Risk Assessment of Phytochemicals in Food: Novel Approaches (Forschungsberichte (DFG))

Diese Seite wurde absichtlich leer gelassen

Risk Assessment of Phytochemicals in Food Novel Approaches Symposium

Diese Seite wurde absichtlich leer gelassen

Risk Assessment of Phytochemicals in Food Novel Approaches

Symposium Editors: DFG Senate Commission on Food Safety (SKLM) Editorial Committee: Gerhard Eisenbrand (Chairman), Jan Hengstler, Hans-Georg Joost, Sabine Kulling, Ivonne Rietjens, Josef Schlatter, Pablo€Steinberg and Doris Marko Scientists of the SKLM Secretariat: Sabine Guth, Michael Habermeyer and Barbara Kochte-Clemens

4 Deutsche Forschungsgemeinschaft German Research Foundation Kennedyallee 40 · 53175 Bonn, Germany Postal address: 53170 Bonn, Germany Phone: +â•›49 228 885-1 Fax: +â•›49 228 885-2777 [email protected] www.dfg.de

Alle Bücher von WILEY-VCH werden sorgfältig erarbeitet. Dennoch übernehmen Autoren, Herausgeber und Verlag in keinem Fall, einschließlich des vorliegenden Werkes, für die Richtigkeit von Angaben, Hinweisen und Ratschlägen sowie für eventuelle Druckfehler irgendeine Haftung.

Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über abrufbar. ISBN 978-3-527-32929-8 © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Gedruckt auf säurefreiem Papier. Alle Rechte, insbesondere die der Übersetzung in andere Sprachen, vorbehalten. Kein Teil dieses Buches darf ohne schriftliche Genehmigung des Verlages in irgendeiner Form – durch Fotokopie, Mikroverfilmung oder irgendein anderes Verfahren – reproduziert oder in eine von Maschinen, insbesondere von Datenverarbeitungsmaschinen, verwendbare Sprache übertragen oder übersetzt werden. Die Wiedergabe von Warenbezeichnungen, Handelsnamen oder sonstigen Kennzeichen in diesem Buch berechtigt nicht zu der Annahme, dass diese von jedermann frei benutzt werden dürfen. Vielmehr kann es sich auch dann um eingetragene Warenzeichen oder sonstige gesetzlich geschützte Kennzeichen handeln, wenn sie nicht eigens als solche markiert sind. All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form – by photoprinting, microfilm, or any other means – nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not specifically marked as such, are not to be considered unprotected by law. Corporate Design: besscom AG, Berlin Layout und Typografie: Tim Wübben, DFG Satz: Primustype Hurler GmbH, Notzingen Druck: betz-druck GmbH, Darmstadt Bindung: Litges & Dopf GmbH, Heppenheim Printed in the Federal Republic of Germany

5

Inhalt╛/↜Contents Vorwort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Preface . . . . . . . . . . . . . . . . . 尓. . . . . . . . . . . . . . . . . . 尓. . . . . . . . . . . . . . . . . 12 1 Bericht und Schlussfolgerungen . . . . . . . . . . . . . . . . . 尓. . . . . . . . . . . . . 13 1.1

Einleitung . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . . . . 13

1.2

Methodenübergreifende Aspekte . . . . . . . . . . . . . . . . . å°“. . . . . . . . 14

1.3

Methoden . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . . . . 15

1.4

Schlussfolgerungen und Empfehlungen . . . . . . . . . . . . . . . . . å°“. . . 22

1.5

Fazit . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . 25

2 Report and Conclusions . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . 26 2.1

Preface . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . 26

2.2

Transdisciplinary Aspects . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . 27

2.3

Methodologies . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . 28

2.4

Conclusions and Recommendations . . . . . . . . . . . . . . . . . å°“. . . . . . 34

2.5

Concluding Remarks . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . 37

3 Contributions . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . 38 3.1

Visions on Toxicity Testing in the 21st Century: �Reflections on a Strategy Document of the US€�National Research Council Marcel Leist, Thomas Hartung, and Pierluigi Nicotera . . . . . . . . . . . . . . . 38

3.2

Safety Assessment of Botanicals and Botanical Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach Gerrit Speijers, Bernard Bottex, Birgit Dusemund et al. . . . . . . . . . . . . . . 57

3.3

In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure– Activity Relationship Analysis Luis G. Valerio Jr., Naomi L. Kruhlak, and R. Daniel Benz . . . . . . . . . . . 78

3.4

Testing Computational Toxicology Models with Phytochemicals Luis G. Valerio Jr., Kirk B. Arvidson, Emily Busta et al. . . . . . . . . . . . . . 93

3.5

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data Benoît Schilter, Manuel Dominguez Estevez, Myriam Coulet et al. . . . . . 110

6

Inhalt╃/╛Contents

3.6

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and DetoÂ�xification of Coumarin and Estragole: Implications for Risk Assessment Ivonne M.╃C.╃M. Rietjens, Ans Punt, Benoît Schilter et al. . . . . . . . . . . . . 124

3.7

In Vitro Models for Carcinogenicity Testing – Reality or Fantasy? Pablo Steinberg, Carsten Müller, Kristina Ullmann et al. . . . . . . . . . . . . 149

3.8

Carcinogen Specific Expression Profiling: Prediction of Carcinogenic Potential? Hans-Jürgen Ahr, and Heidrun Ellinger-Ziegelbauer . . . . . . . . . . . . . . . 160

3.9

Safety and Biological Efficacy Testing of Phyto�chemicals: An Industry Approach Anette Thiel, Jochen Bausch, Mareike Beck et al.╃ . . . . . . . . . . . . . . . . . . . 178

3.10 Metabolite Profiling in Rat Plasma as a Potential New Tool for the Assessment of Chemically Induced Â�Toxicity Hennicke Kamp, Roland Buesen, Eric Fabian et al. . . . . . . . . . . . . . . . . . 189 3.11 Profiling Techniques in Nutrition and Food Research Hannelore Daniel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 3.12 The Complex Links between Dietary Phytochemicals and Human Health Deciphered by Metabolomics Claudine Manach, Jane Hubert, Rafael Llorach et al. . . . . . . . . . . . . . . . 212 3.13 Anti-Oxidative and Antigenotoxic Properties of Â�Vegetables and Dietary Phytochemicals: The Value of Genomics Biomarkers in Molecular Epidemiology Theo M.╃C.╃M. de Kok, Pim de Waard, Lonneke C. Wilms et al.╃ . . . . . . . . 236 3.14 The Japanese Toxicogenomics Project: Application of Toxicogenomics – Utilizing Toxicogenomics into Drug Safety Screening Takeki Uehara, Atsushi Ono, Toshiyuki Maruyama et al. . . . . . . . . . . . . 254 3.15 Toxicology and Risk Assessment of Coumarin: Focus on Human Data Klaus Abraham, Friederike Wöhrlin, Oliver Lindtner et al. . . . . . . . . . . . 272 3.16 Risk from Furocoumarins in Food? An Exposure Â�Assessment Dieter Schrenk, Sabine Guth, Nicole Raquet et al.╃ . . . . . . . . . . . . . . . . . . 295 3.17 Transcriptome Analysis in Benefit–Risk Assessment of Micronutrients and Bioactive Food Components Jaap Keijer, Yvonne G.╃J. van Helden, Annelies Bunschoten et al.╃ . . . . . . 309 3.18 Colorectal and Prostate Cancer: The Role of Â�Candidate Genes in Nutritional Pathways Ulrike Peters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 3.19 Glucosinolates: DNA Adduct Formation In Vivo and Mutagenicity In Vitro Chimgee Baasanjav Gerber, Wolfram Engst, Simone Florian et al.╃ . . . . . 333

Inhalt╃/╛Contents

3.20 Defence Mechanisms against Toxic Phytochemicals in the Diet of Domestic Animals Johanna Fink-Gremmels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343 4 Posters . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . 361 4.1

Coumarin Risk Assessment: Lessons from Human Data Klaus Abraham, Klaus-Erich Appel, and Alfonso Lampen . . . . . . . . . . . 361

4.2

Coffee and Coffee Compounds are Effective �Antioxidants in Human Cells and In Vivo Tamara Bakuradze, Matthias Baum, Gerhard Eisenbrand et al.╃ . . . . . . 364

4.3

Studying Absorption, Distribution, Metabolism, and Excretion of a Complex Extract Mareike Beck, Martine Bruchlen, Volker Elste et al.╃ . . . . . . . . . . . . . . . . . 369

4.4

Polyphenolic Apple Extracts and their Constituents Modulate DNA Strand Breaks and Oxidation �Damage in Human Colon Carcinoma Cells Phillip Bellion, Frank Will, Helmut Dietrich et al.╃ . . . . . . . . . . . . . . . . . . 371

4.5

Comparative Evaluation of Experimental Data on α-Amylase Inhibition by Flavonoids Using Molecular Modelling Lisa M. Bode, Thomas Homann, Harshadrai M. Rawel et al.╃ . . . . . . . . . 376

4.6

Potential Risk of Furan in Foods J. Brück, Dieter Schrenk, U. Schauer et al.╃ . . . . . . . . . . . . . . . . . . . . . . . . 378

4.7

Comparative Study on the Toxicity of Alternariol and Alternariol Monomethyl Ether in Human Tumour Cells of Different Origin Julia Burkart, Markus Fehr, Gudrun Pahlke et al.╃ . . . . . . . . . . . . . . . . . 379

4.8

A Role for Resveratrol and Curcumine in Sensitization of Glioblastoma Cells to Genotoxic Stress Induced by Alkylating Chemotherapeutics Markus Christmann, N. Berdelle, G. Nagel et al.╃ . . . . . . . . . . . . . . . . . . . 381

4.9

BfR Risk Assessment of Alkaloids as Ingredients and Contaminants of Food: Quinine, Opium Alkaloids, and Senecio Pyrrolizidine Alkaloids Birgit Dusemund, Klaus-Erich Appel, and Alfonso Lampen . . . . . . . . . . 382

4.10 Elucidation of the Genotoxic Activity of the Alkaloid Ellipticine in Human Cell Lines Eva Frei, Jitka Poljaková, Lucie Borˇek-Dohalská et al.╃ . . . . . . . . . . . . . . 391 4.11 Dietary Supplements and Herbal Medicinal Products – for a Clear Differentiation. Statement of the Â�Society for Phytotherapy (GPT) to the “Article€13 Health Claim List” of the EFSA Frauke Gaedcke, Bernd Eberwein, Olaf Kelber et al.╃ . . . . . . . . . . . . . . . . 393

7

8

Inhalt╃/╛Contents

4.12 Assessment of Genotoxicity of Herbal Medicinal Preparations According to the Guideline EMEA/HMPC/107079/2007 – A Model Project of Â�Kooperation Phytopharmaka, Bonn, Germany Frauke Gaedcke, Olaf Kelber, Karin Kraft et al.╃ . . . . . . . . . . . . . . . . . . . . 403 4.13 Implications for an Adverse Effect of Vitamin C in Photodynamic Therapy Stefanie Grimm, Nicolle Breusing, and Tilman Grune . . . . . . . . . . . . . . . 408 4.14 Using the Nematode Caenorhabditis elegans to Identify Mode of Action of the Flavonoid Â�Myricetin Gregor Grünz, B. Spanier, and Hannelore Daniel . . . . . . . . . . . . . . . . . . 409 4.15 Low-Temperature Plasma – Mild Preservation Â�Technology for Minimal Processed Fresh Food? Franziska Grzegorzewski, O.╃Schlüter, J.╃Ehlbeck et al.╃ . . . . . . . . . . . . . . 410 4.16 Influence of Fumonisin B1 on Gene Expression and Cytokine Production Dorothee C. Hecker, Christian Salzig, and Dieter Schrenk . . . . . . . . . . . . . 411 4.17 Effects of Quercetin on the Detoxification of the Food Contaminant Benzo[a]pyrene in the Human Intestinal Caco-2 Cell Model Stefanie Hessel, Andrea John, Albrecht Seidel et al.╃ . . . . . . . . . . . . . . . . . 412 4.18 Risk Assessment of T-2 and HT-2 Toxin Using Human Cells in Primary Culture Dennis Mulac, Maika Königs, Gerald Schwerdt et al.╃ . . . . . . . . . . . . . . . 420 4.19 Pyrrolizidine Alkaloids in Honey Bee Products Michael Kempf, Till Beuerle, Annika Reinhard et al.╃ . . . . . . . . . . . . . . . 421 4.20 Identification of Molecular Determinants for Cytotoxicity of Isoliquiritigenin from Liquorice (Glycyrrhiza glabra) towards Leukemia Cell Lines V. Badireenath Konkimalla, Anne Kramer, Yujie Fu et al.╃ . . . . . . . . . . . 429 4.21 Functional Effects of Polyphenol Metabolites Produced by Colonic Microbiota in Colon Cells In Vitro Claudia Miene and Michael Glei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430 4.22 Lifelong Exposure to Isoflavones Results in a Â�Reduced Responsivity of the Mammary Gland in Female Rats towards Oestradiol Almut Molzberger, Torsten Hertrampf, Frank Möller et al.╃ . . . . . . . . . . . 432 4.23 Derivation of Maximum Amounts for the Addition of Functional Ingredients to Foods Sina Tischer, Oliver Lindter, Almut Bauch et al.╃ . . . . . . . . . . . . . . . . . . . . 433 4.24 Constituents of Ginger Induce Micronuclei in Two Mammalian Cell Systems In Vitro Erika Pfeiffer, Julia S. Dempe, Marina J. Gary et al.╃ . . . . . . . . . . . . . . . . 434

Inhalt╃/╛Contents

4.25 Relative Photomutagenic Potency of Furocoumarins and Limettin Christiane Lohr, Dieter Schrenk, and Nicole Raquet . . . . . . . . . . . . . . . . . 435 4.26 Degradation of Green Tea Catechins Markus Schantz, Thomas Erk, and Elke Richling . . . . . . . . . . . . . . . . . . . 436 4.27 Evaluation of the Cytotoxic Effects of Herbal Homeopathic Extracts in Primary Human Â�Hepatocytes In Vitro Ulrike Sobeck, B. Rüdinger, F. Stintzing et al.╃ . . . . . . . . . . . . . . . . . . . . . . 437 4.28 Modulation of Antioxidant Gene Expression by Â�Apple Juice in Rats Bülent Soyalan, J. Minn, Hans-Joachim Schmitz et al.╃ . . . . . . . . . . . . . . 442 4.29 Predictivity Comparison between Screening Assays for Bacterial Mutagenicity for Natural Compounds: Micro-Ames vs. Ames Fluctuation Method Gerlinde Pappa, Tina Wöhrle, Anette Thiel et al.╃ . . . . . . . . . . . . . . . . . . . 449 4.30 Automated In Vitro Micronucleus Testing of Natural Compounds in Correlation with Hydrogen Peroxide Gerlinde Pappa, Tina Wöhrle, Anette Thiel et al.╃ . . . . . . . . . . . . . . . . . . . 450 4.31 Permeability of Apple Polyphenols in T84 Cell Model and their Influence on Tight Junctions Hannah Bergmann, Dorothee Rogoll, Wolfgang Scheppach et al.╃ . . . . . . 451 4.32 Influence of Apple Polyphenols on Inflammatory Gene Expression Sven Triebel, Ralph Melcher, Gerhard Erkel et al.╃ . . . . . . . . . . . . . . . . . . 452 4.33 Diethylstilbestrol-Like Effects of Genistein on Gene Expression of Wnt-Signalling Components in the Endometrial Ishikawa Cell Line Jörg Wagner and Leane Lehmann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453 4.34 Effect of Dietary Flavonoids in Different Cell Lines: Comparison of Uptake, Modulation of Oxidative Stress and Cytotoxic Effects Wim Wätjen, Sven Ruhl, Ricarda Rohrig et al.╃ . . . . . . . . . . . . . . . . . . . . 460 4.35 Risk–Benefit Considerations of Isoflavone Supplements in the Treatment of Menopausal Vasomotor Symptoms Uta Wegewitz, Klaus Richter, A. Jacobs et al.╃ . . . . . . . . . . . . . . . . . . . . . . 461 4.36 Effect of Different Catechins on the Growth of HT-29 Cells Stefanie Wiese, Tuba Esatbeyoglu, Peter Winterhalter et al.╃ . . . . . . . . . . . 463 4.37 Determination of the Isoflavone Content of Soy-Based Infant Formula of the German Market Using a Box-Behnken Experimental Design for Optimizing the Analytical Conditions Stefanie Witte, Hans-Peter Kruse, and Sabine E. Kulling . . . . . . . . . . . . . 465

9

10

Inhalt╃/╛Contents

5 Appendix . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . 467 Participants of the Symposium Risk Assessment of Phytochemicals in Food – Novel Approaches . . . . . . . . . . . . . . . . . å°“. . . . 467 Members of the DFG Senate Commission on Food Safety: Mandate 2007–2010 . . . . . . . . . . . . . . . . . å°“. . . . . . . . . . . . . . . . . . å°“. . . . . 475



Vorwort

Das Symposium „Risk Assessment of Phytochemicals in Food – Novel Approaches“ der Senatskommission zur gesundheitlichen Bewertung von Lebensmitteln (SKLM) der Deutschen Forschungsgemeinschaft (DFG) wurde vom 30.╃März bis 01.╃April 2009 in Kaiserslautern abgehalten. Die Senatskommission hat dabei mit international anerkannten Experten aus Akademie, Industrie und Behörden die Perspektiven innovativer “OMICS”-Techniken unter Einbezug verschiedener in-silico-, in-vitro- und in-vivo-Verfahren diskutiert. Diese neuen Techniken befinden sich in unterschiedlichen Stadien der Entwicklung und ihre Anwendbarkeit auf die Sicherheitsbewertung von Lebensmitteln ist derzeit noch offen. Die SKLM hat im Sinne ihres Beratungsauftrags für die DFG Schlussfolgerungen und Empfehlungen zum Forschungsbedarf erarbeitet, die gemeinsam mit den Einzelbeiträgen der Redner und den Posterbeiträgen in diesem Symposiums-Band veröffentlicht werden. Die SKLM dankt der DFG für die nachhaltige Unterstützung der Symposienreihe. Diese Symposien bieten ein exzellentes Forum zur Beratung aktueller Themen von besonderer wissenschaftlicher Bedeutung mit besonders ausgewiesenen Wissenschaftlern. Ich danke den Teilnehmern des Symposiums für ihre wissenschaftlichen Beiträge und den Mitgliedern und Gästen der Senatskommission für ihre Mithilfe bei der Abfassung der vorliegenden Veröffentlichung. Ebenso danke ich den Mitgliedern des Redaktionskomitees sowie den Vorsitzenden und Rapporteuren des Symposiums, Prof. Hengstler, Prof. Joost, Prof. Kulling, Prof. Rietjens, Prof. Schlatter, Prof. Steinberg und Prof. Marko für ihre Mitarbeit bei der Formulierung der Schlussfolgerungen und Empfehlungen. Das wissenschaftliche Sekretariat der SKLM mit Dr. Sabine Guth, Dr. Michael Habermeyer und Dr. Barbara Kochte-Clemens hat wesentlich zum Zustandekommen dieses Bandes beigetragen. Ihnen gilt mein herzlicher Dank. Besonders danke ich der Leiterin des Fachreferates Lebenswissenschaften I der DFG, Frau Dr. Heike Strelen, für ihre engagierte Unterstützung der Arbeit der SKLM. Die SKLM gibt der Hoffnung Ausdruck, dass dieser aktuelle Bericht mit Symposiumsbeiträgen, Schlussfolgerungen und Empfehlungen im forschungsund gesundheitspolitischen Raum Beachtung findet.

Prof. Dr. Gerhard Eisenbrand Vorsitzender der DFG-Senatskommission zur Bewertung der gesundheitlichen Unbedenklichkeit von Lebensmitteln

11

12



Preface

The Symposium “Risk Assessment of Phytochemicals in Food – Novel Approaches”, organized by the Senate Commission on Food Safety (SKLM) of the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG), was held from March 30th to April 1st 2009 in Kaiserslautern, Germany. The SKLM discussed with international experts from academia, industry and authorities the promise of innovative “OMICS” methodologies including various in silico, in vitro and in vivo approaches. These new methodologies are at different stages of development and their applicability for food safety assessment at present is still open. The SKLM, on behalf of the DFG, has prepared conclusions and recommendations for further research, published in this volume together with oral and poster contributions. The SKLM is grateful to the DFG for sustained support of the SKLM symposia series. These symposia represent an excellent forum to discuss with recognized scientific experts topics of particular importance. I would like to thank the participants for their scientific contributions, as well as the members and guests of the Senate Commission for their support in preparing this publication. I am also grateful to the members of the editorial committee, chairs and rapporteurs, Prof. Hengstler, Prof. Joost, Prof. Kulling, Prof. Rietjens, Prof. Schlatter, Prof. Steinberg and Prof. Marko, for their contributions to the conclusions and recommendations. Thanks are also due to Dr. Sabine Guth, Dr. Michael Habermeyer and Dr. Barbara Kochte-Clemens of the Scientific Office of the SKLM who substantially contributed to the preparation of this volume. I am indebted to Dr. Heike Strelen, Head of the DFG Life Sciences Division 1, for her sustained support of the Senate Commissions activities. The SKLM trusts that this report, encompassing contributions, conclusions and recommendations will find due attention in research and health policy.

Prof. Dr. Gerhard Eisenbrand Chair of the DFG Senate Commission on Food Safety

13

1

Bericht und Schlussfolgerungen

1.1 Einleitung Pflanzliche Lebensmittel sind ein wichtiger Bestandteil der Ernährung, einige Pflanzen jedoch enthalten Stoffe, die bestimmte gesundheitliche Risiken bergen. Solche „sekundären Pflanzenstoffe“ können unerwünschte Wirkungen über verschiedene Mechanismen auslösen. So wirken einige Stoffe auf das Hormonsystem, wie beispielsweise die in Soja enthaltenen Isoflavone. Andere Stoffe wirken lebertoxisch (Cumarin), neurotoxisch (Solanin), phototoxisch (Furocumarine) oder kanzerogen (Estragol). Bei einer normalen Aufnahme als natürliche Bestandteile von Obst, Gemüse, Kräutern und Gewürzen geht man von einem geringen Risiko aus. Potenziell problematisch jedoch ist eine erhöhte Exposition, z.╃B. bei einseitiger Ernährung oder bei Einnahme über Nahrungsergänzungsmittel in isolierter und konzentrierter Form. Für eine Risikoabschätzung sind Stoffe zu identifizieren, die aufgrund ihrer chemischen Struktur potenziell gesundheitsschädlich sind und deren Wirkungen im Organismus unter Bezug auf die Dosis zu klären sind. Hierfür werden zunehmend neuartige Profilierungstechniken und rechnergestützte Methoden eingesetzt – mit vielversprechenden Möglichkeiten. Die Senatskommission zur gesundheitlichen Bewertung von Lebensmitteln (SKLM) der Deutschen Forschungsgemeinschaft (DFG) hat ein Symposium zum Thema „Risk Assessment of Phytochemicals in Food – Novel Approaches“ organisiert, das vom 30.╃März bis 1.╃April 2009 in Kaiserslautern, Deutschland, stattfand. Potenziale, Auswirkungen und Perspektiven neuartiger Methoden für die Risikoabschätzung unter Einschluss von in-silico-, in-vitro- und in-vivoAnsätzen wurden diskutiert und der Stand der Technik unter Berücksichtigung spezifischer Beispiele aus dem Bereich der sekundären Pflanzenstoffe ermittelt. Ziel des Symposiums war, die Bedeutung neuer Methoden für die Risikoabschätzung dieser Stoffe herauszuarbeiten. Die SKLM hat hieraus Schlussfolgerungen und Empfehlungen abgeleitet und Wissenslücken sowie Forschungsbedarf identifiziert. Dieser Bericht basiert auf den Präsentationen von Marcel Leist (DE), Gerrit Speijers (NL), Luis Valerio (US), Benoît Schilter (CH), Ivonne Rietjens (NL), Pablo Steinberg (DE), Hans-Jürgen Ahr (DE), Anette Thiel (CH), Hennicke Kamp (DE), Hannelore Daniel (DE), Augustin Scalbert (FR), Theo de Kok (NL), Takeki Uehara (JP), Alfonso Lampen (DE), Dieter Schrenk (DE), Jaap Keijer (NL), Ulrike Peters (US), Hansruedi Glatt (DE), Johanna Fink-Gremmels (NL) sowie den Diskussionen in den anschließenden Workshops.

14

Bericht und Schlussfolgerungen

1.2 Methodenübergreifende Aspekte Nahrung und Lebensmittelinhaltsstoffe gelten, ganz gleich ob natürlichen oder synthetischen Ursprungs, als expositions-/ ernährungsbedingte Faktoren, die die menschliche Gesundheit über ihre Einwirkung auf das Genom, das Transkriptom, das Proteom und das Metabolom beeinflussen. Da solche Wechselwirkungen das ganze Leben hindurch bestehen, ist auch der physiologische und Entwicklungszustand eines Individuums zum Zeitpunkt der Exposition von Bedeutung. Neue Profilierungstechniken (profiling techniques) bieten die Möglichkeit, Wechselwirkungen zwischen Lebensmittelbestandteilen und Rezeptoren, Signalwegen und anderen Teilen des Genom-/ Transkriptom-/ Proteom-/ Metabolom-Systems zu identifizieren, die das physiologische Gleichgewicht sicherstellen. Die Risikobewertung von Lebensmitteln und ihren Inhaltsstoffen hat eine große Zahl natürlicher oder synthetischer Stoffe, die die menschliche Gesundheit beeinflussen können, in Betracht zu ziehen. Da für die meisten dieser Substanzen keine ausreichenden toxikologischen Daten zur Verfügung stehen, erhofft man sich von den neuen molekularen Techniken schnelle, effiziente und verlässliche Informationen, die es ermöglichen, sichere Expositionsgrenzwerte zu etablieren oder für (pflanzliche) Stoffe eine adäquate Risikoabschätzung auf der Basis angemessener Sicherheitsprüfungen durchzuführen. Eine Vielfalt neuer Methoden befindet sich derzeit in unterschiedlichen Stadien der Entwicklung und Anwendung. Die SKLM hält deren Einsatz zum Erreichen der folgenden Ziele für wesentlich: ►⌺ Aufklärung von Wirkmechanismen ►⌺ Entwicklung neuer Techniken, die eine umfassende Auswertung großer Datensätze erlauben ►⌺ Vergleich experimenteller Daten aus Tierversuchen oder in-vitro-Modellsystemen mit Daten aus Humanstudien ►⌺ Wissenschaftsgeleitete Entwicklung von Lebensmitteln und Lebensmittelinhaltsstoffen mit definierter Biofunktionalität ►⌺ Berücksichtigung individueller Variabilität und Prädisposition. Solche neuen Methoden könnten in klinischen und tierexperimentellen Studien dazu dienen, Daten höherer Aussagequalität zu generieren bzw. ein besseres Verständnis von scheinbar inkonsistenten Daten zu gewinnen. Durch die Einführung molekularer Marker können in künftigen epidemiologischen Studien biomolekulare Angriffspunkte und Effekte sekundärer Pflanzenstoffe eingehender untersucht werden. Hierdurch können Wechselwirkungen sekundärer Pflanzenstoffe mit biomolekularen Prozessen, aber auch die Rolle der genetischen Variabilität humaner Populationen bei der Auslösung heterogener Reaktionen durch sekundäre Pflanzenstoffe besser verstanden werden.

Methoden

1.3 Methoden 1.3.1 Rechnergestützte Toxikologie / in-silico-Modelle In-silico-Technologien verwenden rechnergestützte Informationen und Methoden, um das toxikologische bzw. Wirk-Profil einer Substanz zu prognostizieren. Sie ermöglichen die Überprüfung einer großen Zahl an Substanzen innerhalb einer relativ kurzen Zeit und sind sehr kosteneffizient verglichen mit konventionellen toxikologischen Tierstudien. Sie können darüber hinaus experimentell vielseitig eingesetzt werden, nicht nur für das Screening, sondern auch, um mittels adäquater Lernalgorithmen ihre Vorhersagekraft zu verfeinern. Dies gilt beispielsweise für Ansätze auf der Basis quantitativer Struktur-Aktivitäts-Beziehungen (quantitative structure–activity relationship, QSAR) oder des physiologiebasierten Biokinetik-Modellierens (physiologically-based biokinetic modelling, PBBK). Ein weiterer Vorteil von in-silico-Methoden zur Voraussage von Toxizität liegt im potenziellen Einsparungseffekt bei der Zahl der Versuchstiere, so dass Tierversuche eingeschränkt bzw. sogar ersetzt und damit ein Beitrag zum 3R-Prinzip (refinement, reduction, replacement) geleistet werden kann. Darüber hinaus erscheint der Einsatz von in silico rechnergestützten Toxikologie-Methoden (computational toxicology) für einzelne pflanzliche Substanzen zur Priorisierung der Risikoabschätzung auf chemischer Basis sehr vielversprechend.

1.3.1.1 Quantitative Struktur-Aktivitäts-Beziehungen (QSAR)

Definition: QSAR-basierte Methoden korrelieren quantitativ Parameter, die strukturelle chemische Eigenschaften von Molekülreihen beschreiben, mit deren biologischer Aktivität oder chemischer Reaktivität. Rechnergestützte prädiktive Modellierung auf der Basis von QSAR verwendet statistische Verfahren zur Korrelation von biologischer Aktivität von Molekülen mit Deskriptoren, die für eine Molekülstruktur repräsentativ sind. Stand der Technik: Rechnergestützte prädiktive Modellierung auf der Basis von QSAR liefert ein evidenzbasiertes Werkzeug zur Priorisierung und zur effizienten Gefährdungseinschätzung auf der Grundlage bereits vorhandener Testdaten zu unterschiedlichen Endpunkten. Solche Endpunkte umfassen z.╃B. das mutagene Potenzial von synthetischen und natürlichen Molekülen sowie pflanzlichen Stoffe, die in Pflanzenextrakten, Kräutern und natürlichen Nahrungsquellen zu finden sind. Prädiktive QSAR-Methodiken sind vielversprechende Entscheidungshilfen in der Sicherheits- und Risikobewertung. In dringenden Fällen könnten sie sich für eine schnelle Entscheidungsfindung als wertvoll erweisen und auch die Prioritätensetzung für zusätzliche Toxizitätstests unterstützen.

15

16

Bericht und Schlussfolgerungen

Limitierungen, Wissenslücken und Forschungsbedarf: Die Stoffevaluierung auf der Basis von QSAR erlaubt keine absolute Sicherheit der Aussage in Bezug auf die gesundheitliche Unbedenklichkeit des betreffenden Stoffes. Auch weisen die Programme nicht alle notwendigen Funktionen auf und bieten keine völlig umfassende Information zu den Stoffeigenschaften oder eine 100â•›%ige Spezifität. Die meisten QSAR-basierten prädiktiven Modelle der Stoffevaluierung sind außerdem begrenzt durch ihre inhärent geringe Empfindlichkeit in Bezug auf die Fähigkeit, echte Positive (true positives) richtig zu identifizieren. Exposition oder Spezifität des Angriffspunktes von Stoffen werden nicht gewichtet und mechanistische Gesichtspunkte, unter Einbezug klassenspezifischer Mechanismen, wie z.╃B. bei der Kanzerogenese die vielfachen Einzelschritte der Krebsstehung, Tumorpromotion oder Tumorart, spielen keine Rolle. Eine Kombination von Softwareprogrammen wäre die beste Möglichkeit, die Sensitivität und Spezifität der Vorhersage zu verbessern, da Datenbasen, Algorithmen und Methoden zur strukturellen Interpretation von Programm zu Programm variieren. Derzeit können Prognosen mit relativ guter Empfindlichkeit und Spezifität für bestimmte Endpunkte, wie etwa für die Mutagenität erstellt werden. Letztere hängt von einer begrenzten Zahl an Mechanismen ab, wie etwa der Fähigkeit von Stoffen, kovalent an die DNA zu binden. Dies gilt auch für die akute Toxizität in aquatischen Spezies, insbesondere dann, wenn ein „narkotischer“ Mechanismustypus verantwortlich ist. Sehr viel schwieriger ist es hingegen, komplexere Endpunkte wie Organtoxizität oder Kanzerogenität vorherzusagen, die auf einer Vielzahl möglicher Mechanismen beruhen. Derzeit sind Datenbanken limitiert und Daten über eine ausreichende Zahl an Substanzen nicht vorhanden. Diese Einschränkungen beruhen aber vor allem auf fehlenden rechnerischen Möglichkeiten, weniger auf den Einschränkungen von Datenbanken. Zum jetzigen Zeitpunkt kann man mit Hilfe rechnergestützter QSAR-basierter Toxikologie-Methoden einzelne chemische Substanzen analysieren. Komplexe Gemische, wie etwa pflanzliche Extrakte, können jedoch noch nicht auf diese Weise analysiert werden. Gleichwohl könnten sich in Zukunft rechnergestützte Techniken besonders bei der Bewertung pflanzlicher Gemische und der Voraussage additiver oder synergistischer Effekte als gewinnbringend erweisen.

1.3.1.2 Physiologie-basierte biokinetische Modelle (PBBK)

Definition: Das Physiologie-basierte biokinetische Modell (PBBK) umfasst einen Satz mathematischer Gleichungen, die auf der Grundlage dreier Parametertypen gemeinsam die Charakteristik von Absorption, Verteilung, Metabolismus und Exkretion (ADME; absorption, distribution, metabolism and excretion) eines Stoffes innerhalb eines Organismus beschreiben. Diese Parameter umfassen Physiologie (z.╃B. Herzleistung, Gewebevolumen und Gewebedurchblutung), Physiko-Chemie (z.╃B. Blut/Gewebe-Verteilungskoeffizienten) und Kinetik (z.╃B. kinetische Konstanten für metabolische Reaktionen).

Methoden

Stand der Technik: Ein generelles Problem bei der Risikoabschätzung ist die Notwendigkeit, tierÂ� experimentelle Daten bei hohen Dosen auf die humane Niedrigdosis-Situation zu extrapolieren. Solche Extrapolationen werden erschwert durch Unsicherheiten bezüglich der Dosis-Wirkungs-Kurve im Dosisbereich, wie er für die menschliche Ernährung relevant ist, sowie durch speziesspezifische Unterschiede im Metabolismus. PBBK-Gleichungen können z.╃B. den zeitlichen Verlauf der Gewebskonzentration eines Stoffes oder seiner Metaboliten in jedem Gewebe bei jeder Dosierung vorausberechnen und damit die Analyse von Effekten im niedrigen Dosisbereich, wie er für die humane in-vivo-Situation realistisch ist, ermöglichen. Modellvorhersagen können genutzt werden, um eine stärker mechanismusgetriebene Grundlage zur Einschätzung der Wirkungen in Tieren und Menschen bei niedrigen, nahrungsbezogenen Aufnahmemengen zu liefern, selbst wenn diese nur auf in-vitro-Daten beruhen. Darüber hinaus können PBBKModelle für verschiedene Spezies entwickelt werden, was die Extrapolation von Spezies zu Spezies erleichtert. Auch ist es möglich, durch Einbezug von Gleichungen und kinetischen Konstanten für metabolische Umsetzungen, die aus Proben einzelner humaner Individuen und/ oder spezifischer Isoenzyme gewonnen werden, eine Modellierung interindividueller Variationen und genetischer Â�Polymorphismen durchzuführen. Internationale Standardverfahren und Datenbanken, die eine Standardisierung und Transparenz der Erstellung von PBBK-Modellen gewährleisten, sind bereits verfügbar. Limitierungen, Wissenslücken und Forschungsbedarf: Einen besonderen Schwerpunkt in der Risikobewertung beansprucht die Beurteilung von Langzeiteffekten, wie z.╃B. Kanzerogenität. Tierexperimentelle Untersuchungen hierzu werden in der Regel mit einer einzigen definierten Substanz durchgeführt. Der Mensch ist aber gegenüber sekundären Pflanzenstoffen über die Nahrung exponiert, d.╃h. ein bestimmter Stoff wird innerhalb eines komplexen Gemisches mit anderen Inhaltsstoffen aufgenommen. In der Lebensmittelmatrix können verschiedene Interaktionen stattfinden, die die Bioverfügbarkeit bestimmter Lebensmittelbestandteile beeinflussen. Zudem können auf der Ebene der metabolischen Aktivierung und/ oder Entgiftung Wechselwirkungen mit anderen pflanzlichen Inhaltsstoffen stattfinden. Grundsätzlich sind PBBK-Modelle in der Lage, solche modulierenden Effekte pflanzÂ� licher Inhaltsstoffe in komplexen Gemischen mit einzubeziehen. Gegenwärtige Modelle berücksichtigen vorausberechnete Daten zu dosisabhängigen Effekten, Speziesunterschieden und interindividuellen Unterschieden in der Bioaktivierung. Für eine adäquate Abschätzung des Krebsrisikos bei Expositionen, die für den Menschen relevant sind, sind aber nach wie vor noch Zusatzinformationen erforderlich. Beispielsweise beeinflusst auch die Toxikodynamik die Risikobewertung. Dies lässt sich durch Erweiterung der PBBK-Modelle in so genannte Physiologie-basierte biodynamische Modelle (PBBD-Modelle, physiologically-based biodynamic models) untersuchen. Hierbei werden

17

18

Bericht und Schlussfolgerungen

Dosishöhe und Biotransformation verknüpft mit Biomarkern der Exposition oder der Toxizität bzw. letztlich der Krebsrate. Eine der stärksten Einschränkungen des PBBK-Modellierens besteht derzeit darin, dass ein Modell für jeden einzelnen Stoff aufwendig etabliert werden muss. Diese Validierung ist aber für eine ganze Reihe von Stoffen notwendig. Es lässt sich trotzdem absehen, dass die Bedeutung des PBBK-Modellierens mit sich allmählich erhöhender Verfügbarkeit von Modellen guter Qualität zunehmen wird. Insbesondere in der Kombination mit in-vitro-Tests kann die PBBK-Modellierung bei der Extrapolation von in-vitro-Konzentrations-Wirkungs-Kurven hin zu in-vivo-Dosis-Wirkungs-Kurven sehr hilfreich sein. Dies lässt sich dadurch bewerkstelligen, dass Konzentrations-Wirkungs-Kurven aus einem geeigneten in-vitro-Toxizitätstest als interne Konzentrationen im Modell verwendet werden. Mit Hilfe des PBBK-Modells lassen sich dann die in-vivoDosishöhen errechnen, die zum Erreichen der internen (toxischen) Konzentrationen erforderlich sind. Die so vorausberechneten Dosis-Wirkungs-Kurven können bei der Sicherheitsbewertung von Stoffen für die Bestimmung sicherer Expositionswerte verwendet werden. Auf diese Weise können PBBK-Methoden helfen, externe Dosen mit internen Konzentrationen (Zielgewebedosis) zu verknüpfen und diese internen Konzentrationen mit EC50-Werten (effective concentration 50â•›%) aus in-vitro-Studien zu verbinden und umgekehrt.

1.3.2 Techniken zur Profilerstellung: „OMICS“, in-vitro- und in-vivo-Modelle Der Ausdruck „OMICS“ bezieht sich auf eine Reihe von Disziplinen, die biologische Informationen zu Wechselwirkungen von Stoffen mit dem Genom, Proteom und dem Metabolom analysieren. Diese umfassen eine Vielfalt von Unterdisziplinen, die verschiedene Gruppen von Techniken, Reagenzien und Software verwenden, wie z.╃B. die DNA- und Protein-Microarray-Analyse, Gasund Flüssigchromatographie, Massenspektrometrie und eine Reihe anderer Methoden, die Analysen mit hohem Durchsatz ermöglichen. Eine der Stärken der „OMICS“ liegt darin, dass sie gleichermaßen gut sowohl für in-vitro- als auch für in-vivo-Studien angewendet werden können. Die Genomik umfasst sowohl die „Strukturelle Genomik“, DNA-Sequenzanalyse, Genomkartierung eines Organismus, als auch die „Funktionelle Genomik“, die Charakterisierung der Genwirkung, mRNA-Analyse und Proteinexpressionsprofilierung. Eine spezifischere Disziplin, die Proteomik, befasst sich mit der Untersuchung des vollständigen Proteinsatzes, der in einer Zelle, einem Gewebe, einer Körperflüssigkeit oder einem Organismus exprimiert wird. Im Unterschied zum Genom variiert das Proteom von Zelltyp zu Zelltyp. Die Proteomik versucht, Proteinprofile einzelner Zelltypen zu identifizieren und die Unterschiede im Proteinexpressionsmuster zwischen gesunden und kranken Zellen zu bewerten. Jedoch hat das Symposium den Schwerpunkt auf die Â�Transkriptomik und Metabolomik gelegt und die Proteomik nicht explizit behandelt.

Methoden

1.3.2.1 Transkriptomik

Definition: Die Transkriptomik umfasst die globale Analyse der Genexpression. Man nennt sie auch die „Genomweite Expressionsprofilerstellung“. Sie erfasst den relaÂ�tiven Gehalt an Boten RNA (mRNA), um Muster der Genexpression und der GenÂ� expressionshöhe zu bestimmen, und um die Genregulation zu analysieren. Stand der Technik: Zwar ist die m-RNA nicht das endgültige Produkt der Genexpression, die Transkription stellt aber den ersten Schritt in der Genregulation dar. Informationen bezüglich der Transkriptionshöhe werden benötigt, um Genregulationsnetzwerke besser zu verstehen und Ähnlichkeiten in Genexpressionsmustern aufzudecken, die funktional verbunden sind und dem gleichen genetischen Kontrollmechanismus unterliegen können. Die Analyse des gesamten Genoms/ Transkriptoms stellt dabei einen unvoreingenommenen Ansatz (unbiased approach) dar, der die Identifizierung aller potenziellen (erwarteten/ unerwarteten) Effekte erlaubt. RNA-Expressionssignaturen, die eine Unterscheidung zwischen verschiedenen Stoffklassen, wie etwa genotoxischen oder nicht genotoxischen Leberkarzinogenen erlauben, sind bereits identifiziert. Für eine Risiko-Nutzen-Analyse von sekundären Pflanzenstoffen können auch spezifische genetische Polymorphismen einbezogen werden. Darüber hinaus wird die Modulation der Expression von Genen, die an biologischen und genetischen Signalwegen beteiligt sind, welche für die Krebsentstehung Bedeutung haben, als eine wichtige Komponente des antikanzerogenen Effekts von Gemüse oder sekundären Pflanzenstoffen angesehen. Die Transkriptomik kann bei der Untersuchung solcher Effekte sehr hilfreich sein. Limitierungen, Wissenslücken und Forschungsbedarf: Die Modulation der Transkriptionshöhe ist nicht unbedingt mit entsprechenden Änderungen auf Proteinebene verbunden. Zumindest für spezifische Gene sollten Daten zur RNA-Expression durch die Bestimmung der entsprechenden Proteinkonzentrationen oder -aktivitäten ergänzt werden. Je nach dem Gewebskontext kann auch alternatives Spleißen die Funktionalität beeinflussen. Das Erstellen von Transkriptionsprofilen führt zu extrem großen Datensätzen und benötigt effektive Datenbankressourcen für Interpretation, Management und Analyse. Ob genomweite Expressionsdaten für Routineanwendungen erforderlich sind, muss noch untersucht werden. Möglicherweise benötigt man die Genomik nur für die Identifikation von Gengruppen, die die Etablierung eines Klassifikationsalgorithmus erlauben. Falls die Zahl an Genen, die für die Klassifikation benötigt werden, nicht zu hoch ist, könnten diese mittels quantitativer Techniken analysiert werden, wie z.╃B. qRT-PCR (quantitative real-time polymerase chain reaction).

19

20

Bericht und Schlussfolgerungen

Die Interpretation der Genexpressionsdaten sollte durch validierte Biomarker unterstützt werden. Zum besseren Verständnis der komplexen Information, die man normalerweise erhält, sollten integrierte systembiologische Ansätze entwickelt werden.

1.3.2.2 Metabolomik

Definition: Die Metabolomik hat zum Ziel, das Profil aller Metaboliten in einem einfachen System, z.╃B. einer Zelle, oder im komplexen System, z.╃B. Organ oder Organismus, zu untersuchen. Stand der Technik: Die Metabolomik verwendet hoch entwickelte instrumentell-analytische Techniken, wie Kernmagnetresonanzspektroskopie (NMR) und Massenspektrometrie (MS), meistens in Kopplung mit verschiedenen Trenntechniken wie Gaschromatographie (GC) oder Flüssigchromatographie (LC). Mit Hilfe dieses Instrumentariums kann das endogene und das mit einer Exposition verbundene Metabolom eines Organismus charakterisiert und die Bestandteile der verschiedenen Stoffwechselwege identifiziert werden. Eine Stärke der Metabolomik liegt darin, dass sie das Potenzial bestimmter Nährstoffe und von Xenobiotika erfasst, die Stoffwechseleigenschaften eines Organismus zu verändern. Dabei kann es sich beispielsweise um metabolische Leberfunktionen, Insulinempfindlichkeit von Geweben oder die Sekretion von Schilddrüsenhormonen handeln. Ein weiterer Vorteil liegt in der Möglichkeit der Analyse von Stoffwechselspektren in menschlichem Blut und Urin. So ist es möglich, mittels Metabolomik den Einfluss von Nahrungsbestandteilen im Menschen zu untersuchen. Limitierungen, Wissenslücken und Forschungsbedarf: Die Metabolomik deckt den Anteil am gesamten Metabolom ab, der mittels instrumenteller Analyse zugänglich ist. Allerdings bestehen trotz großer Fortschritte im Verständnis metabolischer Profile oft Unklarheiten bezüglich der Interpretation der komplexen Daten. Da viele Inhaltsstoffe über mehrere Mechanismen agieren, kann die Situation komplex werden. Darüber hinaus sind Effekte von Gemischen auf das Metabolom bisher nicht untersucht. Es besteht Bedarf, die quantitative Erfassung metabolischer Daten zu standardisieren. Zusätzlich sollte eine standardisierte Plattform für den Datenaustausch etabliert werden. Die statistische Auswertung muss soweit verbessert werden, dass induzierte Effekte klar von der physiologischen Streubreite im Metabolom unterschieden werden können. Für die Voraussage von Wirkmechanismen müssen generell einsetzbare Algorithmen etabliert und validiert werden.

Methoden

1.3.2.3 Toxikogenomik und Nutrigenomik

Definition: Toxikogenomik und Nutrigenomik sind Sammelbegriffe, welche die drei Unterdisziplinen Transkriptomik, Proteomik und Metabolomik mit einschließen. Diese Techniken zur Profilerstellung können aus ernährungsbezogenem oder aus toxikologischem Blickwinkel heraus betrachtet werden. Der Ausdruck Toxikogenomik wird verwendet, wenn die Toxikologie mit diesen neuen Methoden kombiniert wird, um die toxischen Wirkungen spezifischer Stoffe besser zu verstehen und zu erfassen, während die Nutrigenomik den Einfluss bestimmter Nährstoffe oder Ernährungsweisen auf die Genexpression untersucht. Dies sollte nicht mit dem Ausdruck „Nutrigenetik“ verwechselt werden, bei der untersucht wird, wie die genetische Variabilität die Reaktion des Körpers auf Nahrungsbestandteile bzw. auf die Ernährung beeinflusst. Stand der Technik: Die Toxikogenomik birgt großes Potenzial, spezifische toxische Effekte von Wirkstoff-Kandidaten und Stoffen vorherzusagen. Ziele der Toxikogenomik sind, Biomarkergene zu identifizieren und Zusammenhänge zwischen Änderungen in Genexpressionsmustern und bestimmten toxikologischen Endpunkten festzulegen. Man geht davon aus, dass eine Kurzzeitexposition Änderungen in den RNA-, Protein- und Metaboliten-Expressionsmustern verursacht, welche sich auch als hilfreich bei der Vorhersage von Langzeiteffekten erweisen könnten. Darüber hinaus konnte bereits überzeugend gezeigt werden, dass diese Techniken starke Hinweise auf toxische Mechanismen liefern können, die durch die zu prüfenden Verbindungen ausgelöst werden. Die Nutrigenomik birgt das Potenzial, innerhalb wohl definierter Versuchsbedingungen Hunderte von Messgrößen zu charakterisieren, die auf einen bestimmten Nährstoff oder Nichtnährstoff, auf eine Behandlung oder Ernährungsweise reagieren. Einschränkungen, Wissenslücken und Forschungsbedarf: Toxikogenomische Untersuchungen und daraus abgeleitete KlassifikationsÂ�Algorithmen beschränken sich bisher nur auf eine sehr begrenzte Zahl an Stoffen. Dosis-Wirkungs-Beziehungen sind bisher nur in sehr wenigen Fällen etabliert worden. Dies könnte sich als problematisch erweisen, da die gleiche Verbindung bei hoher bzw. niedriger Konzentration jeweils über unterschiedliche Mechanismen wirken kann. Für die Zukunft ist es wichtig, „OMICS“-Daten einer großen Zahl von Stoffen, die über gut definierte Mechanismen wirken, zu etablieren. Nur wenn Daten zu einer ausreichend großen Zahl von Stoffen derselben Stoffklasse verfügbar werden, können koordinierte Änderungen in den Transkripten, Proteinen oder Metaboliten identifiziert und mit gemeinsamen molekularen (toxischen) Wegen in Verbindung gebracht werden. Solche Schlüsselwege und deren Zusammenhänge mit biochemischen Daten müssen noch identifiziert werden. Ebenso müssen Änderungen im Expressionsmuster von Genen, Proteinen und Metaboliten sowohl mit nachteiligen, als auch posi-

21

22

Bericht und Schlussfolgerungen

tiven Effekten auf die menschliche Gesundheit verknüpft werden, um die bislang noch limitierte Aussagekraft für die Risikobewertung zu verbessern. Von wesentlicher Bedeutung ist der Aufbau einer öffentlich zugänglichen, umfassenden, gut strukturierten und harmonisierten Toxikogenomik-Datenbank mit Informationen zu Expressions- und toxikologischen Parametern. Für adäquate Ernährungsstudien (Nutrigenomik-Studien) bedarf es gut geplanter experimenteller Untersuchungen an Freiwilligen unter standardisierten Bedingungen in Bezug auf Ansatz, Ernährung und Profilerstellung. Die gegenwärtige Problematik gleicht derjenigen der Toxikogenomik und besteht vor allem darin, die durch die Nutrigenomik generierten großen Datensätze angemessen zu interpretieren. Koordinierte Änderungen bei Transkripten, Proteinen und Metaboliten können nur in wenigen Fällen und auf bekannte Regulationswege projiziert werden. Eine weitere Einschränkung rührt daher, dass sich die Expressionsprofile verschiedener Gewebe unterscheiden und oft nicht klar ist, welches Gewebe relevant ist. Während in Tiermodellen die Gewebe recht umfassend analysiert werden können, ist dies in Humanstudien nicht möglich.

1.4 Schlussfolgerungen und Empfehlungen In-silico- und in-vitro-Techniken zur Identifizierung von Gefährdungen können ebenso wie in-vitro-Techniken zur Prüfung bestimmter mechanistischer Endpunkte die Sicherheitsbewertung von pflanzlichen Stoffen mittels gezielten Studiendesigns verbessern. Darüber hinaus lässt sich mittels solcher in-silico- und in-vitro-Untersuchungen die Anzahl der für Wirksamkeits- und Sicherheitsstudien erforderlichen Tiere reduzieren. Hierzu können auch neue in-vivo-Techniken beitragen und ebenso die Versuchsdauer verkürzen, wenn auf der Basis von Ergebnissen aus „Kurzzeit“-Studien „Langzeit“-Effekte wie Kanzerogenität vorhergesagt werden können. Die SKLM vertritt die Ansicht, dass die neuen Methoden bislang noch nicht aussagekräftig genug sind, um die derzeitigen Verfahren der Stoffprüfung und Risikobewertung zu ersetzen. Sie werden aber als sehr nützlich angesehen, um umfassendere Daten zu erhalten und somit die Auswertung von klinischen und tierexperimentellen Studien zu verbessern bzw. inkonsistente Daten besser zu verstehen. Darüber hinaus können sie mechanistische Einblicke ermöglichen, die in die Risikobewertung einbezogen werden können. Gleichwohl ist eine stringente Validierung unter Einbezug des Vergleichs mit klassischen Verfahren erforderlich, um die Verwendbarkeit für einen bestimmten Zweck abzuÂ� sichern. Um größtmöglichen Erkenntnisgewinn zu erhalten, wird eine neue Teststrategie empfohlen, bei der verschiedene Methoden in einem kombinierten Ansatz integriert werden. Um Standardisierung, Vergleichbarkeit, Reproduzierbarkeit sowie die Zusammenarbeit zwischen verschiedenen Laboratorien zu gewährleisten, ist Planung, Design und Aufbau einer öffentlich zugänglichen, umfassenden, gut strukturierten und harmonisierten Datenbank notwendig. Dies beschleunigt den Validierungsprozess und stellt sicher, dass die neuen Â�Methoden

Schlussfolgerungen und Empfehlungen

angemessen angewendet werden. Bioinformatikexperten sind unerlässlich, um diese Ziele zu erreichen. Bei der Durchführung von Humanstudien muss der genetische Hintergrund der analysierten Population berücksichtigt werden. Der Aufbau einer umfassenden Datenbank zur physiologischen Hintergrundsituation von Individuen sowie der Allgemeinbevölkerung ist eine Voraussetzung für die Untersuchung des Einflusses von Stoffen auf humane Stoffwechselmuster. Die genetische Variabilität kann so besser berücksichtigt werden, und man kann möglicherweise Schlüsselfaktoren identifizieren, die für die Variabilität in der zu untersuchenden Population verantwortlich sind. Menschen sind normalerweise gegenüber pflanzlichen Stoffen in komplexen Stoffgemischen exponiert. Zum gegenwärtigen Zeitpunkt sind die meisten modernen Techniken aber nicht in der Lage, solche Gemische zu handhaben. Bis jetzt wurden „OMICS“-Techniken fast ausschließlich zur Analyse einzelner Substanzen und nicht von Gemischen angewendet. Das ist sinnvoll, da zunächst die bereits sehr komplexen Effekte einzelner Bestandteile identifiziert und verstanden werden müssen. Dennoch ist ein Vorteil der auf Mustern beruhenden „OMICS“-Technologien, dass es irrelevant ist, ob die Änderungen in komplexen Expressionsmustern durch Einzelsubstanzen oder Gemische verursacht werden. Daher wird mit fortschreitender Entzifferung der komplexen Sprache der „OMICS“-Muster für Einzelstoffe diese Methodik auch für Gemische anwendbar werden, was rasche Fortschritte auf dem Forschungsgebiet insgesamt mit sich bringen wird. Schließlich liegt ein einzigartiger Vorzug der „OMICS“-Techniken mit ihrem Zufalls- oder gar genomweiten Ansatz darin, auch solche Effekte zu entdecken, die gar nicht gesucht wurden. Daher ist es wahrscheinlich, dass sie bei zusätzlicher Anwendung gemeinsam mit konvenÂ� tionellen Toxizitätstests die Risikobewertung verbessern werden.

1.4.1 Spezifischer Forschungsbedarf in Schlagworten QSAR:

Verstärkte Erforschung der Möglichkeit einer Kombination von Softwareprogrammen zur Verbesserung von Sensitivität und Spezifität der Vorhersage ►⌺ Weiterentwicklung der Software, um Informationen aus Datenbanken intelligenter nutzbar zu machen ►⌺ Entwicklung von rechnergestützten Verfahren zur Bewertung von StoffÂ� gemischen. ►⌺

PBBK: ►⌺

Erweiterung der PBBK-Modelle in so genannte Physiologie-basierte biodynamische Modelle (PBBD-Modelle, physiologically based biodynamic models)

23

24

Bericht und Schlussfolgerungen

Entwicklung weniger aufwendiger Methoden, um eine große Zahl von Stoffen modellieren zu können ►⌺ Kombination von in-vitro-Toxizitätstests mit in-vivo- Dosis-Wirkungskurven. ►⌺

Transkriptomik:

Ergänzung von Daten zur RNA-Expression durch Bestimmung der entsprechenden Proteinkonzentrationen oder -aktivitäten, auch unter Berücksichtigung des alternativen Spleißens ►⌺ Entwicklung effektiver Datenbankressourcen zur Unterstützung von Interpretation, Management und Auswertung für die Erstellung von TranskripÂ� tionsprofilen; Klärung, ob genomweite Expressionsdaten für Routineanwendungen erforderlich sind ►⌺ Etablierung eines Klassifikationsalgorithmus ►⌺ Entwicklung von validierten Biomarkern zur Interpretation der Genexpressionsdaten ►⌺ Entwicklung von integrierten systembiologischen Ansätzen zum besseren Verständnis komplexer Daten. ►⌺

Metabolomik:

Standardisierung der Bestimmung metabolischer Daten Etablierung einer Plattform für den Datenaustausch ►⌺ Untersuchung der Effekte von Gemischen auf das Metabolom ►⌺ Verbesserung der statistischen Auswertung, so dass induzierte Effekte klar vom normalen Streubereich unterschieden werden können ►⌺ Etablierung generell einsetzbarer Algorithmen für eine Prognose von Wirkmechanismen. ►⌺ ►⌺

Toxikogenomik and Nutrigenomik:

Gewinnung und Nutzung von Daten zu toxikogenomischen Untersuchungen, abgeleiteten Klassifikations-Algorithmen und Dosis-Wirkungs-Beziehungen von einer großen Zahl von Stoffen zur Identifizierung toxikologischer Schlüsselwege und deren Zusammenhänge mit biochemischen Daten ►⌺ Beziehung zwischen Änderungen im Expressionsmuster von Genen, Proteinen und Metaboliten und nachteiligen oder positiven Effekten auf die menschliche Gesundheit ►⌺ Aufbau einer öffentlich zugänglichen, umfassenden, gut strukturierten und harmonisierten Toxikogenomik-Datenbank mit Informationen zu Expressions- und toxikologischen Parametern ►⌺ Durchführung gut definierter experimenteller Untersuchungen an Freiwilligen unter standardisierten Ernährungsbedingungen mit einem standardisier►⌺

Fazit

ten Ansatz zur Durchführung adäquater Ernährungsstudien (NutrigenomikStudien), ►⌺ Standardisierung der Anwendung der Profilerstellungstechniken.

1.5 Fazit Die hier diskutierten neuen Methoden sind erfolgversprechend und können Prioritäten- und Entscheidungsfindung im Prozess der Risikobewertung stützen. Sie können aber derzeit die klassischen Methoden noch nicht ersetzen. In-silico-Methoden, wie etwa QSAR- und PBBK-Modelle, sind in Bezug auf Â�regulatorische Auswirkungen und praktischen Einsatz am weitesten fortÂ� geschritten, da sie in-vivo-Dosis-Wirkungs-Beziehungen für die Toxizität prognostizieren können. Angesichts der immensen Datensätze, die anfallen, liegt eine Hauptaufgabe in der Interpretation. Eine hypothesen- und zweckgeleitete Methodenwahl ist daher, zusammen mit einer stringenten Validierung auf der Basis klassischer Methoden, unerlässlich. Die weitere Entwicklung dieser vielversprechenden neuen Technologien erfordert umfassende Langzeitforschung auf der Basis einer nachhaltigen Förderung.

25

26

2

Report and Conclusions

2.1 Preface Foods of plant origin are an essential part of the diet; however, some plants contain substances that hold certain health risks. Such “phytochemicals” can induce adverse effects by different mechanisms. Thus, some substances, for example soy isoflavones, affect the endocrine system. Other substances are hepatotoxic (coumarin), neurotoxic (solanine), phototoxic (furocoumarins) or carcinogenic (estragole). Normal intake of phytochemicals as natural components of fruits, vegetables, herbs, and spices is regarded to be of low risk. Increased exposure, however, poses a potential problem, for example in cases of unbalanced diets or uptake of dietary supplements in isolated and concentrated form. For risk assessment, it is necessary to identify potentially harmful substances based on their chemical structure. Also, their dose-dependent effects on the organism have to be described. For this purpose, novel profiling techniques and advanced computational methods are increasingly used, with promising possibilities. The Senate Commission on Food Safety (SKLM) of the Deutsche ForÂ� schungsgemeinschaft (DFG) organized a symposium entitled „Risk Assessment of Phytochemicals in Food – Novel Approaches“ that took place in Kaiserslautern, Germany from March 30th to April 1st 2009.╃Power, implications and promises of novel methodologies for risk assessment, including various in silico, in vitro and in vivo approaches were discussed and an evaluation of the state of the art for the various approaches was performed. Specific examples from the field of phytochemicals such as coumarin/ furocoumarins, beta-carotene, quercetin, glucosinolates and isoflavones were considered. The symposium addressed the relevance of the novel approaches for risk assessment of phytoÂ� chemicals. The SKLM has elaborated conclusions and recommendations, also identifying gaps in knowledge and research needs. This report is based on the presentations of Marcel Leist (DE), Gerrit Speijers (NL), Luis Valerio (US), Benoît Schilter (CH), Ivonne Rietjens (NL), Pablo Steinberg (DE), Hans-Jürgen Ahr (DE), Anette Thiel (CH), Hennicke Kamp (DE), Hannelore Daniel (DE), Augustin Scalbert (FR), Theo de Kok (NL), Takeki Uehara (JP), Alfonso Â�Lampen (DE), Â�Dieter Schrenk (DE), Jaap Keijer (NL), Ulrike Peters (US), Hansruedi Glatt (DE), Â�Johanna Fink-Gremmels (NL) and subsequent workshops.

Transdisciplinary Aspects

2.2 Transdisciplinary Aspects Diet and food components, irrespective of whether naturally occurring or manmade, are nowadays perceived as exposure/ nutrition related factors of influence on human health by affecting the genome, transcriptome, proteome and metabolome. In addition, since such interactions persist throughout life, the developmental and physiological status of an individual exposed at a given time in life is considered to be also of importance. New profiling techniques offer the possibility to identify interactions of food constituents with receptors, signalling pathways and other parts of the genome/ transcriptome/ proteome/ metabolome system involved in keeping our physiology in a balance. Risk assessment of food and its ingredients has to consider a large number of man-made chemicals, as well as various naturally occurring compounds which might influence human health. Since for the vast majority of these substances the available toxicological data are insufficient, novel molecular techniques are expected to provide rapid, efficient and reliable information to establish safe levels of exposure or to perform an adequate risk assessment on (phyto-) chemicals based on appropriate safety testing. In view of the large variety of new methodologies, which are at different stages of development and applicability at the present time, the SKLM underlines the importance of using these methodologies when pursuing a number of aims such as: The elucidation of mechanisms of action The development of novel techniques to enable comprehensive validation of large datasets ►⌺ The comparison of experimental data in laboratory animals or in vitro model systems with those obtained from studies on humans ►⌺ The science-driven development of food and food ingredients with a defined biofunctionality ►⌺ The consideration of individual variability and susceptibility. ►⌺ ►⌺

Such novel methodologies might be useful to generate data of improved quality in clinical and animal studies or to achieve a better understanding of seemingly inconsistent data. Biomolecular targets and biological effects of specific phytochemicals might be more adequately investigated by introducing molecular markers in future epidemiological studies. By doing so, the interaction of phytochemicals with biomolecular processes as well as the role of genetic variability in the heterogeneous response of different human populations towards certain phytochemicals may be better understood.

27

28

Report and Conclusions

2.3 Methodologies 2.3.1 Computational Toxicology╃/╃In Silico Models In silico technologies use computational information and strategies to generate a predictable toxicological and adverse effect profile for a given compound. They allow the rapid screening of high numbers of chemicals in a relatively short period of time, and are very cost-effective when compared to conventional animal-based toxicology studies. Moreover, they can also be used in multiple experiments, not only to screen large numbers of chemicals but also to refine their predictive power through adequate learning algorithms including for example quantitative structure–activity relationship (QSAR)-based strategies or so-called physiologically-based biokinetic (PBBK) modelling. A further benefit of in silico methods for predicting toxicity and aiding in risk assessment of chemicals is the potential net effect of reducing, refining or even replacing the number of animals used for laboratory testing, thus contributing to the 3R principle (refinement, reduction, replacement). Furthermore, the use of in silico computational toxicology methods with individual phytochemicals appears of great promise to generate priorities for risk assessment on a chemical basis.

2.3.1.1 Quantitative Structure–Activity Relationship (QSAR)

Definition: QSAR-based approaches quantitatively correlate parameters describing structural chemical characteristics of series of molecules with their biological activity or chemical reactivity. Computational QSAR-based predictive modelling applies statistical tools correlating biological activity of molecules with descriptors representative of a molecular structure. State of the Art: Computational QSAR-based predictive modelling provides an evidence-based tool for prioritization and an efficient estimate of hazard based on pre-existing test data regarding various end points. Such end points comprise e.╃g. the mutagenic potential of synthetic and naturally occurring molecules, including phytochemicals present in botanicals, herbs, and natural dietary sources. Predictive QSAR methodologies hold promise as a decision support tool in safety and risk assessment. They may be valuable in emergency situations to assist in fast decision making and may also help in defining priorities for additional toxicity testing. Limitations, Gaps in Knowledge and Research Needs: QSAR evaluation cannot ensure safety of a given substance with absolute certainty. Also, the programmes do not have all the needed functionalities nor do they give comprehensive information on the chemical properties or 100╛%

Methodologies

coverage or specificity. Furthermore, like most QSAR-based predictive models, the method suffers from inherent poor sensitivity, i.╃e. the ability to correctly identify true positives. It does not weigh exposure, or target specificity of compounds nor is it guided by mechanistic considerations such as class-specific mechanisms, for instance in carcinogenesis the multiple steps in tumorigenesis, tumor promotion, or tumor type. The use of a combination of computational software programs would be the best approach to maximize sensitivity and specificity of the prediction, since databases, algorithms and methods for structural interpretation vary between programs. Presently, predictions can be made with relatively good sensitivity and specificity for end points such as mutagenicity, which depends on a limited number of mechanisms, for instance on the capacity to covalently bind to DNA, or for acute toxicity in aquatic species, especially when a narcotic type of mechanism is responsible. However, more complex end points depending on a multitude of possible mechanisms, such as organ toxicity or carcinogenicity are much more difficult to predict. Currently, databases are limited and data on a sufficient number of compounds are missing. However, limitations are primarily due to current computational possibilities rather than to limited databases. At the present time computational QSAR-based toxicology methods can be used to analyze individual chemicals, but they do not address the screening of complex mixtures such as botanical extracts. However, particularly for the evaluation of mixtures and prediction of additive or synergistic effects computational techniques may be of value in the near future.

2.3.1.2 PBBK-Modelling

Definition: A physiologically-based biokinetic (PBBK) model is a set of mathematical equations that together describe the absorption, distribution, metabolism and excretion (ADME) characteristics of a compound within an organism on the basis of three types of parameters. These parameters include physiological parameters (e.╃g. cardiac output, tissue volumes, and tissue blood flows), physico-chemical parameters (e.╃g. blood/tissue partition coefficients), and kinetic parameters (e.╃g. kinetic constants for metabolic reactions). State of the Art: An overall problem in risk assessment is the need to extrapolate experimental data obtained in animal experiments at high dose levels to a low dose human situation. Uncertainties regarding the shape of the dose-response curve at dose levels relevant for dietary human intake and species differences in metabolism make such extrapolations difficult to be performed. PBBK equations predict, for example, the tissue concentration of a compound or its metabolites in any tissue over time at any dose, also allowing analysis of effects at low dose levels that are more realistic with respect to the human in vivo situation. Model predictions can be used to provide a more mechanism-

29

30

Report and Conclusions

driven basis for the assessment of the effects in animals and humans at low dose dietary intake levels, even based on in vitro data only. Furthermore, such PBBK models can be developed for different species, facilitating interspecies extraÂ� polation. In addition, by incorporating equations and kinetic constants for Â�metabolic conversions obtained from samples of individual humans and/or specific isoenzymes, modelling of interindividual variations and genetic polymorphisms is possible. International standardized procedures and databases to ensure standardization and transparency of the PBBK model building are already available. Limitations, Gaps in Knowledge and Research Needs: Risk assessment puts particular emphasis on long-term effects such as carcinogenicity. Whereas animal carcinogenicity experiments usually are conducted with a single defined compound, humans are exposed to phytochemicals via the diet, i.╃e. within a complex mixture of other ingredients. In a food matrix, various interactions might occur, thereby affecting the bioavailability of particular food components. In addition, interactions with other herbal ingredients might occur at the level of metabolic activation and/or detoxification. In principle, PBBK models can take into consideration such modulating effects of herbal ingredients for complex mixtures. Current models take into account the predicted data on dose-dependent effects, species differences and interindividual differences in bioactivation. However, for a complete assessment of the cancer risk under human-relevant intake conditions, additional information, such as toxicodynamic processes that might also affect the risk assessment, is still needed. This could be investigated by extending the PBBK models into so-called physiologically based biodynamic (PBBD) models, in which dose levels and biotransformation are coupled to biomarkers of exposure or toxicity and – ultimately – cancer incidence. One of the major limitations of PBBK modelling is that nowadays models have to be established in a laborious way for each individual compound. However, validation of a broad range of compounds is required. Nevertheless, the importance of PBBK modelling will certainly augment in the future, when high quality models become increasingly available. Particularly in combination with in vitro tests PBBK modelling might be of great help in extrapolating in vitro concentration–response curves to in vivo dose-response curves. This is achieved by using the concentration–response curves, acquired in an appropriate in vitro toxicity test, as internal concentrations in the model. By using the PBBK model, the in vivo dose levels that are needed to reach the internal (toxic) concentrations can then be calculated. The predicted dose–response curves thus obtained can be used to determine safe exposure levels in chemical safety assessment. Thus, PBBK approaches will help to link external doses to internal concentrations (target tissue dose) and to link these internal concentrations to EC50 values from in vitro studies and vice versa.

Methodologies

2.3.2 Profiling Techniques: “OMICS”, In Vitro and In Vivo Models The term “OMICS” refers to a field of disciplines analyzing biological information on the interaction of compounds with the genome, proteome and metabolome all of them ending in -omics (e.╃g. genomics, proteomics, metabolomics). It comprises a variety of subdisciplines using different sets of techniques, reagents and software like DNA and protein microarrays, gas and liquid chromatography, mass spectrometry and a number of other methodologies enabling high-throughput analyses. One of the strengths of “OMICS” lies in the fact that they can equally well be applied in studies being performed in vitro or in vivo. Genomics includes “structural genomics”, DNA sequence analysis and mapping of the genome of an organism, as well as “functional genomics”, the characterization of gene responses, analyzing mRNA and protein expression profiles. As a more specific discipline, the study of the complete set of proteins expressed in a cell, tissue, body fluid or organism is referred to as proteomics. Unlike the genome, the proteome varies between cell types. Proteomics attempts to identify the protein profile of each cell type, and to assess differences in protein expression patterns between healthy and diseased cells. However, proteomics was not explicitly discussed at the symposium which focussed on Â�transcriptomics and metabolomics.

2.3.2.1 Transcriptomics

Definition: Transkriptomics refers to the global analysis of gene expression, also called “genome-wide expression profiling”, measuring relative amounts of messenger RNA (mRNA) in order to determine patterns and levels of gene expression, and to analyze gene regulation. State of the Art: Although mRNA is not the ultimate product of a gene, transcription is the first step in gene regulation, and information regarding transcript levels is needed to better understand gene regulatory networks and to detect similarities in expression patterns of genes, which may be functionally related and under the same genetic control mechanism. Whole genome transcriptome analysis provides an unbiased approach to the identification of all possible effects, intended (expected) effects as well as unexpected ones. RNA expression signatures that allow differentiation between certain classes of compounds, such as genotoxic and non-genotoxic liver carcinogens, have already been identified. Specific genetic polymorphisms might be taken into account for risk–benefit analysis of phytochemicals. Furthermore, modulation of the expression of genes involved in biological and genetic pathways that are relevant to carcinogenesis is regarded as an important component of the anticarcinogenic effect of vegetables or phytochemicals. Transcriptomics can be helpful to study such effects.

31

32

Report and Conclusions

Limitations, Gaps in Knowledge and Research Needs: Modulation of transcription levels is not necessarily correlated to corresponding changes on the protein level. At least for specific genes, data on RNA expression levels should be complemented by determination of the respective protein levels or activity. Furthermore, alternative splicing may influence functionality, depending on the tissue context. Transcript profiling produces extremely large datasets and requires effective database resources for interpretation, management and analysis. It remains to be studied whether genome-wide expression data will be required for routine applications. Possibly, genomics will only be required to identify sets of genes that allow the establishment of a classification algorithm. If the number of genes required for classification is not too high, they could be analyzed by a quantitative technique, such as qRT-PCR (quantitative real-time polymerase chain reaction). Interpretation of gene expression data should be supported by validated �biomarkers. To better understand the complex information usually obtained, integrated systems biology approaches should be developed.

2.3.2.2 Metabolomics

Definition: Metabolomics intends to investigate the profile of all metabolites present in a simple system, e.╃g. a cell, or in a complex system, e.╃g. entire organ or organism. State of the Art: Metabolomics makes use of advanced analytical techniques including nuclear magnetic resonance (NMR) spectroscopy as well as mass spectrometry (MS) which is mostly coupled with various separation techniques like gas chromato� graphy (GC) and liquid chromatography (LC). These tools allow to characterize the endogenous and exposure related metabolome of an organism and to identify compounds of different metabolic pathways. A strength of metabolomics is its ability to identify the capacity of specific nutrients as well as xenobiotics to alter the metabolic properties of an organism such as for example metabolic liver functions, insulin sensitivity of tissues as well as thyroid hormone secretion. A further advantage is the possibility to analyze metabolic spectra in human blood and urine. Therefore, metabolomics can be applied to study the influence of dietary compounds in humans. Limitations, Gaps in Knowledge and Research Needs: Metabolomics covers the fraction of the entire metabolome accessible to instrumental analysis. However, although already much progress has been made in understanding metabolic profiles, it often is not clear how to interpret the complex data. Since many compounds act by several mechanisms the situation

Methodologies

may become complex. Moreover, analysis of the effect of mixtures on the metabolome has not been addressed so far. There is a need to standardize quantification of metabolomic data. In addition, a standardized platform for data exchange should be established. The statistical assessment needs to be improved to distinguish effects from the normal range of variations within the metabolome. Generalized algorithms for the prediction of the modes of action should be established and validated.

2.3.2.3 Toxicogenomics and Nutrigenomics

Definition: Toxicogenomics and nutrigenomics are collective terms that cover the three sub-disciplines of transcriptomics, proteomics and metabolomics. These profiling techniques can be looked at from a nutritional or toxicological point of view: The term toxicogenomics is used when toxicology is combined with these new methods to better understand and assess the toxic effects of specific compounds, whereas nutrigenomics addresses the impact of specific nutrients or diets on gene expression. It is not to be confused with the term nutrigenetics which investigates how genetic variability influences the body’s response to a nutrient or diet. State of the Art: Toxicogenomics has great potential to predict specific toxic effects of drug candidates and chemicals. Toxicogenomics aims to identify biomarker genes and to establish relationships between changes in gene expression patterns and certain toxicological end points. Short term exposure is expected to cause alterations in RNA, protein or metabolite expression patterns that could also help to predict long term effects. In addition, it has already been convincingly shown that these techniques can provide strong evidence regarding toxic mechanisms being activated by the test compounds. Nutrigenomics has the potential to easily identify hundreds of entities that respond to a given nutrient or non-nutrient, to a treatment or diet in a well defined experimental setting. Limitations, Gaps in Knowledge and Research Needs: Toxicogenomic studies and derived classification algorithms are as yet only based on a limited number of compounds. Dose–response relationships have only been established in a very limited number of cases. This could be critical because different mechanisms might be activated by low and high concentrations of one and the same compound. In future it will be important to establish “OMICS” data of a high number of compounds acting by well defined mechanisms. Only in the case that data on a sufficiently high number of compounds belonging to the same compound classes become available, coordinated changes in transcripts, proteins and metabolites could be identified and associated with common molecular (toxic) pathways. Such key pathways and their correlation

33

34

Report and Conclusions

with biochemical data still have to be identified. Gene, protein and metabolite expression pattern alterations also have to be linked to adverse and beneficial human health effects to increase its as yet limited value for risk assessment. It is essential to establish a harmonized, large-scale, well designed toxicogenomics database with information on gene expression and toxicological parameters that is publicly available. In order to adequately perform nutritional investigations (nutrigenomic studies), well defined experimental studies with volunteers kept under controlled dietary regimens and in a standardized setting are needed. Furthermore, there is a necessity to standardize the application of the profiling techniques. The current problem is in line with that of toxicogenomics, i.╃e. the huge datasets generated by nutrigenomics are difficult to interpret adequately. Coordinated changes in transcripts, proteins and metabolites can only be projected on common pathways of regulation in a limited number of cases. Another limitation is that expression profiles vary between different tissues and it often is uncertain which tissue may be relevant. Whereas in animal models tissues may be analyzed quite comprehensively, this is not possible for studies in humans.

2.4 Conclusions and Recommendations In silico and in vitro techniques for hazard identification as well as in vitro techniques for testing of certain mechanistic end points may facilitate a more efficient safety evaluation of phytochemicals by means of targeted study design. Furthermore, in silico and in vitro investigations may help to reduce the number of animals necessary for efficacy and safety studies. New in vivo techniques may also help to reduce animal numbers and the duration of experiments by predicting “long term” effects such as carcinogenicity based on the results obtained in “short term” studies. In the opinion of the SKLM the novel approaches are not yet powerful enough to replace the current testing and risk assessment procedures. They are, however, regarded as useful to obtain more comprehensive data and to improve the evaluation of clinical and animal studies or to understand inconsistent data. Furthermore, they might provide mechanistic insights to be included in the risk assessment procedure. However, these novel techniques need a stringent validation, including a comparison with classical methods, in order to ascertain their applicability for a given purpose. In order to achieve the maximum outcome it is suggested to develop a novel testing strategy by integrating different methodologies into a combined approach. The planning, design and establishment of harmonized, publicly available, high quality, large-scale databases is required to ascertain standardization, comparability and reproducibility and to facilitate cooperation between different laboratories. This will accelerate the validation process and make sure that the new approaches are adequately applied. Experts in bioinformatics are essential to achieve these goals.

Conclusions and Recommendations

It is essential to consider the genetic background of the analyzed population if human studies are performed. A precondition for studying the influence of compounds on metabolite patterns in humans is a comprehensive database on the physiological background situation of individuals and the general population. By doing so, genetic variability can be better taken into account and it might be possible to identify key factors responsible for variability in the study population. Humans are usually exposed to phytochemicals in complex mixtures of compounds. At present, most of the novel techniques are not able to deal with such mixtures of compounds. So far “OMICS” techniques have been applied almost exclusively to the analysis of individual compounds and not to mixtures. This is reasonable, because we first have to identify and understand the already very complex effects of single compounds. However, an advantage of pattern based “OMICS” technologies is that it is irrelevant whether alterations in complex expression patterns are caused by single compounds or by mixtures. Therefore, as soon as further research will have deciphered the complex language of “OMICS” patterns for individual compounds, this methodology will also be applicable to mixtures and may rapidly advance this field of research. Finally, the unique advantage of “OMICS” techniques due to their random or even genome-wide approach is that we may also find effects, which we were not looking for. Therefore, when applied in addition to our conventional toxicity tests, they probably will improve the risk assessment.

2.4.1 Specific Research Needs in Keywords QSAR:

Increased research into the possibility of combining software programmes to improve sensitivity and specificity of predictions ►⌺ Development of less complex, laborious software to make more intelligent use of databases ►⌺ Development of computational methods to assess mixtures of compounds. ►⌺

PBBK:

Extension of the PBBK models into so-called physiologically based biodynamic (PBBD) models ►⌺ Development of less laborious methods to allow for the modelling of a higher number of substances ►⌺ Combination of in vitro toxicity tests with in vivo dose–response curves. ►⌺

35

36

Report and Conclusions

Transcriptomics:

Complementation of data on RNA expression levels by determination of the respective protein levels or activities, also taking into account alternative splicing ►⌺ Development of effective databank resources to support interpretation, management and analysis for transcript profiling; Determination whether genome-wide expression data will be required for routine applications ►⌺ Establishment of a classification algorithm ►⌺ Development of validated biomarkers for the interpretation of gene expression data ►⌺ Development of integrated systems biology approaches to better understand complex data. ►⌺

Metabolomics:

Standardize the determination of metabolic data Establish a platform for data exchange ►⌺ Study the effects of mixtures on the metabolome ►⌺ Improve the statistical evaluation to distinguish effects from the normal range of variations ►⌺ Establish generalized algorithms for the prediction of modes of action. ►⌺ ►⌺

Toxicogenomics and Nutrigenomics:

Generation and use of data on toxicogenomic studies, derived classification algorithms and dose response relationships for a high number of compounds to identify toxicological key pathways and their correlation with biochemical data ►⌺ Relationships between gene, protein or metabolite expression pattern alterations with adverse or beneficial effects on human health ►⌺ Establish a harmonized, large-scaled, well designed public toxicogenomic database with information on gene expression and toxicological parameters ►⌺ Perform well designed experimental studies with volunteers under controlled dietary regimens to perform adequate nutritional studies (nutrigenomic studies) ►⌺ Standardization of the application of the profiling techniques. ►⌺

Concluding Remarks

2.5 Concluding Remarks At present, the above discussed novel methodologies are of great promise for prioritization and decision making in the process of risk assessment. However, these techniques are not yet able to replace classical methods. In silico methods such as QSAR and PBBK models are most advanced in terms of regulatory implications and use, since they may predict in vivo dose–response curves for toxicity. In view of the huge datasets generated interpretation becomes the major task. Therefore, hypothesis – and purpose-driven choice of methodology is essential, together with a stringent validation referring to classical methods. Further development of these largely promising novel technologies has to be based on comprehensive long-time research requiring sustained support.

37

38

3

Contributions

3.1 Visions on Toxicity Testing in the 21st Century: �Reflections on a Strategy Document of the US€�National Research Council Marcel Leist1,2, Thomas Hartung2, and Pierluigi Nicotera3

Abstract Toxicology faces enormous challenges in a world in which we are exposed to thousands of chemicals and especially millions of mixtures thereof. Radically new approaches to this problem need to be developed. A milestone in this direction is the vision of the US National Research Council (NRC) “Toxicity testing in the 21st century: A Vision and a Strategy”. Currently, an alliance formed by the National Toxicology Program (NTP) and the Chemical Genomics Centre (NCGC) of the National Institutes of Health (NIH) and the Computational Toxicology Centre (NCCT) of the Environmental Protection Agency (EPA) is testing whether this new strategy can realistically form the basis of future public health decisions. The vision requires a radical paradigm shift in the approach to safety assessments, and turns the traditional procedures upside down. Where animal experiments used to be the most important technology, the future is seen in the strength of in vitro and in silico approaches based on human material. Today’s toxicity testing starts with an initial black box screen on animals, sometimes followed by mechanistic studies, while the new vision approaches hazard assessment bottom-up. The procedure would begin with in vitro tests to define the affected pathways. To fill remaining gaps of knowledge, limited and targeted testing in animals would then be performed as a possible second step.

1

Correspondence to: Marcel Leist, University of Konstanz, Doerenkamp-Zbinden Chair for Alternative In Vitro Methods, D-78457 Konstanz, Germany, Tel: +╃49 7531 885037, Fax: +╃49 7531 885039, [email protected].

2

Johns Hopkins University, School of Environmental Health Sciences, Centre for Alternatives to Animal Testing, 615╃N. Wolfe St., W7035 Baltimore, MD, 21205, USA.

3

University of Leicester, MRC Toxicology Unit, Lancaster Road, Leicester LE1 9HN, UK / Scientific director of the German Centre of Neurodegenerative Disease (DZNE), LudwigErhard-Allee€2, 53175 Bonn, [email protected].

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

3.1.1 Introduction Toxicology is an exciting discipline that brings together specialists from vastly different areas. A picture that springs to mind is one of a body with three souls [1]: As for many other medical disciplines, one important aspect of toxicology is that its procedures and the specific knowledge are applied like a craft. In this first domain, which contains the translational aspects of the science, careful documentation, process-optimization and routine are of high importance. A second focus area is regulatory toxicology at the interface of industry and authorities, involved in setting and meeting guidelines and providing a basis for political decisions and legal requirements concerned with environmental health and consumer safety. The third soul of toxicology is its scientific basis. This area is concerned with the generation of new knowledge and is linked to other natural sciences. It appears as if the three souls have lost connection over the past decades and that a large part of toxicology became frozen in time, using and accepting the same old animal models again and again, often without stringent examination of their validity [2, 3]. In this situation, the overall discipline is strongly driven by the demand for protocols and data for regulatory action. Only few resources remain for generation of fresh, fundamental toxicological knowledge and scientific output. A lot of the remaining scientific progress of toxicology depends strongly on import from other biomedical fields [4]. The consequences are reduced innovation, followed by a loss of attractiveness of the field for talented workers, and finally an inability to meet newly arising challenges. Such new challenges are for instance the safety evaluation of compound mixtures in food or the environment, of biologics, of nanomaterials, of irradiated or genetically-altered food or of mobile phone radiation. None of them can be tackled adequately by classical animal-based methods. Huge challenges lie also in finding more predictive systems for developmental neurotoxicants [5] or non-genotoxic carcinogens [6, 7, and 8]. However, these current problems are also a huge opportunity for the future, to bring the domains of toxicology together again, to link the field more closely to progress in other areas of biomedical sciences, and to give it a new basis [1]. There is a vast body of evidence from mechanistic toxicology studies suggesting that the thousands of known noxious substances act by interfering with only a few (i.╃e. dozens) regulatory pathways of cells [9]. For instance, a variety of hepatotoxins act by enhancing TNF-induced apoptosis [10], various compounds are neurotoxic because of perturbed cellular calcium metabolism [11, 12, and 13], various immunotoxicants affect the cell cycle of lymphocytes via the Ah receptor [14], endocrine disrupters often bind to steroid receptors [15, 16, and 17], and interaction with the P450 system has been extensively examined as the basis of the toxicity of thousands of diverse compounds [18, 19, and 20]. Information on such affected pathways can nowadays be obtained rapidly by high-throughput screening systems using human cells, and then be further analyzed with modern methods of systems biology and bioinformatics. Such a new approach has recently been suggested by the US National Toxicol-

39

40

Contributions

ogy Roadmap “A national toxicology program for the 21st century” (http://ntp. niehs.nih.gov/files/NTPrdmp.pdf) and the NRC [9], and testing of its feasibility by major safety authorities has begun [21].

3.1.2 A New Vision of Toxicity Testing The NRC, the most prestigious scientific council of the USA, was funded some years ago by the EPA and the NTP to develop a long range vision and implementation strategy for modern toxicology (Fig.╃1; see list at the end of this text for abbreviations). The heart of the new vision of toxicity testing proposed by the NRC is the concept of “toxicity pathways” (Fig.╃2). As shown in Figure 2â•›a, the vision takes its starting point from the presumption that most toxicants will eventually act by interfering with pivotal cellular structures and regulatory pathways. This would result in a limited number of toxicity pathways (e.╃g. disturbed calcium regulation, triggering of apoptosis, cell cycle derangement …). It is then further presumed that knowledge of these pathways and knowledge of the action of toxicants on these pathways would allow predictions of toxicity on the level of the whole organism. This is a simple concept, but with huge implications. The practical consequence for toxicity testing would be no less than a turn-around of the currently used process from top to bottom (Fig.╃2â•›b, Fig.╃3). Currently, animal models are frequently used as black box system to identify problematic compounds. Only in few cases (e.╃g. for valuable compounds, or compounds leading to high human exposure) will toxicity data ever be followed up to understand why a compound is toxic and whether the effect is relevant to humans. The vision laid out by the NRC suggests a radical paradigm shift. The

Figure 1:€Parents and godfathers of the vision. At the end of 2007, the NRC published its report after the initial trigger by two important regulatory agencies a couple of years earlier. A pivotal strength of the procedure, compared to similar approaches, is the early involvement of and support by major stakeholders (academia, regulators, industry) and the coupling of a vision to an implementation strategy.

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

start of a safety evaluation would begin with the chemical properties of a compound and then proceed to the biological characterization in multiple in vitro systems (Fig.╃2). Bioinformatic procedures would transform this information into a hazard estimate. This procedure would prioritize a few compounds (e.╃g. unclear hazard estimate or biokinetic predictions and high exposure) for further animal testing, and be sufficient on its own to eliminate many compounds and mixtures. This would be a revolutionary approach if it was actually applied in practice, but is the idea really new? There is a saying that “success and good news have many parents, uncles, godfathers╃…, once they are apparent to everybody, while failure is an orphan, with an ugly mother-in-law, at best”. Accordingly,

Figure 2:€Approach to toxicity testing suggested by the NRC (USA). (A) Toxicity pathways lie at the heart of the approach of hazard evaluation and are examined with the help of in vitro models. Gaps of knowledge and uncertainties are addressed by targeted animal testing. Risk estimates are then based on the hazard evaluation, exposure data and the risk context. For evaluation of this approach, a number of important questions need to be addressed. (B) The new vision follows a bottom-up approach in contrast to the present approach.

41

42

Contributions

many will claim now, that they have worked on the same idea as promoted by the NRC [9] for years, or even decades. It is indeed true that in vitro toxicology is a firmly established and well-organized discipline which has produced similar ideas and also already some applications in the regulatory field and in applied research [22, 23, 24, 25, 26, and 27]. There has been a continuously good output over decades from laboratories interested in mechanistic toxicology, and many companies and regulatory toxicologists are deeply involved in the development of alternative methods, such as in silico and in vitro screens. For instance, G. Zbinden showed already 20 years ago the trend towards mechanistic models and the necessity for international regulators to follow this line and incorporate the ideas into the regulatory context [28, 29]. So again: What’s new? It is the way it is done. The determination to “think big”, the broad basis, the wide scope, the involvement of many stakeholders and drive by major authorities, the generation of open interfaces to the interested public (including accessible data bases) and the coupling of the vision to an implementation strategy that is robust enough to have a chance for success.

3.1.3 Testing the Vision A condensed overview of the initial phases of the implementation strategy was given recently by the involved US authorities [21]. Here, we want to outline the essential features (Fig.╃3), mainly as stimulation for the interested European readers and to provide a basis for potential interactions. Presently, the implementation strate5gy is being explored by three major players on the basis of a memorandum of understanding clarifying the roles and duties (Fig.╃2). One of the contributing institutions is the NCCT [30] under the roof of the EPA. The two other players are funded by the NIH: The NCGC contributes its screening infrastructure (robots, compound management, highthroughput measurement devices) and performs quantitative high-throughput screens (qHTS). The final player is the NTP which contributes with classical toxicological expertise, non-rodent animal models (for instance zebra fish embryos) and especially a screen programme for about 300 selected compounds run through hundreds of assays (Fig.╃3).

3.1.4 Steps toward a New Toxicology What happens with the data obtained? Here the idea of open public interfaces and generally accessible databases comes into play. This sounds like a relatively trivial issue, but it should by no means be underestimated. We all have witnessed how the free internet availability of literature references via NCBI’s PubMed has revolutionized the way scientific information is retrieved, and how Google has entirely changed the way general information is retrieved. Toxicology urgently needs a parallel effort. At present a number of interconnected databases is being developed (Fig.╃4) and expanded, but their user-

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

Figure 3:€Testing the feasibility of a new way of toxicity testing and reduction to practice. The vision and theoretical strategy were laid down by the NRC. Top: the paradigm shift according to this vision is outlined. Centre and bottom: In order to test whether the vision holds in the face of reality three major players agreed in a memorandum of understanding on a common test strategy. The three players are institutes and programmes of the EPA and the NIH, and contribute expertise as indicated.

friendliness is far from perfect. Much of the screening data will eventually end up in PubChem, which already harbours over 900 bioassays and will be fed directly with data from screens of the NCGC. Some of this data will also appear in classical journal publications, but in order to understand such publications one will have to be able to retrieve information from PubChem. An example is a publication describing the test of the cytotoxicity of about 1400 compounds on 13 different cell lines [31]. The publication compiles data from different screens and extracts information from comparisons of cell lines and compounds. However, the compounds themselves and the original data from the screens will have to be extracted from the database [32] – and, conversely, the database information may eventually be used again for new analyses and journal publications. DSSTox is another database with generally richer datasets than PubChem. Here, reviewed and quality-controlled classical toxicological information is added to the compounds. The DSSTox website provides a public forum for publishing downloadable, structure-searchable, standardized chemical structure files associated with toxicity data [33].

43

44

Contributions

Figure 4:€Databases to help computational toxicology and in vitro toxicity testing. Different interlinked databases allow public access to cpd (compound) and assay information. RefToxDB is presently not publicly accessible. Links to the other databases are indicated.

One important input for DSSTox is ToxCast [34]. This programme was designed to predict toxicity pathways and to characterize the hazard of a relatively small learning set of tool compounds (n=╃300) run through 400 different assays. ToxCast™ signatures will be evaluated by their ability to predict outcomes from existing mammalian toxicity testing and to identify toxicity pathways that are relevant to human health effects. High added value will be generated when this is linked to ToxRefDB, a database designed to contain data from huge historical animal testing efforts, including compounds selected for Toxcast. ToxRefDB is integrated into a more comprehensive data management system developed by NCCT called ACToR (Aggregated Computational Toxicology Resource) that manages the large-scale datasets of ToxCast™. The above databases are mainly compound and assay-focussed. For hazard assessment further dimensions are essential. We need to understand how the human body handles a given chemical, what the important toxicity pathways are, and how we deal with human genetic variability (Fig.╃5). An in vitro test strategy requires more than the test system and data analysis. It cannot function without a prediction model to make use of the data. This also applies to complex integrated test strategies, and here pharmacokinetic information and dose–response modelling become highly important issues for the construction of prediction models. During establishment of the test strategy, variations of the following problem are frequently encountered: “Pesticide X induces signs of toxicity (e.╃g. muscle paralysis) at a dose of Y mg/kg. Which concentrations should induce a positive readout in a corresponding in vitro toxicity test system in order to consider the test system relevant? In other words, which in vitro cytotoxic concentration would one predict from the in vivo data? Which would be a biologically relevant prediction model for in vitro concentrations, when in vivo doses are given?” Databases that translate such information are

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

Figure 5:€Identification of toxicity pathways and in vitro–in vivo extrapolation. To support the overall project to test a new vision for toxicity testing, compound information alone is not sufficient. An important accessory programme is the initiative to model pharmacokinetics and in vitro–in vivo extrapolations and dose–response relationships with the help of physiologicallybased pharmacokinetic modelling (PBPK). Another initiative makes use of the HapMap project. This is a multi-country effort to identify and catalogue genetic similarities and differences in human beings. Using this information, researchers will be able to find genes that affect health and individual responses to environmental factors. The tool library used by the NCGC (about 2800 compounds (cps) will be screened on cell lines with known haplotypes (i.╃e. known genetic variation). These compounds will also be tested within the HSP on transgenic mice and cells derived from them. This is expected to yield information on which genes have a major impact on adverse effects of environmental agents. The biological approach is complemented by the Virtual Liver Project, which plans to develop a database and algorithms able to predict liver toxicity and forms of liver carcinogenicity.

urgently required. During the application of an established test strategy to unknown compounds, a related problem occurs: “compound A triggers toxicity in vitro at concentrations higher than B micromolar. How much of the compound can be ingested safely?” PBPK databases will need to contain all the essential data on metabolism, protein binding and barrier permeation of compounds, in addition to suitable algorithms that will allow at least rough conversions of in vitro concentrations to in vivo doses. The setup of these databases is still in a very early phase. The HapMap project is attempting to map and understand human haplotypes (i.╃e. variants of a given gene that are found in different proportions of the population). This project can also be linked to toxicity testing strategies. Interesting information is expected from testing a set of 2800 compounds on human cell lines with known haplovariants. In a parallel approach taken by the host susceptibility programme (HSP), compounds are compared through a large number of transgenic mouse models and derived cell lines. These two

45

46

Contributions

programmes can contribute to the clarification of toxicity pathways and susceptibility genes and their respective effects. A different approach is taken by the EPA with the VirtualLiver project, which attempts to model the most important target organ of toxicity in its interaction with compounds. On its website it is stated ambitiously that “…the 5-year plan for the Virtual Liver Project is to develop a knowledgebase for qualitatively describing species-specific toxicity pathways due to exposure to chemicals, and to develop a virtual liver tissue that lays the foundation for quantitatively predicting the risk of non-genotoxic neoplastic lesions due to activation of certain genetic regulatory elements (i.╃e., nuclear receptors and other transcription factors) in humans”.

3.1.5 The Precautionary Principle Toxicological studies are designed to provide a basis for consumer protection by identifying hazardous compounds. The test systems will necessarily also produce false positives (compounds that are not hazardous to humans, but look hazardous in the test system) and false negatives (compounds that are hazardous to humans, but are not correctly identified by the test system) [1]. The latter class has been of particular concern. Therefore, the test systems and prediction models were tuned in a way to minimize this class as far as possible at cost of a largely increased class of false positives. This tuning of toxicity testing is called the precautionary principle and is one of the corner stones of toxicological thinking. Major changes in toxicity testing will always provoke fears in the public, in regulatory authorities and in other stakeholders that the precautionary principle may be violated. Therefore, one of the major tasks of the implementation strategy of a new vision is to address these worries and to generate confidence that the safety level will not be compromised. A first important issue to be considered is the understanding of the concept of “applicability domains”. All toxicological methods are not generally applicable, but have applicability domains, i.╃e. limitations as to for which part of the chemical universe their predictive value has been shown. For instance, “drugs” or “pesticides” are typical applicability domains. Test guidelines, legislation, authorities, and the questions asked are vastly different in these areas. Other applicability domains would be industrial chemicals, cosmetics, biologics, and food additives. The concept was taken from the field of (Q)SAR and translated to test methods first in ECVAM’s Modular Approach [35]. The vision discussed here applies mainly to the domain of environmental agents (i.╃e. pesticides or chemicals with relevant human exposure, for instance through the food chain). This is also reflected by different risk context scenarios that are explored and that are an important feature of the implementation strategy. Whether it can be translated to other domains without compromising the precautionary principle is one of the open questions for the future, and will certainly involve additional stakeholders. The key issue to consider is, what new methods of toxicity testing should be used for comparison? Can we expect a 100â•›% failsafe method? We know that

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

present animal-based testing does not guarantee absolute safety [36]. This is an obvious fact that is often forgotten in discussions on new approaches. New alternative tests are validated stringently [37], while many animal tests have never been formally validated [2, 3]. Even studies that address the question whether animal studies are of any toxicological use at all with respect to human safety are extremely scarce [38]. At least some doubt comes from the extreme variation of results when one and the same compound is used in different animal studies, and from the partially poor correlations between one species and the next, for instance between mouse and rat [2]. Thus, a fair and honest approach to alternative testing strategies would imply that one does not require a 100╛% safety level, but rather a safety level that is in the range of (or at least as good as) that of standard animal experimentation. This also implies that showing the one or other insufficiency of in vitro approaches, and of the cell culture technology in particular [39], does not invalidate the usefulness of a technology. The strengths and weaknesses of animal and non-animal test approaches will just lie in different areas. Only looking at the comparison of the overall performance with regards to human safety will allow a reasonable judgement of the value. In this context it is important to reconsider what the ultimate aim of the precautionary principle is: Human safety. Sometimes, more exact knowledge on toxicity does not contribute to higher safety, but precautionary measures, e.╃g. regarding the transport of chemicals, take this function. Extensive animal testing will often generate redundant information, and, in addition, we accumulate more false-positive results [40]. To trigger a certain and adequate set of measures, sometimes limited in vitro and in silico information may be sufficient [41].

3.1.6 The European Side The 3R principle (reduce, replace, refine), which already envisaged a combination of in vitro and in vivo approaches in the 1950’s was originally developed in Europe [42]. Is European toxicology less visionary now? What could be learned from the NIH/EPA approach? Europe has a different, more diversified, but also more fragmented political landscape and different countries have found their own ways. For instance, the MRC in the UK decided almost 10 years ago to restructure its entire central toxicology institute in Leicester. Already at that time the guiding principle was to promote research on bottom-up toxicology, taking its starting point from understanding toxicity pathways and common processes like apoptosis. In Germany, ZEBET, a federal institute, was established nearly 20 years ago to develop, test, and validate alternative methods to animal experimentation, and has been a major driver in the design of the first OECD toxicity testing guidelines based on in vitro testing only (for phototoxicity). On the EU level, the first major driver for a new vision of toxicity testing comes from a different applicability domain than in the US – from cosmetics products. Here, the vision was immediately reduced to practice by law. The 7th amendment of the Cosmetics Directive set a strict timeline, finally banning the

47

48

Contributions

use of cosmetics if their ingredients were tested on animals. The implementation strategy implies that industry will need to establish animal-free test methods or change the business model. This is an interesting test case for the whole world to follow. In order to guide the development of methods and to ensure their validity, the EU founded ECVAM in 1992, a research institute entirely devoted to the validation of alternative methods. Now corresponding agencies and institutes are also found in the USA, Japan and other countries [43]. ECVAM harbours also an important database, DB-Alm, which is a high-quality source for in vitro test protocols and alternative methods [44]. At present, the major driver for a rethinking of toxicity testing in Europe is the REACH legislation [45]. Over the last two years a revolution of the concept of how safety of chemicals is evaluated took place in Europe in this context: While in the past a (tonnage-triggered) set of mainly animal tests had to be provided in a tick-box manner, now (for both existing and new chemicals) integrated testing strategies making use of all information opportunities must be applied. A group of more than 200 experts from regulatory bodies, European Commission and industry developed these strategies (http://ecb.jrc.it/reach/ rip/) in REACH Implementation Project 3.3 under the coordination of CEFIC and ECVAM. New and existing approaches were combined in order to optimize information generation for REACH, making use also of in vitro, in silico and read-across data from similar compounds. This law is at the basis of an enormous effort to re-evaluate about 30,000 chemicals already marketed in the EU and generates major financial and logistic pressures in addition to the ethical problem of the requirement for millions or tens of millions of animals to fulfil the test requirements. Faced with this enormous challenge, industry and the European Commission formed a partnership in the form of the EPAA [46], which is working on new visions and implementation strategies. In parallel, the Directorate General of Research (DG Research) is heavily funding research consortia within the sixth and seventh framework programme to develop new in vitro test systems and strategies. The key feature of REACH in the context of new visions of toxicology is that it has been influenced by an important postulate of the European animal legislation from 1986 (Directive 609/86), which can be summarized as “when alternatives to animal experimentation are available, they must be used”; “more of these alternatives need to be developed”. More precisely, article 7.╃2.╃states: “An experiment shall not be performed if another scientifically satisfactory method of obtaining the result sought, not entailing the use of an animal, is reasonably and practicably available.” And in Article 23.1.: “The Commission and Member States should encourage research into the development and validation of alternative techniques which could provide the same level of information as that obtained in experiments using animals but which involve fewer animals or which entail less painful procedures, and shall

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

take such other steps as they consider appropriate to encourage research in this field.” Article 1.1 of the REACH regulation reads: “Aim and scope 1.╃The purpose of this regulation is to ensure a high level of protection of human health and the environment, including the promotion of alternative methods for assessment of hazards of substances, as well as the free circulation of substances on the internal market while enhancing competitiveness and innovation.” REACH is thus the first major legislation in the huge application domain of industrial chemicals that gives some space for “intelligent test strategies”, readacross between different information domains, the use of validated alternative methods, and also the use of non-validated alternative methods at least in a preliminary hazard evaluation [40, 47, 48, and 49]. Nevertheless, REACH will still require millions of animal experiments, and the free space given by legislation is still far away from the vision of toxicity testing in the 21st century laid out by the NRC. Whether this heavy animal testing effort will lead to a parallel increase of human safety with respect to chemicals already on the market has been doubted [50]. Thus, a new movement is presently forming that focuses on a more stringent validation of animal models and promotes an evidence-based toxicology, in which the best given test strategy is used instead of stringent adherence to only historically-legitimated animal models [51, 52, and 3]. A multitude of bottom-up movements are emerging at present, which include for instance ASAT, the NTC, and InViTech, to name a few.

3.1.7 Tasks Ahead We have tried here to survey exciting new developments and movements. Proofof-concept studies need to clearly demonstrate the predictive power gained from these new approaches. More researchers need to be attracted to join the efforts, and regulatory authorities must show a willingness to embrace the new approaches as they gain scientific acceptance. The next few years should witness the early fruits of such efforts, but the paradigm shift will require a long-term investment and commitment to reach full potential. In a brief last paragraph we want to summarize critical issues to be addressed by the scientific community, granting agencies and authorities: Databases: These require a change of attitude as they move more into the centre of the process instead of being a final end product. It often appears from the lack of care and the limited analysis and accessibility options that they are more or less considered a tiresome duty to those who have generated the data. It is not sufficient to simply “dump” the data somewhere, even if they are flexibly retriev-

49

50

Contributions

able and adequately quality-controlled. The science of visualization of data and especially visualization of large complex data spaces needs to be applied much more strongly here. The importance of this process and the need for users and developers to hold a constant dialogue during the design of analysis and visualization algorithms are still heavily underestimated. Another important issue is the cross-linking of information. For instance a number of databases have been generated in Europe on in vitro acute toxicity data. For instance the MEIC database (see list of abbreviations) covers a very large number of toxicity assays, and the Halle Registry [53] over 300 in vitro–in vivo comparisons, but cross-linking is limited, as is general and easy accessibility. Libraries: To fill the databases with information, real compounds are required. Especially in the proof-of-principle phase of testing, the selection of these compound libraries plays an important role and contributes to the success, the validity, and the general acceptance of validation efforts. Not only will the right “theoretical” composition of the libraries be of high importance, but also the physical composition and availability. Compound stability and purity, the general accessibility and continuous quality control are non-trivial issues, especially in the field of environmental chemicals, industrial chemicals and pesticides. Here, one solution to be considered is chemical reference laboratories making defined library copies available to others. This has been conceptualized on the European level in form of CORRELATE [54] and should also be considered as a great opportunity in the context of REACH (see [3]). Process: Many areas of basic biomedical research have experienced bumpy rides with periods of hype and disappointment. Toxicology has a continuous high responsibility for human safety and cannot, even transiently, simply drop the precautionary principle. However, it can ask critical questions on how it should best be applied in different situations, exposure scenarios and applicability domains. This provides a basis for a continuous, long-term effort to let toxicology evolve to a higher level than now. This process needs essentially to be global and involve all stakeholders [43]. Despite all enthusiasm, rapid success is not to be expected and all hype should rather be avoided as initial setbacks are likely to happen. This has to be accepted in the strategy. However, the determination to move on needs to be strong enough to attack problems with the right critical mass and impact right from the beginning and as they emerge. Chances: The process of putting regulatory toxicology and the process of toxicity testing on a more mechanistic basis provides a chance for toxicology to evolve as a discipline, and also contribute general biomedical knowledge. This closes the circle started at the beginning of this article. In the past, toxicology had the chance to promote the advance of biomedical sciences in general, for instance by discovering and driving the fields of apoptosis, toxinology or stress response.

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

However, these opportunities were not seized, and other sciences drove these fields instead. Now, new chances are arising, possibly in the fields of systems biology, DNA repair or pathological aging. Possibly also in the fields of chemical genetics and the introduction of chemical screens to non-pharmaceutical areas. To grasp any of these chances, it is important to dare to take the lead and not to lose touch with basic science. Application of HTS or qHTS as described above sounds fancy, but it is at the moment only a technology, not a science. This technology has brought a lot of disappointment in drug discovery, which one can learn from. It will be important in the future to avoid the mistakes of the past, and to incorporate the “technology” into a robust “scientific concept”, which combines brain with the muscles. Fairness and honesty: An unbiased approach, based on scientific evidence only, will be the best way to find solutions acceptable for all stakeholders. Presently one may wonder what the scientific basis for some animal experiments is. The lousy output and poor information from acute toxicity studies with lethality endpoint has been criticized for a long time [55, 56, 57, and 58], and now, at least in the application domain of drugs, there seems to be a broad agreement that the assay could easily have been abolished [59]. Why hasn’t this already happened? A similar situation can be found for two-generation studies for developmental toxicity testing, where the second generation apparently does not contribute with significant information [60]. Here, non-scientific reasons seem to prevail, and the argument may be expanded to more examples of animal toxicity testing [2]. It is also a sign of poor science that so little pharmacokinetic information is available from acute toxicity tests. This makes the present in vivo–in vitro comparisons very difficult and thus prevents a potential substitution of animal experiments by alternative methods. To be honest, the field of alternative methods also needs to look at obvious weaknesses of its own methods and establish itself as an academic discipline [61]. Many assays are still just as much black box systems as animal experiments and pharmacokinetic information has been terribly neglected. If all sides focus on a vision of best science for best toxicology, then the progress is sure.

Acknowledgements We gratefully acknowledge the valuable input and proofreading of S. Kadereit and S. Schildknecht, the secretarial help by B. Schanze and grant support by the Doerenkamp-Zbinden Foundation, the State of Baden-Württemberg and the European Union (ESNATS).

51

52

Contributions

Abbreviations and Glossary of Terms (Q)SAR: (quantitative) structure–activity relationship. A way to correlate chemical structural information with biological endpoints (e.╃g. receptor binding or toxicity). ASAT: EU initiative on “assuring safety without animal testing”; http://www. asat-initiative.eu. CEFIC: European Chemical Industry Council. DG RESEARCH: Directorate General of the EU for research (http://ec.europa. eu/dgs/research/index_en.html). In national terms this would correspond to the Ministry for Research. It is the major funding body for the large EU framework programme research projects. ECVAM: European Centre for the Validation of Alternative Methods. EPA: Environmental Protection Agency (of the USA). EPAA: European Partnership (of the European Commission and industry organisations) for Alternative Approaches to Animal Testing. InViTech: the EU high-throughput-high content centre; http://bms.jrc. ec.europa.eu/projects/InViTech.htm. MEIC: In 1989, Björn Ekwall and the Scandinavian Society for Cell Toxicology organised the Multicentre Evaluation of In Vitro Cytotoxicity (MEIC). Fifty compounds were evaluated in dozens of cytotoxicity assays and the results were published in a series of papers in 1998 in ATLA. MRC: Medical Research Council of the UK; runs own research institutes, e.╃g. MRC Toxicology Unit in Leicester. NCBI: National Center for Biotechnology Information, a division of the National Library of Medicine (NLM) at the NIH. NCCT: National Center for Computational Toxicology (of the USA). NCGC: NIH Chemical Genomics Centre . NIH: National Institutes of Health (of the USA). NRC: National Research Council (of the USA), the principal operating agency of the National Academies of Sciences of the USA, the National Academy of Engineering and the Institute of Medicine. The National Academy of Sciences is known by many as publisher of the Proceedings of the National Academy of Sciences, USA. NTC: the Netherlands Toxicogenomics Centre; http://toxicogenomics.nl. NTP: National Toxicology Program (of the USA). PubChem: PubChem provides information on the biological activities of small molecules. It is a component of the NIH’s Molecular Libraries Roadmap Initiative. PubMed: Biomedical literature database at the NCBI. qHTS: quantitative high throughput screening. This technology allows the testing of thousands to ten-thousands of compounds in a single experiment. This compound number is 1–2 orders of magnitude lower than what would be used in industrial drug discovery screens. However, the data output is relatively rich, as compounds are screened at about 10 different concentrations and the shape of the resultant response curves yields additional information.

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

REACH: European Regulation (EC) No.╃1907/2006 on the Registration, Evaluation, Authorisation and Restriction of Chemicals, which entered into force on the 1st of June 2007. ZEBET: Zentralstelle zur Erfassung und Bewertung von Ersatz- und Ergänzungsmethoden zum Tierversuch am BfR; Centre for Documentation and Evaluation of Alternatives to Animal Experiments at the BfR (Federal Institute for Risk Assessment).

References 1. Leist, M., Hartung T., Nicotera P. (2008) The dawning of a new age of toxicology. ALTEX 25, 103–114. 2. Hartung, T. (2008â•›a). Food for thought… on animal tests. ALTEX 25, 3–10. 3. Hartung, T. (2008â•›b). Food for thought…on the evolution of toxicology and phasing out of animal testing. ALTEX 25, (this issue). 4. Lotti, M., Nicotera, P. (2002). Toxicology: a risky business. Nature 416, 481. 5. Grandjean, P., Landrigan, P.╃J. (2006). Developmental neurotoxicity of industrial chemicals. Lancet 368, 2167–2178. 6. Ashby, J. (1996). Alternatives to the 2-species bioassay for the identification of potential Â�human carcinogens. Hum. Exp. Toxicol. 15, 183–202. 7. Trosko, J.╃E., Upham, B.╃L. (2005). The emperor wears no clothes in the field of carcinogen risk assessment: ignored concepts in cancer risk assessment. Mutagenesis 20, 81–92. 8. Williams, G.╃M., Whysner, J. (1996). Epigenetic carcinogens: evaluation and risk assessment. Exp. Toxicol. Pathol. 48, 189–195. 9. NRC (2007). Committee on Toxicity Testing and Assessment of Environmental Agents, Â�National Research Council. Toxicity Testing in the 21st Century: A Vision and a Strategy. The national academies press. http://www.nap.edu/ catalog.php?record_id=╃11970. 10. Leist, M., Gantner, F., Naumann, H. et al. (1997). Tumor necrosis factorinduced apoptosis during the poisoning of mice with hepatotoxins. Gastroenterology 112, 923–934. 11. Nicotera, P. (1996). The Gerhard Zbinden Memorial Lecture. Alteration of cell signalling in chemical toxicity. Arch. Toxicol. 18 Suppl., 3–11. 12. Orrenius, S., Zhivotovsky, B., Nicotera, P. (2003). Regulation of cell death: the calcium-apoptosis link. Nat. Rev. Mol. Cell Biol. 4, 552–565. 13. Leist, M, Nicotera, P. (1998). Calcium and neuronal death. Rev. Physiol. Biochem. Pharmacol. 132, 79–125. 14. Kolluri, S.╃K., Weiss, C., Koff, A., Göttlicher, M. (1999). p27(Kip1) induction and inhibition of proliferation by the intracellular Ah receptor in developing thymus and hepatoma cells. Genes Dev.╃13, 1742–1753. 15. Vedani, A., Dobler, M., Lill, M.╃A. (2005). Virtual test kits for predicting harmful effects triggered by drugs and chemicals mediated by specific proteins. ALTEX 22, 123–134.

53

54

Contributions

16. Vedani, A., Dobler, M., Spreafico, M. et al. (2007). VirtualToxLab – in silico prediction of the toxic potential of drugs and environmental chemicals: evaluation status and internet Â�access protocol. ALTEX 24, 153–161. 17. Waring, R.╃H., Harris, R.╃M. (2005). Endocrine disrupters: a human risk? Mol. Cell Endocrinol. 244, 2–9. 18. Krebsfaenger, N., Mürdter, T.╃E., Zanger, U.╃M. (2003). V79 Chinese hamster cells genetically engineered for polymorphic cytochrome P450 2D6 and their predictive value for humans. ALTEX 20, 143–154. 19. Nussler, A.╃K., Wang, A., Neuhaus, P. et al. (2001). The suitability of hepatocyte culture models to study various aspects of drug metabolism. ALTEX 18, 91–101. 20. Ioannides, C., Lewis, D.╃F. (2004). Cytochromes P450 in the bioactivation of chemicals. Curr. Top. Med. Chem. 4, 1767–1788. 21. Collins, F.╃S., Gray, G.╃M., Bucher, J.╃R. (2008). Toxicology. Transforming environmental health protection. Science 319, 906–907. 22. Andersen, M.╃E., Dennison, J.╃E., Thomas, R.╃S. and Conolly, R.╃B. (2005). New directions in incidence-dose modeling. Trends Biotechnol. 23(3), 122– 127. 23. Hendriksen, C.╃F. (2006). Towards eliminating the use of animals for regulatory required vaccine quality control. ALTEX 23, 187–190. 24. Gruber, F.╃P., Hartung, T. (2004). Alternatives to animal experimentation in basic research. ALTEX 21 Suppl. 1, 3–31. 25. Hartung, T. (2001). Three Rs potential in the development and quality control of pharmaceuticals. ALTEX 18 Suppl. 1, 3–13. 26. Seiler, A., Buesen, R., Hayess, K. et al. (2006). Current status of the embryonic stem cell test: the use of recent advances in the field of stem cell technology and gene expression analysis. ALTEX 23 Suppl., 393–399. 27. Whitlow, S., Bürgin, H., Clemann, N. (2007). The embryonic stem cell test for the early selection of pharmaceutical compounds. ALTEX 24, 3–7. 28. Zbinden, G. (1988). Reduction and replacement of laboratory animals in toxicological testing and research. Interim report 1984–1987.╃Biomed. Environ. Sci.╃1, 90–100. 29. Zbinden, G. (1990). Alternatives to animal experimentation: developing invitro methods and changing legislation. Trends Pharmacol. Sci.╃11, 104–107. 30. Kavlock, R.╃J., Ankley, G., Blancato, J. et al. (2007). Computational Toxicology – A State of the Science Mini Review. ToxSci. Advance Access published on December 7, 2007.╃doi:10â•›1093/toxsci/kfm297. 31. Xia, M., Huang, R., Witt, K.╃L. et al. (2008â•›a). Compound cytotoxicity profiling using quantitative high-throughput screening. Environ. Health Perspect. 116, 284–291. 32. Xia, M. et al. (2008â•›b). The 1408 compounds of Environ Health Perspect 116, 284.╃http://www.epa.gov/ncct/dsstox/sdf_ntphts.html#DownloadTable. 33. Houck, K., Dix, D., Judson, R. et al. (2008). DSSTox EPA ToxCast High Throughput Screening Testing Chemicals Structure-Index File: SDF File and Documentation, Updated version: TOXCST_v2b_320_08Feb2008, http:// www.epa.gov/ncct/dsstox/sdf_toxcst.html.

Visions on Toxicity Testing in the 21st Century: Reflections on a Strategy Document of the US€�National Research Council

34. Dix, D.╃J., Houck, K.╃A., Martin, M.╃T. et al. (2007). The ToxCast program for prioritizing toxicity testing of environmental chemicals. Tox. Sci.╃95, 5–12. 35. Hartung, T., Bremer, S., Casati, S. (2004). A Modular Approach to the ECVAM Principles on Test Validity. ATLA 32, 467–472. 36. Zbinden, G. (1991). Predictive value of animal studies in toxicology. Regul. Toxicol. Pharmacol. 14, 167–177. 37. Hartung, T. (2007â•›a). Food for thought on… validation. ALTEX 24, 67–73. 38. Mathews, R.╃A.╃J. (2008). Medical progress depends on animal models – doesn’t it? J.╃R. Soc. Med.╃101, 95–98. 39. Hartung, T. (2007â•›b). Food for thought … on cell culture. ALTEX 24, 143– 147. 40. Bremer, S., Pellizzer, C., Hoffmann, S. et al. (2007). The development of new concepts for assessing reproductive toxicity applicable to large scale toxicological programmes. Curr. Pharm. Des.╃13(29), 3047–3058. 41. Rogers, M.╃D. (2003). Risk analysis under uncertainty, the precautionary principle, and the new EU chemicals strategy. Regul. Toxicol. Pharmacol. 37, 370–381. 42. Russell, W.╃M. and Burch, R.╃L. (1959). The Principles of Humane Experimental Technique. London: Methuen. 43. Bottini, A.╃A., Amcoff, P., Hartung, T. (2007). Food for thought… on globalization. ALTEX 24, 255–261. 44. DB-Alm (2007). Invittox protocol number 101.╃http://ecvam-dbalm.jrc. ec.europa.eu/public_view_doc2.cfm?id=╃6E7E72104B2DEFD6BE979B3B139176C67180BB0BC12CB10496CDA74B54630A05A3291B895581F634. 45. REACH (2006). REACH legislation under directive (EC) No€ 1907/2006, http://eur-lex.europa.eu/JOHtml.do?uri=OJ:L:2006:396:SOM:en:HTML. 46. EPAA (2006). European Partnership to Promote Alternative Approaches to Animal Testing http://ec.europa.eu/enterprise/epaa/conf_2006_presentationvdgraaf_unilever.pdf. 47. Combes, R., Grindon, C., Cronin, M.╃T. et al. (2008). Integrated decision-tree testing strategies for acute systemic toxicity and toxicokinetics with respect to the requirements of the EU REACH legislation. ATLA 36(1), 45–63. 48. Grindon, C., Combes, R., Cronin, M.╃T. et al. (2008â•›a). An integrated decisiontree testing strategy for repeat dose toxicity with respect to the requirements of the EU REACH legislation. ATLA 36(1), 93–101.╃PMID: 18333717. 49. Grindon, C., Combes, R., Cronin, M.╃T. et al. (2008â•›b). Integrated decisiontree testing strategies for developmental and reproductive toxicity with respect to the requirements of the EU REACH legislation. ATLA 36(1), 65– 80. 50. Knight, A. (2007). Animal experiments scrutinised: systematic reviews demonstrate poor human clinical and toxicological utility. ALTEX 24(4), 320–325. 51. Hoffmann, S., Hartung, T. (2006). Toward an evidence-based toxicology. Hum. Exp. Toxicol. 25, 497–513.

55

56

Contributions

52. Guzelian, P.╃S., Victoroff, M.╃S., Halmes, N.╃C. et al. (2005). Evidence-based toxicology: a comprehensive framework for causation. Hum. Exp. Toxicol. 24, 161–201. 53. Halle, W. (2003). The Registry of Cytotoxicity: toxicity testing in cell cultures to predict acute toxicity (LD50) and to reduce testing in animals. ATLA 31, 89–198. 54. Correlate (2007). http://projects-2007.jrc.ec.europa.eu/show.gx?Object. object_id=PROJECTS0000000003008C51. 55. Tamborini P, Sigg H, Zbinden G. (19990). Acute toxicity testing in the nonlethal dose range: a new approach. Regul Toxicol Pharmacol. 12, 69–87. 56. Zbinden, G. (1986). Invited contribution: acute toxicity testing, public responsibility and scientific challenges. Cell Biol. Toxicol. 2, 325–335. 57. Paget, E. (1983). The LD50 test. Acta Pharmacol. Toxicol. (Copenh.) 52 Suppl.€2, 6–19. 58. Zbinden, G., Flury-Roversi, M. (1981). Significance of the LD50-test for the toxicological evaluation of chemical substances. Arch. Toxicol. 47, 77–99. 59. Robinson, S., Delongeas, J.╃L., Donald, E. et al. (2007). A European pharmaceutical company initiative challenging the regulatory requirement for acute toxicity studies in pharmaceutical drug development. Regulatory Toxicol. Pharmacol. (in press). doi:10â•›1016/j.yrtph.2007.11â•›009 60. Janer, G., Hakkert, B.╃C., Slob, W. et al. (2007). A retrospective analysis of the two-generation study: what is the added value of the second generation? Reprod. Toxicol. 24, 97–102. 61. Leist, M. (2006). What can a chair on alternatives to animal experimentation effectuate? ALTEX 23, 211–213.

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

3.2 Safety Assessment of Botanicals and Botanical Â�Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety AuthorityTiered Approach Gerrit Speijers1, Bernard Bottex2, Birgit Dusemund3, Andrea Lugasi4, Â� Jaroslav Toth5, Judith Amberg-Müller6, Corrado Galli7, Vittorio Silano8, and Ivonne M.╃C.╃M. Rietjens9 This manuscript was originally published in Mol. Nutr. Food Res., 2010, 54(2): page 175–185.

Abstract The present paper describes results obtained by testing the EFSA tiered guidance approach for safety assessment of botanicals and botanical preparations intended for use in food supplements. Main conclusions emerging are: (1) Botanical ingredients must be identified by their scientific (binomial) name, in most cases down to the subspecies level or lower. (2) Adequate characterization and description of the botanical parts and preparation methodology used is needed. Safety of a botanical ingredient cannot be assumed only relying on the long-term safe use of other preparations of the same botanical. (3) Because of possible adulterations, misclassifications, replacements or falsifications and restorations, establishment of adequate quality control is necessary. (4) The strength of the evidence underlying concerns over a botanical ingredient should be included in the safety assessment. (5) The matrix effect should be taken into

1

General Health Effects Toxicity and Safety Food, Winterkoning 7, NL-3435 RN Nieuwegein, Netherlands.

2

European Food Safety Authority, Scientific Committee and Advisory Forum Unit, Largo N. Palli 5A, I-43100 Parma, Italy.

3

Federal Institute for Risk Assessment, Department of Food Safety, Thielallee 88–92, D-14195 Berlin, Germany.

4

National Institute for Food and Nutrition Science, Gyáli út 3/a, H-1097 Budapest, Hungary.

5

Comenius University, Faculty of Pharmacy, Department of Pharmacognosy and Botany, Odbojárov 10, SK-83232 Bratislava, Slovak Republic.

6

Federal Office of Public Health, Food Safety Division, Nutritional and Toxicological Risks Section, Stauffacherstrasse 101, CH-8004 Zuerich, Switzerland.

7

University of Milan, Department of Pharmacological Sciences, Laboratory of Toxicology, Via Balzaretti 9, I-Milan, Italy.

8

National Institute for the Promotion of Migrant’s Health and the Control of Poverty-related diseases, I-Rome, Italy.

9

Correspondence to: Prof. Dr. ir. Ivonne M.╃C.╃M. Rietjens, Wageningen University, Division of Toxicology, Tuinlaan 5, NL-6703 HE Wageningen, The Netherlands, Tel: +╃31 317 483971, Fax: +╃31 317 484931, [email protected].

57

58

Contributions

account in the safety assessment on a case-by-case basis. (6) Adequate data and adequate methods for appropriate exposure assessment are often missing. (7) Safety regulations concerning toxic contaminants have to be complied with. The application of the guidance approach can result in the conclusion that safety can be presumed, that the botanical ingredient is of safety concern or that further data are needed to assess safety.

3.2.1 Introduction In June 2004 the Scientific Committee of the European Food Safety Authority (EFSA) adopted a discussion paper on botanicals and botanical preparations widely used as food supplements and related products [1]. First it was noted that the expanding market volume raises the need for a better characterization of botanicals and botanical preparations, and for harmonization of the scientific assessment of risks from exposure of consumers to these products. In addition EFSA launched, via its Advisory Forum, a questionnaire to the national food safety authorities of the European countries, to get a clearer picture of the extent of the issue in Europe. After responses to the questionnaire were received work was undertaken to (1) analyze the information provided by 25 European countries in response to the questionnaire; (2) prepare a guidance document on how to assess the safety of botanicals and botanical preparations intended for use in food supplements; and (3) establish a list of main categories of botanicals and used parts thereof (compendium) in order to prioritize the botanical preparations to be considered for a safety assessment. The draft safety guidance document and compendia thus prepared were revised following a public consultation and the updated draft guidance document was published on the EFSA website [2]. After the draft guidance document was published, EFSA concluded that it was necessary to test the proposed approach for the safety assessment of botanicals and botanical preparations to be used as ingredients in food supplements with a selected number of cases and to further update the compendium. To this end, on 15 April 2008, an EFSA Scientific Cooperation (ESCO) Working Group, composed of experts identified by EFSA and by the European Member States, was established, in order to (1) enlarge the information basis underlying the compendium of botanicals reported to contain toxic, addictive, psychotropic or other substances of concern; (2) test the proposed tiered approach for the safety assessment of botanicals and botanical preparations with a selected number of botanicals as real-case examples; and (3) provide a report summarizing the outcome of the case studies as well as to advise on the adequacy of the proposed approach for the safety assessment of botanicals and botanical preparations. The present paper aims at providing an overview of the outcome of the second above-mentioned task of this ESCO working group and describes especially the major issues emerging when testing the proposed tiered approach through a selected number of cases. Table 1 summarizes the case studies selected and

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach Table 1: Overview of the selected cases tested using the guidance document and the possible safety issues expected to be linked with these examples. Botanical

Preparation

Possible safety issue

Triticum aestivum L.

Wheat Bran

Low concern – presumption of safety

Citrus aurantium L. ssp. aurantium L.

Hydroalcoholic extract of dried peel

Misidentification/ adulteration

Camellia sinensis (L.) O. Kuntze

Dried green tea extract

Liver toxicity

Foeniculum vulgare Mill. ssp vulgare var. vulgare

Seeds and oil from the seeds

Carcinogenicity

Ocimum tenuiflorum L.

Dry leaves extract

Reproductive toxicity

Linum usitatissimum L.

Dried ripe seeds

Phytoestrogenic activity

presents an overview of the possible safety issues expected to be linked with these real case examples.

3.2.2 Materials and Methods Safety evaluation of the selected botanicals was performed using the proposed draft guidance document published on the EFSA website [2]. This guidance document indicates that data underlying a safety assessment of a botanical or botanical preparation should include technical data on the identity and nature of the source material, the manufacturing process, the chemical composition, specifications, stability of the botanical (preparation) used as ingredient in food supplements, proposed uses and use levels, information on existing assessments, exposure data including anticipated exposure and cumulative exposure, modality of use, as well as information on historical use and toxicological data. Data used when testing the guidance document for the safety evaluation of the selected botanicals were collected from the open literature and were not intended to be complete. The work aimed at testing the proposed tiered approach for the safety assessment of botanicals and botanical preparations with selected cases considering relevant constituents of concern within the botanical or botanical preparation. The evaluation was not aiming at providing a formal safety assessment of the botanical or its preparations, since each example focused on one type of preparation only. Once the outcome of this testing exercise has been considered for updating the draft guidance document for the safety assessment of botanicals and botanical preparations intended for use as ingredients in food supplements, EFSA published the reports summarizing the outcome of the case studies, together with the updated guidance document and the compendium on its website. The conclusions and recommendations of the present paper reflect those of its authors as individual scientists and not necessarily represent the views of EFSA.

59

60

Contributions

3.2.3 Results Triticum aestivum L. (wheat bran): Triticum aestivum L. (wheat bran) was chosen as an example of a botanical of low safety concern. Wheat bran is a by-product obtained in the manufacture of wheat flour from the grain. It consists mainly of the outer layers of the wheat kernel, including the aleuron layer i.╃e. the husk, seed, coat and germ. This example already reveals that when evaluating botanicals and botanical preparations it is essential though not always easy to adequately define and characterize the actual preparation being evaluated. Some selected wheat cultivars and varieties as well as genetically engineered wheat varieties exist [3]. The latter would fall under specific already existing European regulatory framework [4, 5]. Furthermore, selection of better tolerated wheat varieties by patients affected by gluten-induced enteropathy (celiac disease) has been described [6]. This illustrates that the botanical needs to be identified by its scientific name (binomial name, i.╃e. genus, species, subspecies, author), including the part of the plant used. Wheat bran may be obtained, in addition to hexaploid wheats, also from tetraploid wheats, namely Triticum turgidum L. ssp. durum (Desf.) Husn. (= syn. T. durum Desf., English common names: durum wheat, hard wheat), which further supports the need to adequately define the botanical evaluated by its full scientific name. Furthermore, there may be changes induced in bran composition by wheat growing conditions including high temperature stress and solar radiation [7, 8], and the composition of bran can vary depending on the milling process as well. Information on the milling process and the resulting size of the bran particles is of interest since these factors affect the biological properties, as illustrated by a particle size-dependency of the laxative effect and colonic fermentation [9]. This is why not only the scientific name but also information on the manufacturing procedure and chemical specifications of the botanical or botanical preparation are essential to adequately define the preparation to be evaluated. Another important issue emerging when evaluating this first example was the possible presence of contaminants in botanical preparations. Wheat bran must conform to the provisions of food regulations (Council Regulation 315/93/ EEC) [10], especially in terms of mycotoxins [11] arising from external fungal contamination (Fusarium spp.) [12], microbiology and pesticides. For instance, a maximum level for the trichothecene mycotoxin deoxynivalenol of 500╃µg/kg in cereal products was proposed by the Codex alimentarius [13], and the maximum level for sclerotia of Claviceps purpurea is set at 0.05â•›% m/m for wheat [14]. Fungal contamination and mycotoxin production cannot be totally eliminated at present [15]. In particular, mycotoxin contamination from Fusarium spp. is the result of a minor infection of grains and their envelopes by the fungi which may be transferred during the milling process [16–21]. Whereas for wheat bran the situation with respect to these contaminants may be well recognized and is even regulated at some extent, this may not hold true for other botanicals and botanical preparations.

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

Citrus aurantium L. ssp. aurantium L. (bitter orange): Citrus aurantium L. ssp. aurantium L. (bitter orange) is an example of a botanical for which it is necessary to define the botanical or botanical preparation down to the subspecies level or even lower given that different subspecies may vary in the constituents and the level of substances of concern. Citrus aurantium L. ssp. aurantium L. (bitter orange) as compared to Citrus aurantium L. ssp. bergamia (Risso & Poit.) Engl. (bergamot orange) produces different fruits that contain different levels of biologically active principles such as furanocoumarins and para-synephrine (p-synephrine) [22, 23]. The case of Citrus aurantium L. ssp. aurantium L. was chosen since it represents the issue of misidentification and/ or adulteration which are matters for considerable concern with respect to the safety of botanicals and botanical preparations. Exposure to bitter orange peel and its constituents occurs primarily via ingestion of the fruit itself or its products (e.╃g. orange juice, marmalade, and dietary supplements). Bitter orange peel is added to various foods (e.╃g. beer, liquors and other beverages and cakes). Moreover, bitter orange juice may be added in limited amounts to sweet orange juice. Exposure can also result from peel oil used in aromatherapy and flavouring. Several evaluations of Citrus aurantium L. ssp. aurantium L. have concluded that there is no safety concern related to the regular food use of bitter orange [24, 25]. However, more recent evaluations concerning preparations containing high amounts of the sympathomimetic alkaloid p-synephrine concluded that there may be a possible safety concern [26, 27]. Bitter orange extracts used in food supplements, such as weight-loss pills, are possibly enriched in p-synephrine, typically to an amount of 6–10â•›% (but even extracts with a content of 95â•›% p-synephrine are documented) [23, 28– 32]. Thus, extracts used in many dietary supplements and herbal weight-loss formulas as an alternative to Ephedra have concentrations of p-synephrine that are often much higher than the p-synephrine concentrations reported for traditional extracts of the dried fruit or peel. This reflects another important issue to be taken into account when assessing the safety of botanical ingredients, i.╃e. that some preparations of a botanical may be marketed containing significantly higher levels of active (toxic) principles than those normally occurring in historical food uses of the same botanical. Furthermore, the position isomer of synephrine found in bitter orange peel is pâ•‚synephrine, not mâ•‚synephrine. Metaâ•‚synephrine (mâ•‚synephrine) and neo-synephrine are relatively rare synonyms of the compound named phenylephrine in the International Non-proprietary Name (INN) list of the WHO. Phenylephrine is used as a decongestant synthetic drug [33]. At least one product purportedly containing synephrine alkaloids from Citrus aurantium has been reported to contain both pâ•‚synephrine and mâ•‚synephrine [34, 35]. There is no evidence that octopamine or other phenethylamine alkaloids are present in bitter orange peel in any appreciable levels, although their increased content has been reported in some extracts and herbal products on the market [29–32]. The presence of any amounts of m-synephrine, higher amounts of the (+)-psynephrine stereoisomer or higher amounts of octopamine in food supplements

61

62

Contributions

supposedly containing only extracts or alkaloid fractions of Citrus aurantium L. ssp. aurantium L. should be considered undesirable, and suspicious of adulteration. The origin of these compounds is unlikely the natural botanical source (Citrus aurantium L. ssp. aurantium L.), thus strongly suggesting a requirement for a more efficient quality control. Camellia sinensis (L.) O. Kuntze (green tea): Dried green tea extracts prepared from young leaves and leave buds from Camellia sinensis (L.) O. Kuntze are used as food, including beverages and food supplements, and as pharmaceuticals. Uses as a food include a stimulant drink in the form of ready-to-drink beverages or of beverages prepared by the consumer from instant green tea powder. While the worldwide long-term consumption of traditional green tea infusions is assumed to be safe, a weight-loss product containing a high-dosed hydroalcoholic extract of green tea was marketed only until April 2003, when the French and Spanish authorities suspended the market authorization given its hepatotoxic side-effects [36, 37]. Some data point at epigallocatechingallate (EGCG) as the ingredient of concern in relation to the hepatotoxicity of green tea extracts but this relationship is not firmly established. In green tea, EGCG is a major constituent in terms of quantity and a constituent useful to characterize the quality of the preparation, besides caffeine, theanine and other catechins [38, 39]. Thus, the case of Camellia sinensis revealed that when evaluating the safety of a botanical or botanical preparation it can be difficult to identify the constituent or group of constituents of concern. In other cases it may be difficult to identify the active principle responsible for an effect, and therefore it is concluded that the strength of the evidence underlying the concerns over a compound being reason for concern should be given in a safety assessment of the respective botanical. Furthermore, the case of green tea reflects that different preparations from the same botanical source material can have a different outcome in the safety evaluation, especially since use of the different preparations may result in difference in composition and consequently in consumer exposure. Thus, regular intake of dried green tea extracts with food supplements or related products differs from the intake resulting from use of traditional green tea infusions (or beverages with identical composition). Dried aqueous green tea extracts, which are manufactured under the same extraction conditions as applied in the traditional preparation of green tea infusions and which are used to prepare solid or liquid food supplements may be evaluated based on their EGCG content and the daily exposure resulting from their proposed uses and use levels. In food supplements and related products the active green tea ingredients and particularly EGCG, which is associated with hepatotoxic concern, are available in a more concentrated form making higher dosage and bolus administration more likely than with the aforementioned beverages. Cases of liver disorders associated with intake of products containing dried aqueous green tea extract [37, 40, and 41] have to be taken into consideration. Moreover, the green tea example indicates that the matrix effect should be taken into account. When given in a green tea extract to rats EGCG appears to

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

be eliminated less readily from the body [42] and to have a higher toxicity than when given as a pure compound [43]. In addition, studies in healthy volunteers point at a reduced bioavailability of EGCG in the presence of a food matrix, showing that administration of concentrated green tea extracts under fasting conditions lead to a significant increase of plasma concentrations of EGCG compared to administration with food [44]. Thus the example of Camellia sinensis reflects the importance of the matrix effect that should be taken into account in the safety assessment of botanicals and botanical preparations. The use of dried green tea extracts in beverages or food supplements for weight reduction purposes under fasting conditions or reduced food intake might require adequate safety data accounting for the increased bioavailability in the absence of the food matrix effect. This applies as well to products containing dried green tea extracts as a part of the ingredient. Foeniculum vulgare Mill. ssp. vulgare var. vulgare: Foeniculum vulgare Mill. ssp. vulgare var. vulgare (bitter fennel) was selected as one of the real cases to be evaluated given that it contains estragole, an ingredient that in animal experiments and at certain concentrations showed both genotoxic and carcinogenic activity. Fruits from Foeniculum vulgare Mill. ssp. vulgare var. vulgare contain 2–6â•›% essential oil [45, 46]. The major constituent of the essential oil is trans-anethole at levels between 50–75â•›%, and the essential oil contains estragole at levels amounting to 3.5 to 12â•›% [46]. Estragole is an alkenylbenzene that is of safety concern given its reported carcinogenic effect at high dose levels [47]. In the safety assessment of botanicals and botanical ingredients a major issue is the question of how to deal with botanicals and botanical ingredients that contain chemicals that are both genotoxic and carcinogenic. The EFSA draft guidance document [2] states that in cases where the botanical ingredient contains substances that are both genotoxic and carcinogenic, the “Margin of Exposure” (MOE) approach [48] could be applied covering the botanical(s) under examination and any other dietary sources of exposure. The MOE approach compares animal toxic effect levels with human exposure levels. The guidance document states that alternatively, it could be evaluated whether the expected exposure to the genotoxic and carcinogenic ingredient will not be significantly increased compared to the intake from multiple sources. This implies that further data are required with respect to the assessment of the risk posed by the estragole levels present in bitter fennel fruits and their extracts including an estimate of the MOE. The MOE approach uses a reference point, usually taken from data from an animal experiment that represents a dose causing a low but measurable cancer response denoted the Benchmark Response (BMR). It can be for example the BMDL10, the lower confidence bound of the Benchmark Dose that gives 10â•›% (extra) cancer incidence (BMD10). The MOE is defined as the ratio between the BMDL10, and the estimated dietary intake (EDI) in humans. To date, carcinogenicity data for estragole from which a BMDL10, and thus a MOE, can be derived result from a long-term carcinogenicity study conducted in mice [49]. An accompanying paper of the present special issue reports a BMD analysis of these data using

63

64

Contributions

BMDS version 1.4.1â•›c software resulting in a BMDL10 value for estragole that varies between 9 and 33 mg/kg bw and day [50]. This value can be compared to, for example, estimated intake levels resulting from the use of bitter fennel fruits for the preparation of fennel tea. The exposure to estragole from bitter fennel fruits can be estimated based on the assumption that 4.5 to 7.5 gram (3 times 1.5 to 2.5â•›g) of fennel fruits per day would be used for the preparation of fennel tea. Assuming that fruits contain 5â•›% essential oil, that the extraction efficiency of the essential oil is 25–35↜狀%, and that there is 3.5 to 12â•›% estragole in the oil, this would imply an intake of 1.9 to 15.8 mg estragole per day. For a 60 kg person this amounts to an estragole exposure from tea consumption that amounts to 33 to 263╃µg/kg bw and day. Using the BMDL10 values of 9 to 33 mg/kg bw and day for female mice as derived from the Miller et al. study [49, 50] one can calculate a MOE in the range of 34 to 1000 which indicates that use of bitter fennel fruits for the preparation of fennel tea could be considered a high priority for risk managers [48]. In addition, the example of bitter fennel reflects the possibility to use, in cases where the botanical ingredient of concern is not genotoxic and carcinogenic, the acceptable daily intake (ADI) for the safety assessment. The safety of the intake of trans-anethole from use of bitter fennel fruits can be judged using the temporary ADI of 0–2.0 mg/kg bw and day for trans-anethole derived by JECFA [51]. The exposure to trans-anethole from bitter fennel fruits can be estimated based on the assumption that 4.5 to 7.5 gram (3 times 1.5 to 2.5â•›g) of fennel fruits per day would be used for the preparation of fennel tea. Assuming that fruits contain 5â•›% essential oil, that the extraction efficiency of the essential oil is 25–35â•›%, and that there is 50–75â•›% trans-anethole in the oil, this would imply an intake of 28 to 98 mg trans-anethole per day. For a 60 kg person this amounts to an intake of 0.5 to 1.6 mg trans-anethole/kg bw and day. This is below the above mentioned ADI established by JECFA. However, as the exposure to trans-anethole resulting from the use of bitter fennel fruits for the preparation of fennel tea already amounts to 25 to 80â•›% of the ADI, a possibility exists for exceeding the ADI due to other sources of trans-anethole. The case of bitter fennel further highlights the uncertainties associated with the kinetics as well as the expression of the inherent toxicity of a naturally occurring substance, i.╃e. estragole, possibly related to effects induced by the matrix. The question may be raised, whether studies with pure compounds dosed by gavage without the normal food matrix being present, represent a good starting point for the risk assessment of botanical ingredients. An illustrative example can be given for sweet basil which contains high amounts of estragole in the essential oil. Jeurissen et al. [52] demonstrated that the level of DNA binding of the proximate carcinogenic metabolite 1’-hydroxyestragole to DNA in vitro but also to DNA in intact HepG2 human hepatoma cells could be inhibited by a methanolic basil extract. It was demonstrated that the inhibition by the basil extract occurs at the level of the sulfotransferase mediated bioactivation of 1’-hydroxyestragole to 1-sulfoxyestragole [52]. Although it remains to be established whether a similar inhibition will occur in vivo, the inhibition of sulfotransferase mediated bioactivation of 1’-hydroxyestragole by basil ingredi-

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

ents suggests that the possibilities for bioactivation and subsequent adverse effects may be lower when estragole is dosed in a matrix of other basil ingredients than what would be expected on the basis of experiments dosing estragole as a single compound. Where a matrix effect is advocated to support the safety of specific levels of compounds (e.╃g. that data from a pure compound may overestimate effects of the compound in the botanical matrix), testing and/ or other data should be provided to demonstrate the occurrence of the matrix effect of the preparation and its magnitude. It is important to realize that when a matrix effect is demonstrated for an essential oil this matrix effect will not be similar for the intact botanical. Thus, the example of bitter fennel containing estragole supports that research on individual substance–matrix interactions cannot be used to draw general conclusions about intact botanicals, herbs and spices under all conditions of use, ingestion and metabolism and that the matrix effect should be judged on a case-by-case basis. Ocimum tenuiflorum L. (holy basil): Ocimum tenuiflorum L. (holy basil) was included in the evaluations representing an example of a botanical that may be of concern given its possible reproductive toxicity. There is, however, no information on actual constituents likely responsible for this effect. Only a few scientists attempted to look into the various changes in the reproductive system in detail after feeding Ocimum tenuiflorum L. leave extract and there is considerable debate regarding the histopathological changes in reproductive organs following the feeding of Ocimum tenuiflorum L. leaves [53–55]. In addition to the concerns over possible reproductive toxicity which need further testing, Ocimum tenuiflorum L. contains methyleugenol, an alkenylbenzene known to be both genotoxic and carcinogenic [56]. An Ocimum tenuiflorum L. leaf extract may contain up to 86â•›% methyleugenol [57–59]. Further details on an MOE assessment for methyleugenol, in line with what was done for estragole in the real-case example on Foeniculum vulgare Mill. ssp. vulgare var. vulgare, can be found in the literature [60]. Finally, the case of Ocimum tenuiflorum L. indicates once more the importance of defining the correct scientific name of a botanical to be evaluated. Ocimum tenuiflorum L. is the correct scientific name, but most publications still make use of the synonym Ocimum sanctum. Linum usitatissimum L. (flaxseed): Flax is known to be the richest food source of plant lignans including secoisolariciresinol diglucoside. This plant lignan is a precursor of the mammalian lignans, enterodiol and enterolactone and converted into these forms via the activity of colonic facultative aerobes [61]. Other lignans such as matairesinol and lariciresinol are also found in flaxseed. 100â•›g dry flaxseed contain about 300 mg lignans, including pinoresinol (∼870╃µg), syringaresinol (∼48╃µg), lariciresinol (∼1780╃µg), secoisolariciresinol (SEC, ∼165 mg), matairesinol (MAT, ∼529╃µg), and hydroxymatairesinol (HMR, ∼35╃µg), all expressed as aglycons [62, 63]. Phyto-oestrogens represent a family of plant compounds that have been shown

65

66

Contributions

to have both oestrogenic and anti-oestrogenic properties. Lignans, similarly to isoflavonoids and coumestans, are often referred to as phyto-oestrogens, and may possess oestrogen receptor agonistic or antagonistic properties, with unclear effects on hormone-sensitive cancers such as breast, uterine, and prostate cancer. Pharmacodynamic studies suggest that there might be an oestrogenic or anti-oestrogenic effect of flaxseed [64]. Some authors therefore call mammalian lignans modulators of endogenous sex steroid hormones. Since 1981, when mammalian lignans were identified in human urine, evidence supporting their role as modulators of endogenous sex steroid hormones has increased. However, the most convincing results have come from in vitro, animal and epidemiological studies, whereas results of the few intervention studies that have been conducted have been mixed [65–69]. Therefore, further research, including in particular long-term intervention trials, is needed to provide clarification for this relationship [70, 71]. The case of flax seed demonstrates that even when the compounds of concern are clearly identified the actual evidence for the effects may be controversial and may require further testing. So in this case it is clear that the strength of the evidence underlying the concerns over a botanical ingredient should be included in the safety assessment and that an evaluation based on the available knowledge can result in the conclusion that further data are requested.

3.2.4 Discussion Testing the EFSA draft guidance document for the safety assessment of botanicals and botanical preparations intended for use as food supplements [2], through its application to several selected real cases, has revealed many specific issues to be taken into account in the safety evaluation of botanicals and botanical preparations intended to be used as ingredients in food supplements and has led to a set of suggested amendments of the guidance document, in addition to its validation. The scheme proposed for safety assessment of botanicals and botanical preparations intended for use as ingredients in food supplements other than novel foods and GMOs (for which specific sectoral regulations exist) has been amended as shown in Figure 1.╃The safety assessment approach is a tiered approach starting with the evaluation on available knowledge (level A) in compliance with the criteria described in the EFSA guidance document as amended on the basis of the results of the tests described in the present paper. A level A assessment can result in the conclusion that safety can be presumed based on available knowledge (like for Triticum aestivum bran or some of the Camellia sinensis extracts), but it could also lead to the conclusion that the ingredient is of safety concern. If needed, the assessment should continue with further experimental studies, following guidance provided in the EFSA document, to obtain additional data required to reach a conclusion on safety (level B). The level B assessment may result in the conclusion that either the product is of safety concern or that the botanical or botanical preparation is not of safety concern.

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

Figure 1: Scheme proposed for the safety assessment of botanicals and botanical preparations not regulated in the framework of specific regulations such as those on novel foods and GMOs. The safety assessment could include a tiered approach starting with a safety assessment based on available knowledge (level A) and the need to continue with further testing to obtain additional data (level B).

This guidance document is of importance to harmonize such an approach across Europe: In fact, in spite of the extensive harmonization that occurred through the EU Food Law, the safety assessment of food supplements based on botanicals and botanical preparations has remained a competence of each EU Member State. The outcomes of the safety evaluations of the selected cases using the EFSA guidance document will be published on the EFSA website as annexes of the advice of the ESCO Working Group on the adequacy of the proposed EFSA approach for the safety assessment of botanicals and botanical preparations. The most important issues emerging when performing the safety assessment on the selected botanicals can be summarized as follows (Tab.╃2). The botanical ingredient needs to be identified by its scientific name (binomial name, i.╃e. genus, species, subspecies, author), and the part of the plant used. In most cases it will be necessary to define the botanical down to the subspecies level or even lower given that different subspecies or varieties mostly vary in the constituents and the level of toxic principles. Examples are Foeniculum vulgare Mill ssp. vulgare var. dulce (sweet fennel) versus var. vulgare (bitter fennel) with the essential oil of the former containing about 10 times lower levels of estragole than the latter, and Citrus aurantium L. ssp. aurantium L. (bitter orange) versus Citrus aurantium L. spp. bergamia (Risso & Poit.) Engl. (bergamot orange) that produce fruits containing different levels of active principles such as furanocoumarins and p-synephrine. In other cases, however, it is possible to

67

68

Contributions Table 2: Overview of issues emerging when testing the tiered approach for the safety assessment of botanicals and botanical preparations used as ingredients in food supplements. Recommendation

Rationale

Botanical example

Botanical ingredients must be identified by their scientific (binomial) name, in most cases down to the subspecies level or lower.

Different subspecies or �varieties mostly vary in the constituents and the level of toxic principles.

Foeniculum vulgare Mill ssp. vulgare var. dulce versus var. vulgare Citrus aurantium L. ssp. aurantium L. versus Citrus aurantium L. spp. bergamia (Risso & Poit.) Engl.

Adequate characterisation and description of the botanical parts and preparation methodology used is needed. Each safety evaluation should focus on a well-defined species (or subspecies or variety), a well-defined part of the plant, and a well defined preparation.

Different preparations can be obtained from the chosen parts of a specific botanical. Composition of a botanical may vary significantly due to factors that cannot be easily controlled. Safety of a botanical ingredient cannot be assumed only relying on the long-term safe use of other preparations of the same botanical.

Camellia sinensis; different green tea preparations result in different outcome of the safety evaluation.

Establishment of adequate quality control is necessary.

Adulterations, misclassifications, replacements or falsifications and restorations may occur.

Citrus aurantium L. ssp. aurantium L. containing meta-synephrine which is not naturally occurring in bitter orange fruits.

The strength of the evidence underlying concerns over a botanical ingredient should be included in the safety assessment.

It is often difficult to identify the constituent or group of constituents in a botanical or botanical preparation that is responsible for the safety concern.

No firm link between hepatotoxicity and EGCG from leaves of Camellia sinensis. Controversial evidence adverse effects of phyto-oestrogens from seeds of Linum usitatissimum L. (flaxseed).

The matrix effect should be taken into account on a caseby-case basis.

The kinetics and toxicity of a naturally occurring substance can be modified by the surrounding.

Foeniculum vulgare Camellia sinensis.

Adequate data and methods for exposure assessment are needed.

Exposure assessment often plays a decisive role in the outcome of the safety assessment of the botanical or botanical preparations.

Margin of exposure (MOE) to the intake of estragole and margin of safety (MOS) to ADI for intake of trans-anethole from Foeniculum. vulgare

Safety regulations concerning toxic contaminants have to be complied with. Specifications should include maximum levels for possible contaminants, e.╃g. pesticide residues, mycotoxins, heavy metals and PAHs, according to existing guidelines for foods.

Some contaminants may arise from the manufacturing process and need to be kept within safety limits.

Presence of polycyclic aromatic hydrocarbons (PAHs) in dried preparations,

evaluate a variety of subspecies on the basis of one representative species. For example, this would be the case for rose hips, the spurious fruits of dog rose

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

(Rosa canina L.), alpine rose (Rosa pendulina L.) and other Rosa species, most commonly Rosa rugosa Thumb; the ripe hips of the different species are collected in late autumn and differ only slightly in their form as well as in the content of their main active constituent, ascorbic acid. Many different preparations can be obtained from the chosen parts of a specific botanical, depending on a number of factors including, for example, the solvents used and the extraction process. The example of green tea preparations from Camellia sinensis demonstrates that while the consumption of traditional infusions is assumed to be safe, toxicological concerns have been associated with certain extracts intended for weight-loss purposes. The composition of a botanical may vary significantly due to other factors that cannot be easily controlled; e.╃g. concentrations of active ingredients measured in the plant material may show significant variation with geographical origin, plant maturity at harvest, harvesting techniques, storage conditions, processing (e.╃g. drying) and method of detection. Therefore, adequate description is needed, not only of the botanical subspecies or variety evaluated, but also of the harvesting and manufacturing process. It is concluded that safety of a botanical ingredient cannot be assumed only relying on the long-term safe use of different preparations of the same botanical source, but that it is necessary to rely on well characterized preparations. Each safety evaluation should focus on a well-defined species (or subspecies or variety), a well-defined part of the plant, and a well defined preparation. Adulteration may occur. Manufacturers may add for example to Citrus aurantium L. ssp. aurantium L. preparations synthetic p-synephrine or isomers like meta-synephrine (also called phenylephrine or neosynephrine) which is not naturally occurring in bitter orange fruits. This will not become evident when in the specifications only known ingredients are listed and quantified. Furthermore, in some countries restoration of botanical preparations is allowed and may be part of the manufacturing process, i.╃e. addition of volatile ingredients lost in the manufacturing process to a dry extract. Given these aspects, the establishment of adequate quality control methods is necessary. It is often difficult to identify the constituent or group of constituents in a botanical or botanical preparation that is responsible for the safety concern. An example is EGCG from leaves of Camellia sinensis which is quantitatively a major constituent and useful to characterize the quality of the preparation, but for which no firm link has been established with the hepatotoxicity of the dried green tea extracts. Furthermore, the case of the seeds of Linum usitatissimum L. (flaxseed) indicates that, even when compounds of concern have been clearly identified, the evidence for their effects may be controversial and require further testing. The strength of evidence underlying the concerns over a botanical ingredient should be, therefore, included in the safety assessment. The matrix effect should be taken into account in the safety assessment of botanicals and botanical preparations. It is plausible that the kinetics as well as the expression of the inherent toxicity of a naturally occurring substance can be modified by the surrounding matrix. Depending on the mechanism of action of the substance and the nature of the matrix, this could result in the toxicity of

69

70

Contributions

the specific substance being unchanged, reduced or even increased. Research data on individual substance–matrix interactions, mainly available in vitro on specific botanical preparations, cannot be used to draw general conclusions applicable to intact botanicals or other preparations or in vivo. Where a matrix effect is advocated to support the safety of a botanical ingredient of specific levels of compounds, ad hoc test data should be provided to demonstrate the real occurrence of the matrix effect in that preparation and its magnitude. A matrix effect should be judged on a case-by-case basis as described in this paper for Foeniculum vulgare and Camellia sinensis. While working through the real cases the outcomes of the exposure assessment most often appeared to play a decisive role in the outcome of the safety assessment of the botanical or botanical preparations. For example, the margin of exposure (MOE) to the intake of estragole from the consumption of Foeniculum vulgare as well as judging whether the intake of trans-anethole from the use of bitter fennel fruits for the preparation of fennel tea would remain below the ADI for trans-anethole depends on the outcome of the exposure assessment. However, data on present uses and use levels of a botanical or botanical preparation may be sparse or lacking and other uncertainties may be in some cases unavoidable. Botanical food ingredients must be obviously in compliance with regulations on contaminants. Some contaminants such as for example polycyclic aromatic hydrocarbons (PAHs) in dried preparations may arise from the manufacturing process and need to be kept within safety limits. Therefore, specifications should include maximum levels for possible contaminants, e.╃g. pesticide residues, mycotoxins, heavy metals and PAHs, according to existing guidelines for foods. It should be pointed out that, although being outside the EFSA mission, there are other important issues that need to be considered to ensure safety of food supplements. These include: (1) the over-the-counter availability of food supplements through internet sites from countries where regulations are not in place or not aligned to European standards; and (2) the fact that the control systems in place to guarantee the safety and quality of botanical supplements are not well harmonized among different countries. The latter is of particular concern as some products on the market are known to be of variable quality with high variation in the content of the active and/ or the toxic principles, and due to the fact that already examples of replacement of a harmless variety with a toxic alternative have occurred [72–75]. Misidentification of plants harvested from the wild may add to the problem. The growing volume of products and sales call for a more formal pre-marketing assessment and better and stricter controls than at present. Regulatory bodies have become aware of the problem and are increasing their efforts to ensure the safety of botanical supplements€[1]. A last consideration is related to consumer information and empowerment which would make it possible to reduce phenomena such as over-consumption of food supplements by particular groups and the fact that many consumers equate “natural” with “safe” when considering botanical food supplements.

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

Acknowledgement The work reported in the paper was carried out in the framework of the EFSA ESCO working group on Botanicals and Botanical Preparations with the support of the EFSA Scientific Committee & Advisory Forum Unit. The ESCO working group was composed of: Experts nominated by the EFSA Scientific Committee: Robert Anton, Angelo Carere, Luc Delmulle, Corrado L. Galli, Ivonne Rietjens, Vittorio Silano and Gerrit Speijers. Experts nominated by the members of the EFSA Advisory Forum: Ilze Abolina, Judith Amberg-Müller, Ulla Beckman-Sundh, Birgit Dusemund, MarieHélène Loulergue, Andrea Lugasi, Martijn Martena, Maria Nogueira, Kirsten Pilegaard, Mauro Serafini, Jaroslav Toth, Arnold Vlietinck and Magdalini Zika. The conclusions and recommendations of the present paper reflect those of its authors as individual scientists and not necessarily represent the views of EFSA.

List of Abbreviations ADI: Acceptable Daily Intake BMD: Benchmark Dose BMDL: Lower confidence bound of the Benchmark Dose BMR: Benchmark Response ECGC: Epigallocatechingallate EDI: Estimated daily Intake EFSA: European Food Safety Authority ESCO: EFSA Scientific Cooperation JECFA: Joint FAO/WHO Expert Committee on Food Additives MOE: Margin of Exposure PAHs: polycyclic aromatic hydrocarbons.

Conflict of Interest Statement The authors declare that there are no financial/commercial conflicts of interest.

References â•⁄ 1. Scientific Committee of the European Food Safety Authority, Discussion paper on “botanicals and botanical preparations widely used as food supplements and related products: coherent and comprehensive risk assessment and consumer information approaches”. Brussel 2004, pp.╃1–6. http://www.efsa.eu.int/science/sc_commitee/sc_documents/616/scdoc_advice03_botanicals_en1.pdf.╃

71

72

Contributions

â•⁄ 2. EFSA (European Food Safety Authority), Draft guidance document of the Scientific Committee for the safety assessment of botanicals and botanical preparations intended for use as ingredients in food supplements, 2008.╃http:// www.efsa.europa.eu/EFSA/efsa_locale-1178620753812_1178717026833. htm.╃ â•⁄ 3. Peterson, R.╃K., Shama, L.╃M., A comparative risk assessment of genetically engineered, mutagenic, and conventional wheat production systems. Transgenic Res., 2005, 14, 859–875. â•⁄ 4. EFSA (European Food Safety Authority), Guidance document of the Scientific Panel on Genetically Modified Organisms for the risk assessment of genetically modified plants and derived food and feed, EFSA Journal, 2006, 99, 1–100. http://www.efsa.europa.eu/EFSA/efsa_locale-1178620753812_ 1178620775747.htm â•⁄ 5. Regulation (EC) 1829/2003 on GM food and feed, and Directive 2001/18/EC on the release of GMOs into the environment. http://eur-lex.europa.eu/JOIndex.do?year=╃2003&serie=L&textfield2╃=╃26 8&Submit=Search. â•⁄ 6. Spaenij-Dekking, L., Kooy-Winkelaar, Y., Van Veelen, P., Drijfhout, J.╃W., Jonker, H., Van Soest, L., Smulders, M.╃J., Bosch, D., Gilissen, L.╃J., Koning, F., Natural variation in toxicity of wheat: potential for selection of nontoxic varieties for celiac disease patients. Gastroenterology 2005, 129, 797–806. â•⁄ 7. Zhou, K., Su, L., Yu, L.╃L., Phytochemicals and antioxidant properties in wheat bran. J. Agric. Food Chem. 2004, 52, 6108–6114. â•⁄ 8. Zhou, K., Yu L., Antioxidant properties of bran extracts from Trego wheat grown at different locations. J. Agric. Food Chem. 2004, 52, 1112–1117. â•⁄ 9. Jenkins, D., Kendall, C., Vuksan, V., Augustin, L., Li, Y-M., Lee, B., Mehling, C., Parker, T., Faulkner, D., Seyler, H., Vidgen, E., Fulgoni III, V., The effect of wheat bran particle size on laxation and colonic fermentation. J.╃Am. Col. Nutr. 1999, 18, 339–345. 10. Council Regulation (EEC) No.╃315/93 of 8 April 1993 laying down Community procedures for contaminants in food. Official Journal of the European Union L 37, 13.╃2. 1993, p.╃1.╃http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=CONSLEG:1993R0315:20031120:EN:PDF 11.╃ WHO, Selected mycotoxins: ochratoxins, trichothecenes, ergot. Environmental Health Criteria, 1990, 105, 263 pages. 12.╃ Mueller, H.╃M., Metzger, K.╃U., Modi, R., Reimann, J., Ergosterol and Fusarium toxins in wheat bran and wheat. J. Anim. Physiol. Anim. Nutr. 1994, 71, 48–55. 13.╃Visconti, A., Haidukowski, E.╃M., Michelangelo Pascale, M., Silvestri, M., Reduction of deoxynivalenol during durum wheat processing and spaghetti cooking. Toxicol. Lett. 2004, 153, 181–189. 14.╃ Codex Alimentarius 1995, Codex standard for wheat and durum wheat. CODEX STAN 199–1995.╃www.codexalimentarius.net/download/standards/62/CXS_199â•›e.pdf. 15. Codex Alimentarius 2003, Code of practice for the prevention and reduction of mycotoxin contamination in cereals, including annexes on ochra-

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

toxin a, zearalenone, fumonisins and tricothecenes. CAC/RCP 51–2003: 1–8. 16. Brera, C., Catano C., De Santis, B., Debegnach, F., De Giacomo, M., Pannunzi, E., Miraglia, M. Effect of industrial processing on the distribution of aflatoxins and zearalenone in corn-milling fractions. J. Agricul. Food Chem. 2006, 54, 5014–5019. 17.╃ Brera, C., Debegnach F., Grossi S., Miraglia, M., Effect of industrial processing on the distribution of fumonisin B-1 in dry milling corn fractions. J. Food Protec. 2004, 67, 1261–1266. 18.╃ Broggi, L.╃E., Resnik S.╃L., Pacin, A.╃M., GonzaÂlez, H.╃H.╃L., Cano, G., Taglieri, D., Distribution of fumonisins in dry-milled corn fractions in Argentina. Food Add. Contam. 2002, 19, 465–469. 19.╃ Ryu, D., Jackson, L.╃S., Bullerman, L.╃B., Effects of processing on zearalenone. Mycotoxins Food Safety 2002, 504, 205–216. 20.╃ Trigo Stockli, D.╃M., Deyoe, C.╃W., Satumbaga, R.╃F., Pedersen, J.╃R., Distribution of deoxynivalenol and zearalenone in milled fractions of wheat. Cereal Chem.1996, 73, 388–391. 21.╃ Rafai, P., Bata, A., Jakab, L., Vanyi, A., Evaluation of mycotoxin-contaminated cereals for their use in animal feeds in Hungary. Food Add. Contam. 2000, 17, 799–808. 22.╃ Gardana, C., Nalin, F., Simonetti, P. Evaluation of flavonoids and furanocoumarins from Citrus bergamia (bergamot) juice and identification of new compounds. Molecules 2008, 13, 2220–2228. 23.╃ Avula, B., Upparapalli, S.╃K., Navarrete, A., Khan, I.╃A., Simultaneous quantification of adrenergic amines and flavonoids in C. aurantium, various Citrus species, and dietary supplements by liquid chromatography. J. AOAC Intern. 2005, 88, 1593–1606. 24.╃ Fugh-Berman, A., Myers, A., Citrus aurantium, an ingredient of dietary supplements marketed for weight loss: current status of clinical and basic research. Exp. Biol. Med.╃2004, 229, 698–704. 25.╃ Haaz, S., Fontaine, K.╃R., Cutter, G., Limdi, N., Perumean-Chaney, S., Allison, D.╃B., Citrus aurantium and synephrine alkaloids in the treatment of overweight and obesity: an update. Obesity Rev.╃2006, 7, 79–88. 26.╃ Calapai, G., Firenzuoli, F., Saitta, A., Squadrito, F., Arlotta, M.╃R., Costantino, G., Inferrera G., Antiobesity and cardiovascular toxic effects of Citrus aurantium extracts in the rat: A preliminary report. Fitoterapia 1999, 70, 586–592. 27.╃ Jack, S., Desjarlais-Renaud, T., Pilon, K. Bitter orange or synephrine: update on cardiovascular adverse reactions. Canadian Adverse Reaction Newsletter 2007, 17, 2–3.╃ 28.╃Avula, B., Upparapalli, S.╃K., Khan, I.╃A., Simultaneous analysis of adrenergic amines and flavonoids in Citrus peel jams and fruit juices by liquid chromatography: part 2.╃J. AOAC Intern. 2007, 90, 633–640. 29.╃ Pellati, F., Benvenuti, S., Melegari, M., Firenzuoli, F., Determination of adrenergic agonists from extracts and herbal products of Citrus aurantium L. var. amara by LC. J. Pharmac. Biomed. Anal. 2002, 29, 1113–1119.

73

74

Contributions

30.╃ Pellati, F., Benvenuti, S. Chromatographic and electrophoretic methods for the analysis of phenethylamine alkaloids in Citrus aurantium. J. Chromatog. A 2007, 1161, 71–88. 31.╃ NTP/NIEHS. Bitter orange (Citrus aurantium var. amara) extracts and constituents (±)-p-synephrine [CAS No.╃94–07–5] and (±)-p-octopamine [CAS No.╃104–14–3]. Review of toxicological literature. National Toxicology Program (NTP), 2004, National Institute of Environmental Health Sciences (NIEHS), National Institutes of Health, US Department of Health and Human Services, Research Triangle Park, North Carolina, VIII +╃73â•›p. http:// ntp.niehs.nih.gov/ntp/htdocs/Chem_Background/ExSumPdf/Bitterorange. pdf. 32. Blumenthal, M., Bitter orange peel and synephrine: Part 1 & Part 2.╃Herbal Gram 2005, 66, 0 (web-only exclusive). http://content.herbalgram.org/abc/ herbalgram/articleview.asp?a=╃2833&p=Y.╃ 33. Gaglia, C.╃A. Jr., Phenylephrine hydrochloride. Analyt. Prof. Drug Subst. 1974, 3, 483–512. 34.╃ Allison, D.╃B., Cutter, G., Poehlman, E.╃T., Moore, D.╃R., Barnes, S., Exactly which synephrine alkaloids does Citrus aurantium (bitter orange) contain? Intern. J. Obesity 2005, 29, 443–446. 35.╃ Santana, J., Sharpless, K.╃E., Nelson B.╃C., Determination of para-synephrine and meta-synephrine positional isomers in bitter orange-containing dietary supplements by LC/UV and LC/MS/MS. Food Chem. 2008, 109, 675–682. 36.╃ AFSSAPS, Agence Française de Sécurité Sanitaire des Produits de Santé, 2003, Communiqué de Presse. Suspension de l’autorisation de mise sur le ® marché de la spécialité pharmaceutique EXOLISE (gallate d’épigallocatéchol). 37.╃ Sarma, D.╃N., Barrett, M.╃L., Chavez, M.╃L., Gardiner, P., Ko, R., Mahady, G.╃B., Marles, R.╃J., Pellicore, L.╃S., Giancaspro, G.╃I., Dog, T.╃L., Safety of green tea extracts. A systematic review by the US pharmacopeia. Drug Safety 2008, 31, 469–484. 38.╃ Isbrucker, R.╃A., Edwards, J.╃A., Wolz, E., Davidovich, A., Bausch, J., Safety studies on epigallocatechin gallate (EGCG) preparations. Part 2: dermal, acute and short-term toxicity studies. Food Chem. Toxicol. 2006, 44, 636– 650. 39.╃ USDA database for the flavonoid content of selected foods. 2007, US Department of Agriculture, Agricultural Research Service. 40.╃ Kantelip, J.╃P., Laroche, D., Green tea and liver disorders. National Drug Surveillance survey submitted to the Technical Committee. Besancon: Besancon regional drug surveillance centre, 11 February, 2003. 41.╃ Canadian Adverse Reaction Newsletter Green tea extract (Green Lite): suspected association with hepatotoxicity. 2007,17, 1–3. 42.╃Chen, L., Lee, M-J., Li, H., Yang, C.╃C., Absorption, distribution, and elimination of tea polyphenols in rats. Drug Metab. Dispos. 1997, 25, 1045–1050. 43.╃ Johnson, W.╃D., Morrissey, R.╃L., Crowell, J.╃A., McCormick, D.╃L., Subchronical oral toxicity of green tea polyphenols in rats and dogs. The Toxicologist 1999, 48, 57–58.

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

44.╃ Chow, H.╃H., Hakim, I.╃A., Vining, D.╃R., Effects of dosing condition on the oral bioavailability of green tea catechins after single-dose administration of Polyphenon E in healthy individuals. Clin. Cancer Res.╃2005, 11, 4627– 4633. 45.╃ European Pharmacopoeia (2005) Fennel, Bitter – Foeniculi amari fructus. Council of Europe. 5th ed., 01/2005:0824. 46.╃ Council of Europe, 2006, Active principles (constituents of toxicological concern) contained in natural sources of flavourings. Approved by the Committee of Experts on Flavouring Substances, October 2005, Health Protection of the Consumer Series. Council of Europe Press, Strasbourg. http://www.coe.int/t/e/social_cohesion/soc-sp/public_health/Flavouring_ substances/Active%20principles.pdf.╃ 47. SCF, 2001, Opinion of the Scientific Committee on Food on estragole (1-allyl-4-methoxybenzene). http://europa.eu.int/comm/food/fs/sc/scf/ out104_en.pdf. 48. EFSA, 2005, Opinion of the scientific committee on a request from EFSA related to a harmonised approach for risk assessment of substances which are both genotoxic and carcinogenic. http://www.efsa.europa.eu/en/science/sc_commitee/sc_opinions/1201.html. 49.╃ Miller, E.╃C., Swanson, A.╃B., Phillips, D.╃H., Fletcher, T.╃L., Liem, A., Miller, J.╃A., Structure-activity studies of the carcinogenicities in the mouse and rat of some naturally occurring and synthetic alkenylbenzene derivatives related to safrole and estragole. Cancer Res.╃1983, 43, 1124–1134. 50.╃ Rietjens, I.╃M.╃C. M., Punt, A., Schilter, B., Scholz, G., Delatour, T., Van Bladeren, P.╃J., In-silico methods for physiologically based biokinetics (PBBK models) describing bioactivation and detoxification of coumarin and estragole; implications for risk assessment. Mol. Nutr. Food Res., 2010, 54 (2), 195–207. 51.╃JECFA, 1998, trans-Anethole (addendum), In: Safety evaluation of certain food additives, prepared by the 51st meeting of JECFA, FAS 42-JECFA 51/5, p.╃5–32.╃http://www.inchem.org/documents/jecfa/jeceval/jec_137.htm. 52. Jeurissen, S.╃M.╃F., Punt, A., Delatour, Th., Rietjens, I.╃M.╃C. M., Basil extract inhibits the sulfotransferase mediated formation of DNA adducts of the procarcinogen 1’-hydroxyestragole by rat and human liver S9 homogenates and in HepG2 human hepatoma cells. Food Chem. Toxicol. 2008, 46, 2296–2302. 53.╃ Ahmed, M., Khan, M.╃Y. Khan, A.╃A. Effects of Ocimum sanctum (Tulsi) on the reproductive system, an updated review. Biomed. Res.╃2002, 13, 63–67. 54.╃ Ahmed, M., Ahamed, N., Aladakatti, R.╃H. Ghosesawar, M.╃G., Reversible anti-fertility effect of benzene extract of Ocimum sanctum leaves on sperm parameters and fructose content in rats. J. Basic Clin. Physiol. Pharmacol. 2002, 13, 51–59. 55.╃ Reghunandan, R., Sood, S., Reghunandan, V., Arora, B.╃B., Gopinathan, K., Mahajan, K.╃K., Effects of feeding Ocimum sanctum (Tulsi) leaves on fertility in rabbits. Biomed. Res. Aligarh, 1997, 8, 187–191.

75

76

Contributions

56.╃ SCF, 2001, Opinion of the Scientific Committee on Food on methyleugenol (4-allyl-1,2-dimethoxybenzene). http://ec.europa.eu/food/fs/sc/scf/out102 _en.pdf. 57. WHO, 2002, Folium Ocimi Sancti, WHO monographs on selected medicinal plants, 2 , 206–216.╃World Health Organization, Geneva, Switzerland. 58.╃ Blaschek, W., Hänsel, R., Keller, K., Reichling, J., Rimpler, H. , Schneider, G. (Eds.), 1998.╃Hager’s Handbuch der pharmazeutische Praxis. Folgeband 2: Drogen A-K, 5th ed. Springer-Verlag, Berlin, Germany. 59.╃ Lal, R.╃N., Sen, T.╃K., Nigam, M.╃C., Gas chromatography of the essential oil of Ocimum sanctum L., Parfümerie und Kosmetiks 1978, 59, 230–231. 60.╃ Rietjens, I.╃M.╃C. M., Slob, W., Galli, C., Silano, V., Risk assessment of botanicals and botanical preparations intended for use in food and food supplements: Emerging issues. Toxicol. Lett., 2008, 180, 131–136. 61.╃ Thompson, L.╃U., Robb, P., Serraino, M., Cheung, F., Mammalian lignan production from various foods. Nutr. Cancer 1991, 16, 43–52. 62.╃ Smeds, A.╃I., Eklund, P.╃C., Sjöholm, R.╃E., Willför, S.╃M., Nishibe, S., Deyama, T., Holmbom, B.╃R., Quantification of a broad spectrum of lignans in cereals, oilseeds, and nuts. J. Agric. Food. Chem. 2007, 55, 1337–1346. 63.╃ Milder, I.╃E.╃J., Arts, I.╃C.╃W., Van de Putte, B., Venema, D.╃P., Hollman, P.╃C.╃H., Lignan contents of Dutch plant foods: a database including lariciresinol, pinoresinol, secoisolariciresinol, and matairesinol. Br. J. Nutr., 2005, 93, 393–402. 64.╃ Adlercreutz, H., Phyto-oestrogens and cancer. Lancet Oncol, 2002, 3, 364– 373. 65.╃ Hutchins, A.╃M., Martini, M.╃C., Olson, B.╃A., Thomas, W., Slavin, J.╃L., Flaxseed consumption influences endogenous hormone concentrations in postmenopausal women. Nutr. Cancer 2001, 39, 58–65. 66.╃ Phipps, W.╃R., Martini, M.╃C., Lampe, J.╃W., Slavin, J.╃L., Kurzer, M.╃S., Effect of flax seed ingestion on the menstrual cycle. J. Clin. Endocrinol. Metab. 1993, 77, 1215–1219. 67.╃ Kurzer, M.╃S., Lampe, J.╃W., Martini, M.╃C., Adlercreutz, H., Fecal lignan and isoflavonoid excretion in premenopausal women consuming flaxseed powder. Cancer Epidemiol. Biomarkers Prev. 1995, 4, 353–358. 68.╃ Thompson L.╃U., Chen J.╃M., Li T., Strasser-Weippl K., Goss P.╃E., Dietary flaxseed alters tumor biological markers in postmenopausal breast cancer. Canada Clin. Cancer Res, 2005, 11, 3828–3835. 69.╃ Dodin S., Lemay A., Jacques H., Legare F., Forest J-C., Masse B., The effects of flaxseed dietary supplement on lipid profile, bone mineral density, and symptoms in menopausal women: a randomized, double-blind, wheat germ placebo-controlled clinical trial. J. Clin. Endocrin. Metab. 2005, 90, 1390–7. 70. Adlercreutz, H., Hoeckerstedt, K., Bannwart, C., Bloigu, S., Hämäläinen, E., Fotsis, T., Ollus, A., Effect of dietary components, including lignans and phytoestrogens, on enterohepatic circulation and liver metabolism of estrogens and on sex hormone binding globulin (SHBG). J. Steroid Biochem. 1987, 27, 1135–1144.

Safety Assessment of Botanicals and Botanical �Preparations Used as Ingredients in Food Supplements: Testing an European Food Safety Authority-Tiered Approach

71.╃ Adlercreutz, H., Does fiber-rich food containing animal lignan precursors protect against both colon and breast cancer? An extension of the „fiber hypothesis“. Gastroenterology 1984, 86, 761–764. 72.╃ Vanherweghem, J.-L., Depierreux, M., Tielemans, C., Abramowicz, D., Dratwa, M., Jadoul, M., Richard, C., Vandervelde, D., Verbeelen, D., Vanhaelen-Fastre, R., et al., Rapidly progressive interstitial renal fibrosis in young women: association with slimming regimen including Chinese herbs. Lancet 1993, 341, 387–391. 73.╃ Vanhaelen, M., Vanhaelen-Fastre, R., But, P., Vanherweghem, J.-L., Identification of aristolochic acid in Chinese herbs. Lancet 1994, 343, 174. 74.╃ Oudesluys-Murphy, A.╃M., Oudesluys, N., Tea: not immoral, illegal, or fattening, but is it innocuous? Lancet 2002, 360, 878. 75.╃ Johanns, E.╃S.╃D., van der Kolk, L.╃E., van Gemert, H.╃M.╃A., Sijben, A.╃E., et al., An epidemic of epileptic seizures after consumption of herbal tea. Ned. Tijdschr. Geneesk. 2002, 146, 813–816.

77

78

Contributions

3.3 In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure–Activity Relationship Analysis Luis G. Valerio Jr.1, Naomi L. Kruhlak2, and R. Daniel Benz2

Abstract In silico toxicology employs evidence-based methods using chemoinformatics and advanced computational analyses in evaluations by regulatory agencies done for safety assessments of drug-related substances (active pharmaceutical ingredients, metabolites, impurities), the industry drug discovery process, safety assessment of indirect food additives, environmental agents, and other applied uses of value for protecting public health. At the US Food and Drug Administration (FDA), in silico approaches harness vast experimental chemical toxicity test data that exist in regulatory archives to predict toxicological end points of regulatory interest on the basis of the classic structure–activity (SAR) paradigm, probabilistic evidence-based reasoning, machine learning methods, data mining, and human rules integrated into computational software programmes. The in silico approach can provide more affordable and time efficient alternatives to traditional studies (e.╃g., to support a decision in the event of equivocal evidence of toxic potential from laboratory results), and may reduce the use of animals in some circumstances and aid in risk management for the prioritization of chemicals requiring safety testing. The specific use of advanced in silico methods, including predictive quantitative structure–activity relationship (QSAR) analysis and chemoinformatic data mining software, for estimating the rodent carcinogenic potential of phytochemicals present in botanicals, herbs, and natural dietary sources, is addressed in the context of an external validation study. External validation is the most stringent scientific method of measuring QSAR predictive performance. The external validation statistics measuring performance for predicting rodent carcinogenic potential of a dataset of phytochemicals is presented based on two different computational software programmes in current use at the FDA. How the FDA, Center for Drug Evaluation and Research, Office of Pharmaceutical Science, uses chemoinformatics and computational toxicology software including QSAR modelling to predict the ability of phar-

1

Correspondence address and presentation: Luis G. Valerio, Jr., Ph.╃D., US Food and Drug Administration, Center for Drug Evaluation and Research, Science and Research Staff, Office of Pharmaceutical Science, White Oak 51, Room 4128, 10903 New Hampshire Ave., Silver Spring, MD 20993–0002, USA, Fax: +╃1 301 796 9997, [email protected].

2

US Food and Drug Administration, Center for Drug Evaluation and Research, Office of Testing and Research, Office of Pharmaceutical Science, White Oak 64, 10903 New Hampshire Ave., Silver Spring, MD 20993–0002,╃USA.

In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure–Activity Relationship Analysis

maceuticals, their metabolites, impurities, and degradation products to cause toxicity in animals is also covered.

3.3.1 Introduction The Science and Research Staff (SRS) is a component of the Science and Research Staff of the Office of Pharmaceutical Science at the US FDA/ Center for Drug Evaluation and Research (CDER). SRS provides specialized support in the development and applied regulatory use of in silico (computer-based) toxicology methods at CDER. SRS has established a consortium of collaborators who are engaged in harvesting data from FDA archives and creating quantitative structure–activity relationship (QSAR) computational toxicology models of animal toxicological and human health effect end points. Computational databases of animal toxicology studies and human clinical trial and surveillance data have been compiled and the non-proprietary portions are being made publicly available through the collaborators. SRS provides computational toxicology safety evaluations of drugs, metabolites, contaminants, and degradants to CDER reviewers to support regulatory decision-making. SRS supports CDER regulatory decision-making by using computational information from chemically similar substances to generate a predicted toxicological and adverse effect profile for a compound to help support regulatory decision-making. Such practice provides additional scientific evidence to help in safety assessment and identify and eliminate compounds with potentially significant adverse effect properties early in the drug discovery and development process. Current SRS research is focused on developing strategies for the use of computational toxicology models for regulatory purposes, evaluating new computational software approaches, and validating predictive performance of the models. For a recent review of SRS computational toxicology activities at the FDA please refer to the article by Yang [1].

3.3.2 Why Use In Silico Predictive Models at FDA? One of SRS’ important roles at CDER is to provide predictions based on computational analyses of chemically similar substances that can help identify toxicity in the early stages of the drug review process. These responsibilities and activities are consistent with the FDA Critical Path Initiative [2]. Launched in March 2004, the FDA Critical Path Initiative is the agency‘s effort to stimulate and facilitate, via modernized approaches, the scientific process through which a potential human drug, biological product, medical device, or food additive is transformed from a discovery or “proof of concept” into a medical product or food ingredient (http://www.fda.gov/oc/initiatives/criticalpath/). Computational toxicology has been acknowledged as playing a role in FDA’s Critical Path.

79

80

Contributions

In 2007, the US National Research Council (NRC) published a report entitled, “Toxicity Testing in the 21st Century: A Vision and Strategy” at the request of the US Environmental Protection Agency (EPA) [3]. The report has sparked tremendous interest as it reviewed established toxicology methodologies, and discussed the use of alternative approaches and emerging technologies as a vision and strategy to increase efficiency and relevance of toxicity testing in risk assessment. In Chapter 4 of the report, computational toxicology and in silico SAR-based methods are specifically discussed as a future and long-term strategy for toxicity testing in risk assessment. Although addressing environmental agents, the report may also be applied across substances of regulatory importance including pharmaceutical molecules, food ingredients, and phytochemicals. Further rationale for use of in silico approaches includes the advantage of high efficiency. In silico technologies permit the rapid screening of thousands of chemicals in minutes, and are very cost-effective compared to conventional animal toxicology studies. For example, the cost to conduct the rodent 2-year cancer bioassay in two species and genders can run upwards of $╃3 million dollars, and this does not even take into consideration the time and resources involved in generating reports of results derived from a standardized protocol such as that of the US National Toxicology Program. By comparison, the acquisition of a battery of computational toxicology software may cost but a fraction of the amount of one 2-year bioassay and, moreover, can be used in multiple experiments to screen more chemicals. The other attraction of in silico methods for predicting toxicity and aiding in risk assessment of chemicals is the potential net effect of saving animals used for laboratory testing. The European Commission’s legislation known as REACH (Registration, Evaluation, Authorisation, and restriction of Chemicals) is a current and widely referred to example of the reality in reducing the use of animals in toxicity testing [4]. Under REACH, no animal test should be performed if it can be replaced with other techniques such as reliable in silico predictions. Through this law, millions of test animals may be spared if (Q)SAR methods performed by computers are accepted for REACH purposes. There are international standards in place in order to determine whether a QSAR method is acceptable for regulatory purposes. The Organisation for Economic Co-operation and Development (OECD) has established five principles that need to be addressed for a QSAR model to be acceptable: (1) A defined end point, (2) a defined (or unambiguous) algorithm, (3) a defined domain of applicability, (4) appropriate measures of goodness of fit, robustness, and predictivity, and (5) mechanistic interpretation, if possible [4]. Because of the aforementioned regulatory initiatives, and the increased usage and development of in silico approaches, improvements in predictive accuracy have been evolving. This is evident in recent external validation studies for predicting rodent carcinogenicity which have reported performance statistics above 90â•›% for sensitivity [5], while other models have been built and reported to perform more optimally with high specificity (>80â•›%) [6]. Rarely do models accomplish both high sensitivity and specificity, so a choice must be made about

In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure–Activity Relationship Analysis

which predictive parameters are most important for a particular regulatory application. Among other topics, this is being addressed by a newly formed CDER committee, the Pharmacology and Toxicology Coordinating Committee (PTCC) Computational Toxicology Subcommittee (CTCS). The purpose of the PTCC CTCS is to disseminate the appropriate guidance to CDER review staff and the pharmaceutical industry on the use and assessment of computational toxicology studies. The PTCC CTCS serves as an internal resource to the CDER Office of New Drugs and Office of Pharmaceutical Science on scientific and regulatory aspects of computational toxicology issues. Another reason to use in silico methods is that recently, it has been proposed that computational modelling can be a tool consistent with the principles and concept of evidence-based toxicology due to the objectivity with which chemoinformatic approaches can be applied [7]. The fact that computational approaches can provide a systematic analysis of high quality test data is recognized [7]. However, although in silico methods may be a useful tool for evidence-based toxicology for specific causation analysis, these methods cannot in themselves establish causation.

3.3.3 What In Silico Predictive Models does the FDA Use? In building computational models, non-proprietary and proprietary laboratory toxicity testing results have been mathematically transformed to construct global QSAR models that use 2-dimensional molecular fragment and descriptorbased approaches in the analysis of chemical structure to predict a battery of pre-clinical in vitro and animal end points important to regulatory safety assessments. In addition, recent work has centred on models designed to predict clinical end points and human adverse effects of pharmaceuticals [8]. The toxicological and clinical end points modelled and planned for development at CDER are listed in Table 1. The model validation methods that have been used in the put begin with conducting 100â•›% internal cross-validation. In internal cross-validation experiments, a 10╃×╃10â•›% procedure is conducted where chemicals that are part of the training dataset are randomly assigned to 10 equally-sized validation test Table 1: Computational modelling suites available and planned for development at the US FDA/CDER Office of Pharmaceutical Science. Non-clinical effect models

Human clinical adverse effect models

Carcinogenicity Genetic toxicity Reproductive toxicity Behavioural toxicity Phospholipidosis Quantitative MTD (vs. time)1 Subchronic toxicity2 Chronic rodent studies2

Hepatobililary Renal/ Bladder Cardiological Pulmonary1 Immunological1 Other organ systems2 Maximum recommended daily dose Molecular mechanism of action Human drug metabolism

1 under current development 2 planned for development

81

82

Contributions

sets (10â•›% each). In each iteration of the validation process, one of these sets of compounds is removed from the training dataset and a new model is constructed with the remaining 90â•›%. The compounds (10â•›%) are then used as a test set to run against the new model and the process is repeated 10 times until all of the validation compounds are used once as a test chemical. When additional data are available for a given end point, an external validation study is performed using a balanced set of compounds that were never part of the model. It is widely recognized that external validation studies are of the highest rigor and stringency in testing predictive performance of a QSAR model [5], but it is also acknowledged that previously unmodelled compounds can sometimes be poorly represented by the domain of applicability of the model. This reinforces the need for a suitable measure of molecular coverage. Data from external validation study of the predictive performance of models for the highly sought after toxicological end point of rodent carcinogenicity are addressed in this presentation.

3.3.4 What In Silico Predictive Software does the FDA Use? Through FDA-approved Cooperative Research and Development Agreements (CRADAs), Material Transfer Agreements (MTAs), Research Collaboration Agreements (RCA), and licensing, CDER is involved in long-term evaluation of and research with computational toxicology software for development of predictive (Q)SAR models, as well as their use in internal computational toxicology consultation services to CDER reviewers [1]. CDER’s current in silico toolbox of (Q)SAR predictive toxicity screens includes the software programmes listed in Table 2.╃It is important to point out that the FDA does not endorse any computational software product including the ones listed in Table 2.╃These software products are listed only because the FDA has entered into agency-approved collaborative research agreements such as RCAs, CRADAs, MTAs, or licensing with the companies supporting these software programmes. These agreements are intended to evaluate the utility of the software programmes and do not endorse any of them. Table 2: Computational toxicology software used in applied research at the US FDA/CDER Office of Pharmaceutical Science. Computational Toxicology Software Programme1

Product Developer

BioEpisteme® Derek for Windows and Meteor Leadscope Model Applier and Predictive Data Miner MC4PC and Meta MetaDrug™ and MetaTox™ SciQSAR (MDL-QSAR)

Prous Institute for Biomedical Research Lhasa Limited Leadscope® Inc.

1â•… No endorsement implied. Listed in alphabetical order.

MultiCASE Inc. GeneGo Inc. Scimatics Inc.

In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure–Activity Relationship Analysis Table 3: Computational toxicology software platforms1 to estimate rodent carcinogenic potential. Leadscope Model Applier

MC4PC

SciQSAR

BioEpisteme

Derek for Windows

QSAR Methodology

Partial Logistic Statistical Regression (PLR) Analysis Algorithm/Statistical Analyses

Discriminant Analysis

Genetic Algorithm/ Statistical Analyses

Human Expert Rules

Structural Interpretation

Fingerprint Molecular Features/ Scaffolds

2–10 Atom Molecular Fragments

Molecular Descriptors

Molecular Descriptors

Structural Alert Molecular Features (Fragments)

2D/3D Molecular Descriptors

2D (n~10)

2D (n~6)

2D (n~200, Kier and Hall)

3D (n~126, 2D (n~4) 2D set of descriptors; 3D is a future functionality)

Training Data� set

FDA

FDA

FDA

FDA and PIBR

Private Industry, Government, Literature, and FDA

Coverage Measure

Presence in Molecular Feature Domain

Presence of 2–3 Atom Unknown Fragments

Descriptorbased Membership in Class

None

None

Operating System

Windows Desktop

Windows Desktop

Windows Desktop

Windows Windows Desktop Desktop (client server work station)

1â•… No endorsement implied.

The basis for selection and use of these computational platforms is principally driven by the scientific technical approach by each software programme as it relates to QSAR methodology and structural interpretation of chemical molecular features. Multiple approaches to the assessment of molecular features are necessary in making computational toxicology predictions that are reliable [6]. The more diverse, the more opportunity in general that computational analysis of a chemical or drug will lead to the correct prediction since chemical space itself is diverse. Thus, the use of multiple computational toxicology software applications is anticipated to be complementary from the technical standpoint. To help illustrate this point, Table 3 describes the varying computational approaches employed by the software that have been evaluated at the FDA. The main differences between the computational platforms are in the QSAR methodology and structural interpretation of molecular features in a chemical that is screened in silico. An important consideration is whether a computational software programme offers a means to assess molecular coverage. One common feature among all the computational software that FDA uses and for a large portion of other software programmes available otherwise, is that the molecular descriptors used to assess structure are two-dimensional. For some software

83

84

Contributions

programmes however, there are plans to add three-dimensional molecular descriptors. Another common feature of the computational software programmes is that they operate under a Windows Desktop environment.

3.3.5 Why Use In Silico Strategies as a Novel Approach to Assess Toxicity of Phytochemicals? Phytochemicals are often encountered as “active” (desirable or undesirable) constituents in mixtures for which there are little or no toxicological data. Yet given their high potential for human exposure through dietary sources in conventional food and supplement products, there is a practical need for assessing their toxicity using efficient and reliable methods [5, 9]. Moreover, natural products show a vast structural diversity and are being discovered at an accelerating pace. Phytochemicals are a class of substances that present a data-poor situation in assessment of toxicity. Thus, phytochemicals are a prime example of a need for efficient prioritization in testing and screening of chemical toxicity. In terms of chronic toxicity, one of the most important end points to consider for regulatory purposes and protection of human health is the carcinogenic potential of a chemical. Since carcinogenicity is an end point that cannot be tested in humans and is pivotal in regulatory decision-making for many types of substances, including phytochemicals, the in silico methods used by FDA/ CDER were tested to assess their utility for accuracy in screening naturally occurring carcinogens and non-carcinogens. In previous research studies by the FDA Center for Food Safety and Applied Nutrition (CFSAN) and CDER, a dataset of phytochemicals was tested in external validation studies to determine the performance characteristics and overall accuracy of computational toxicology software programmes for predicting rodent carcinogenicity [7, 9]. Results were very encouraging with one computational software programme demonstrating a high level of sensitivity (97â•›%) for predicting rodent carcinogenic phytochemicals, albeit marginal performance was reported for specificity (53â•›%) [7]. Given the practical interest and search for valid methods for high-throughput toxicity screening of phytochemicals, the desire to test multiple computational software platforms, and the lack of toxicological data in the public domain that is needed for safety and risk assessment of phytochemicals, this presentation provides the results from in silico screening of a dataset of non-proprietary phytochemicals for rodent carcinogenicity.

3.3.6 Prediction of Rodent Carcinogenicity of Phytochemicals in an External Validation Study 3.3.6.1 Background

Based on previous external validation testing for predicting rodent carcinogenicity of natural products using computational QSAR modelling at the FDA

In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure–Activity Relationship Analysis

[5, 9], our interest was to evaluate the performance of two statistical-based computational software platforms for their utility as a predictive tool of toxicity for phytochemicals. As a class, phytochemicals are plant-derived substances that are frequently lacking animal toxicology study data, thus presenting a challenge for risk assessors needing to evaluate their potential human health hazards and safety in consumer products. The hypothesis to be tested is that QSAR modelling could be useful as a decision support tool in safety assessment by providing accurate predictions on the rodent carcinogenic activity of phytochemicals. In addition, the in silico screening approach may serve as a useful tool for prioritizing which phytochemicals should undergo experimental testing for carcinogenicity in the absence of empirical data. The most rigorous method for determining if a QSAR model is performing satisfactorily in terms of its predictive performance for toxicity is to conduct an external validation study. External validation refers to in silico screening a dataset of chemicals that were never used to make the QSAR training dataset, yet the toxicity of the chemicals is known from experimental testing. Furthermore, there are retrospective and prospective types of external validation testing. In this presentation, data are presented from a retrospective external validation study of the rodent carcinogenic activity of phytochemicals.

3.3.6.2 External Validation Test Set

Using the same criteria for selection of phytochemicals as was described in a recent study using external validation [5], a dataset of 43 phytochemicals was screened as an external validation test set using two statistical-based computational software programmes, the Leadscope Model Applier (Leadscope Inc.) and MC4PC (MultiCASE Inc.). The predictive paradigms of these two computational software programmes have been described previously [6]. The external validation test set of 43 phytochemicals that were screened using these approaches is presented in Table 4, where a total of 24 active and 19 inactive phytochemicals are listed. It is notable that the phytochemicals may be found in the human diet as part of conventional foods, spices, flavouring agents, botanical extracts, and ingested ethnotraditional medicinal remedies, thus representing a wide variety of plant-derived natural products from different sources. Results of the external validation testing were used to evaluate the performance of the two software programmes to accurately screen the 43 phytochemicals for rodent carcinogenicity. Rodent cancer bioassay data exist for each of these chemicals and were identified from literature in the public domain as previously described [5]. In addition, a small positive control set of synthetic chemicals which can be found in the human diet mainly as contaminants was also added to the external validation test in order to follow predictive performance. These 10 synthetic chemicals were also not used to construct the QSAR model and are known rodent carcinogens based on bioassay results [5]. The synthetic chemicals are: 4-aminoazobenzene, dibutyltin diacetate, permanent orange, quinoline, trin-butyl phosphate, ammonium perfluorooctanoic acid, 1,3-dichloropropanol,

85

86

Contributions Table 4: Phytochemicals tested with computational toxicology software programmes in external validation. Phytochemical

Natural Occurrence

1‘-Hydroxyestragole

Basil, Ocimum basilicum; Metabolite of estragole

2-Ethyl-1-hexanol

Sassafras, Sassafras albidum; Metabolite of safrole

5-Methoxypsoralen

Parsley, Petroselinum sativum

6-Methylcoumarin

Oregano, Origanum vulgare

Capsaicin

Hot peppers, Capsicum annum

Dehydromonocrotaline

Russian comfrey, Symphytum uplandicum

Estragole

Basil, Ocimum basilicum

Heliotrine

Russian comfrey, Symphytum uplandicum

Vanillin

Vanilla, Vanilla planifolia

Ptaquilosin

Bracken fern, Pteridum aquilinum

Ptaquilosin (APT) dienone

Bracken fern, Pteridum aquilinum

Hydroxysenkirkine

Medicinal herb, Crotalaria laburnifolia

4-Methylphenylhydrazine

Edible mushroom, Agaricus bisporus; metabolite

Allyl hexanoate

Tea Tree oil, Melaleuca alternifolia; flavour

Anethole

Fennel, Foeniculum vulgare

β-apo-8’-carotenal

Carrot, Paucus carota

Citrate

Lemon, Citrus limon

Crotonaldehyde

Potato, Solanum tuberosum

Curcumin

Turmeric, Curcuma longa

Epicatechin

Green Tea, Camellia sinensis

Formic acid

Carrot, Paucus carota

Gallic acid

Mango, Mangifera indica

Indole

Corn, Zea mays

Indole-3-acetic acid

Strawberry fruit, Fragaria vesca

Linalool

Apricots, Prunus armeniaca

Lipoic acid

Spinach, Spinacia oleracea

Maltol

Roasted coffee, Coffea arabica

Piperonal

Vanilla, Vanilla planifolia

Piperine

Black pepper, Piper nigrum

Intermedine

Comfrey, Symphytum officinale

Isosafrole

Oil of Sassafras, Sassafras albidum

Jacobine

Ragwort herb, Senecio jacobaea

In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure–Activity Relationship Analysis

Lycopsamine

Comfrey, Symphytum officinale

Methylglyoxal

Roasted coffee, Coffea arabica

Parasorbic acid

Rowan berry, Sorbus aucubaria

Propionic acid

Tomato, Lycopersicon esculentum

Retronecine

Medicinal herb, Crotalaria laburnifolia

Senecionine

Ragwort herb, Senecio jacobaea

Seneciphyllinine

Ragwort herb, Senecio jacobaea

Tannic acid

Tea, Camellia sinensis

Hydroxymethylphenylhydrazine

Edible mushroom, Agaricus bisporus

1-Octacosanol

Perilla seeds, Perilla frutescens

Protocatechuic acid

Shallot onions, Allium cepa

2-chloro-1,2-propanediol, 4-hydroxyphenylacetamide, and diazoaminobenzene. The procedures used for in silico screening of the two-dimensional structures in the validation test set have been described previously [5, 6].

3.3.6.3 Rodent Carcinogenicity Models

A total of seven predictive QSAR models for the rodent carcinogenicity end point per software were used to in silico screen the phytochemical test set. The carcinogenicity QSAR models used in the external validation test are defined according to the species/ gender data that were used to construct the model as follows: rat, male rat, female rat, mouse, male mouse, female mouse, and rodent composite. Specific details as to how these QSAR models were built have been described previously [6, 10–11]. Generally, the models are comprised of over 24,500 study records, and contain over 1500 QSAR modellable chemicals from sources including the US National Toxicology Program, FDA/CDER archives, International Agency for Research on Cancer, the Gold Carcinogenicity Potency Database, and the published literature.

3.3.6.4 External Validation Predictive QSAR Performance Statistics

The results of the statistical analysis of the predictive performance of the Leadscope Model Applier software programme in the external validation test are presented in Table 5.╃The results show only marginal performance for predicting carcinogens (sensitivity) and non-carcinogens (specificity) using this data test set of 43 phytochemicals. Overall concordance followed suit with the performance for sensitivity and specificity. Table 6 presents the results of the statistical analysis of the predictive performance of the MC4PC software programme using the same external validation test set. Predictive performance for specificity

87

88

Contributions Table 5: External validation statistics for the Leadscope Model Applier computational software programme for predicting rodent carcinogenicity of 43 phytochemicals and 10 synthetic dietary chemicals based on QSAR analysis of seven models. Statistical Performance Parameters €

Experimental

€

+

–

+

16

7

–

14

9

€

Coverage

100.00â•›%

€

Specificity

56.25â•›%

€

Sensitivity

53.33â•›%

€

Concordance

54.35â•›%

€

False Positives

43.75â•›%

€

False Negatives

46.67â•›%

€

Positive Predictivity

69.57â•›%

Prediction€

€

Table 6: External validation statistics for the MC4PC computational software programme for predicting rodent carcinogenicity of 43 phytochemicals and 10 synthetic dietary chemicals based on QSAR analysis of seven models. Statistical Performance Parameters €

Experimental

€

+

–

+

14

1

–

16

15

€

Coverage

100.00â•›%

€

Specificity

93.75â•›%

€

Sensitivity

46.67â•›%

€

Concordance

63.04â•›%

€

False Positives

6.25â•›%

€

False Negatives

53.33â•›%

€

Positive Predictivity

93.33â•›%

Prediction€

€

In Silico Toxicology Screening of the Rodent Carcinogenic Potential of Phytochemicals Using Quantitative Structure–Activity Relationship Analysis

was very high at 94â•›%, but there was poor sensitivity (47â•›%) with the phytochemical test set. Because of the high degree of specificity, a high false-negative rate occurred as expected. With the low performance for sensitivity, a low falsepositive rate was observed. However, high positive predictivity was observed in the statistical performance data suggesting that if there is a positive prediction it is likely to be accurate. Overall concordance was marginal with a value of 63â•›%. Coverage for both programmes was excellent meaning that the software programmes were able to screen all 43 phytochemicals and 10 synthetics run against the models, thus suggesting that the chemical space of the phytochemicals and synthetics was adequately represented by the rodent carcinogenicity QSAR training datasets that were used in the predictive modelling.

3.3.6.5 Discussion

The results from the external validation test of QSAR predictive performance for the rodent carcinogenicity of 43 phytochemicals and 10 synthetic chemicals using two different computational toxicology software programmes show that it was possible to have excellent predictive performance for specificity using one software programme but there was poor performance in sensitivity with both programmes. Taking a close look at the internal validation statistics for the software which performed well in predicting non-carcinogens, the predictive performance based on internal validation is also high for specificity (>90â•›%) but low for sensitivity (85â•›% of compounds tested within +/- 1 log). When taken together in an integrated approach, the models available provide a good coverage of the “world of chemicals”, including food additives, pesticides, environmental contaminants and biologically active substances (human and veterinary drugs). Their application in an integrated strategy is likely to allow establishing levels of safety concern of compounds for which no hard toxicological information is available. The application of such an approach can be valuable in emergency situations to support fast decision-making. It may also help in defining priorities for additional toxicity testing. Its application for botanical extract has still to be defined.

3.5.1 Introduction Over recent years, there has been mounting concern about food as a source of exposure to potentially toxic chemicals. It has been estimated that there are over five million man-made chemicals known, of which 70,000 are in industrial use today [1]. The application of continuously improving analytical methods has revealed that many of these chemicals can enter the food chain and result in human exposure, albeit often at low levels. Since for the vast majority of these chemicals, toxicological information is absent or limited, the assessment of their health significance is therefore difficult or impossible. Nevertheless, the detection of such chemicals in food may trigger not only heavy management actions (e.╃g. public recall) but also alarm resulting in loss of consumer confidence for the food supply. In such situations, the availability of reliable tools

1

Nestlé Research Centre, Quality and Safety Department, P.╃O. Box 44, CH-1026 Lausanne, Switzerland, [email protected].

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data

allowing establishing levels of safety concern appear of particular importance to ensure adequate consumer protection without undue over-conservatism. Solutions to this general issue are not straightforward. Experimental toxicology is not a practical tool to deal with emergency situations requiring fast decisionmaking. In this context, in silico predictive models have obvious advantages in terms of time, cost and also animal protection. Ideally, such models should predict safe levels of exposure. A second potential use of predictive models under investigation relates to naturally occurring chemicals in foods. Because of potential health benefits, the application of traditional plants and plant extracts in food products is attracting a growing interest. For many of these ingredients, very limited toxicological information is available and their safety assessment is often based on traditional history of medicinal use, which may not be suitable to cover food applications. One challenge is to address the safety significance of inherent bioactives or toxicants occurring naturally in such plant materials [2]. A battery of in silico toxicology tools may be envisaged to establish the level of safety concern associated with key constituents of plant extracts. Such an approach may be valuable for early decision-making in new product development and to decide on the need for further toxicological studies. In addition, it could be applied to set up specifications and to define analytical requirements for compositional characterization and standardization. Specific in silico approaches for research and development purposes have been already designed and are successfully applied to pre-clinical screening of potential drugs in pharmaceutical discovery pipelines, where an early identification of toxicological hazard offers a clear competitive advantage [3]. Such efforts allow the exclusion of chemicals that could potentially reveal unacceptable based on further mandatory regulatory toxicological tests. In the present paper, in silico models developed to establish levels of safety concern of food chemicals are described. Their potential use in the safety assessment strategy of functional botanical products is briefly discussed.

3.5.2 Computational Toxicology Models Relevant for the Food �Sectors: Requirements As compared to the pharmaceutical area, the situation of the food sector requires the development of alternative models with the following specific characteristics. Quantitative rather than Qualitative Predictions: In the food context, the most likely application of computational toxicology models would be in the establishment of the level of safety concern associated with the inadvertent/accidental presence of a chemical in products. This requires not only qualitative information on the potential hazardous properties of the chemical (e.╃g., carcinogenicity) but also quantitative information (e.╃g., carcinogenic potency), allowing the derivation of a margin of exposure (MOE) with the estimated intake. The interpretation of the size of the MOE (e.╃g., al-

111

112

Contributions

lowing for various uncertainties such as inter- and intra-species differences) would likely help to make decisions at the management level. Reliable, High Sensitivity: Most quantitative structure–activity relationship (QSAR) predictive models suffer from inherent poor sensitivity, i.╃e., the ability to correctly identify true positives. Modellers, partially because they are often confronted with non-representative datasets, have focused their attention on identifying toxicophores that are overly general, and as a result, models tend to have many false positives. This has made computational toxicology a useful tool for high-throughput screening but different strategies should be developed if the aim is to have a low number of false negatives or a high concordance. Global Chemical Diversity: Compounds found in foods and food ingredients present a wide structural diversity and complexity that may be greater than synthetic pharmaceuticals targeted for a particular purpose, and, therefore, require the development of global in silico models (rather than local, referring to particular classes of chemical structures). Relevance and Transparency: Ideally, in silico toxicology strategies for food safety assessment should be able to predict adverse health effects in the human population. Because the toxicological training databases currently available are based on mainly in vitro and animal data with high limitations (species differences etc.) to directly predict the human situation, the development of such models will always constitute a significant challenge. Currently, their practical application in the food sectors will depend upon their potential to accurately predict biological end points/ hazards that are used in food chemical risk assessment. This includes the need to establish confidence limits. The acceptance of these models will be possible only if the analysis is fully transparent. Therefore, the promotion of validated, freely available tools based on open-source codes, such as those developed by the European Chemical Bureau and the US Environmental Protection Agency, is warranted.

3.5.3 Available Computational Toxicology Models for Food �Applications For most food-related compounds for which a complete toxicological database is available, chronic toxicity studies provide the most sensitive end point and usually the pivotal data to establish safe levels of exposure such as the acceptable or tolerable daily intakes [1, 4]. To predict chronic toxicity is therefore considered a first priority in the development of computational toxicology models. In this context the following models appear to be relevant:

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data

3.5.3.1 Chronic oral Toxicity

3.5.3.1.1╇ Threshold of Toxicological Concern (TTC) TTC is a concept widely used in chemical food safety and is based on the decision tree developed by Cramer [5] for the estimation of chronic toxic hazard and the de minimis concept. Depending on its chemical features, a chemical can be assigned into one of three classes reflecting a presumption of low, moderate or serious toxicity corresponding to human exposure threshold values below which, there are no significant risks to human health [1, 4]. Although the TTC concept is not per se a computational predictive toxicity system, it has been the first attempt to formally combine structural chemical information with statistical processing of available oral toxicity data to establish levels of no safety concern for chemicals for which no hard toxicological data are available. Recently, additional applications have been proposed for botanical extracts [6], impurities in pharmaceutical preparations [7–9], occupational [10–12] as well as dermal [13–16] exposures. 3.5.3.1.2╇ Lowest Observed Adverse Effect Level (LOAEL) The effect of long-term exposure to chemicals is generally addressed by feeding rodents with various doses of test materials over long periods of time up to lifetime. These chronic studies are designed to obtain a dose-response including a dose producing severe toxicity, the lowest dose exhibiting overt toxic effects (Lowest Adverse Effect Level, LOAEL), and a dose without any toxic effects (No Observed Adverse Effect Level, NOAEL). These tests are fundamental for risk assessment, but the correlation between the chemical structures and the toxicological outputs has received only little attention. This is partly due to the complexity of such experimental tests that embrace a plethora of different biological effects and mechanisms of action, making QSAR studies extremely challenging. Often, it is considered that chronic toxicity is a far too heterogeneous end point to be encoded in a single predictive model. However, several models aimed at predicting chronic toxicity (LOAEL) have recently been developed. Published validation studies have indicated sufficient performance to justify applications in well selected and controlled situations. Commercially available models: Several commercially available systems can help risk assessors to predict a number of end points; however, the provision of a numerical value for chronic exposure is currently only supported by the TOPKAT package developed by Table 1: Performance of the TOPKAT chronic rat toxicity model. Percent of predictions within factors of:

10

100

Venkatapathy et al. [17]

72

98

Tilaoui et al. [18]

59

86

Cadmus Group, Inc. [19]

68

95

113

114

Contributions

Accelrys. This system, aimed at predicting Rat-LOAELs, has been challenged by several authors using different test sets. Results are summarized in Tab.╃1. Rat-LOAEL Model: We recently reported a predictive in silico study of more than 400 compounds based on two-dimensional chemical descriptors and multivariate analysis [20]. The analysis used a highly homogenous LOAEL dataset restricted to chronic (defined as longer than 180 days), oral (gavage, diet and drinking water) rat studies. The root mean squared error of the predictive model was found to be 0.73 (in a logarithmic scale) on a leave-one-out cross-validation and is close to the observed variability of actual experimental values (0.64). More than 65â•›% of predictions fall within a factor of 5; 85â•›% within a factor of 10 and 99â•›% within a factor of 100 of empirical data. The analysis of the model revealed that the bioavailability of the compound drives chronic toxicity effects, constituting a baseline effect where additional toxicity is possibly described by a few specific chemical moieties. The results obtained give confidence that this model can be useful in supporting the prioritization of issues in food chemical toxicology research. Model Based on Human Data. (Maximum Recommended Therapeutic Dose, MRTD): The above mentioned models suffer from several limitations including the low number of chemicals in the database and consequently the limited coverage of the “world of chemical structures”, as well as the questionable relevance of the end point modelled to the human situation (i.╃e. the end point is not relevant for humans). Conversely, end points relevant for humans such as headache and nausea cannot be detected in animal tests or predicted from corresponding in silico models. To overcome these limitations, a model was developed [21] using Maximum Recommended Therapeutic Doses (MRTD) of a large number (over 1300) of drugs [22, 23]. The MRTD is an estimated upper dose limit beyond which a drug’s efficacy is not increased and/ or undesirable adverse effects begin to outweigh beneficial effects. The MRTD is essentially equivalent to the NOAEL in humans, a dose beyond which adverse (toxicological) or undesirable pharmacological effects are observed. It has been considered that with the exception of chemotherapeutics and immunosuppressants, the MRTD/10 would correspond to a dose exerting neither therapeutic nor chronic adverse effects in human [22, 23]. For non-pharmaceutical chemicals, there is no desired pharmacological effect and any compound-related effect could be interpreted as adverse or non-desirable effect. The MRTD is empirically derived from human clinical trials and is a direct measure of the dose-related effects of pharmaceuticals in humans. These data are of particular interest because they refer to the effect on humans of biologically active molecules, mainly marketed drugs. The model used spans nearly 9 orders of magnitude of dose variation and predicts 70â•›% of the compounds with q2╃=╃64â•›% and more than 82â•›% within 1 log unit of all experimental values (89â•›% within the applicability domain defined by a confidence interval >0.2) and has an overall mean log error of 0.╃51.╃These performances are comparable to those of published models based on the same

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data

database and different commercial software: Matthews et al. [22] (mean log error=╃0.56, concordance=╃86â•›% on a two-class, high–low toxicity prediction) and Contrera et al. [23] (mean log error=╃0.58, concordance=╃71â•›% on a two-class, high–low toxicity prediction). The good performances of these models are an indication that it is indeed possible to predict human end points reliably with QSAR techniques, and suggest the importance of the quality of the dataset over the statistical techniques used to mine it.

3.5.3.2 Mutagenicity

The Ames test in Salmonella typhimurium is a bacterial short-term in vitro assay aimed at detecting the mutagenicity caused by chemicals. Mutagenicity has been considered as an early alert for carcinogenicity caused by (direct and indirectly acting) DNA damaging agents. Based on experimental data generated over decades, several QSAR studies on this end point yielded enough information to make feasible the construction of reliable computational models for prediction of mutagenicity from the molecular structure. Combining a fragment-based (SA) model and an inductive database, it was possible to develop a predictive model [24] which is quantitatively similar to the experimental error of Ames test data (error on external test set compounds=╃15╛%, sensitivity=╃15╛%, specificity=╃15╛%). Sensitivity is defined as the ability to identify correctly true positives, while specificity is the ability to identify true negatives). Moreover, each single prediction is provided with a specific confidence level. The results obtained give confidence that this system can be applied to support early and rapid evaluation of the level of mutagenicity concern.

3.5.3.3 Carcinogenicity

Carcinogenicity is among the toxicological end points raising the highest concern and scientific debate. Standard carcinogenicity bioassays in rodents used to study the carcinogenic effects of chemicals are long and costly, requiring the sacrifice of large numbers of animals. In addition, the extrapolation of animal carcinogenicity data to the human situation is highly challenging. Several attempts to develop alternative predictive models for carcinogenicity are available [3, 25]. Most local QSAR models for congeneric chemical classes agree with, and/ or support the available scientific knowledge, and exhibit good statistics. Local models that discriminate active and inactive chemical (qualitative prediction) are 70 to 100â•›% correct, whereas QSARs that estimate the potency of chemicals have considerably lower accuracy (30 to 70â•›%). In addition, the commonly used statistical internal cross-validation procedures resulted to be poorly correlated with external validation statistics. SA models have an accuracy of about 65â•›% for rodent carcinogens; however, these models do not discriminate well between active and inactive chemicals within individual chemical classes, suffering from poor sensitivity (specificity much higher than sensitiv-

115

116

Contributions

ity). Overall, the present QSAR-based predictions of carcinogenicity are more suited for preliminary (organization and rationalization of data, elucidation of mechanisms of action), or large-scale screenings and need to be complemented by data from other sources [3, 25].

3.5.4 Model Integration and Application Attempts to integrate currently available predictive toxicology models to optimize food applications are ongoing (see Fig.╃1). Preliminary thoughts are provided below. The minimum information required is chemical structure and an estimate of potential or actual exposures. Any new molecule entering the system is first tested for mutagenicity [24]. In absence of an alert for mutagenicity/ genotoxicity, the calculated exposure is compared to the relevant Threshold of Toxicological Concern (TTC) [1, 4]. Exposures lower than the relevant TTC are considered of low safety concern [1, 4]. In case of estimated exposure higher than the relevant TTC, margins of exposure (MOE) between the predicted toxicity values obtained from the TOPKAT, Rat-LOAEL and MRTD models are calculated. The interpretation of the different MOEs is not straightforward and should be done on a case-by-case basis. To conclude on low concern, the MOEs based on Rat-LOAELs should at least be large enough to account for potential inter- (UF=╃10) and intra- (UF=╃10) species differences and to allow for the conversion of LOAELs into a NOAELs (UF=╃3–10). Other inter-species uncertainty factors could be proposed based on allometric scaling. An additional factor would increase the confidence to fully cover the potential error of the models. The MOEs obtained with predicted

Figure 1: Model integration.

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data

MRTD should be large enough to allow for the conversion of the MRTD into a safe level of exposure (UF=╃10). Additional factors (UF=╃2–10) may be necessary to cope with potential intra-human differences [22, 23]. Final interpretation will have to consider and compare the resulting three different and independent MOEs. In case of alerts for mutagenicity/ genotoxicity, the estimated exposure is compared to a TTC of 0.15 μg/pers [4]. Exposure below this TTC would be considered unlikely to be of any concern, even for compounds with mutagenic properties [1, 4]. Time adjustment may be envisaged in case of established shortduration exposure [8]. Additional development is necessary to handle chemicals with mutagenicity alerts at exposure levels significantly higher than the TTC of 0.15 μg/pers. Models predicting carcinogenicity are currently evaluated. A chemical with mutagenicity alert but negative in carcinogenicity predictive models would then enter the chronic toxicity prediction scheme as described above. A chemical positive in both mutagenicity and carcinogenicity predictions could theoretically be managed through the calculation of a MOE between a predicted carcinogenic potency (e.╃g. TD50) and the estimated exposure. However, no tools are currently available to undertake such an analysis.

3.5.5 Computational Toxicology and Safety Assessment of �Botanical Extracts 3.5.5.1 General Considerations

The use of botanical-derived functional/ medicinal ingredients in foods has raised several scientific and regulatory issues [2]. These products often fall into grey and undefined regulatory areas between foods and medicines. Furthermore, there is still debate regarding the information required to document beneficial effects and how to communicate such benefits through claims. In addition, there have been a number of recorded cases of intoxications with botanical products raising the issue of their safety [2]. However, whatever the regulatory status or claim, products must be safe for their intended use. The application of functional plant-derived ingredients in food products raises serious and complex questions. Plant misidentifications have been the source of severe intoxications, while the extreme compositional variability of botanical ingredients has often been highlighted. The biological activity of functional botanicals gives rise to other specific safety considerations. The doses producing either health benefits or adverse effects may not be so different and consequently the safety margins associated with the use of functional botanicals are often expected to be reduced as compared to for example food additives [2]. For many botanical extracts, safety assessments are based on history of medicinal use together with limited, often acute, toxicological data on the extracts and/ or extract constituents. One difficulty is to evaluate the applicability of medicinal data to back up the safety of food applications. Amongst the major differences between food and medicinal applications are the duration of exposure

117

118

Contributions

(often expected to be longer for food applications) and the doses (usually lower for food applications), and the acceptability of any risk–benefit consideration. A computational toxicology strategy as outlined above may be of value to address long-term concerns associated with exposure to the main identified constituents of the extracts. Although toxicological data on the extract themselves are considered more relevant for safety assessment (because of potential matrix effects, interaction of the different components), information on individual components may play a significant role to establish the level of safety concern and decide on further need for hard toxicological data.

3.5.5.2 Case Study

To study the potential value of the in silico strategy outlined above for establishing the level of safety concern associated with the chronic use of botanical extracts and phytochemicals, Ginkgo biloba leaf extract was chosen for a learning exercise. This material was selected because of the availability of a relatively large amount of both compositional and toxicological data allowing appropriate comparisons. Composition: For the present exercise, the application of a standardized commercial extract of the following composition was considered (26): ►⌺ 24â•› % flavonoid glycosides containing quercetin, kaempferol and isorhamnetin ►⌺ 6â•› % terpenoids, in which 3.0â•›% are ginkgolides A, D, C, M, and J, and 2.9â•›% bilobalide. The chemical structures are provided in Table 2. Exposures: In pharmaceutical applications 80 to 240 mg of standardized leaf extract are taken while food fortifications have proposed lower amounts, such as 10 to 60 mg/day. Considering the compositional data described above and extract intakes ranging from 10–240 mg/day, exposures to the different constituents were calculated assuming a standard consumer of 60 kg (Tab.╃3). Predictions: Mutagenicity predictions were negative for all chemicals tested. When available in both models, predictions based on TOPKAT and Rat-LOAELs were relatively similar (within a factor of 3). The Rat-LOAEL provides more conservative values. The predicted values were relatively high but compatible with available rat experimental data indicating low chronic toxicity for quercetin and the extract itself [26]. As expected, predicted MRTDs were significantly lower than predicted Rat-LOAELs.

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data Table 2: Structures of key Ginkgo Biloba constituents. Quercetin

Kaempferol

Isorhamnetin

Bilobalide

Ginkgolide A

Ginkgolide B

Ginkgolide C

Ginkgolide J

Ginkgolide M

To compare both types of parameters, animal LOAEL (from TOPKAT and RatLOAEL models) and human MRTD, is not simple. Levels of low safety concern based on MRTD could be estimated by dividing the predicted value by a factor of 10 (conversion of MRTD into no effect). An additional factor (2 to 10) may be applied for possible intra-specific differences [22, 23]. Low level of concern based on Rat-LOAEL predictions would require the application of a factor of 100 to cover inter- and intra-species differences and an additional factor of 3–10 to deal with the conversion of the LOAEL to NOAEL. For the flavonoids, the levels of low concern based on the three independent models were all within the same order of magnitude providing additional confidence for the predictions. Margins of Exposures: With the highest dose envisaged in human applications, MoEs lower than values considered necessary to provide a good confidence of safety were obtained. This is somehow expected since this dose is documented to be pharmacologically active. Much lower doses which have been sometimes proposed for food fortification are characterized by larger MoEs considered sufficient to provide a good guaranty of safety. In such case safety would likely be ensured at the expense of efficacy. The MoEs derived from this exercise in computational analysis are in good agreement with MoEs that can be derived from actual clinical and experimental data (26). Ginkgo biloba extract up to 240 mg/day is considered to be pharmacologically active with a good record of safety, although some possible side effects were reported in a limited number of exposed individuals.

119

120

Contributions Table 3: Exposure estimates, chronic toxicity and MRTD predictions and calculated margins of exposure (MOEs) for key Ginkgo biloba extract constituents. Name (%2)

Exposure1

TOPKAT3 (MOE)

Rat-LOAEL3 (MOE)

MRTD3 (MOE)

Quercetin (8)

13–320

371.8 (29,000–1200)

199.5 (15,000–625)

2.0 (154–6)

Kaempferol (8)

13–320

273.2 (21,000–875)

166.4 (13,000–540)

1.9 (146–6)

Isorhamnetin (8)

13–320

334.4 (26,000–1083)

127.4 (9800–400)

4.0 (307–13)

Bilobalide (2.9)

5–116

OPS(-)

252.2 (50,000–2100)

NA

Ginkgolide A (0.6)

1–24

OPS(-)

103.5 (103,500–4312)

NA (-)

Ginkgolide B (0.6)

1–24

OPS(-)

134.4 (134,400–5600)

NA (-)

Ginkgolide C (0.6)

1–24

OPS(-)

159.1 (159,100–6630)

NA (-)

Ginkgolide J (0.6)

1–24

OPS(-)

126.7 (126,700–5280)

NA (-)

Ginkgolide M(0.6)

1–24

OPS(-)

142.6 (142,600–5940)

NA (-)

Exposure in µg/kg/d based on intake of 10 to 240 mg/extract/day (assessing a 60╃kg bodyweight) Default contribution to the total extract mass (based on compositional data) 3 in mg/kg bw/d MOE: margin of exposure; OPS-: outside optimum prediction space; NA: not applicable, structure not recognized. 1 2

3.5.6 Discussion and Conclusion Computational models are available to predict quantitatively chronic rodent and human toxicity ►⌺ Validation studies indicate a reasonable performance of these models (>85â•› % of compounds tested within +/- 1 log) ►⌺ Ongoing research suggests a good inter-model correlation ►⌺ When taken together in an integrated approach, the models available provide a good coverage of the “world of chemicals”, including food additives, pesticides, veterinary drugs, environmental contaminants and biologically active substances (human drugs) ►⌺ This integrated approach is likely to allow establishing levels of safety concern of compounds for which no hard toxicological information is available ►⌺ The application of such an approach can be valuable in emergency situations to support fast decision-making. It may also help in defining priorities for additional toxicity testing ►⌺ As shown in the present paper, such an approach may also provide some valuable information to assess the level of chronic safety concern of functional botanical extracts ►⌺

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data

The application of such an approach for botanical extract is mainly limited by the quality of the available compositional data. In addition, it does not allow addressing matrix effects ►⌺ Overall limitations of such models depend more on the experimental data than on the available computational methods which are now considered as mature ►⌺ Further work is necessary to optimize the interpretation of the information provided by the models ►⌺ The prediction of carcinogenicity and carcinogenic potency appear as the next big challenges ►⌺ Future efforts will have to focus on modelling mechanistic end points of higher relevance for low levels of human chemical exposures. ►⌺

References â•⁄ 1. Kroes R., Galli C., Munro I., Schilter B., Tran L.-A., Walker R., Würtzen G. (2000) Threshold of toxicological concern in the diet: a practical tool for assessing the need for toxicity testing. Fd. Chem. Toxicol. 38, 255–312. â•⁄ 2. Schilter B., Andersson C., Anton A. Constable R., Kleiner J., Korver O., O’Brien J., Renwick A., Smit F., Walker R. (2003) Guidance for the safety assessment of botanicals and botanical preparations with use in food and food supplements. Fd. Chem. Toxicol. 41, 1625–1649. â•⁄ 3. Benfenati, E., Benigni, R., Marini, D., Helma, C., Kirkland, D., Martin, T.╃M., Mazzatorta, P., Meunier, J.-R., Ouédraogo-Arras, G., Richard, A., Schilter, B., Schoonen, W.╃G.╃E.╃J., Snyder, R., Yang, C., Youne, D.╃M.(2009) Predictive models for carcinogenicity and mutagenicity. Frameworks, state-of-the art and perspectives. (submitted). â•⁄ 4. Kroes R., Renwick, A., Cheeseman, M., Kleiner, J., Mangelsdorf, I., Piersma, A., Schilter, B., Schlatter, J., van Schothorst, Vos J.╃G., Würtzen G. (2004) Structure-based thresholds of toxicological concern: guidance for application to substances present at low levels in the diet. Fd. Chem. Toxicol. 42, 65–83. â•⁄ 5. Cramer G.╃M.; Ford R.╃A. (1978) Estimation of toxic hazard – a decision tree approach. Fd. Cosmet. Toxicol. 16, 255–276. â•⁄ 6. Rietjens I.╃M., Slob W., Galli C., Silano V. (2008) Risk assessment of botanicals and botanical preparations intended for use in food and food supplements: emerging issues. Toxicol. Lett. 15, 131–6. â•⁄ 7. Bercu J.╃P., Hoffman W.╃P., Lee C., Ness D.╃K. (2008) Quantitative assessment of cumulative carcinogenic risk for multiple genotoxic impurities in a new drug substance. Regul. Toxicol. Pharmacol. 51(3), 270–7. â•⁄ 8. Müller L., Mauthe R.╃J., Riley C.╃M., Andino M.╃M., Antonis D.╃D., Beels C., DeGeorge J., De Knaep A.╃G., Ellison D., Fagerland J.╃A., Frank R., Fritschel B., Galloway S., Harpur E., Humfrey C.╃D., Jacks A.╃S., Jagota N., Mackinnon J., Mohan G., Ness D.╃K., O’Donovan M.╃R., Smith M.╃D., Vudathala G., Yotti L. (2006) A rationale for determining, testing, and controlling specific

121

122

Contributions

impurities in pharmaceuticals that possess potential for genotoxicity. Regul. Toxicol. Pharmacol. 44(3), 198–211. â•⁄ 9. Delaney E.╃J. (2007) An impact analysis of the application of the threshold of toxicological concern concept to pharmaceuticals. Regul. Toxicol. Pharmacol. 49(2), 107–24. 10. Carthew P., Clapp C. Gutsell S. (2009) Exposure based waiving: The applicaÂ� tion of the toxicological threshold of concern (TTC) to inhalation exposure for aerosol ingredients in consumer products. Fd. Chem. Toxicol. 46(6), 1287– 1295. 11. Hardin B.╃D., Robbins C.╃A., Fallah P., Kelman B.╃J. (2009) The concentration of no toxicologic concern (CoNTC) and airborne mycotoxins. J. Toxicol. Environ. Health A.╃72(9), 585–98. 12. Drew R., Frangos J. (2007) The concentration of no toxicological concern (CoNTC): a risk assessment screening tool for air toxics. J. Toxicol. Environ. Health A.╃70(19), 1584–93. 13. Safford R.╃J. (2008) The dermal sensitisation threshold- a TTC approach for allergic contact dermatitis. Regul. Toxicol. Pharmacol. 51(2), 195–200. 14. Kroes R., Renwick A.╃G., Feron V. Galli C.╃L., Gibney M., Greim H., Guy R.╃H., Lhuguenot J.╃C., van de Sandt J.╃J. (2007) Application of the threshold of toxicological concern (TTC) to the safety evaluation of cosmetic ingredients. Fd. Chem. Toxicol. 45(12), 2533–62. 15. Re T.╃A., Mooney D., Antignac E., Dufour E., Bark I., Srinivasan V., Nohynek G. (2009) Application of the threshold of toxicological concern approach for the safety evaluation of calendula flower (Calendula officinalis) petals and Â�extracts used in cosmetic and personal care products. Fd. Chem. Toxicol. 47(6), 1246–1254. 16. Blackburn K., Stickney J.╃A., Carlson-Lynch H.╃L., McGinnis P.╃M., Chappell L., Felter S.╃P. (2005) Application of the threshold of toxicological concern approach to ingredients in personal and household care products. Regul. Toxicol. Pharmacol. 2005 43(3), 249–59. 17. Venkatapathy R., Moudgal C., Swartout J., Bruce, R.╃M. (2004) Assessment of the rat chronic LOAEL model in TOPKAT, a QSAR software for toxicity prediction. J. Chem. Inf. Comput. Sci.╃44, 1623–1629. 18. Tilaoui L., Schilter B., Tran L.-A., Mazzatorta P., Grigorov M. (2997). Integrated computational methods for prediction of the lowest observable adverse effect level of food-borne molecules. QSAR Comb. Sci.╃26, 102–108. 19. The Cadmus Group, Inc. (2003) Evaluation of the use of QSAR models to generate data for use in screening the CCL universe to the PCCL. Discussion draft for NDWAC CCL workshop. 20. Mazzatorta P., Dominguez Estevez M., Coulet M., Schilter B. (2008) Modelling oral rat chronic toxicity. TJ. Chem. Inf. Mod.╃48, 1949–1954. 21. Maunz A., Helma C. (2008) Prediction of chemical toxicity with local support vector regression and activity-specific kernels. SAR QSAR Environ. Res.╃19, 413–431. 22. Matthews E.╃J., Kruhlak N.╃L., Benz R.╃D., Contrera J.╃F. (2004) Assessment of the health effects of chemicals in humans: I. QSAR estimation of

In Silico Models to Establish Level of Safety Concern in Absence of Sufficient Toxicological Data

the maximum recommended therapeutic dose (MRTD) and no effect level (NOEL) of organic chemicals based on clinical trial data. Curr. Drug Disc. Tech. 1(1), 61–76. 23. Contrera J.╃F., Matthews E.╃J., Kruhlak N.╃L., Benz R.╃D. (2004) Estimating the safe starting dose in phase I clinical trials and no observed effect level based on QSAR modelling of the human maximum recommended daily dose. Reg. Tox. Pharm. 40, 185–2006. 24. Mazzatorta P., Tran L.-A., Schilter B., Grigorov M. (2007) Integration of structure-activity relationship and artificial intelligence systems to improve in silico prediction of Ames test mutagenicity. J. Chem. Inf. Mod.╃47, 34–38. 25. Benigni R., Bossa C. (2008) Predictivity and reliability of QSAR models: the case of mutagens and carcinogens. Toxicol. Mech. Met.╃18, 137–147. 26. Chan P.-C., Xia Q., Fu P.╃P. (2007) Ginkgo Biloba leave extract: biological, medicinal, and toxicological effects. J. Environ. Sci. Health Part C 25, 211–244.

123

124

Contributions

3.6 In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and DetoÂ�xification of Coumarin and Estragole: Implications for Risk Assessment Ivonne M.╃C.╃M. Rietjens1,2, Ans Punt2, Benoît Schilter3, Gabriele Scholz3, Thierry Delatour3, and Peter J. van Bladeren2,3 This manuscript was originally published in Mol. Nutr. Food Res., 2010, 54(2): page 195–207.

Abstract In chemical safety assessment, information on adverse effects after chronic exposure to low levels of hazardous compounds is essential for estimating human risks. Results from in vitro studies are often not directly applicable to the in vivo situation, and in vivo animal studies often have to be performed at unrealistic high levels of exposure. Physiologically based biokinetic (PBBK) modelling can be used as a platform for integrating in vitro metabolic data to predict dose- and species- dependent in vivo effects on biokinetics, and can provide a method to obtain a better mechanistic basis for extrapolations of data obtained in experimental animal studies to the human situation. Recently we have developed PBBK models for the bioactivation of the alkenylbenzene estragole to its DNA binding ultimate carcinogenic metabolite 1’-sulfooxyestragole in both rat and human, as well as rat and human PBBK models for the bioactivation of coumarin to its hepatotoxic o-hydroxyphenylacetaldehyde metabolite. The present paper presents an overview of the results obtained sofar with these in silico methods for physiologically based biokinetics, focusing on the possible implications for risk assessment, and some additional considerations and future perspectives.

1

Correspondence to: Prof. Dr. ir. Ivonne M.╃C.╃M. Rietjens, Division of Toxicology, Wageningen University, Tuinlaan 5, NL-6703 HE Wageningen, The Netherlands, Tel: +╃31 317 483971, Fax: +╃31 317 484931, [email protected].

2

Division of Toxicology, Wageningen University, Tuinlaan 5, NL-6703 HE Wageningen, The Netherlands.

3

Nestlé Research Centre, Quality and Safety Department, P.╃O. Box 44, CH-1026 Lausanne, Switzerland.

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

3.6.1 Introduction Estragole

Estragole is an alkenylbenzene that occurs in different herbs such as tarragon, basil, and fennel and is present in products derived from these herbs such as pesto and essential oils [1, 2]. Average daily intake of estragole was estimated to be 10–70╃µg/kg bw per day [1, 3]. There is interest in the safety assessment of estragole as a food constituent, since estragole has been identified to be genotoxic in vitro and carcinogenic in rodent studies performed at high dose levels [4–6]. Based on disposition studies of 14C-methoxy-labelled estragole in rats, mice, and humans and identification of the metabolites excreted, the principal metabolic pathways of estragole have been established [7, 8]. Figure 1 presents an overview of estragole metabolism including pathways for bioactivation to the proximate and ultimate carcinogenic metabolite and pathways for detoxification. The main phase I metabolic pathways include 1’-hydroxylation, O-demethylation, epoxidation and 3’-hydroxylation of estragole. The main metabolic pathways of the proximate carcinogen 1’-hydroxyestragole are sulfonation to the ultimate carcinogen 1’-sulfooxyestragole, and detoxification through glucuronidation to 1’-hydroxyestragole glucuronide, or oxidation to 1’-oxoestragole. Several evaluations have been performed to assess the safety of human exposure to estragole at low dietary intake levels. In an evaluation performed by the Scientific Committee on Food of the European Committee (SCF) in 2001, it was concluded that estragole is genotoxic and carcinogenic and restrictions in use were indicated [3]. The Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) classified estragole in 1965 as GRAS (Generally Recognized as Safe) under conditions of intended use as flavouring substance in food [9]. In 2002, the FEMA re-evaluated the data available for estragole and concluded again that exposure to estragole from food, mainly as spices or added as such, does not pose a significant cancer risk to humans [1]. In this conclusion it was taken into account that there are experimental data suggesting a non-linear relationship between dose and profiles of metabolism and metabolic activation. In a more recent evaluation performed by the Joint FAO/WHO Expert Committee on Food Additives (JECFA) in 2008, it was indicated that although evidence of carcinogenicity to rodents given high doses of estragole exists, further research is needed to assess the potential risk to human health from low-level dietary exposure to estragole present in foods and essential oils and used as flavouring agents [10]. Overall, these different expert judgments reflect a general problem in cancer risk assessment studies, which is a lack in scientific consensus on how to translate carcinogenicity data obtained in experiments with rodents at high levels of exposure to the situation for humans exposed to low levels. Determining the cancer risk in humans at low dose dietary intake levels requires extrapolation of the animal carcinogenicity data obtained with respect to species and dose.

125

126

Contributions

Figure 1: Metabolism of estragole with the bioactivation pathway proceeding by formation of the proximate carcinogen 1’-hydroxyestragole and the ultimate carcinogen 1’-sulfooxyestragole. Formation of the other metabolites eventually leads to detoxification and excretion.€

Uncertainties exist about the shape of the dose–response curve below the range of the animal experimental data, and about possible species differences in metabolism including metabolic activation and detoxification. The aim of our PBBK studies for estragole was to obtain quantitative insight into dose- and species-dependent differences in the bioactivation and detoxification of estragole.

Coumarin

Coumarin is a naturally occurring compound that was first isolated from Tonka beans, and is found at high levels in some essential oils, particularly cassia leaf

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

oil, cinnamon leaf oil, cinnamon bark oil and in lavender oil and peppermint oil. Coumarin is also found in fruits (bilberry), green tea and other foods, such as chicory, and in personal care products [11–13]. Chronic exposure to coumarin by the oral route has been reported to result in liver adenomas and carcinomas in rats and liver adenomas in mice [11–15]. Recently, the European Food Safety Authority (EFSA), based on results of a study on DNA adduct formation in kidney and liver of rats demonstrating that coumarin does not bind covalently to DNA, concluded that coumarin induces liver tumours by a non-genotoxic mode of action. A tolerable daily intake (TDI) of 0.1 mg coumarin/kg bw was established [13]. The theoretical maximum daily intake of coumarin was calculated to be about 4.1 mg/day or 0.07 mg kg bw per day for a 60 kg person [11, 13]. Figure 2 presents an overview of coumarin metabolism. The major route of coumarin bioactivation is 3,4-epoxidation to coumarin-epoxide, which is followed by subsequent rearrangement of the epoxide to o-hydroxyphenylacetaldehyde (oHPA) [16] which is considered to be the hepatotoxic intermediate [11, 17–19]. Coumarin epoxide may also be conjugated to glutathione both chemically and enzymatically, the latter route being especially efficient in rats and mice [18]. oHPA can be detoxified by reduction to o-hydroxyphenyletha-

Figure 2: Metabolism of coumarin with the bioactivation pathway proceeding by formation of oHPA. Formation of the other metabolites eventually leads to detoxification and excretion.€

127

128

Contributions

nol (oHPE), but especially by oxidation to o-hydroxyphenylacetic acid (oHPAA) [11, 18, 20–22]. Significant species differences between rat and human exist in coumarin bioactivation via the 3,4-epoxide pathway. In rats and mice the 3,4-epoxidation pathway appears to be the major route of coumarin biotransformation, whereas in humans the detoxifying coumarin 7-hydroxylation predominates, a reaction catalyzed by CYP2A6 [11, 16, 19, 20, 23–27]. Furthermore, detoxification of oHPA to oHPAA was shown to be more efficient in humans than in rats [18]. Based on these species differences in biotransformation Felter et al. [19] argued that the uncertainty factor for inter-species variation used for definition of the TDI could be reduced from 10 to 2.5 leaving only the factor 2.5 for toxicodynamic differences but taking out the factor 4 for toxicokinetics. However, also of importance is that in man a genetic polymorphism has been identified for CYP2A6, the P450 enzyme catalyzing the detoxifying 7-hydroxylation of coumarin [28–30]. The aim of our PBBK studies for coumarin was to quantify the metabolic pathway(s) replacing the 7-hydroxycoumarin formation in homozygous CYP2A6 deficient subjects and to estimate computationally the expected consequences of the CYP2A6 deficiency for oHPA formation in the liver of humans.

Physiologically Based Biokinetic (PBBK) Models

As outlined above an overall problem in current risk assessment strategies is the need to extrapolate experimental data obtained in animal experiments at high dose levels to a low dose human situation. Uncertainties about the shape of the dose–response curve at dose levels relevant for dietary human intake, and about species differences in metabolic activation and detoxification, make it difficult to perform such extrapolations. Physiologically based biokinetic (PBBK) modelling can provide a method to obtain a better mechanistic basis for extrapolations of data obtained in experimental animal studies to the human situation [31–33]. A PBBK model is a set of mathematical equations that together describe the absorption, distribution, metabolism and excretion (ADME) characteristics of a compound within an organism on the basis of three types of parameters [34–37]. These parameters include physiological parameters (e.╃g. cardiac output, tissue volumes, and tissue blood flows), physicochemical parameters (e.╃g. blood/tissue partition coefficients), and kinetic parameters (e.╃g. kinetic constants for metabolic reactions) [34–37]. Solution of the PBBK equations produces outcomes that are an indication of, for example, the tissue concentration of a compound or its metabolite in any tissue over time at any dose, allowing analysis of effects at both high but also more realistic low dose levels. Furthermore, such PBBK models can be developed for different species, which can facilitate inter-species extrapolation. In addition, by incorporating equations and kinetic constants for metabolic conversions by individual human samples and/

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

or specific isoenzymes, modelling of inter-individual variations and genetic polymorphisms becomes feasible [38]. For the development of a PBBK model for a specific compound, model parameters need to be obtained. The physiological parameters of a species (e.╃g. blood flow rates and tissue volumes) can be obtained from the literature [39]. Tissue-blood partition coefficients might be obtained experimentally in vitro using vial equilibration techniques or equilibrium dialysis techniques [40, 41], but can also be obtained using in silico methods. Several in silico models have been published by which tissue-blood partition coefficients of a compound can be calculated based on their octanol-water partition coefficients [42]. Biochemical parameters for PBBK models, including metabolic parameters, are most often obtained by making preliminary assumptions about metabolic routes and optimizing the kinetic constants by fitting the model to available in vivo data [34, 35]. Alternatively, metabolic parameters might also be derived from in vitro experiments with tissue fractions, primary cell cultures, or tissue slices of organs involved in the metabolism of the compound [43]. Lipscomb and Poet [43] have pointed out some advantages of using in vitro metabolic parameters to define PBBK models, which include the ability to separately define and analyze individual metabolic processes, such as phase I metabolism and phase II metabolism, or bioactivation and detoxification, and to compare contributions from individual conversions to the overall metabolism across species and between individuals, when limited in vivo data are available as is often the case for humans [43].

3.6.2 Methods In the present studies the metabolic parameters for the relevant biotransformation reactions for estragole and coumarin, depicted in Figure 1 and 2, were determined using in vitro experiments with tissue fractions [18, 20, 44–46]. PBBK models for estragole and coumarin in rat and human were developed based on these in vitro metabolic data. For estragole in rat and human the models defined consisted of seven compartments including blood, liver, kidney, lung, fat, richly perfused tissue and slowly perfused tissue [44, 47]. For coumarin in rat and human the models defined consisted of five compartments including blood, liver, fat, richly perfused tissue and slowly perfused tissue [45, 47]. A schematic diagram of both PBBK models is shown in Figure 3.╃The physiological parameters and partition coefficients used in the models can be found in the literature [44–46]. The physiological parameters were obtained from the literature [39]. The partition coefficients were estimated from the log Kow based on a method of DeJongh et al. [48]. Log Kow values were estimated with the software package ClogP version 4.0 (Biobyte, Claremont, CA). Model equations were coded and numerically integrated in Berkeley Madonna 8.0.1 (Macey and Oster, UC Berkeley, CA, USA), using the Rosenbrock’s algorithm for stiff systems. The Vmax values for the different phase I metabolic pathways in the liver, expressed as nmol/ min.mg microsomal protein, were scaled to the liver using

129

130

Contributions

Figure 3: Schematic diagram of the PBBK model for (a) estragole and (b) coumarin [44, 45].

a microsomal protein yield of 35 mg per g liver [49]. The Vmax values for the different phase I metabolic pathways in the lung and kidney were scaled accordingly using a microsomal protein yield of 20 mg per g lung, and 7 mg per g kidney [50–52]. The Vmax values for sulfonation, oxidation and glucuronidation of 1’-hydroxyestragole, expressed as nmol/ min.mg S9 or microsomal protein, were scaled to the liver using a S9 protein yield of 143 mg per g liver and a microsomal protein yield of 35€mg per g liver [50]. The apparent in vitro Km values were assumed to correspond to the apparent in vivo Km values. The uptake of estragole and coumarin from the gastro-intestinal tract was described by a firstorder process, assuming direct entry from the intestine to the liver compartment. The absorption rate constant (Ka) was set to 1.0â•›h-1, resulting in a rapid absorption of estragole or coumarin from the gastro-intestinal tract [7].

3.6.3 Results Estragole

As an example Figure 4 presents the estragole concentration dependent rate of formation of the different estragole phase I metabolites by rat and human liver microsomes. From these curves Vmax and Km values could be derived [44]. Vmax and Km values were also determined for glucuronidation, oxidation and sulfation of 1’-hydroxyestragole [44, 45] using rat as well as human samples [44, 47]. Based on the in vitro kinetic data for the different bioactivation and detoxification reactions catalyzed by rat and human tissue fractions PBBK models for estragole metabolism in rat and human were developed [45, 47]. With these models predictions were made on formation of different metabolites in human liver in time and at different oral doses. In male rat O-demethylation of estragole appeared to be a major metabolic route at low doses of estragole, occurring mainly in the lung and kidney. Due

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

Figure 4: Estragole concentration dependent rate of formation of 4-allylphenol (a), hydroxyestragole (b), estragole-2’,3’-oxide (c), and 3’-hydroxyestragole (d) by rat (▲) and human () liver microsomes. In the plots each point represents the mean (± SD) of three replicates.

to saturation of the O-demethylation pathway in lung and kidney, formation of the proximate carcinogenic metabolite 1’-hydroxyestragole, which was shown to occur mainly in the liver of male rat, becomes relatively more important at higher doses of estragole. The PBBK model predicted that formation of this metabolite increased from 16â•›% of the dose at a dose of 0.07 mg/kg bw to 29â•›% of the dose at a dose of 300 mg/kg bw. This relative increase in formation of 1’-hydroxyestragole leads to a relative increase in formation of 1’-hydroxyestragole glucuronide, 1’-oxoestragole, and 1’-sulfooxyestragole, the latter being the ultimate carcinogenic metabolite of estragole. The formation of 1’-sulfooxyestragole predicted by the PBBK model increased from 0.08â•›% of the dose at a dose of 0.07 mg/kg bw to 0.16â•›% of the dose at a dose of 300 mg/kg bw. Overall these results indicate that the relative importance of different metabolic pathways of estragole may vary in a dose-dependent way, leading to a relative increase in bioactivation of estragole at higher doses. The findings of the PBBK model for male rat were in good agreement with observations in the literature, revealing dose-dependent effects on the biokinetics for estragole in female Wistar rats in vivo. In these in vivo studies the proportion of O-demethylation was observed to decrease with increasing doses (as determined by the percentage of exhalation as 14CO2), whereas the proportion of the dose excreted as 1’-hydroxyestragole glucuronide in the urine increased from 1.3–5.4â•›% of the dose in the range of 0.05–50 mg/kg bw to 11.4–13.7â•›% in the dose range of 500–1000 mg/kg bw [7]. The PBBK model provided insight in

131

132

Contributions

the mechanism underlying this dose-dependent effect observed in vivo, which was identified to be a result of saturation of the O-demethylation pathway in the lung and kidney. Based on the PBBK model for estragole in human dose-dependent effects in bioactivation and detoxification of estragole in humans could be studied as well. In humans no relative increase in formation of 1’-sulfooxyestragole was identified to occur with increasing dose levels. The PBBK model even predicted that the relative formation of this metabolite decreased from 0.19â•›% of the dose at a dose of 0.07 mg/kg bw to 0.08â•›% of the dose at a dose of 300 mg/kg bw, due to saturation of the 1’-hydroxylation pathway in the liver. Further analysis revealed that this difference between the rat and human model, showing respectively an increase versus a decrease in the relative formation of 1’-sulfooxyestragole with increasing dose, was due to the fact that in the human model efficient O-demethylation in lung and kidney was absent, whereas in the rat these conversions reduced the relative formation of 1’-sulfooxyestragole at low dose levels. The human PBBK model also revealed that at a dose-range within one order of magnitude of the estimated average dietary human intake of 0.07 mg/kg bw, these dose-dependent effects on the relative percentage of the dose converted to 1’-sulfooxyestragole were not significant. The performance of the PBBK model defined for estragole in human could, to some extent, be evaluated against available in vivo data on the disposition of 0.001 mg/kg bw [methoxy-14C]-labelled estragole in two human volunteers obtained from Sangster et al. [8]. The PBBK model predicted formation of 1’-hydroxyestragole glucuronide, corresponding to 2.0â•›% of the dose after 24â•›h, is comparable to the reported in vivo level of this metabolite being ~0.5â•›% of the dose [8]. The predicted formation of 4-allylphenol, corresponding to 2.4â•›% of the dose after 8â•›h, is 4-fold lower than the reported in vivo level of ~10â•›% of the dose after 8â•›h [8]. These results indicate that the PBBK model predicts the formation of these metabolites within the same order of magnitude as the reported levels. Figure 5 presents an overview of the PBBK based predictions for the dose-dependent formation of 4-allylphenol, resulting from O-demethylation, 1’-hydroÂ� xyestragole, the proximate carcinogenic metabolite, 1’-sulfooxyestragole, the ultimate carcinogenic metabolite, 1’-hydroxyestragole glucuronide, and 1’-oxoestragole in the liver of rat and human at dose levels up to 300 mg/Â�kg bw. The results obtained clearly reflect significant species dependent differences in the relative importance of O-demethylation, being more important in male rat than in human (Fig.╃5â•›a), as well as in the major pathway for detoxification of 1’-hydroxyestragole, being glucuronidation to 1’-hydroxyestragole glucuronide in male rat (Fig.╃5â•›d) but oxidation to 1’-oxoestragole in human (Fig.╃5â•›e). These results also indicate that lower levels of urinary excretion of 1’-hydroxyestragole glucuronide in human than in male rat do not necessarily reflect lower levels of formation of the proximate and ultimate carcinogenic metabolites 1’-hydroxyestragole and 1’-sulfooxyestragole. The PBBK results obtained indicate that in spite of marked species differences in O-demethylation of estragole and in glucuronidation and oxidation of 1’-hydroxyestragole, the resulting species

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

Figure 5: PBBK model based predictions for the dose-dependent formation of (a) 4-allylphenol, (b) 1’-hydroxyestragole, (c) 1’-sulfooxyestragole, (d) 1’-hydroxyestragole glucuronide, and (e) 1’-oxoestragole in the liver of rat (─) and human (– – –) at dose levels up to 300 mg/kg bw.

differences in formation of 1’-hydroxyestragole and 1’-sulfooxyestragole up to dose levels of 50 mg/kg bw are moderate and amount to less than a 2-fold species dependent variation in bioactivation. Formation of 1’-oxoestragole has not been considered to be an important metabolic route of 1’-hydroxyestragole before, mainly because in rat only relatively small amounts of derivatives of this metabolite have been detected in the urine after exposure to estragole [53]. Based on the approach of identifying principal metabolic pathways of estragole in incubations with tissue fractions of relevant organs, it could be revealed that in human oxidation of 1’-hydroxyestragole to 1’-oxoestragole is a major metabolic pathway, which was predicted by the PBBK model to account for 62.7â•›% of the dose. Altogether it is concluded that the species dependent variation in bioactivation of estragole to 1’-sulfooxyestragole is smaller than the default factor of

133

134

Contributions

4€generally assumed to reflect inter-species variation in kinetics (assuming that the default factor of 10 can be divided into a factor of 4 for kinetics and 2.5 for dynamics) [54].

Coumarin

Figure 6 presents the HPLC chromatograms of incubations of coumarin with microsomes from pooled human and rat liver reflecting the species dependent

Figure 6: HPLC chromatograms of 30 min incubations of 1000╃µM coumarin with (a) pooled human liver microsomes and (b) pooled rat liver microsomes. An unidentified metabolite peak marked with a question mark [45].

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

Figure 7: PBBK model-predicted dose-dependent concentration of oHPA in the liver of (a) rat and (b) a human wild-type CYP2A6 subject (dotted line) and a human homozygous CYP2A6 deficient subject (Vmax for coumarin 7-hydroxylation set to zero, solid line).€

differences in formation of 7-hydroxycoumarin, the major metabolite formed by human liver microsomes (Fig.╃6╛a) and oHPA, the major metabolite formed by rat liver microsomes (Fig.╃6╛b). In rat microsomal incubations there was no formation of 7-hydroxycoumarin whereas in human microsomal incubations formation of oHPA was not observed to any significant extent. The PBBK model defined for coumarin included: (1) uptake of coumarin from the intestine by passive diffusion, (2) transport to the liver, fat and all other organs lumped together as either rapidly perfused tissue or slowly perfused tissue, (3) hepatic metabolism of coumarin to 7-hydroxycoumarin, oHPA, 3-hydroxycoumarin and 4-hydroxy-3-glutathionyl-coumarin (CE-SG) and (4) conversion of oHPA to oHPAA and oHPE. The PBBK model thus defined provided relative estimates of liver levels of oHPA, in man and rat, but also in humans deficient in coumarin 7-hydroxylation, at increasing levels of coumarin exposure (Fig.╃7). For rat liver a dose-dependent increase in the Cmax for oHPA formation is observed (Fig.╃7╛a). For human liver of wild-type CYP2A6 subjects (Fig.╃7╛b dotted line) a dose-dependent increase in oHPA formation is only observed at dose levels above 15 mg/kg bw when 7-hydroxylation of coumarin becomes saturated and additional amounts of coumarin start to be metabolized through alternative biochemical pathways. For homozygous CYP2A6 deficient subjects, with Vmax for coumarin 7-hydroxylation set to zero, there is a dose-dependent increase in the Cmax for oHPA in the liver without an apparent threshold (Fig.╃7╛b solid line). Nevertheless, comparison of Figure 7╛b to figure 7╛a reveals that along the whole dose range modelled the predicted oHPA levels in liver of CYP2A6 deficient subjects remain at least 10-fold lower than the Cmax values predicted for oHPA in rat liver at similar dose levels.

3.6.4 Discussion The results presented show that integrating in vitro metabolic parameters, using a PBBK model as a framework, provides a good method to evaluate the occurrence of dose-dependent effects and species differences in bioactivation and

135

136

Contributions

detoxification of estragole and coumarin. Using this approach, mechanisms underlying dose-dependent effects in bioactivation were revealed. Furthermore, insight was obtained in the occurrence of species differences in metabolism and metabolic activation.

Implications for Risk Assessment Coumarin

For coumarin significant species differences exist in metabolism. In man, the 3,4-epoxidation of coumarin leading to the hepatotoxic oHPA is a minor route, whereas in rats the detoxifying 7-hydroxylation appears to be a minor route. Whether these species differences in toxicokinetics may provide an argument for reduction of the inter-species safety factor when extrapolating from the animal studies to the human situation, as previously suggested by Felter et al. [19], is dependent on how these differences in kinetics together influence the levels of oHPA in the liver of rat and human. To provide some insight into this question a PBBK model for coumarin for both rat and human was developed, taking into account coumarin 7-hydroxylation, coumarin 3-hydroxylation, formation of the glutathione conjugate of coumarin 3,4-epoxide, formation of oHPA, detoxification of oHPA to oHPAA and conversion of oHPA to oHPE. The PBBK models presented may not give insight in the absolute formation of oHPA in the liver of rat and human in vivo since the models were not validated against in vivo data. Nevertheless, the model simulations give insight in the relative differences in oHPA formation in the liver of rat and human, in order to assist extrapolation of rat data to the human situation. The predicted Cmax for oHPA in the liver of the average CYP2A6 wild-type human subject was predicted to be about three orders of magnitude lower than the Cmax predicted for the liver of rats, representing a species sensitive to coumarin liver toxicity [19]. The PBBK models developed also allowed modelling of the kinetics in CYP2A6 deficient human subjects that are homozygous for the CYP2A6*2 allele (Vmax for coumarin 7-hydroxylation set to zero). The PBBK model thus defined revealed that for these CYP2A6 deficient human subjects the Cmax and AUC0–24â•›h for oHPA formation could amount to values that were respectively 70- and 500-fold higher than those predicted for the average CYP2A6 wild-type human subject. The increased AUC is in line with observations reported before for an individual who was homozygous for the CYP2A6*2 allele in which approximately 50â•›% of a 2 mg dose of coumarin was excreted as oHPAA [29], reflecting a significant increase in the percentage of coumarin excreted over time via the coumarin 3,4-epoxide pathway upon homozygous CYP2A6 deficiency. Assuming that all oHPAA formed will be excreted in the urine, the PBBK model predicts that upon dosing 2 mg (0.03 mg/kg bw for a 60 kg person), corresponding to the dose applied in the Hadidi et al. study [29] and within the range of estimated normal dietary exposure [15], the excretion of oHPAA will be 0.05â•›% of the dose for a wild-type human subject and

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

increases to 28â•›% of the dose for€a homozygous CYP2A6 deficient human subject. This increase in the predicted value of the dose excreted as oHPAA to 28â•›% approximately explains the 50â•›% reported by Hadidi et al. [29] and reveals that the model for the CYP2A6 deficient human subjects is in good agreement with the in vivo data obtained by Hadidi et al. [29]. Cmax values predicted for oHPA formation in the liver of CYP2A6 wild-type human and of homozygous CYP2A6 deficient subjects were, respectively, about 1000- and 10-fold lower than Cmax values predicted for rat liver. For wild-type human subjects and the subjects with completely deficient coumarin 7-hydroxylation the AUC0–24â•›h values for oHPA in the liver were also, respectively, about 1000- and 10-fold lower than that for rat liver. This points at reduced chances on oHPA liver toxicity in humans as compared to rat even for homozygous CYP2A6 deficient subjects. The results obtained also demonstrated that this holds over a dose range from 0.1 mg/kg bw (the TDI) to 50 mg/kg bw (Fig.╃7). It is concluded that even in human subjects with complete deficiency in detoxifying 7-hydroxylation the chances on formation of the hepatotoxic coumarin metabolite oHPA will be lower than those expected in the liver of rats when exposed to a similar dose on a body weight basis. Clinical data and case reports have been interpreted to indicate that a subgroup within the human population might be especially sensitive to the hepatotoxic effects of coumarin, occurring a few weeks after treatment [55–57]. Our modelling results corroborate that the CYP2A6 polymorphism is unlikely to be the factor underlying the higher sensitivity of these individuals. This outcome agrees with the lack of a correlation between hepatotoxic responses and the CYP2A6 genotype status of the patients [58] because the frequency of homozygous individuals with two defective alleles in the general population is estimated to be much lower than the frequency of study patients with elevated liver enzymes. The higher sensitivity in these individuals may be due to other as yet unsolved factors which may include the fact that these studies were generally not performed in healthy subjects, such as patients with chronic lymphedema [55], chronic venous insufficiency [56, 57], individuals with a history of hepatitis [56, 57], and/ or upon concomitant exposure to troxerutin [56, 57]. Increased toxicity could also be due to bolus dosing rather than dietary administration. Therefore, it can be concluded that the human studies include several confounding factors and that the reason for the increased susceptibility to liver damage of some individuals within the groups of patients treated with coumarin remains to be established. This could be a reason for taking not only these patient studies, but also animal studies into account in the safety assessment on coumarin. The PBBK results reveal that in human subjects, even when 7-hydroxylation is deficient, the chances on formation of the toxic oHPA metabolite will be significantly lower than those expected in the liver of rats when exposed to a similar dose on a body weight base. This conclusion should be taken into account when extrapolating data from experimental studies in sensitive animals, i.╃e. rats, to the general human population, and could be a reason to reduce the uncertainty factor for inter-species variation used for definition of the TDI from 10 to 2.5

137

138

Contributions

leaving only the factor 2.5 for toxicodynamic differences but taking out the factor 4 for toxicokinetics, as suggested before [60]. Estragole

Worldwide different approaches exist to assess the risk of compounds that are both genotoxic and carcinogenic. Numerical estimates of the risk associated with human exposure might be derived by extrapolation of carcinogenicity data obtained in animals at high dose levels to low dose levels relevant for the human situation. Many mathematical models have been proposed by which such an extrapolation below the available experimental data can be performed, of which linear extrapolation is the simplest form [59]. Extrapolating from animal tumour data at high doses using mathematical modelling in order to obtain estimates of the risk to humans at low dose exposure levels has been much debated, since it is not known whether or not the model chosen actually reflects the underlying biological processes. In addition, it is argued that species differences are not taken into account and that obtaining numerical estimates may be misused or misinterpreted in further risk management and risk communication, where the uncertainties and inaccuracy connected to the model may not be communicated [60]. Considering these disadvantages the Scientific Committee of the European Food Safety Authority (EFSA) recommends using a different approach, known as the Margin of Exposure (MOE) approach [60]. The MOE approach uses a reference point, usually taken from data from an animal experiment that represents a dose causing a low but measurable cancer response. It can be for example the BMDL10, the lower confidence bound of the Benchmark Dose that gives 10â•›% (extra) cancer incidence (BMD10). The MOE is defined as the ratio between this reference point, the BMDL10, and the estimated dietary intake (EDI) in humans. When this MOE is higher than 10,000, the compound is considered to be of low priority for risk management actions [60–62]. This safety margin of 10,000 is applied to adequately allow for various uncertainties in the MOE approach, including: 1. A factor of 100 for species differences and human variability in biokinetics and biodynamics. 2. A factor of 10 for inter-individual human variability in cell cycle control and DNA repair. 3. A factor of 10 for uncertainties in the shape of the dose-response curve outside the observed dose range. To date carcinogenicity data for estragole from which a BMDL10, and thus a MOE, can be derived result from a long-term carcinogenicity study conducted in mice [5]. Table 1 presents the carcinogenicity data obtained for estragole in female mice in this study. A BMD analysis of these data using BMDS version 1.4.1â•›c software was performed of which the results are given in Table 2.╃Based on the results presented in Table 2 it is concluded that the BMDL10 value varies between 9 and 33 mg/kg bw per day.

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment Table 1: Overview of the data from Miller et al. [5] on the incidence of haematomas in female mice exposed for 12 months via the diet to estragole. Dose

Estimated dose No. of animals No. of mice with mg/kg bw per day haematomas

Incidence

0

0

43

0

0

0.23â•›% in diet

150–300

48

27

56

0.46â•›% in diet

300–600

49

35

71

Table 2: Results of a BMD analysis of the data from Miller et al. [5] on the incidence of hepatomas in female mice exposed for 12 months via the diet to estragole (Tab.╃1), using BMDS version 1.4.1╛c and the default settings of extra risk, a Benchmark Response (BMR) of 10╛% and a 95╛% confidence limit. To make a worst case estimate the lowest dose levels of the range were used (i.╃e.╃150 and 300 mg/kg bw per day respectively). Mice gender

Model

No. of parameters

Log likeli- Accepted hood

BMD10 mg/kg bw per day

BMDL10 mg/kg bw per day

Females

null

1

–96â•›1243

Females

full

3

–62â•›2103

Females

two-stage

1

–62â•›7403

yes

22.4

18.1

Females

gamma

1

–62â•›7403

yes

22.4

18.1

Females

log-logistic

1

–62â•›2124

yes

13.1

9.2

Females

log-probit

1

–62â•›7928

yes

40.7

32.7

Females

Weibull

1

–62â•›7403

yes

22.4

18.1

The average per capita daily intake of estragole was estimated by the Scientific Committee on Food of the European Union (SCF) to amount to about 4.3 mg per day (corresponding to 0.07 mg/kg bw per day for a 60 kg person) [3]. This estimation is based on a relative conservative method using theoretical maximum use levels of estragole in 28 food categories and consumption data for these food categories based on seven days dietary records of adult individuals [3]. Using a different method, a lower average per capita daily intake of estragole was estimated by the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) [1]. This estimation was performed using production volume data of herbs, essential oils, and flavour substances containing estragole in the US [1]. The FEMA estimated the daily per capita intake to be less than 10╃µg/ kg bw per day [1]. Using the exposure assessment provided by the SCF [3] of 0.07 mg/kg bw per day and the BMDL10 of 9 to 33 mg/kg bw per day, the MOE value would amount to 129 to 471, which is lower than 10,000, indicating that the consumption of estragole at these intake levels might be a high priority for risk management. Using the exposure assessment provided by Smith et al. [1] of 0.01/kg bw per day and the BMDL10 of 9 to 33/kg bw per day, the MOE value would amount to 900 to 3300.╃Comparison of this MOE value to the value of 10,000 indicates

139

140

Contributions

that at these intake levels the use of estragole containing spices and their essential oils might also be considered a priority for risk management. In the opinion of the EFSA it has been stated that the default MOE of 10,000 can be reduced or increased when appropriate chemical specific data are available [60]. The results of our PBBK modelling can provide insight in especially the applicability of the default safety factor for species differences in biokinetics used to define the value of 10,000.╃The outcomes of the PBBK models of the present work reveal that species differences in bioactivation of estragole were observed to be about 2-fold and thus smaller than the default factor of 4 generally assumed to reflect inter-species variation in kinetics. However, a 2-fold reduction of the default value of 10,000 would not lead to a different conclusion on the priority for risk management. A similar conclusion emerges from the approach in which linear extrapolation from a defined point of departure is used to derive a so-called virtual safe dose (VSD) at which the additional cancer risk upon life-time exposure would be one in a million and considered negligible [63]. Using the data and BMD analysis of the study of Miller et al. [5] (Tab.╃1 and 2) it can be concluded that in mice, a BMR (Benchmark Response) of 10â•›% extra tumour risk is observed at a BMD10 value of 13 to 41 mg/kg bw per day. By linear extrapolation from this point of departure, the VSD that results in an additional cancer risk of 1 in a million is calculated to amount to 0.13 to 0.41╃µg/kg bw per day. Comparison of this estimated VSD to the estimated dietary human intake of 10–70╃µg/kg bw per day [1, 3] indicates that dietary intake levels are about two orders of magnitude above the VSD, indicating a priority for risk management. The results of the PBBK models developed for estragole for male rat and human indicate that kinetic data do not provide a reason to argue against such a linear extrapolation from the rat tumour data to the human situation. This is illustrated in Figure 8, in which the PBBK model-predicted dose-dependent formation of 1’-sulfooxyestragole in the liver of rat and human is displayed. Both curves appear to be quite linear from doses as high as the BMD10 at which actual increased tumour incidences are observed in rodent bioassays, down to as low as the VSD, when plotted on a log-log scale as done in Figure 8 as well as on a linear scale (Figure not shown). Since the BMD10 appears to be within the linear part of the curve and since the rat and human curves do not differ substantially, the PBBK results of the present study support that possible non-linear kinetics and species differences in kinetics should not be used as arguments against using this linear low dose extrapolation from high dose animal data to the low dose human situation. Altogether, the results presented demonstrate that PBBK models provide a useful tool in risk assessment of food-borne chemicals when evaluating human risks.

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

Figure 8: PBBK model-predicted dose-dependent formation of 1’-sulfooxyestragole in the liver of rat (─) and human (– – –).

Additional Considerations

Whereas the risk assessments outlined above for estragole and coumarin take into account the predicted data on dose-dependent effects, species differences and inter-individual differences in bioactivation, it should be noted that other factors might affect the risk assessment as well. The carcinogenic effects of estragole and coumarin will for instance also depend on toxicodynamic processes (i.╃e. processes of importance for the ultimate formation and development of tumours). This could be investigated in further detail by extending the PBBK models to so-called physiologically based biodynamic (PBBD) models in which dose levels and 1’-sulfooxyestragole or oHPA formation should be coupled to DNA adduct formation, considered a biomarker of exposure, or to toxicity and – ultimately – cancer incidence. In addition, it should be noted that whereas animal carcinogenicity experiments are conducted with a pure compound, human dietary exposure to estragole or coumarin occurs in a complex food matrix containing other (herbal) ingredients. In a complex food matrix, interactions can occur that can affect the bioavailability of food components [46, 64]. For example, a slow or incomplete release of estragole or coumarin from the matrix could result in a reduced bioavailability as compared to the bioavailability when dosed as a pure compound by oral gavage. In addition to the effect of the food matrix on the bioavailability, interactions with other herbal ingredients might occur at the level of metabolic activation and/ or detoxification [46, 64]. It was for instance observed by Jeurissen et al. [65] that a methanolic basil extract is able to efficiently inhibit the sulfotransferase mediated DNA adduct formation in HepG2 human hepatoma cells exposed to 1’-hydroxyestragole. These results suggest that the bioactivation of estragole and subsequent adverse effects of estragole are prob-

141

142

Contributions

ably lower when estragole is consumed in a matrix of other basil ingredients than would be expected on the basis of experiments using estragole as a single compound. Whether this inhibition of DNA adduct formation by matrix ingredients could also occur in vivo was, however, not yet established and should be further explored. In conclusion, the data presented show that PBBK modelling provides a good method to evaluate the occurrence of dose-dependent effects, species differences, and human variability in bioactivation and detoxification. The model predictions obtained can be used to provide a more mechanistic basis for the assessment of the effects in humans at low dose dietary intake levels based on data obtained in experiments with rodents at high dose levels. However, for a complete assessment of the cancer risk at low dose human intake scenarios additional information is still needed. For example more insight will be needed in toxicodynamic processes that can affect the risk assessment, and more insight is needed in the modulating effects of herbal ingredients on the carcinogenic process resulting in a so-called matrix effect.

Acknowledgements Part of the work on coumarin was supported by the Dutch Ministry of Economic Affairs (Innovation Vouchers G071064 and G062238), J.╃S. Polak Koninklijke Specerijenmaalderij b.╃v., and the Vereniging voor de Bakkerij- en Zoetwarenindustrie (VBZ). Part of the work on estragole was supported by the Nestlé Research Center Lausanne, Switzerland.

Abbreviations BMD: Benchmark Dose BMDL: lower confidence bound of the Benchmark Dose BMR: Benchmark Response CE-SG: 4-hydroxy-3-glutathionyl-coumarin EDI: Estimated daily Intake EFSA: European Food Safety Authority FEMA: Flavor and Extract Manufacturers Association JECFA: Joint FAO/WHO Expert Committee on Food Additives MOE: Margin of Exposure oHPA: o-hydroxyphenylacetaldehyde oHPAA: o-hydroxyphenylacetic acid oHPE: o-hydroxyphenylethanol PBBK model: physiologically based biokinetic model SCF: Scientific Committee on Food VSD: Virtual Safe Dose

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

Conflict of Interest Statement The authors declare that there are no financial/ commercial conflicts of interest.

References â•⁄ 1. Smith, R.╃L., Adams, T.╃B., Doull, J., Feron, V.╃J., Goodman, J.╃I., Marnett, L.╃J., Portoghese, P.╃S., Waddell, W.╃J., Wagner, B.╃M., Rogers, A.╃E., Caldwell, J., Sipes, I.╃G., Safety assessment of allylalkoxybenzene derivatives used as flavouring substances – methyl eugenol and estragole. Food Chem. Toxicol. 2002, 40, 851–870. â•⁄ 2. Siano, F., Ghizzoni, C., Gionfriddo, F., Colombo, E., Servillo, L., Castaldo, D., Determination of estragole, safrole and eugenol methyl ether in food products. Food Chem. 2003, 81, 469–475. â•⁄ 3. SCF (Scientific Committee on Food, Opinion of the scientific committee on food on estragole (1-allyl-4-methoxybenzene). 2001 http://ec.europa.eu/ food/fs/sc/scf/out104_en.pdf.╃ â•⁄ 4. Drinkwater, N.╃R., Miller, E.╃C., Miller, J.╃A., Pitot, H.╃C., Hepatocarcinogenicity of estragole (1-allyl-4-methoxybenzene) and 1’-hydroxyestragole in the mouse and mutagenicity of 1’-acetoxyestragole in bacteria. J. Natl. Cancer Inst. 1976, 57, 1323–1331. â•⁄ 5. Miller, E.╃C., Swanson, A.╃B., Phillips, D.╃H., Fletcher, T.╃L., Liem, A., Miller, J.╃A., Structure–activity studies of the carcinogenicities in the mouse and rat of some naturally occurring and synthetic alkenylbenzene derivatives related to safrole and estragole. Cancer Res.╃1983, 43, 1124–1134. â•⁄ 6. Wiseman, R.╃W., Miller, E.╃C., Miller, J.╃A., Liem, A., Structure–activity studies of the hepatocarcinogenicities of alkenylbenzene derivatives related to estragole and safrole on administration to preweanling male C57BL/6J x C3H/HeJ F1mice. Cancer Res.╃1987, 47, 2275–2283. â•⁄ 7. Anthony, A., Caldwell, J., Hutt, A.╃J., Smith, R.╃L., Metabolism of estragole in rat and mouse and influence of dose size on excretion of the proximate carcinogen 1’-hydroxyestragole. Food Chem. Toxicol. 1987, 25, 799–806. â•⁄ 8. Sangster, S.╃A., Caldwell, J., Hutt, A.╃J., Anthony, A., Smith, R.╃L., The metabolic disposition of [methoxy-14C]-labelled trans-anethole, estragole and p-propylanisole in human volunteers. Xenobiotica 1987, 17, 1223–1232. â•⁄ 9. Hall, R.╃L., Oser, B.╃L., Recent progress in the consideration of flavoring ingredients under the food additives amendment III. GRAS substances. Food Technol. 1965, 253, 151–197. 10.╃ JECFA Sixty-ninth meeting, 2008 Rome, Italy, 17–26 June 2008.╃http:// www.who.int/entity/ipcs/food/jecfa/summaries/summary69.pdf.╃ 11. Lake, B.╃G., Coumarin metabolism, toxicity and carcinogenicity: relevance for human risk assessment. Food Chem. Toxicol. 1999, 37, 423–453. 12. SCF (Scientific Committee on food), Opinion on coumarin (a constituent of natural flavouring source materials limited by annex II of flavourings

143

144

Contributions

directive 88/388/EC) expressed on 16 December 1994.╃Reports of the Scientific Committee for Food. Thirty-sixth series, 1997, 13–20. 13. EFSA, Opinion of the Scientific Panel on Food additives, flavourings, processing aids and materials in contact with food (AFC) on a request from the commission related to coumarin. Question number EFSA-Q-2003–118.╃EFSA J.╃2004, 104, 1–36. 14. NTP (National Toxicology Program), Toxicology and carcinogenesis studies of coumarin (CAS No.╃91–64–5) in F344/N rats and B6C3F1 mice (gavage studies). Technical Report No. NTP TR 422.╃NIH Publication No.╃93–3153, US Department of Health and Human Services, NIH, Research Triangle Park, NC 1993. 15. Carlton, B.╃D., Aubrun J-C., Simon G.╃S., Effects of coumarin following perinatal and chronic exposure in Sprague-Dawley rats and CD-1 mice. Fundam. Appl. Toxicol. 1996, 30, 145–151. 16. Born, S.╃L., Rodriguez, P.╃A., Eddy, C.╃L., Lehman-McKeeman, L.╃D., Synthesis and reactivity of coumarin 3,4-epoxide. Drug Metab. Dispos. 1997, 25, 1318–1324. 17. Lake B.╃G., Gray, T.╃J.╃B., Evans, J.╃G., Lewis, D.╃F.╃V., Beamand, J.╃A., Hue, K.╃L., Studies on the mechanism of coumarin-induced toxicity in rat hepatocytes: comparison with dihydrocoumarin and other coumarin metabolites. Toxicol. Appl. Pharmacol. 1989, 97, 311–323. 18. Vassallo, J.╃D., Hicks, S.╃M., Daston, G.╃P., Lehman-McKeeman, L.╃D., Metabolic detoxification determines species differences in coumarin-induced hepatotoxicity. Toxicol. Sci.╃2004, 80, 249–257. 19. Felter, S.╃P., Vassallo, J.╃D., Carlton, B.╃D., Daston, G.╃P., A safety assessment of coumarin taking into account species-specificity of toxicokinetics. Food Chem, Toxicol. 2006, 44, 462–475. 20. Born, S.╃L., Caudill, D., Smith, B.╃J., Lehman-McKeeman, L.╃D., In vitro kinetics of coumarin 3,4-epoxidation: application to species differences in toxicity and carcinogenicity. Toxicol. Sci.╃2000, 58, 23–31. 21. Fentem J.╃H., Fry, J.╃R., Whiting, D.╃A., o-Hydroxyphenylacetaldehyde: A major novel metabolite of coumarin formed by rat, gerbil and human liver microsomes. Biochem. Biophys. Res. Commun. 1991, 179, 197–203. 22. Lake, B.╃G., Osborne, D.╃J., Walters, D.╃G., Price R.╃J., Identification of o-hydroxyphenylacetaldehyde as a major metabolite of coumarin in rat hepatic microsomes Food Chem. Toxicol. 1992, 30, 99–104. 23. Born, S.╃L., Api, A.╃M., Ford, R.╃A., Lefever, F.╃R., Hawkins, D.╃R., Comparative metabolism and kinetics of coumarin in mice and rats. Food Chem. Toxicol. 2003, 41, 247–258. 24. Van Iersel, M., Walters, D.╃G., Price, R.╃J., Lovell, D.╃P., Lake B.╃G., Sex and strain differences in mouse hepatic coumarin 7-hydroxylase activity. Food Chem. Toxicol. 1994, 32, 387–390. 25. Van Iersel, M.╃L., Henderson, C.╃J., Walters, D.╃G., Price, R.╃J., Wolf, C.╃R., Lake, B.╃G., Metabolism of [3–14C]coumarin by human liver microsomes. Xenobiotica 1994, 24, 795–803.

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

26. Draper, A.╃J., Madan, A., Parkinson, A., Inhibition of coumarin 7-hydroxylase activity in human liver microsomes. Arch. Biochem. Biophys. 1997, 341, 47–61. 27. Bogan, D.╃P., Deasy, B., O’Kennedy, R., Smyth, M.╃R., Interspecies differences in coumarin metabolism in liver microsomes examined by capillary electrophoresis. Xenobiotica 1996, 26, 437–445. 28. Fernandez-Salguero, P., Hoffman, S.╃M.╃G., Cholerton, S., Mohrenweiser, H., Raunio, H., Rautio, A., Pelkonen, O., Huang J.╃D., Evans, W.╃E., Idle, J.╃R., Gonzalez, F.╃J., A genetic polymorphism in coumarin 7-hydroxylation: sequence of the human CYP2A genes and identification of variant CYP2A6 alleles. Am. J. Human Genet. 1995, 57, 651–660. 29. Hadidi, H., Zahlsen, K., Idel, J.╃R., Cholerton, S., A single amino acid substitution (Leu160His) in cytochrome P450 CYP2A6 causes switching from 7-hydroxylation to 3-hydroxylation of coumarin. Food Chem.Toxicol. 1997, 35, 903–907. 30. Hadidi, H., Irshaid, Y., Vågbø, C.╃B., Brunsvik, A., Cholerton, S., Zahlsen, K., Idle, J.╃R., Variability of coumarin 7- and 3-hydroxylation in a Jordanian population is suggestive of a functional polymorphism in cytochrome P450 of CYP2A6. Eur. J. Clin. Pharmacol. 1998, 54, 437–441. 31. Andersen, M.╃E., Krishnan, K., Physiologically-based pharmacokinetics and cancer risk assessment. Environ. Health. Persp. 1994, 102, 103–108. 32. Clewell, H.╃J., Gentry, P.╃R., Gearhart, J.╃M., Allen, B.╃C., Andersen, M.╃E., Comparison of cancer risk estimates for vinyl chloride using animal and human data with a PBPK model. Sci. Total Environ. 2001, 274, 37–66. 33. Clewell, H.╃J., 3rd, Andersen, M.╃E., Barton, H.╃A., A consistent approach for the application of pharmacokinetic modeling in cancer and noncancer risk assessment. Environ. Health Perspect. 2002, 110, 85–93. 34. Krewski, D., Withey J.╃R., Ku L.╃F., Andersen M.╃E., Applications of physiologic pharmacokinetic modeling in carcinogenic risk assessment. Environ. Health. Perspect. 1994, 102 (Suppl 11), 37–50. 35. Krishnan, K., Andersen, M.╃E., Physiologically-based pharmacokinetic modeling and toxicology. In: Principles and methods of toxicology (A.╃W. Hayes, Ed.), Raven Press, N.╃Y.╃2001, 193–241. 36. Chiu, W.╃A., Barton, H.╃A., DeWoskin, R.╃S., Schlosser, P., Thompson, C.╃M., Sonawane, B., Lipscomb, J.╃C., Krishnan, K., Evaluation of physiologically based pharmacokinetic models for use in risk assessment. J. Appl. Toxicol. 2007, 27, 218–237. 37. Clewell, R.╃A., Clewell, H.╃J., 3rd, Development and specification of physiologically based pharmacokinetic models for use in risk assessment. Regul. Toxicol. Pharmacol. 2008, 50, 129–143. 38. Bogaards, J.╃J.╃P., Hissink, E.╃M., Briggs, M., Weaver, R., Jochemsen, R., Jackson, P., Bertrand, M., and van Bladeren, P.╃J. Prediction of interindividual variation in drug plasma levels in vivo from individual enzyme kinetic data and physiologically based pharmacokinetic modeling. Eur. J. Pharm. Sci. 2000, 12, 117–124.

145

146

Contributions

39. Brown, R.╃P., Delp, M.╃D., Lindstedt, S.╃L., Rhomberg, L.╃R., Beliles, R.╃P., Physiological parameter values for physiologically based pharmacokinetic models. Toxicol. Indust. Health 1997, 13, 407–484. 40. Gargas, M.╃L., Burgess, R.╃J., Voisard, D.╃E., Cason, G.╃H., Andersen, M.╃E., Partition coefficients of low molecular-weight volatile chemicals in various liquids and tissues. Toxicol. Appl. Pharmacol. 1989, 98, 87–99. 41. Jepson, G.╃W., Hoover, D.╃K., Black, R.╃K., McCafferty, J.╃D., Mahle, D.╃A., Gearhart, J.╃M., Partition coefficient determination for non-volatile and intermediate volatility chemicals in biological tissues. Toxicologist 1992, 12, 262. 42. Payne, M.╃P., Kenny, L.╃C., Comparison of models for the estimation of biological partition coefficients. J. Toxicol. Environ. Health A 2002, 65, 897–931. 43. Lipscomb, J.╃C., Poet, T.╃S., In vitro measurements of metabolism for application in pharmacokinetic modeling. Pharmacol. Ther. 2008, 118, 82–103. 44. Punt, A., Delatour, T., Scholz, G., Schilter, B., van Bladeren, P.╃J., Rietjens, I.╃M.╃C. M., Tandem mass spectrometry analysis of N2-(trans-isoestragole3’-yl)-2’deoxyguanosime as a strategy to study species differences in sulfotransferase conversion of the proximate carcinogen 1’-hydroxyestragole. Chem. Res. Toxicol. 2007, 20, 991–998. 45. Punt, A. Freidig, A., Delatour, T., Scholz, G., Schilter, B., van Bladeren, P.╃J., Rietjens, I.╃M.╃C. M., A physiologically based biokinetic (PBBK) model for estragole bioactivation and detoxification in rat. Toxicol. Appl. Pharmacol. 2008, 231, 248–259. 46. Rietjens, I.╃M.╃C. M., Zaleska, M., Boersma.M.╃G., Punt A., Differences in simulated liver concentrations of toxic coumarin metabolites in rats and different human populations evaluated through physiologically based biokinetic (PBBK) modeling. Toxicol. in Vitro 2008, 22, 1890–1901. 47. Punt, A., Paini, A.╃A., Boersma, M.╃G., Freidig, A.╃P., Delatour, T., Scholz, G., Schilter, B., van Bladeren, P.╃J., Rietjens, I.╃M.╃C. M., Use of physiologically based biokinetic (PBBK) modeling to study estragole bioactivation and Â�detoxification in human as compared to male rat. Toxicol. Sci. 2009, 110, 255–269. 48. Ramsey, J.╃C., Andersen, M.╃E., A physiologically based description of the inhalation pharmacokinetics of styrene in rats and humans. Toxicol. Appl. Pharmacol. 1984, 73, 159–175. 49. DeJongh, J., Verhaar, H.╃J., Hermens, J.╃L., A quantitative property–property relationship (QPPR) approach to estimate in vitro tissue-blood partition coefficients of organic chemicals in rats and humans. Arch. Toxicol. 1997, 72, 17–25. 50. Barter, Z.╃E., Bayliss, M.╃K., Beaune, P.╃H., Boobis, A.╃R., Carlile, D.╃J., Edwards, R.╃J., Houston, J.╃B., Lake, B.╃G., Lipscomb, J.╃C., Pelkonen, O.╃R., Tucker, G.╃T., Rostami-Hodjegan, A., Scaling factors for the extrapolation of in vivo metabolic drug clearance from in vitro data: reaching a consensus on values of human microsomal protein and hepatocellularity per gram of liver. Curr. Drug Metab. 2007, 8, 33–45.

In Silico Methods for Physiologically Based Biokinetic (PBBK) Models Describing Bioactivation and Deto�xification of Coumarin and Estragole: Implications for Risk Assessment

51. Medinsky, M.╃A., Leavens, T.╃L., Csanády, G.╃A., Gargas, M.╃L., Bond, J.╃A., In vivo metabolism of butadiene by mice and rats: a comparison of physiological model predictions and experimental data. Carcinogenesis 1994, 7, 1329–1340. 52. Beierschmitt, W.╃P., Weiner, M., Age-related changes in renal metabolism of acetaminophen in male Fischer 344 rats. Age 1986, 9, 7–13. 53. Solheim, E., Scheline, R.╃R., Metabolism of alkenebenzene derivatives in the rat. I. p-Methoxyallylbenzene (Estragole) and p-methoxypropenylbenzene (Anethole). Xenobiotica 1973, 3, 493–510. 54. WHO, International Programme on Chemical Safety (IPCS): Assessing human health risks of chemicals: Principles for the assessment of risk to human health from exposure to chemicals. Environmental Health Criteria 210, World Health Organisation, Geneva. 1999, http://www.inchem.org/ documents/ehc/ehc/ehc210.htm. 55. Loprinzi, C.╃L., Kugler, J.╃W., Sloan, J.╃A., Rooke, T.╃W., Quella, S.╃K., Novotny, P., Mowat, R.╃B., Michalak, J.╃C., Stella, P.╃J., Levitt, R., Tschetter, L.╃K., Windschitl, H., Lack of effect of coumarin in women with lymphedema after treatment for breast cancer. New Engl. J. Med.╃1999, 340, 346–350. 56. Schmeck-Lindenau, H.╃J., Naser-Hijazi, B., Becker, E.╃W., Henneicke-von Zepelin, H.╃H., Schnitker, J., Safety aspects of a coumarin-troxerutin combination regarding liver function in a double-blind placebo-controlled study. Int. J. Clin. Pharmacol. Therapeut. 2003, 41, 193–199.€ 57. Vanscheidt, W., Rabe, E., Naser-Hijazi, B., Ramelet, A.╃A., Partsch, H., Diehm, C., Schultz-Ehrenburg, U., Spengel, F., Wirsching, M., Gotz, V., Schnitker, J., Henneicke-von Zepelin, H.╃H., The efficacy and safety of a coumarin-/ troxerutin-combination (SB-LOT) in patients with chronic venous insufficiency: a double blind placebo-controlled randomised study. Vasa 2002, 31, 185–190. 58. Burian, M., Freudenstein, J., Tegtmeier, M, Naser-Hijazi, B., Henneicke-von Zepelin, H.╃H., Legrum, W., Single copy of variant CYP2A6 alleles does not confer susceptibility to liver dysfunction in patients treated with coumarin. Int. J. Clin. Pharmacol. Therapeut. 2003, 41, 141–147. 59. COC (Committee on Carcinogenicity of chemicals in food, consumer products and the environment), Guidance on a strategy for the risk assessment of chemical carcinogens. 2004, http://www.advisorybodies.doh.gov.uk/coc/ guideline04.pdf.╃ 60. EFSA, Opinion of the scientific committee on a request from EFSA related to a harmonised approach for risk assessment of substances which are both genotoxic and carcinogenic. EFSA J.╃2005, 282, 1–31. 61. Barlow, S., Renwick, A.╃G., Kleiner, J., Bridges, J.╃W., Busk, L., Dybing, E., Edler, L., Eisenbrand, G., Fink-Gremmels, J., Knaap, A., Kroes, R., Liem, D., Muller, D.╃J., Page, S., Rolland, V., Schlatter, J., Tritscher, A., Tueting, W., Wurtzen, G., Risk assessment of substances that are both genotoxic and carcinogenic, report of an International Conference organized by EFSA and WHO with support of ILSI Europe. Food Chem. Toxicol. 2006, 44, 1636– 1650.

147

148

Contributions

62. Dybing, E., O’Brien, J., Renwick, A.╃G., Sanner, T., Risk assessment of dietary exposures to compounds that are genotoxic and carcinogenic, an overview. Toxicol. Lett. 2008, 180, 110–117. 63. EPA, Guidelines for carcinogen risk assessment and supplemental guidance for assessing susceptibility from early-life exposure to carcinogens. Washington, DC, US Environmental Protection Agency, 2005, http://cfpub.epa.gov/ncea/raf/recordisplay.cfm?deid=╃116283. 64. Schilter, B., Andersson, C., Anton, R., Constable, A., Kleiner, J., O’Brien, J., Renwick, A.╃G., Korver, O., Smit, F., and Walker, R., Guidance for the safety assessment of botanicals and botanical preparation for use in food and food supplements. Food Chem. Toxicol. 2003, 41, 1625–1649. 65. Jeurissen, S.╃M.╃F., Punt, A., Delatour, Th., Rietjens, I.╃M.╃C. M., Basil extract inhibits the sulfotransferase mediated formation of DNA adducts of the procarcinogen 1’-hydroxyestragole by rat and human liver S9 homogenates and in HepG2 human hepatoma cells. Food Chem. Toxicol. 2008, 46, 2296–2302.

In Vitro Models for Carcinogenicity Testing – Reality or Fantasy?

3.7 In Vitro Models for Carcinogenicity Testing – Reality or Fantasy? Pablo Steinberg1,2, Carsten Müller1, Kristina Ullmann1, and René Thierbach1,2

Abstract In the present article two experimental systems to evaluate the carcinogenic potential of chemicals are presented. On the one hand, the BALB/c 3T3 cell transformation has been optimized in such a way that it can detect compounds with tumour initiating and tumour promoting activities within 3 to 3.5 weeks. On the other hand, an automated version of the soft agar assay in a 96-well format is now available that allows determining within one week whether a cell line treated with a chemical is able to grow in an anchorage-independent way. At the present time an assay, in which the two above-mentioned experimental systems are combined, is being developed. Although the in vitro test systems already available or under development nowadays will not substitute the carcinogenicity testing in whole animals, they very well could help to significantly reduce the number of animals needed for the in vivo testing of carcinogenicity in the near future.

3.7.1 Introduction Up to the present time the “gold standard” method to prove whether a chemical is carcinogenic or not is to test the chemical in whole animals. However, this procedure is extremely time-consuming, makes use of a high number of animals and cannot be used to screen a high number of compounds at a time. Because of these limitations in the last few years great efforts have been undertaken to develop test systems that could be used to evaluate the carcinogenic potential of chemicals in vitro. In the present article two important advances in this research field will be described.

3.7.2 BALB/c 3T3 Cell Transformation Assay Several different in vitro transformation assays have been established for the detection of carcinogenic compounds. Three of them, the Syrian hamster embryo cell assay (the so-called SHE assay), the C3H10T1/2 cell assay and the

1

University of Potsdam, Chair of Nutritional Toxicology, Arthur-Scheunert-Allee 114–116, D-14558 Nuthetal, Germany.

2

University of Veterinary Medicine Hannover, Institute for Food Toxicology and Analytical Chemistry, Bischofsholer Damm 15, D-30173 Hannover, Germany.

149

150

Contributions

BALB/c 3T3 cell assay, have recently been reviewed in detail by the OECD [1], whereby the most promising one is the BALB/c 3T3 cell assay. A twostage protocol for the BALB/c 3T3 cell assay was described by Sakai and Sato in 1989 [2] and optimized over the years by Makoto Umeda and colleagues at the Hatano Research Institute in Japan [3–5]. The protocol used nowadays in our laboratory is shown in Figure 1.╃The initiating activity of a compound can be tested by incubating the BALB/c 3T3 clone A31–1-1 cells between day 1 and 4 with the compound to be evaluated and then treating the cells with the strong tumour promoting agent 12-O-tetradecanoyl-phorbol-13-acetate for two weeks (between day 7 and 21). At the end of the 6th week the number of foci formed is determined. When wanting to determine the tumour promoting activity of a compound the cells are first treated with the strong initiating agent 3-methylcholanthrene between day 1 and 4.╃Thereafter, the cells are incubated for two weeks with the compound being evaluated and the number of cell foci is determined at the end of the 6th week. In order to shorten the duration of the cell transformation assay BALB/c 3T3 cells transfected with v-Ha-ras [6], so-called Bhas 42 cells, are now being used [7]. By doing so transformed foci can develop within 3 to 3.5 weeks. Interlaboratory studies have clearly shown that the Bhas 42 cell-based transformation assay is able to detect chemicals with tumour initiating and/ or tumour promoting activity [5]. A scheme of the Bhas 42 cell-based transformation assay for the detection of compounds with tumour initiating and/ or tumour promoting activity is depicted in Figure 2.╃In order to detect compounds with initiating activity in this assay, Bhas 42 cells must be seeded at a low density (2·103 cells/ ml). At this density it is guaranteed that cells carrying DNA adducts can divide, thereby “perpetuating” the DNA damage in form of mutations in the daughter cells. These cells in turn will proliferate and later give rise to the foci. The results of a cell transformation assay with Bhas 42 cells, in which the initiating activity of aflatoxin B1 was tested in the absence and in the presence of a rat liver homogenate, are exemplarily shown in Table 1.╃Aflatoxin B1 exerts its initiating activity only after having been metabolized to aflatoxin B1–8,9epoxide (e.╃g. by hepatic cytochromes P450). In line with this concept aflatoxin B1 only induced the formation of transformed foci in the presence of a rat liver homogenate (i.╃e. the S9 mix; Tab.╃1), Furthermore, the initiating activity of aflatoxin B1 was first observed when incubating the cells with concentrations of aflatoxin B1╃≥╃1 mg/ml.

Figure 1: Diagram of the two-stage protocol for the BALB/c 3T3 cell transformation assay according to Sakai and Sato [2].

In Vitro Models for Carcinogenicity Testing – Reality or Fantasy?

Figure 2: Scheme of the Bhas 42 cell-based transformation assay for the detection of tumour initiating (A) and tumour promoting chemicals (B) according to Asada et al. [5].

The data of a Bhas 42 cell-based transformation assay, in which the tumour promoting activity of 12-O-tetradecanoyl-phorbol-13-acetate and lithocholic acid were tested, are shown in Table 2.╃A concentration-dependent increase in the tumour promoting activity of both compounds was observed. An assay to test the tumour-initiating activity of the two chemicals was performed in parallel and revealed, as expected, that 12-O-tetradecanoyl-phorbol-13-acetate and lithocholic acid do not possess tumour initiating activity at all (data not shown) [5]. The BALB/c 3T3 cell transformation assay can also be used to analyze the cancer-preventive activity of natural compounds and extracts. The protocol used in our laboratory to test this activity is shown in Figure 3. As shown in Figure 4 resveratrol (trihydroxystilbene), a polyphenol with anti-oxidative and cancer-preventive activities present in red wine, is able to inhibit the chemically-induced malignant transformation of BALB/c 3T3 cells. Table 1: Initiating activity of a flatoxin B1 in the Bhas 42 cell-based transformation assay (data from [5]). Aflatoxin B1 concentration (µg/ml)

S9 mix

Initiation assay (foci/well)

0 0.05 0.1 0.2 0.5 1 2 0 0.05 0.1 0.2 0.5 1 2

– – – – – – – + + + + + + +

â•⁄ 3.0╃±â•ƒ1.1 â•⁄ 2.5╃±â•ƒ1.1 â•⁄ 1.8╃±â•ƒ1.2 â•⁄ 3.0╃±â•ƒ1.7 â•⁄ 3.5╃±â•ƒ1.5 â•⁄ 2.5╃±â•ƒ0.5 â•⁄ 0.8╃±â•ƒ1.0 â•⁄ 3.2╃±â•ƒ1.5 â•⁄ 3.7╃±â•ƒ1.6 â•⁄ 4.2╃±â•ƒ1.5 â•⁄ 4.8╃±â•ƒ3.3 â•⁄ 4.7╃±â•ƒ1.8 14.3╃±â•ƒ2.4* 11.7╃±â•ƒ2.8*

*Significantly different from the solvent control (p0.99 (preferably >0.995) and a PCR efficiency of 100â•›%±â•ƒ10â•›%. More qRT-PCR details and thresholds are recently put forward to be included in analysis and publications (MIQE) similar to the MIAME compliant strategy for microarray analysis [10]. Initially, qRT-PCR validation of microarray results was an absolute necessity, because microarray analysis was insufficiently robust. Present day microarray platforms in the hands of experienced groups are highly robust and technical validation is more a matter of making sure, than of necessity. Indeed, recent experiments in our lab and others have given identical results in a quantitative and qualitative manner, provided that microarray selected reference genes were used in qRT-PCR. It can be argued that data obtained using microarrays are more dependable, rather than less, compared to qRT-PCR, which is particularly important when small differences in gene expression are considered. The microarray whole genome transcriptome profiling technique is a powerful tool to measure differences in mRNA quantity. Differences in mRNA expression are of interest, but are only of relevance if they match to functional biology, and thus are linked to changes in either protein or metabolite levels, or Table 1: Differential gene expression of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) in adipose tissue of wild-type C57BL/6J male adult mice following nutritional intervention (n=╃12 per group) as analyzed by DNA microarray analysis (for details see text and legend of Fig.╃2). Gene Name

GAPDH C: CR: LF: E:

CR versus C

LF versus C

E versus C

ratio

p-value

ratio

p-value

Ratio

p-value

3.32

5.4·10–9

1.46

0.001

-1.02

0.69

Control, a purified 30enâ•–% high fat diet. 30â•›% caloric restriction of C. A purified low fat diet (10en% fat, fat substituted by starch). C plus 0.5â•›% EGCg.

313

314

Contributions

to the predicted physiological changes. Differences at the protein level can be confirmed using Western blotting [11] or flowmeter based multiplex analysis [12], which both provide quantitative information. Moreover, immunohistochemistry [13] can be used to provide positional information. However, these methods can be difficult or even impossible to perform when a specific and well characterized antibody is lacking or when protein concentrations are low. Another technique that can be used is 2D gel electrophoresis coupled to mass spectrometry [14]. Using this technique, proteins are separated by their isoelectric point in the first dimension and by the size of protein in the second dimension. Thereafter, proteins are stained and quantified using image analysis software. Differentially expressed proteins can be cut from the gel to be identified by mass spectrometry. This allows for simultaneous analysis of many proteins in one sample, which is referred to as proteome analysis. Proteomics has the advantage over transcriptome analysis that it directly assesses at a functional biochemical level. The diverse characteristics of proteins, including variation in size, hydrophobicity, abundancy and secondary, tertiary, or quaternary modifications imply a smaller window of targets that can be analyzed simultaneously. This precludes replacement of transcriptome analysis for genome wide identification of potential functional effects of micronutrients. This will hold true even if robustness and the window of analysis will be improved by implementing other proteomic approaches that are being developed, which include antibody arrays [15] and chromatographic separation [16]. In many ways proteome analysis is complementary to transcriptome analysis, each targeting a different functional level.

3.17.4 Magnitude of Micronutrient Effects As indicated, benefit–risk assessment will inherently deal with small functional effects, because studies will try to capture stable differences in homeostatic conditions in a relatively short time frame, rather than waiting for a life-time outcome of these changes. To exemplify the magnitude of the effects that can be expected, the number (and magnitude) of genes that changed expression was assessed for a mouse dietary intervention with three different bioactive food components and compared to a strong nutritional intervention (calorie restriction) and a widely examined nutritional intervention (high fat versus low fat). The 30â•›% calorie restriction (CR) intervention resulted in 15,041 genes that were significantly (p