, E.; Chapuzet, E.; Mertens, B.; Klinkenberg, R.; Ceccato, A.; Talbi, M.; Streel, B.; Bouklouze, A.; Boulanger, B.; Hubert, P. Critical Analysis of Several Analytical Method Validation Strategies in the Framework of the Fit for Purpose Concept. Journal of Chromatography A 2010, 1217, 3180–3192. 29. Vander Heyden, Y.; Nijhuis, A.; Smeyers-Verbeke, J.; Vandeginste, B. G. M.; Massart, D. L. Guidance for Robustness/Ruggedness Tests in Method Validation. Journal of Pharmaceutical and Biomedical Analysis 2001, 24, 723–753. 30. Hund, E.; Massart, D. L.; Smeyers-Verbeke, J. Inter-Laboratory Studies in Analytical Chemistry. Analytica Chimica Acta 2000, 423, 145–165. 31. Aebersold, R.; Mann, M. Mass Spectrometry-Based Proteomics. Nature 2003, 422, 198–207. 32. Altelaar, A. F. M.; Munoz, J.; Heck, A. J. R. Next-Generation Proteomics: Towards an Integrative View of Proteome Dynamics. Nat. Rev. Genet. 2013, 14, 35–48. 33. Yates, J. R.; Ruse, C. I.; Nakorchevsky, A. Proteomics by Mass Spectrometry: Approaches, Advances, and Applications. Annual Review of Biomedical Engineering 2009, 11, 49–79. 34. Nesvizhskii, A. Protein Identification by Tandem Mass Spectrometry and Sequence Database Searching; Humana Press, 2007. 35. Nesvizhskii, A. A Survey of Computational Methods and Error Rate Estimation Procedures for Peptide and Protein Identificaiton in Shotgun Proteomics. Journal of Proteomics 2010, 73, 2092–2123. 36. Lange, V.; Picotti, P.; Domon, B.; Aebersold, R. Review Selected Reaction Monitoring for Quantitative Proteomics. Molecular Systems Biology 2008, 4. 37. Picotti, P.; Aebersold, R. Selected Reaction Monitoring-Based Proteomics: Workflows, Potential, Pitfalls and Future Directions. Nat. Meth. 2012, 9, 555–566. 38. Budowle, B.; Schutzer, S. E.; Morse, S. A.; Martinez, K. F.; Chakraborty, R.; Marrone, B. L.; Messenger, S. L.; Murch, R. S.; Jackson, P. J.; Williamson, P.; Harmon, R.; Velsko, S.

220

39. 40.

41. 42.

43.

44.

45.

46.

47. 48.

49.

50.

P. Criteria for Validation of Methods in Microbial Forensics. Applied and Environmental Microbiology 2008, 74, 5599–5607. United States Government Accountability Office. Anthrax: Agency Approaches to Validation and Statistical Analysis Could Be Improved; GAO, December 2014. Lluveras-Tenorio, A.; Vinciguerra, R.; Galano, E.; Blaensdorf, C.; Emmerling, E.; Colombini, M. P.; Birolo, L.; Bonaduce, I. GC/MS and Proteomics to Unravel the Painting History of the Lost Giant Buddhas of Bamiyan (Afghanistan). Plos One 2017, 12, 18. Vinciguerra, R.; De Chiaro, A.; Pucci, P.; Marino, G.; Birolo, L. Proteomic Strategies for Cultural Heritage: From Bones to Paintings. Microchem J. 2016, 126, 341–348. Boros-Major, A.; Bona, A.; Lovasz, G.; Molnar, E.; Marcsik, A.; Palfi, G.; Mark, L. New Perspectives in Biomolecular Paleopathology of Ancient Tuberculosis: A Proteomic Approach. Journal of Archaeological Science 2011, 38, 197–201. Welker, F.; Collins, M. J.; Thomas, J. A.; Wadsley, M.; Brace, S.; Cappellini, E.; Turvey, S. T.; Reguero, M.; Gelfo, J. N.; Kramarz, A.; Burger, J.; Thomas-Oates, J.; Ashford, D. A.; Ashton, P. D.; Rowsell, K.; Porter, D. M.; Kessler, B.; Fischer, R.; Baessmann, C.; Kaspar, S.; Olsen, J. V.; Kiley, P.; Elliott, J. A.; Kelstrup, C. D.; Mullin, V.; Hofreiter, M.; Willerslev, E.; Hublin, J.-J.; Orlando, L.; Barnes, I.; MacPhee, R. D. E. Ancient Proteins Resolve the Evolutionary History of Darwin’s South American Ungulates. Nature 2015, 522, 81. Jarman, K. H.; Heller, N. C.; Jensen, S. C.; Hutchison, J. R.; Deatheridge-Kaiser, B. L.; Payne, S. H.; Wunschel, D. S.; Merkley, E. D. Proteomics Goes to Court: A Statistical Foundation for Forensic Toxin/Organism Identification Using Bottom-up Proteomics. Journal of Proteome Research 2018, 17, 3075–3085. Dworzanski, J. P.; Deshpande, S. V.; Chen, R.; Jabbour, R. E.; Snyder, A. P.; Wick, C. H.; Li, L. Mass Spectrometry-Based Proteomics Combined with Bioinformatic Tools for Bacterial Classification. Journal of Proteome Research 2006, 5, 76–87. Dworzanski, J. P.; Dickinson, D. N.; Deshpande, S. V.; Snyder, A. P.; Eckenrode, B. A. Discrimination and Phylogenomic Classification of Bacillus Anthracis-Cereus-Thuringiensis Strains Based on LC-MS/MS Analysis of Whole Cell Protein Digests. Analytical Chemistry 2010, 82, 145–155. Dworzanski, J. P.; Snyder, A. P. Classification and Identification of Bacteria Using Mass Spectrometry-Based Proteomics. Expert Review of Proteomics 2005, 2, 863–878. Dworzanski, J. P.; Snyder, A. P.; Chen, R.; Zhang, H.; Wishart, D.; Li, L. Identification of Bacteria Using Tandem Mass Spectrometry Combined with a Proteome Database and Statistical Scoring. Analytical Chemistry 2004, 76, 2355–2366. VerBerkmoes, N. C.; Hervey, W. J.; Shah, M.; Land, M.; Hauser, L.; Larimer, F. W.; Van Berkel, G. J.; Goeringer, D. E. Evaluation of “Shotgun” Proteomics for Identification of Biological Threat Agents in Complex Environmental Matrixes: Experimental Simulations. Analytical Chemistry 2005, 77, 923–932. Clowers, B. H.; Wunschel, D. S.; Kreuzer, H. W.; Engelmann, H. E.; Valentine, N.; Wahl, K. L. Characterization of Residual Medium Peptides from Yersinia Pestis Cultures. Analytical Chemistry 2013, 85, 3933–3939.

221

51. Laatsch, C. N.; Durbin-Johnson, B. P.; Rocke, D. M.; Mukwana, S.; Newland, A. B.; Flagler, M. J.; Davis, M. G.; Eigenheer, R. A.; Phinney, B. S.; Rice, R. H. Human Hair Shaft Proteomic Profiling: Individual Differences, Site Specificity and Cuticle Analysis. Peerj 2014, 2. 52. Parker, G. J.; Leppert, T.; Anex, D. S.; Hilmer, J. K.; Matsunami, N.; Baird, L.; Stevens, J.; Parsawar, K.; Durbin-Johnson, B. P.; Rocke, D. M.; Nelson, C.; Fairbanks, D. J.; Wilson, A. S.; Rice, R. H.; Woodward, S. R.; Bothner, B.; Hart, B. R.; Leppert, M. Demonstration of Protein-Based Human Identification Using the Hair Shaft Proteome. PLOS ONE 2016, 11, e0160653. 53. Legg, K. M.; Powell, R.; Reisdorph, N.; Reisdorph, R.; Danielson, P. B. Discovery of Highly Specific Protein Markers for the Identification of Biological Stains. Electrophoresis 2014, 35, 3069–3078. 54. Legg, K. M.; Powell, R.; Reisdorph, N.; Reisdorph, R.; Danielson, P. B. Verification of Protein Biomarker Specificity for the Identification of Biological Stains by Quadrupole Time-of-Flight Mass Spectrometry. Electrophoresis 2017, 38, 833–845. 55. Van Steendam, K.; De Ceuleneer, M.; Dhaenens, M.; Van Hoofstat, D.; Deforce, D. Mass Spectrometry-Based Proteomics as a Tool to Identify Biological Matrices in Forensic Science. International Journal of Legal Medicine 2013, 127, 287–298. 56. Procopio, N.; Buckley, M. Minimizing Laboratory-Induced Decay in Bone Proteomics. Journal of Proteome Research 2017, 16, 447–458. 57. Buckley, M.; Collins, M.; Thomas-Oates, J.; Wilson, J. C. Species Identification by Analysis of Bone Collagen Using Matrix-Assisted Laser Desorption/Ionisation Time-of-Flight Mass Spectrometry. Rapid Communications in Mass Spectrometry 2009, 23, 3843–3854. 58. Teubl, J.; Yang, H.; Siegel, D.; Fenyo, D. Species Identification Using Bayesian Modeling and Mass Spectrometry. In 64th Conference on Mass Spectrometry and Allied Topics; American Society of Mass Spectrometry: San Antonio, Texas, 2016. 59. Fredriksson, S.-Å.; Hulst, A. G.; Artursson, E.; de Jong, A. L.; Nilsson, C.; van Baar, B. L. M. Forensic Identification of Neat Ricin and of Ricin from Crude Castor Bean Extracts by Mass Spectrometry. Analytical Chemistry 2005, 77, 1545–1555. 60. Schieltz, D. M.; McGrath, S. C.; McWilliams, L. G.; Rees, J.; Bowen, M. D.; Kools, J. J.; Dauphin, L. A.; Gomez-Saladin, E.; Newton, B. N.; Stang, H. L.; Vick, M. J.; Thoma, J.; Pirkle, J. L.; Barr, J. R. Analysis of Active Ricin and Castor Bean Proteins in a Ricin Preparation, Castor Bean Extract, and Surface Swabs from a Public Health Investigation. Forensic Sci. Int. 2011, 209, 70–79. 61. Brinkworth, C. S.; Pigott, E. J.; Bourne, D. J. Detection of Intact Ricin in Crude and Purified Extracts from Castor Beans Using Matrix-Assisted Laser Desorption Ionization Mass Spectrometry. Analytical Chemistry 2009, 81, 1529–1535. 62. Kalb, S. R.; Barr, J. R. Mass Spectrometric Detection of Ricin and Its Activity in Food and Clinical Samples. Analytical Chemistry 2009, 81, 2037–2042. 63. Schieltz, D. M.; McWilliams, L. G.; Kuklenyik, Z.; Prezioso, S. M.; Carter, A. J.; Williamson, Y. M.; McGrath, S. C.; Morse, S. A.; Barr, J. R. Quantification of Ricin, Rca and Comparison of Enzymatic Activity in 18 Ricinus Communis Cultivars by Isotope Dilution Mass Spectrometry. Toxicon 2015, 95, 72–83.

222

64. Dupré, M.; Gilquin, B.; Fenaille, F.; Feraudet-Tarisse, C.; Dano, J.; Ferro, M.; Simon, S.; Junot, C.; Brun, V.; Becher, F. Multiplex Quantification of Protein Toxins in Human Biofluids and Food Matrices Using Immunoextraction and High-Resolution Targeted Mass Spectrometry. Analytical Chemistry 2015, 87, 8473–8480. 65. Boyer, A. E.; Moura, H.; Woolfitt, A. R.; Kalb, S. R.; McWilliams, L. G.; Pavlopoulos, A.; Schmidt, J. G.; Ashley, D. L.; Barr, J. R. From the Mouse to the Mass Spectrometer: Detection and Differentiation of the Endoproteinase Activities of Botulinum Neurotoxins a−G by Mass Spectrometry. Analytical Chemistry 2005, 77, 3916–3924. 66. Kalb, S. R.; Barr, J. R. Mass Spectrometric Identification and Differentiation of Botulinum Neurotoxins through Toxin Proteomics, Reviews in Analytical Chemistry. Analytical Chemistry 2013, 32, 189–196. 67. Gilquin, B.; Jaquinod, M.; Louwagie, M.; Kieffer-Jaquinod, S.; Kraut, A.; Ferro, M.; Becher, F.; Brun, V. A Proteomics Assay to Detect Eight Cbrn-Relevant Toxins in Food. Proteomics 2017, 17, 1600357. 68. Fredriksson, S.-A.; Artursson, E.; Bergstrom, T.; Ostin, A.; Nilsson, C.; Astot, C. Identification of Rip-Ii Toxins by Affinity Enrichment, Enzymatic Digestion and Lc-Ms. Analytical Chemistry 2015, 87, 967–974. 69. Merkley, E. D.; Sego, L. H.; Lin, A.; Leiser, O. P.; Kaiser, B. L. D.; Adkins, J. N.; Keim, P. S.; Wagner, D. M.; Kreuzer, H. W. Protein Abundances Can Distinguish between NaturallyOccurring and Laboratory Strains of Yersinia Pestis, the Causative Agent of Plague. PLOS ONE 2017, 12, e0183478. 70. Keim, P. S.; Budowle, B.; Ravel, J., Chapter 2—Microbial Forensic Investigation of the Anthrax-Letter Attacks. In Microbial Forensics, 2nd ed.; Budowle, B., Schutzer, S. E., Breeze, R. G., Keim, P. S., Morse, S. A. , Eds.; Academic Press: San Diego, 2011; pp 15−25. 71. Huber, L. Understanding the Final Fda Guidance for Validation of Analytical Methods; https://vimeo.com/170008718 (accessed Aug 20, 2019). 72. Eng, J.; Searle, B.; Clauser, K.; Tabb, D. A Face in the Crowd: Recognizing Peptides through Database Search. Molecular & Cellular Proteomics 2011, 10, R111.009522. 73. Ma, B.; Johnson, R. De Novo Sequencing and Homology Searching. Molecular & Cellular Proteomics : MCP 2012, 11, O111.014902. 74. Hughes, C.; Ma, B.; Lajoie, G. A. De Novo Sequencing Methods in Proteomics. Methods in Molecular Biology (Clifton, N.J.) 2010, 604, 105–121. 75. Devabhaktuni, A.; Elias, J. E. Application of De Novo Sequencing to Large-Scale Complex Proteomics Data Sets. Journal of Proteome Research 2016, 15, 732–742. 76. Lam, H.; Aebersold, R. Spectral Library Searching for Peptide Identification Via Tandem Ms. Methods in Molecular Biology (Clifton, N.J.) 2010, 604, 95–103. 77. Griss, J. Spectral Library Searching in Proteomics. PROTEOMICS 2016, 16, 729–740. 78. Seidler, J.; Zinn, N.; Boehm, M. E.; Lehmann, W. D. De Novo Sequencing of Peptides by MS/ MS. Proteomics 2010, 10, 634–649. 79. Allmer, J. Algorithms for the De Novo Sequencing of Peptides from Tandem Mass Spectra. Expert Rev. Proteomics 2011, 8, 645–657.

223

80. Potgieter, T.; Nel, A. J.; Tabb, D. L.; Fortuin, S.; Garnett, S.; Blackburn, J.; Mulder, N. Metanovo: A Probabilistic Approach to Peptide and Polymorphism Discovery in Complex Mass Spectrometry Datasets. bioRxiv 2019, 605550. 81. Mani, D. R.; Abbatiello, S. E.; Carr, S. A. Statistical Characterization of Multiple-Reaction Monitoring Mass Spectrometry (Mrm-Ms) Assays for Quantitative Proteomics. BMC Bioinformatics 2012, 13, S9. 82. Alves, G.; Wang, G.; Ogurtsov, A. Y.; Drake, S. K.; Gucek, M.; Sacks, D. B.; Yu, Y. K. Rapid Classification and Identification of Multiple Microorganisms with Accurate Statistical Significance Via High-Resolution Tandem Mass Spectrometry. J. Am. Soc. Mass Spectrom. 2018, 29, 1721–1737. 83. Budowle, B.; Monson, K. L. The Forensic Signficance of Various Reference Population Databases for Estimating the Rarity of Variable Number of Tandem Repeat (Vntr) Loci Profiles. DNA Fingerprinting: State of the Science 1993, 67, 177–191. 84. Muth, T.; Benndorf, D.; Reichl, U.; Rapp, E.; Martens, L. Searching for a Needle in a Stack of Needles: Challenges in Metaproteomics Data Analysis. Molecular BioSystems 2013, 9, 578–585. 85. Timmins-Schiffman, E.; May, D. H.; Mikan, M.; Riffle, M.; Frazar, C.; Harvey, H. R.; Noble, W. S.; Nunn, B. L. Critical Decisions in Metaproteomics: Achieving High Confidence Protein Annotations in a Sea of Unknowns. Isme Journal 2017, 11, 309–314. 86. Oulas, A.; Pavloudi, C.; Polymenakou, P.; Pavlopoulos, G. A.; Papanikolaou, N.; Kotoulas, G.; Arvanitidis, C.; Iliopoulos, I. Metagenomics: Tools and Insights for Analyzing NextGeneration Sequencing Data Derived from Biodiversity Studies. Bioinformatics and Biology Insights 2015, 9, 75–88. 87. Elias, J.; Gygi, S. Target-Decoy Search Strategy for Mass Spectrometry-Based Proteomics. In Proteome Bioinformatics; Hubbard, S. J., Jones, A. R., Eds.; Humana Press, 2010; Vol. 604, pp 55−71. 88. Elias, J. E.; Gygi, S. P. Target-Decoy Search Strategy for Increased Confidence in Large-Scale Protein Identifications by Mass Spectrometry. Nat. Meth. 2007, 4, 207–214. 89. Yadav, A. K.; Kumar, D.; Dash, D. Learning from Decoys to Improve the Sensitivity and Specificity of Proteomics Database Search Results. PLoS One 2012, 7, e50651. 90. Keller, A.; Nesvizhskii, A. I.; Kolker, E.; Aebersold, R. Empirical Statistical Model to Estimate the Accuracy of Peptide Identifications Made by MS/MS and Database Search. Analytical Chemistry 2002, 74, 5383–5392. 91. Käll, L.; Canterbury, J. D.; Weston, J.; Noble, W. S.; MacCoss, M. J. Semi-Supervised Learning for Peptide Identification from Shotgun Proteomics Datasets. Nat. Meth. 2007, 4, 923–925. 92. Spivak, M.; Weston, J.; Bottou, L.; Käll, L.; Noble, W. S. Improvements to the Percolator Algorithm for Peptide Identification from Shotgun Proteomics Data Sets. Journal of Proteome Research 2009, 8, 3737–3745. 93. Käll, L.; Storey, J. D.; Noble, W. S. Qvality: Non-Parametric Estimation of Q-Values and Posterior Error Probabilities. Bioinformatics 2009, 25, 964–966. 94. Kim, S.; Pevzner, P. A. Ms-Gf+ Makes Progress Towards a Universal Database Search Tool for Proteomics. Nat. Commun. 2014, 5. 224

95. Kim, S.; Gupta, N.; Pevzner, P. A. Spectral Probabilities and Generating Functions of Tandem Mass Spectra: A Strike against Decoy Databases. Journal of Proteome Research 2008, 7, 3354–3363. 96. Howbert, J. J.; Noble, W. S. Computing Exact P-Values for a Cross-Correlation Shotgun Proteomics Score Function. Molecular & Cellular Proteomics 2014, 13, 2467–2479. 97. Nesvizhskii, A. I.; Keller, A.; Kolker, E.; Aebersold, R. A Statistical Model for Identifying Proteins by Tandem Mass Spectrometry. Anal Chem 2003, 75. 98. Li, Y. F.; Arnold, R. J.; Tang, H.; Radivojac, P. The Importance of Peptide Detectability for Protein Identification, Quantification, and Experiment Design in Ms/Ms Proteomics. Journal of Proteome Research 2010, 9, 6288–6297. 99. Huang, T.; He, Z. A Linear Programming Model for Protein Inference Problem in Shotgun Proteomics. Bioinformatics 2012, 28, 2956–2962. 100. Audain, E.; Uszkoreit, J.; Sachsenberg, T.; Pfeuffer, J.; Liang, X.; Hermjakob, H.; Sanchez, A.; Eisenacher, M.; Reinert, K.; Tabb, D. L.; Kohlbacher, O.; Perez-Riverol, Y. In-Depth Analysis of Protein Inference Algorithms Using Multiple Search Engines and Well-Defined Metrics. J. Proteomics 2017, 150, 170–182. 101. Alves, G.; Wang, G.; Ogurtsov, A. Y.; Drake, S. K.; Gucek, M.; Suffredini, A. F.; Sacks, D. B.; Yu, Y.-K. Identification of Microorganisms by High Resolution Tandem Mass Spectrometry with Accurate Statistical Significance. Journal of the American Society for Mass Spectrometry 2016, 27, 194–210. 102. Tang, Y.; Srihari, S. N. Likelihood Ratio Estimation in Forensic Identification Using Similarity and Rarity. Pattern Recognition 2014, 47, 945–958. 103. Sampson, A. R.; Smith, R. L. An Information-Theory Model for the Evaluation of Circumstantial Evidence. IEEE Transactions on Systems Man and Cybernetics 1985, 15, 9–16. 104. Page, M.; Taylor, J.; Blenkin, M. Uniqueness in the Forensic Identification Sciences—Fact or Fiction? Forensic Science International 2011, 206, 12–18. 105. Nordgaard, A.; Hedberg, K.; Widen, C.; Ansell, R. Comments on “the Database Search Problem” with Respect to a Recent Publication in Forensic Science International. Forensic Science International 2012, 217, E32–E33. 106. Nordgaard, A.; Hedberg, K.; Widen, C.; Ansell, R. Comments on “the Database Search Problem” with Respect to a Recent Publication in Forensic Science International. Forensic Science International 2011, 217, e32–e33. 107. Lindley, D. V. Subjective-Probability, Decision-Analysis and Their Legal Consequences. Journal of the Royal Statistical Society Series a-Statistics in Society 1991, 154, 83–92. 108. Lenth, R. V. On Identification by Probability. Journal of the Forensic Science Society 1986, 26, 197–213. 109. Kaye, D. H. Beyond Uniqueness: The Birthday Paradox, Source Attribution and Individualization in Forensic Science Testimony. Law Probability & Risk 2013, 12, 3–11. 110. Jayaprakash, P. T. Reply to the Letter to the Editor: On the Limitations of Probability in Conceptualizing Pattern Matches in Forensic Science. Forensic Science International 2014, 239. 111. Jayaprakash, P. T. Practical Relevance of Pattern Uniqueness in Forensic Science. Forensic Science International 2013, 231, 403.e401–403.e416.

225

112. Hicks, T.; Biederman, A.; Koeijer, J. A. d.; Taroni, F.; Champod, C.; Evett, I. W. The Importance of Distinguishing Information from Evidence/Observations When Formulating Propositions. Science & Justice 2015, 55, 520–525. 113. Evett, I. W.; Aitken, C. G. G.; Berger, C. E. H.; Buckleton, J. S.; Champod, C.; Curran, J. M.; Dawid, A. P.; Gill, P.; González-Rodríguez, J.; Jackso, G.; Kloosterman, A.; Lovelock, T.; Lucy, D.; Margot, P.; McKenna, L.; Meuwly, D.; Neumann, C.; Daeid, N. N.; Nordgaard, A.; Puch-Solis, R.; Rasmusson, B.; Radmayne, M.; Roberts, P.; Robertson, B.; Roux, C.; Sjerps, M. J.; Taroni, F.; Tjin-A-Tsoi, T.; Vignaux, G. A.; Willis, S. M.; Zadora, G. Expressing Evaluative Opinions: A Position Statement. Science & Justice 2011, 51, 1–2. 114. Evett, I. W. A Bayesian-Approach to the Problem of Interpreting Glass Evidence in ForensicScience Casework. Journal of the Forensic Science Society 1986, 26, 3–18. 115. Block, A.; Weyermann, C.; Dujourdy, L.; Esseiva, P.; Berg, J. v. d. Different Likelihood Ratio Approaches to Evaluate the Strength of Evidence of Mdma Tablet Comparisons. Forensic Science International 2009, 191, 42–51. 116. Biederman, A.; Taroni, F.; Garbolino, P. Equal Prior Probabilities: Can One Do Any Better? Forensic Science International 2007, 172, 85–93. 117. Biederman, A.; Gittelson, S.; Taroni, F. Recent Misconceptions About the “Database Search Problem”: A Probabilistic Analysis Using Bayesian Networks. Forensic Science International 2011, 212, 51–60. 118. Biederman, A.; Bozza, S.; Taroni, F. Decision Theoretic Properties of Forensic Identification: Underlying Logic and Argumentative Implications. Forensic Science International 2008, 177, 120–132. 119. Berger, J. O.; Sellke, T. Testing a Point Null Hypothesis—the Irreconcilability of P-Values and Evidence. Journal of the American Statistical Association 1987, 82, 112–122. 120. Berger, C. E. H.; Buckleton, J.; Champod, C.; Evett, I. W.; Jackson, G. Evidence Evaluation: A Response to the Court of Appeal Judgement in R V T. Science & Justice 2011, 51, 43–49. 121. Balding, D. J.; Steele, C. D. Weight-of-Evidence for Forensic DNA Profiles, 2nd ed.; John Wiley & Sons, Inc: UK, 2015. 122. Asten, A. v. On the Added Value of Forensic Science and Grand Innovation Challenges for the Forensic Community. Science and Justice 2014, 54, 170–179. 123. Ali, T.; Spreewers, L.; Veldhuis, R.; Meuwly, D. Sampling Variability in Forensic LikelihoodRatio Computation: A Simulation Study. Science & Justice 2015, 55, 499–508. 124. Aitken, C. G. G.; Lucy, D. Evaluation of Trace Evidence in the Form of Multivariate Data. Applied Statistics 2004, 53, 109–122. 125. Sahl, J.; Vazquez, A.; Hall, C.; Busch, J.; Tuanyok, A.; Mayo, M.; Schupp, J.; Lummis, M.; Pearson, T.; Shippy, K.; Colman, R.; Allender, C.; Theobald, V.; Sarovich, D.; Price, E.; Hutcheson, A.; Korlach, J.; LiPuma, J.; Ladner, J.; Lovett, S.; Koroleva, G.; Palaclos, G.; Limmathurotsakul, D.; Wuthlekanun, V.; Wongsuwan, G.; Currie, B.; Keim, P.; Wagner, D. The Effects of Signal Erosion and Core Genomic Reduction on the Identification of Diagnostic Markers. mBio 2016, 7, e00846-16. 126. Slezak, T.; Kuczmarski, T.; Ott, L.; Torres, C.; Medeiros, D.; Smith, J.; Truitt, B.; Mulakken, N.; Lam, M.; Vitalis, E. Comparative Genomics Tools Applied to Bioterrorism Defence. Briefings in Bioinformatics 2003, 4. 226

127. Penders, J.; Verstraete, A. Laboratory Guidelines and Standards in Clinical and Forensic Toxicology. Accreditation and Quality Assurance 2006, 11, 284–290. 128. Society of Forensic Toxicology. Forensic Toxicology Laboratory Guidelines; 2006. 129. Federal Bureau of Investigation. Quality Assurance Standards for Forensic DNA Laboratories. https://www.fbi.gov/about-us/lab/biometric-analysis/codis/ (accessed August 25, 2016). 130. Association for Forensic Quality Assurance Managers. www.afqam.org (accessed August 25, 2016). 131. Bradshaw, R. A.; Burlingame, A. L.; Carr, S.; Aebersold, R. Reporting Protein Identification Data: The Next Generation of Guidelines. Molecular & Cellular Proteomics 2006, 5.5. 132. Deutsch, E. W.; Overall, C. M.; Van Eyk, J. E.; Baker, M. S.; Paik, Y.-K.; Weintraub, S. T.; Lane, L.; Martens, L.; Vandenbrouck, Y.; Kusebauch, U.; Hancock, W. S.; Hermjakob, H.; Aebersold, R.; Moritz, R. L.; Omenn, G. S. Human Proteome Project Mass Spectrometry Data Interpretation Guidelines 2.1. Journal of Proteome Research 2016, 15, 3961–3970. 133. Frank, A. M.; Savitski, M. M.; Nielsen, M. L.; Zubarev, R. A.; Pevzner, P. A. De Novo Peptide Sequencing and Identification with Precision Mass Spectrometry. Journal of Proteome Research 2007, 6, 114–123. 134. Isir, A. B.; Ozkorkmaz, A.; Pehlivan, S. Allele Frequencies for 13 Strs Loci in a Western Anatolia Population and Their Forensic Evaluation. Annals of Human Biology 2015, 42, 494–497. 135. Thevis, M.; Loo, J. A.; Loo, R. R.; Schanzer, W. Recommended Criteria for the Mass Spectrometric Identification of Target Peptides and Proteins (