Statistical Methods For Immunogenicity Assessment Yang Harry Yu
- 1. Statistical Methods For Immunogenicity Assessment Yang Harry Yu download https://ebookbell.com/product/statistical-methods-for- immunogenicity-assessment-yang-harry-yu-5281202 Explore and download more ebooks at ebookbell.com
- 2. Here are some recommended products that we believe you will be interested in. You can click the link to download. Statistical Methods For Immunogenicity Assessment Harry Yang Jianchun Zhang Binbing Yu Wei Zhao Yang https://ebookbell.com/product/statistical-methods-for-immunogenicity- assessment-harry-yang-jianchun-zhang-binbing-yu-wei-zhao-yang-23404954 Statistical Methods For Data Analysis With Applications In Particle Physics Luca Lista https://ebookbell.com/product/statistical-methods-for-data-analysis- with-applications-in-particle-physics-luca-lista-49752494 Statistical Methods For The Social And Behavioural Sciences A Modelbased Approach First Edition David B Flora https://ebookbell.com/product/statistical-methods-for-the-social-and- behavioural-sciences-a-modelbased-approach-first-edition-david-b- flora-54833924 Statistical Methods For Dynamic Disease Screening And Spatiotemporal Disease Surveillance Peihua Qiu https://ebookbell.com/product/statistical-methods-for-dynamic-disease- screening-and-spatiotemporal-disease-surveillance-peihua-qiu-57213800
- 3. Statistical Methods For Geography 1st Edition Peter A Rogerson https://ebookbell.com/product/statistical-methods-for-geography-1st- edition-peter-a-rogerson-2086764 Statistical Methods For Survival Data Analysis Third Edition 3rd Edition Elisa T Lee https://ebookbell.com/product/statistical-methods-for-survival-data- analysis-third-edition-3rd-edition-elisa-t-lee-2119098 Statistical Methods For Disease Clustering 1st Edition Toshiro Tango Auth https://ebookbell.com/product/statistical-methods-for-disease- clustering-1st-edition-toshiro-tango-auth-2122474 Statistical Methods For The Social Sciences Fifth Edition Agresti https://ebookbell.com/product/statistical-methods-for-the-social- sciences-fifth-edition-agresti-21355094 Statistical Methods For The Evaluation Of University Systems 1st Edition Lorenzo Bernardi Auth https://ebookbell.com/product/statistical-methods-for-the-evaluation- of-university-systems-1st-edition-lorenzo-bernardi-auth-2150630
- 5. Immunogenicity assessment is a prerequisite for the successful de- velopment of biopharmaceuticals, including safety and efficacy eval- uation. Using advanced statistical methods in the study design and analysis stages is therefore essential to immunogenicity risk assess- ment and mitigation strategies. Statistical Methods for Immunoge- nicity Assessment provides a single source of information on sta- tistical concepts, principles, methods, and strategies for detection, quantification, assessment, and control of immunogenicity. The book first gives an overview of the impact of immunogenicity on biopharmaceutical development, regulatory requirements, and sta- tistical methods and strategies used for immunogenicity detection, quantification, and risk assessment and mitigation. It then covers anti-drug antibody (ADA) assay development, optimization, valida- tion, and transfer as well as the analysis of cut point, a key assay performance parameter in ADA assay development and validation. The authors illustrate how to apply statistical modeling approach- es to establish associations between ADA and clinical outcomes, predict immunogenicity risk, and develop risk mitigation strategies. They also present various strategies for immunogenicity risk control. The book concludes with an explanation of the computer codes and algorithms of the statistical methods. A critical issue in the development of biologics, immunogenicity can cause early termination or limited use of the products if not managed well. This book shows how to use robust statistical methods for de- tecting, quantifying, assessing, and mitigating immunogenicity risk. It is an invaluable resource for anyone involved in immunogenicity risk assessment and control in both non-clinical and clinical biophar- maceutical development. Statistics ISBN: 978-1-4987-0034-4 9 781498 700344 90000 K24516 w w w . c r c p r e s s . c o m Harry Yang • Jianchun Zhang Binbing Yu • Wei Zhao Yang • Zhang • Yu • Zhao Statistical Methods for Immunogenicity Assessment Statistical Methods for Immunogenicity Assessment K24516_cover.indd 1 6/3/15 10:15 AM
- 7. Editor-in-Chief Shein-Chung Chow, Ph.D., Professor, Department of Biostatistics and Bioinformatics, Duke University School of Medicine, Durham, North Carolina Series Editors Byron Jones, Biometrical Fellow, Statistical Methodology, Integrated Information Sciences, Novartis Pharma AG, Basel, Switzerland Jen-pei Liu, Professor, Division of Biometry, Department of Agronomy, National Taiwan University, Taipei, Taiwan Karl E. Peace, Georgia Cancer Coalition, Distinguished Cancer Scholar, Senior Research Scientist and Professor of Biostatistics, Jiann-Ping Hsu College of Public Health, Georgia Southern University, Statesboro, Georgia Bruce W. Turnbull, Professor, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, New York Published Titles Adaptive Design Methods in Clinical Trials, Second Edition Shein-Chung Chow and Mark Chang Adaptive Designs for Sequential Treatment Allocation Alessandro Baldi Antognini and Alessandra Giovagnoli Adaptive Design Theory and Implementation Using SAS and R, Second Edition Mark Chang Advanced Bayesian Methods for Medical Test Accuracy Lyle D. Broemeling Advances in Clinical Trial Biostatistics Nancy L. Geller Applied Meta-Analysis with R Ding-Geng (Din) Chen and Karl E. Peace Basic Statistics and Pharmaceutical Statistical Applications, Second Edition James E. De Muth Bayesian Adaptive Methods for Clinical Trials Scott M. Berry, Bradley P. Carlin, J. Jack Lee, and Peter Muller Bayesian Analysis Made Simple: An Excel GUI for WinBUGS Phil Woodward Bayesian Methods for Measures of Agreement Lyle D. Broemeling Bayesian Methods in Epidemiology Lyle D. Broemeling Bayesian Methods in Health Economics Gianluca Baio Bayesian Missing Data Problems: EM, Data Augmentation and Noniterative Computation Ming T. Tan, Guo-Liang Tian, and Kai Wang Ng Bayesian Modeling in Bioinformatics Dipak K. Dey, Samiran Ghosh, and Bani K. Mallick Benefit-Risk Assessment in Pharmaceutical Research and Development Andreas Sashegyi, James Felli, and Rebecca Noel Biosimilars: Design and Analysis of Follow-on Biologics Shein-Chung Chow Biostatistics: A Computing Approach Stewart J. Anderson Causal Analysis in Biomedicine and Epidemiology: Based on Minimal Sufficient Causation Mikel Aickin Clinical and Statistical Considerations in Personalized Medicine Claudio Carini, Sandeep Menon, and Mark Chang
- 8. Clinical Trial Data Analysis using R Ding-Geng (Din) Chen and Karl E. Peace Clinical Trial Methodology Karl E. Peace and Ding-Geng (Din) Chen Computational Methods in Biomedical Research Ravindra Khattree and Dayanand N. Naik Computational Pharmacokinetics Anders Källén Confidence Intervals for Proportions and Related Measures of Effect Size Robert G. Newcombe Controversial Statistical Issues in Clinical Trials Shein-Chung Chow Data Analysis with Competing Risks and Intermediate States Ronald B. Geskus Data and Safety Monitoring Committees in Clinical Trials Jay Herson Design and Analysis of Animal Studies in Pharmaceutical Development Shein-Chung Chow and Jen-pei Liu Design and Analysis of Bioavailability and Bioequivalence Studies, Third Edition Shein-Chung Chow and Jen-pei Liu Design and Analysis of Bridging Studies Jen-pei Liu, Shein-Chung Chow, and Chin-Fu Hsiao Design and Analysis of Clinical Trials for Predictive Medicine Shigeyuki Matsui, Marc Buyse, and Richard Simon Design and Analysis of Clinical Trials with Time-to-Event Endpoints Karl E. Peace Design and Analysis of Non-Inferiority Trials Mark D. Rothmann, Brian L. Wiens, and Ivan S. F. Chan Difference Equations with Public Health Applications Lemuel A. Moyé and Asha Seth Kapadia DNA Methylation Microarrays: Experimental Design and Statistical Analysis Sun-Chong Wang and Arturas Petronis DNA Microarrays and Related Genomics Techniques: Design, Analysis, and Interpretation of Experiments David B. Allison, Grier P. Page, T. Mark Beasley, and Jode W. Edwards Dose Finding by the Continual Reassessment Method Ying Kuen Cheung Elementary Bayesian Biostatistics Lemuel A. Moyé Empirical Likelihood Method in Survival Analysis Mai Zhou Exposure-Response Modeling: Methods and Practical Implementation Jixian Wang Frailty Models in Survival Analysis Andreas Wienke Generalized Linear Models: A Bayesian Perspective Dipak K. Dey, Sujit K. Ghosh, and Bani K. Mallick Handbook of Regression and Modeling: Applications for the Clinical and Pharmaceutical Industries Daryl S. Paulson Inference Principles for Biostatisticians Ian C. Marschner Interval-Censored Time-to-Event Data: Methods and Applications Ding-Geng (Din) Chen, Jianguo Sun, and Karl E. Peace Introductory Adaptive Trial Designs: A Practical Guide with R Mark Chang Joint Models for Longitudinal and Time- to-Event Data: With Applications in R Dimitris Rizopoulos Measures of Interobserver Agreement and Reliability, Second Edition Mohamed M. Shoukri
- 9. Medical Biostatistics, Third Edition A. Indrayan Meta-Analysis in Medicine and Health Policy Dalene Stangl and Donald A. Berry Mixed Effects Models for the Population Approach: Models, Tasks, Methods and Tools Marc Lavielle Modeling to Inform Infectious Disease Control Niels G. Becker Modern Adaptive Randomized Clinical Trials: Statistical and Practical Aspects Oleksandr Sverdlov Monte Carlo Simulation for the Pharmaceutical Industry: Concepts, Algorithms, and Case Studies Mark Chang Multiple Testing Problems in Pharmaceutical Statistics Alex Dmitrienko, Ajit C. Tamhane, and Frank Bretz Noninferiority Testing in Clinical Trials: Issues and Challenges Tie-Hua Ng Optimal Design for Nonlinear Response Models Valerii V. Fedorov and Sergei L. Leonov Patient-Reported Outcomes: Measurement, Implementation and Interpretation Joseph C. Cappelleri, Kelly H. Zou, Andrew G. Bushmakin, Jose Ma. J. Alvir, Demissie Alemayehu, and Tara Symonds Quantitative Evaluation of Safety in Drug Development: Design, Analysis and Reporting Qi Jiang and H. Amy Xia Randomized Clinical Trials of Nonpharmacological Treatments Isabelle Boutron, Philippe Ravaud, and David Moher Randomized Phase II Cancer Clinical Trials Sin-Ho Jung Sample Size Calculations for Clustered and Longitudinal Outcomes in Clinical Research Chul Ahn, Moonseong Heo, and Song Zhang Sample Size Calculations in Clinical Research, Second Edition Shein-Chung Chow, Jun Shao and Hansheng Wang Statistical Analysis of Human Growth and Development Yin Bun Cheung Statistical Design and Analysis of Stability Studies Shein-Chung Chow Statistical Evaluation of Diagnostic Performance: Topics in ROC Analysis Kelly H. Zou, Aiyi Liu, Andriy Bandos, Lucila Ohno-Machado, and Howard Rockette Statistical Methods for Clinical Trials Mark X. Norleans Statistical Methods for Drug Safety Robert D. Gibbons and Anup K. Amatya Statistical Methods for Immunogenicity Assessment Harry Yang, Jianchun Zhang, Binbing Yu, and Wei Zhao Statistical Methods in Drug Combination Studies Wei Zhao and Harry Yang Statistics in Drug Research: Methodologies and Recent Developments Shein-Chung Chow and Jun Shao Statistics in the Pharmaceutical Industry, Third Edition Ralph Buncher and Jia-Yeong Tsay Survival Analysis in Medicine and Genetics Jialiang Li and Shuangge Ma Theory of Drug Development Eric B. Holmgren Translational Medicine: Strategies and Statistical Methods Dennis Cosmatos and Shein-Chung Chow
- 10. Harry Yang • Jianchun Zhang Binbing Yu • Wei Zhao MedImmune, LLC Gaithersburg, Maryland, USA Statistical Methods for Immunogenicity Assessment
- 11. CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2016 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group, an Informa business No claim to original U.S. Government works Version Date: 20150515 International Standard Book Number-13: 978-1-4987-0035-1 (eBook - PDF) This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint. Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information stor- age or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copy- right.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that pro- vides licenses and registration for a variety of users. For organizations that have been granted a photo- copy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com
- 12. To our families
- 13. This page intentionally left blank This page intentionally left blank
- 14. Contents Preface xv List of Figures xvii List of Tables xxi 1 Introduction 1 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Immunogenicity . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3 Impact of Immunogenicity . . . . . . . . . . . . . . . . . . . 5 1.4 Regulatory Environment and Guidelines . . . . . . . . . . . 6 1.4.1 FDA Guidelines . . . . . . . . . . . . . . . . . . . . . 7 1.4.1.1 Tiered Approach to ADA Assay Development 7 1.4.1.2 Immunogenicity Risk Assessment . . . . . . 9 1.4.2 European Medicines Agency (EMA) Guidance . . . . 13 1.4.2.1 EMA Guidelines on Immunogenicity Assessment . . . . . . . . . . . . . . . . . . 13 1.4.2.2 Latest Development of EMA Immunogenicity Guidelines . . . . . . . . . . . . . . . . . . . 14 1.4.3 Japanese Regulatory Requirements of Immunogenicity 14 1.5 Statistics in Immunogenicity Risk Assessment . . . . . . . . 15 1.5.1 In Silico Prediction of Immunogenicity . . . . . . . . . 16 1.5.2 ADA Detection and Quantification . . . . . . . . . . . 16 1.5.3 Clinical Characterization of ADA . . . . . . . . . . . . 17 1.5.3.1 Characteristics of ADA Immune Response . 17 1.5.3.2 Correlation between ADA and PK/PD, Clinical Safety and Efficacy . . . . . . . . . . 18 1.5.3.3 Relationship of ADA with Clinical Efficacy and Safety . . . . . . . . . . . . . . . . . . . 18 1.5.3.4 Identification of Risk Factors . . . . . . . . . 18 1.5.4 Control of Immunogenicity Risk . . . . . . . . . . . . 19 1.5.4.1 Control of Process/Product Factors . . . . . 19 1.5.4.2 Biomarkers for Immunogenicity . . . . . . . 19 1.6 Statistical Considerations in Comparative Immunogenicity Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 ix
- 15. x Contents 1.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . 21 2 ADA Assay Development and Validation 23 2.1 ADA Assays . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.1.1 Multi-Tiered Approach . . . . . . . . . . . . . . . . . 24 2.1.1.1 Screening Assay . . . . . . . . . . . . . . . . 25 2.1.1.2 Confirmatory Assay . . . . . . . . . . . . . . 25 2.1.1.3 Neutralizing Assay . . . . . . . . . . . . . . . 26 2.1.2 Assay Platforms . . . . . . . . . . . . . . . . . . . . . 26 2.2 Assay Development and Validation . . . . . . . . . . . . . . 28 2.2.1 Assay Parameters . . . . . . . . . . . . . . . . . . . . 29 2.2.1.1 Cut Point . . . . . . . . . . . . . . . . . . . . 29 2.2.1.2 Sensitivity . . . . . . . . . . . . . . . . . . . 29 2.2.1.3 Drug Tolerance . . . . . . . . . . . . . . . . . 33 2.2.1.4 Precision . . . . . . . . . . . . . . . . . . . . 33 2.2.1.5 Robustness . . . . . . . . . . . . . . . . . . . 35 2.2.1.6 Ruggedness/Reproducibility . . . . . . . . . 35 2.2.2 Life-Cycle Approach . . . . . . . . . . . . . . . . . . . 35 2.3 Design of Experiment . . . . . . . . . . . . . . . . . . . . . . 36 2.3.1 Fractional Factorial Design . . . . . . . . . . . . . . . 38 2.3.2 Response Surface Design . . . . . . . . . . . . . . . . . 41 2.3.3 Split-Plot Design . . . . . . . . . . . . . . . . . . . . . 43 2.3.4 Design Optimality . . . . . . . . . . . . . . . . . . . . 46 2.3.5 An Example of Neutralizing Antibody Assay Development and Validation . . . . . . . . . . . . . . 48 2.4 Method Transfer . . . . . . . . . . . . . . . . . . . . . . . . . 48 3 Determination of ADA Assay Cut Point 57 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.2 Cut Point Experimental Design . . . . . . . . . . . . . . . . 58 3.3 Statistical Methods for Cut Point Determination . . . . . . . 61 3.3.1 White Paper Approach . . . . . . . . . . . . . . . . . 62 3.3.1.1 Investigate Distribution and Exclude Outliers 63 3.3.1.2 Compare Assay Run Means and Variances . 65 3.3.1.3 Calculate the Screening Cut Point . . . . . . 66 3.3.1.4 Confirmatory Cut Point . . . . . . . . . . . . 68 3.3.2 Some Recent Developments . . . . . . . . . . . . . . . 68 3.3.2.1 Data Normalization . . . . . . . . . . . . . . 68 3.3.2.2 Outliers . . . . . . . . . . . . . . . . . . . . . 69 3.3.2.3 Non-Normal Distributions . . . . . . . . . . . 70 3.3.3 Comparison of Cut Point Methods . . . . . . . . . . . 83 3.3.3.1 Data Including Positive Donors . . . . . . . . 83 3.3.3.2 Prediction Interval and Tolerance Interval . . 85
- 16. Contents xi 3.3.3.3 Confirmatory Cut Point . . . . . . . . . . . . 87 4 Clinical Immunogenicity Assessment 89 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 4.2 Monoclonal Antibodies for the Treatment of Rheumatoid Arthritis (RA) . . . . . . . . . . . . . . . . . . . . . . . . . . 92 4.2.1 Hypothetical Clinical Trial of an mAb Drug for RA . 93 4.2.2 ADA Testing Protocol . . . . . . . . . . . . . . . . . . 94 4.2.3 Analysis Diagram . . . . . . . . . . . . . . . . . . . . 94 4.3 Statistical Analysis of ADA Status . . . . . . . . . . . . . . . 95 4.3.1 Confidence Intervals of ADA Response Rates . . . . . 96 4.3.2 Comparison of ADA Response Rates . . . . . . . . . . 97 4.3.3 Statistical Tests of ADA Response Rates . . . . . . . . 98 4.3.4 Dose–Response Model of ADA Response . . . . . . . . 100 4.3.5 Example . . . . . . . . . . . . . . . . . . . . . . . . . . 101 4.4 Effects of ADA on Drug Efficacy . . . . . . . . . . . . . . . . 103 4.4.1 Categorical Endpoints as Efficacy Measure . . . . . . 104 4.4.1.1 Multiple 2 × 2 Tables . . . . . . . . . . . . . 104 4.4.1.2 Multiple R × C Tables . . . . . . . . . . . . 105 4.4.1.3 Example . . . . . . . . . . . . . . . . . . . . 106 4.4.2 Effect of ADA on Survival Outcomes . . . . . . . . . . 106 4.4.2.1 Rank-Based Statistical Tests . . . . . . . . . 109 4.4.2.2 Survival Regression Models . . . . . . . . . . 110 4.4.2.3 Recurrent Event Data . . . . . . . . . . . . . 112 4.4.2.4 Competing-Risks Survival Data . . . . . . . 114 4.4.2.5 Example . . . . . . . . . . . . . . . . . . . . 115 4.4.3 Effect of ADA on Longitudinal Outcomes . . . . . . . 116 4.4.3.1 Generalized Linear Model . . . . . . . . . . . 119 4.4.3.2 Marginal Generalized Linear Model . . . . . 120 4.4.3.3 Generalized Linear Mixed Model . . . . . . . 121 4.4.3.4 Generalized Additive Model . . . . . . . . . 123 4.4.3.5 Example . . . . . . . . . . . . . . . . . . . . 124 4.5 Effect of ADA on AEs . . . . . . . . . . . . . . . . . . . . . . 127 4.5.1 Statistical Methods for AEs . . . . . . . . . . . . . . . 127 4.5.2 Controlling False Discovery Rate . . . . . . . . . . . . 129 4.5.3 Bayesian Hierarchical Model . . . . . . . . . . . . . . 131 4.5.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . 132 4.5.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . 134 4.6 Relationship between ADA and Pharmacokinetics . . . . . . 134 4.6.1 Nonlinear Mixed-Effects Models . . . . . . . . . . . . 137 4.6.2 Pharmacokinetics Model for Multiple Doses . . . . . . 137 4.7 Statistical Analysis ADA Onset and Duration . . . . . . . . 139 4.7.1 Modeling ADA Onset Times . . . . . . . . . . . . . . 139
- 17. xii Contents 4.7.1.1 Hypothesis Testing for Interval-Censored Data . . . . . . . . . . . . . . . . . . . . . . 140 4.7.1.2 Regression Models for ADA Onset Times . . 142 4.7.2 Statistical Issues for ADA Duration . . . . . . . . . . 142 4.7.2.1 Persistent and Transient ADA . . . . . . . . 142 4.7.2.2 Statistical Analysis for Doubly Interval-Censored Data . . . . . . . . . . . . 143 4.7.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . 144 4.8 Statistical Analysis of ADA Titer . . . . . . . . . . . . . . . 146 4.8.1 Traditional Analysis of ADA Titers . . . . . . . . . . . 147 4.8.2 Maximum Likelihood Method for ADA Titer . . . . . 148 4.8.3 Nonparametric Method for Comparing ADA Titers . . 149 4.8.4 Longitudinal Modeling of ADA Titer . . . . . . . . . . 150 4.8.5 Example . . . . . . . . . . . . . . . . . . . . . . . . . . 150 4.9 Meta-Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 152 4.9.1 Fixed-Effects and Random-Effects Models . . . . . . . 153 4.9.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . 154 5 Immunogenicity Risk Control 157 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 5.2 Risk Assessment . . . . . . . . . . . . . . . . . . . . . . . . . 160 5.2.1 Identification of Risk Factors . . . . . . . . . . . . . . 160 5.2.2 Criticality of Risk Factors . . . . . . . . . . . . . . . . 161 5.3 Immunogenicity Risk Control . . . . . . . . . . . . . . . . . . 165 5.3.1 Acceptance Range . . . . . . . . . . . . . . . . . . . . 166 5.3.2 Case Example . . . . . . . . . . . . . . . . . . . . . . . 169 5.3.2.1 Risk of Residual Host Cell DNA . . . . . . . 169 5.3.2.2 Control Strategies . . . . . . . . . . . . . . . 170 5.3.2.3 DNA Inactivation . . . . . . . . . . . . . . . 171 5.3.2.4 Determination of Acceptance Range . . . . . 171 5.3.2.5 Numerical Calculations . . . . . . . . . . . . 173 5.4 Biomarkers for Immunogenicity . . . . . . . . . . . . . . . . 175 5.4.1 General Strategies . . . . . . . . . . . . . . . . . . . . 178 5.4.2 Clustering Methods . . . . . . . . . . . . . . . . . . . 179 5.4.2.1 K-Means Clustering . . . . . . . . . . . . . . 179 5.4.2.2 Hierarchical Clustering . . . . . . . . . . . . 180 5.4.2.3 Comparison of Clustering by K-Means and HC Using Simulated Data . . . . . . . . . . 181 5.4.2.4 Principal Component Analysis . . . . . . . . 181 5.4.3 Prediction Models . . . . . . . . . . . . . . . . . . . . 184 5.4.3.1 Logistic Regression Model . . . . . . . . . . . 185 5.4.3.2 K-Nearest Neighbors (KNN) . . . . . . . . . 185 5.4.3.3 Linear Discriminant Analysis (LDA) . . . . . 186
- 18. Contents xiii 5.4.3.4 Classification Tree and Random Forest Methods . . . . . . . . . . . . . . . . . . . . 189 5.4.3.5 Cross Validation and Bagging . . . . . . . . 190 5.4.3.6 Simple Is a Virtue . . . . . . . . . . . . . . . 190 5.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . 191 6 Computational Tools for Immunogenicity Analysis 193 6.1 Read Data into R . . . . . . . . . . . . . . . . . . . . . . . . 193 6.2 ADA Assay and Cut Point . . . . . . . . . . . . . . . . . . . 194 6.2.1 ADA Assay Development . . . . . . . . . . . . . . . . 194 6.2.2 Implementation of White Paper Approach . . . . . . . 197 6.3 Implementation of Statistical Analysis of Clinical Immunogenicity Assessment . . . . . . . . . . . . . . . . . . 201 6.3.1 Statistical Analysis of ADA Status . . . . . . . . . . . 201 6.3.2 Effects of ADA on Drug Efficacy . . . . . . . . . . . . 202 6.3.3 Statistical Analysis of ADA Onset and Duration . . . 204 6.3.4 Statistical Analysis of ADA Titer . . . . . . . . . . . . 204 6.4 Graphical Tools for Cause and Effect and Design Space Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 6.4.1 Ishikawa Diagram . . . . . . . . . . . . . . . . . . . . 205 6.4.2 Pareto Plot . . . . . . . . . . . . . . . . . . . . . . . . 206 6.4.3 Acceptance Region . . . . . . . . . . . . . . . . . . . . 206 6.5 Immunogenicity Biomarker Discovery . . . . . . . . . . . . . 207 6.6 Report Automation . . . . . . . . . . . . . . . . . . . . . . . 208 6.6.1 Generate Reports in pdf, rtf, and docx Formats . . . . 209 6.6.2 Shiny Server: Online Report Generation . . . . . . . . 211 Bibliography 213 Index 235
- 19. This page intentionally left blank This page intentionally left blank
- 20. Preface Biotechnology-derived therapeutics including monoclonal antibodies, pro- teins, and peptides hold great promise for treating various diseases such as cancer and inflammatory diseases. They also represent an important class of therapeutic interventions. However, because of their large size, complex struc- ture, and complicated manufacture process, biopharmaceutical products can lead to immunogenic responses, resulting in formation of anti-drug antibod- ies (ADAs). Immune responses to non-vaccine biologics have the potential to negatively affect both patient safety and product efficacy. For example, a neutralizing antibody is deleterious if it inhibits the efficacy of the prod- uct, and can be harmful when it cross-reacts with an endogenous counter- part of the therapeutic in patients. Non-neutralizing antibodies may affect the pharmacokinetic properties of the drug, thus may affect dosing regime. These immunologically-based consequences may cause drug developers to ei- ther terminate development or limit the use of otherwise effective therapies. Therefore, immunogenicity assessment is a key component of biopharmaceuti- cal safety and efficacy evaluation, and a prerequisite for the successful develop- ment of biopharmaceuticals. Furthermore, immunogenicity is also a complex phenomenon, owing to myriad factors potentially affecting immunogenicity. For the purposes of this book, these factors are grouped into two categories: product-specific factors such as product origin, glycosylation, aggregation, im- purities and formulation, and patient-related characteristics such as genetic makeup and immune status and competency. These numerous and varied fac- tors impose challenges to immunogenicity risk assessment and development of risk mitigation strategies. The intrinsic complexity of detection, quantifica- tion, characterization, and control or mitigation of ADA argues for advanced statistical methods in both study design and analysis. This book is intended to provide a single source of information on statistical concepts, principles, methods, and strategies for detection, quantification, assessment, and control of immunogenicity. The book consists of six chapters. Chapter 1 provides an overview of im- munogenicity, its impact on biopharmaceutical development, regulatory re- quirements, statistical methods and strategies used for immunogenicity detec- tion, quantification, risk assessment, and mitigation. Chapter 2 deals with ADA assay development, optimization, validation, and transfer based on sound statistical principles, design, and analysis. It discusses statistical con- siderations in many aspects of screening, confirmatory, and neutralizing assay development. Chapter 3 is focused on analysis of cut point, a key assay per- xv
- 21. xvi Preface formance parameter in ADA assay development and validation. It covers a wide range of topics from sample size calculation, data normalization, outlier detection and removal, to selection of proper models for cut point analysis. Challenges and limitations of cut point applied to practical clinical sample testing are also explained. In Chapter 4, we illustrate how to apply statisti- cal modeling approaches to establishing associations between ADA and clin- ical outcomes, and process parameters, predicting immunogenicity risk, and developing risk-mitigation strategies. Various strategies for immunogenicity risk control are presented in Chapter 5. Finally, the majority of computer codes/algorithms of the statistical methods introduced in the book are pro- vided and explained in Chapter 6. In recent years, assessment of immunogenicity has emerged as an impor- tant regulatory initiative as evidenced by a growing number of white papers on the subject, and publication of the FDA and EMA guidelines. It is also a crucial step toward using risk-based strategies in biopharmaceutical product development. To ensure regulatory compliance, gain deep understanding of immunogenicity, and develop effective immunogenicity risk mitigation strate- gies, it is imperative to apply robust statistical methods and thinking in the detection, quantification, assessment, and mitigation of immunogenicity risk. To that end, a single book covering statistical concepts, principles, meth- ods, and strategies in immunogenicity assessment will provide an invaluable resource for practitioners in biopharmaceutical therapy development. As im- munogenicity risk assessment and control are issues faced by professionals who are involved in non-clinical, clinical, and bioprocess development, this book will be helpful to many individuals in various scientific and regulatory disci- plines, including statisticians, pharmacokineticists, toxicologists, clinical assay developers, clinicians, biopharmaceutical engineers, and regulatory reviewers. We are extremely grateful to John Kimmel, executive editor, Chapman & Hall/CRC Press, for giving us the opportunity to work on this book. We would like to express our gratitude to Laura Richman, Dianne Hirsch, and Kicab Castañeda-Méndez for their expert review of the book and helpful comments. Harry Yang Jianchun Zhang Binbing Yu Wei Zhao Gaithersburg, Maryland, USA
- 22. List of Figures 1.1 Tiered approach to immunogenicity assessment. . . . . . . 8 2.1 Schematic of bridging assay format. . . . . . . . . . . . . . 27 2.2 Illustration of a nonlinear dilution curve and interpolation. 30 2.3 Illustration of a simple precision design of experiment. . . . 33 2.4 Factorial design resolution table. . . . . . . . . . . . . . . . 40 2.5 A typical response surface graph with the maximal response in the center of the region. . . . . . . . . . . . . . 41 2.6 Different shape of response surface. . . . . . . . . . . . . . 43 2.7 Central composite design for two factors. . . . . . . . . . . 44 2.8 Split-plot design in JMP. . . . . . . . . . . . . . . . . . . . 47 2.9 Factor setting in JMP custom design. . . . . . . . . . . . . 49 2.10 Response surface model specification in JMP custom design. 50 2.11 Generation of response surface design table. . . . . . . . . . 51 2.12 Illustration of equivalence test. . . . . . . . . . . . . . . . . 52 2.13 Scatter plot vs. Bland–Altman plot . . . . . . . . . . . . . 56 3.1 Example of balanced design for cut point. . . . . . . . . . 59 3.2 A schematic diagram for screening cut point determination 64 3.3 Illustration of impact of outliers removal order . . . . . . . 71 3.4 A flow chart for cut point determination . . . . . . . . . . . 72 3.5 Density of gamma distribution with various shape and scale (threshold is set at 0). . . . . . . . . . . . . . . . . . . . . . 73 3.6 Density of lognormal distribution with location and scale. . 74 3.7 Density of skew-t distribution with location and scale. . . . 76 3.8 Density of log-gamma distribution with location and scale. 77 3.9 Comparison of empirical distribution with fitted distributions for marginal response and estimated random effects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 3.10 The impact of cut point on mixed populations . . . . . . . 85 4.1 Relationship between treatment, immunogenicity and clinical efficacy. . . . . . . . . . . . . . . . . . . . . . . . . . 95 4.2 Kaplan–Meier estimates of % of subjects staying in remission by ADA status. . . . . . . . . . . . . . . . . . . . 116 xvii
- 23. xviii List of Figures 4.3 Kaplan–Meier estimates of % of subjects staying in remission by ADA status for the four dose groups. . . . . . 117 4.4 Trajectory of DAS28 by ADA status for the four treatment groups. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 4.5 Concentration of drug as a function of time. . . . . . . . . 136 4.6 Drug concentration over time after administration of multiple doses. . . . . . . . . . . . . . . . . . . . . . . . . . 138 4.7 Survival functions of the ADA onset time by dose group. . 145 4.8 Histogram of the logarithm of ADA titer. . . . . . . . . . . 151 4.9 Forest plot . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 5.1 Process of quality risk management. . . . . . . . . . . . . . 159 5.2 An example of a fishbone diagram. . . . . . . . . . . . . . . 161 5.3 Pareto plot of potential risk factors . . . . . . . . . . . . . 164 5.4 Acceptance range of drug substance aggregate and impurity when aggregate and impurity do not have any joint effect on product immunogenicity. . . . . . . . . . . . . . . . . . 167 5.5 Specifications of drug substance aggregate and impurity when aggregate and impurity have a joint effect on product immunogenicity. . . . . . . . . . . . . . . . . . . . . . . . . 168 5.6 Diagram of DNA inactivation using enzyme . . . . . . . . . 172 5.7 Plot of risk score of having an immunogenic event caused by unmethylated CpG motifs, based on annual use of 50 million doses of the product. . . . . . . . . . . . . . . . . . 175 5.8 Acceptable range of DNA size and amount . . . . . . . . . 176 5.9 Insufficiency of individual or combined assessment of factors X and Y in predicting immunogenic risk: neither factor X, factor Y , nor their combination. . . . . . . . . . . 177 5.10 A linear combination of the two factors can clearly differentiate Group A from Group B. . . . . . . . . . . . . 178 5.11 A simulated data set with 100 risk factors from 40 samples. 182 5.12 Comparing hierarchical clustering with K-means clustering methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 5.13 Principal component analysis of the simulated biomarker data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 5.14 Simulated segregation of 500 samples into immunogenicity-positive and immunogenicity-negative groups. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 5.15 Immunogenicity distributions based on W1 = (1, 0), W2 = (0, 1), W3 = (0.707, 0.707), W4 = (−0.707, 0.707). . . . . . . 188 5.16 Classification of the simulated 100 biomarker data using rpart() function. . . . . . . . . . . . . . . . . . . . . . . . . 189 6.1 Example of finding optimal Box–Cox transformation. . . . 198 6.2 Example of QQ plot. . . . . . . . . . . . . . . . . . . . . . . 198
- 24. List of Figures xix 6.3 Simple scatter plot . . . . . . . . . . . . . . . . . . . . . . . 210
- 25. This page intentionally left blank This page intentionally left blank
- 26. List of Tables 1.1 Patient-specific factors that affect immunogenicity and risk-mitigating strategies . . . . . . . . . . . . . . . . . . . 11 1.2 Product-specific factors that affect immunogenicity and risk-mitigating strategies . . . . . . . . . . . . . . . . . . . 12 2.1 ANOVA table for nested model . . . . . . . . . . . . . . . 34 2.2 Fractional factorial design for five factors . . . . . . . . . . 39 2.3 Example of split-plot design . . . . . . . . . . . . . . . . . 45 2.4 P-values for corrosion-resistance example . . . . . . . . . 45 2.5 ANOVA table for a split-plot design . . . . . . . . . . . . 46 2.6 Example data for illustration of limits of agreement of Bland–Altman method . . . . . . . . . . . . . . . . . . . . 54 3.1 A numerical example of variance components analysis . . 59 3.2 The minimal required sample size to meet the 0.05 precision for different intraclass correlation and different number of replicates . . . . . . . . . . . . . . . . . . . . . 62 3.3 ANOVA table for testing assay run means . . . . . . . . . 66 3.4 Comparison of cut point estimation(scenario a) . . . . . . 80 3.5 Comparison of cut point estimation(scenario b1) . . . . . 81 3.6 Comparison of cut point estimation(scenario b2) . . . . . 81 3.7 Comparison of cut point estimation(scenario c1) . . . . . 81 3.8 Comparison of cut point estimation(scenario c2) . . . . . 82 3.9 Comparison of cut point estimation from different random effects models . . . . . . . . . . . . . . . . . . . . . . . . . 83 4.1 The 2 × 2 table for ADA responses in two groups. . . . . . 97 4.2 The 2 × K table of ADA dose–response data in multiple dose groups . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.3 Number and percentage of ADA+ patients by treatment group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 4.4 Estimates of OR and RR of ADA responses between the treatment groups and the placebo group . . . . . . . . . . 102 4.5 Statistical tests that compare the ADA response rates between the treatment groups and the placebo group . . . 102 xxi
- 27. xxii List of Tables 4.6 The 2 × 2 table of ADA status and clinical response in the kth dose group . . . . . . . . . . . . . . . . . . . . . . . . 104 4.7 RA recurrence status by ADA status and dose level . . . . 106 4.8 Stratified log-rank test for the effect of ADA on remission after controlling for dose level . . . . . . . . . . . . . . . . 117 4.9 Parameter estimates of the Cox model for RA recurrence . 118 4.10 Common distributions from the exponential family . . . . 119 4.11 Parameter estimates of the marginal linear model for DAS28 from GEE . . . . . . . . . . . . . . . . . . . . . . . 125 4.12 Parameter estimates of the random-intercept model . . . . 126 4.13 The 2 × 2 table of AE and ADA reaction in the jth dose group for the kth AE . . . . . . . . . . . . . . . . . . . . . 128 4.14 The 2 × 2 table of testing the effect of ADA on all types of AE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 4.15 AEs by event type, treatment group and ADA status . . . 133 4.16 Unadjusted mean and median ADA titer by dose group . 151 4.17 Estimates of ADA titers by dose group from the likelihood method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 5.1 Determination of severity scores . . . . . . . . . . . . . . . 162 5.2 Severity definition . . . . . . . . . . . . . . . . . . . . . . 163 5.3 Likelihood definition . . . . . . . . . . . . . . . . . . . . . 163 5.4 Parameter values needed for design of control strategy . . 174 6.1 Summary of linear regression model fitting: lm1 . . . . . . 210
- 28. 1 Introduction CONTENTS 1.1 Background ...................................................... 2 1.2 Immunogenicity .................................................. 4 1.3 Impact of Immunogenicity ....................................... 5 1.4 Regulatory Environment and Guidelines ........................ 6 1.4.1 FDA Guidelines .......................................... 7 1.4.1.1 Tiered Approach to ADA Assay Development 7 1.4.1.2 Immunogenicity Risk Assessment ............ 9 1.4.2 European Medicines Agency (EMA) Guidance ......... 13 1.4.2.1 EMA Guidelines on Immunogenicity Assessment ................................... 13 1.4.2.2 Latest Development of EMA Immunogenicity Guidelines ..................................... 14 1.4.3 Japanese Regulatory Requirements of Immunogenicity . 14 1.5 Statistics in Immunogenicity Risk Assessment ................. 15 1.5.1 In Silico Prediction of Immunogenicity ................. 16 1.5.2 ADA Detection and Quantification ..................... 16 1.5.3 Clinical Characterization of ADA ....................... 17 1.5.3.1 Characteristics of ADA Immune Response .. 17 1.5.3.2 Correlation between ADA and PK/PD, Clinical Safety and Efficacy .................. 18 1.5.3.3 Relationship of ADA with Clinical Efficacy and Safety .................................... 18 1.5.3.4 Identification of Risk Factors ................. 18 1.5.4 Control of Immunogenicity Risk ........................ 19 1.5.4.1 Control of Process/Product Factors ......... 19 1.5.4.2 Biomarkers for Immunogenicity .............. 19 1.6 Statistical Considerations in Comparative Immunogenicity Studies .......................................................... 20 1.7 Concluding Remarks ............................................. 21 1
- 29. 2 Statistical Methods for Immunogenicity Assessment 1.1 Background The discovery of DNA in 1953, and the many advances made afterwards in cellular and molecular biology in the late 1970s brought into existence the biotechnology industry. Of particular importance was the development of re- combinant DNA technology which enabled the creation and production of proteins in a laboratory setting. These technological advances have provided biopharmaceutical companies with the tools needed to develop “targeted ther- apies” aimed at the biological underpinnings of various diseases. The first recombinant biologic therapy licensed in the United States (U.S.) was recom- binant human insulin which was approved by U.S. Food and Drug Admin- istration (FDA) in 1982. Since the approval of recombinant human insulin, more than 200 biological products have been approved over the past sev- eral decades, treating diseases ranging from cancers to rare genetic disorders (Guilford-Blake and Strickland (2008)). As of 2013, more than 900 molecules, targeting over 100 diseases including cancer, multiple sclerosis, and rheuma- toid arthritis, were at various stages of development (PhRMA (2013)). These biotechnology-derived therapeutics hold a great deal of promise for future medicinal innovation and breakthroughs. However, despite the promise of therapeutic proteins and monoclonal anti- bodies to meet unmet medical needs, development of biologics poses a host of unique challenges. Biopharmaceutical products are often large in size, having complex structures which are often modified post-translationally, e.g., glyco- sylation, and/or during manufacturing to improve product quality, e.g. pe- gylation. Additionally, most therapeutic proteins are produced in non-human cell lines and therefore are not identical to the homologous human protein. In light of these complexities, it is not surprising that the manufacture of biolog- ics requires complicated and tightly controlled manufacturing processes. An additional consideration is that most biologics are administered intravenously or subcutaneously. As a result, therapeutic proteins and monoclonal antibod- ies (mAbs) have the potential to induce immune responses when administered to patients. One common immunogenic response to therapeutic proteins and mAbs is the development of anti-drug antibodies (ADA). While the development of ADAs against therapeutic proteins is common and often has no measurable clinical effects, ADA responses have the potential to negatively affect both patient safety and product efficacy (Shankar et al. (2008)). For instance, for a therapeutic protein that has a non-redundant endogenous counterpart, a neutralizing antibody response can cross-react with the endogenous protein, causing serious consequences (FDA (2014)). One example is recombinant hu- man erythropoietin (rhEPO) which is used to treat anemia. It was shown that neutralizing antibodies (NAbs) directed against rhEPO secondary to adminis- tration of the product also blocked the function of endogenous erythropoietin
- 30. Introduction 3 which was causal in the development of pure red cell aplasia (Casadevall et al. (2002)). ADA binding to the therapeutic can also impact product efficacy. For instance, 50% patients treated with the murine monoclonal antibody OKT3 developed human anti-mouse antibodies (HAMAs) that correlated with de- creased efficacy (Kuus-Reichel et al. (1994)). Readers interested in reviews on immunogenicity are referred to van de Weert and Møller (2008) and Baker et al. (2010). In recent years, various methods and strategies have been developed to reduce and manage immunogenicity of biologic products. Early efforts were centered on methods for measuring ADA. Now, in addition to ADA monitor- ing, therapeutic protein manufacturers are increasingly focusing on engineer- ing therapeutics with reduced risk of inducing ADA responses. Approaches include development of humanized proteins, removal of T-cell epitopes, and selection of less immunogenic proteins using in silico, in vitro, and in vivo pre- diction methods. Therapeutic proteins are often produced in non-human cell lines and species, e.g., mice. As such, their protein sequences differ from the human counterpart, thus increasing immunogenic potential of the therapeutic protein when administered to human recipients. Humanization of proteins pro- duced in non-human species is a process to increase the proteins similarity, through modifying non-human sequences to homologous human sequences. In certain cases, humanization has been shown to be effective in reducing the risk of immunogenicity. In one retrospective review of ADA responses to mAbs, murine mAbs were shown to have the highest frequency of ADA re- sponses, and that replacement of the mouse immunoglobin constant regions with human sequences reduced the development of ADAs (Hwang and Foote (2005)). ADA responses to T-cell epitopes is also well recognized. It has been shown that the presence of T-cell epitopes in a therapeutic protein is one driver of ADA responses. When T-cell receptors recognize small fragments derived from protein antigens coupled with major histocompatibility complex (MHC) class II molecules on the surface of antigen-presenting cells (APCs), T-cell responses are activated. Therefore, one way to minimize immunogenic risk is to deactivate T-cell responses to a therapeutic protein. For this pur- pose, several methods have been utilized to identify and design proteins that have a more acceptable immunogenic profile. The strategies include removal of T-cell epitopes through in silico, in vitro, and in vivo prediction, patient im- munosuppression and tolerization (Adair and Ozanne (2002)). Using in-vitro experiments, T-cell epitopes can be screened and then proteins with the least T-cell epitopes can be used for subsequent development. Immunosuppression reduces immunogenicity through treating subjects with drugs that suppress T-cell activities; whereas the tolerance approach focuses on desensitizing the immune system to the therapeutic protein so that the therapeutic protein is no longer recognized as foreign. As pointed out by De Groot and Martin (2009), successful mitigation of immunogenicity potential of a therapy is likely to rely on a combined approach. It uses rational sequence design, and in vitro and in vivo animal testing to
- 31. 4 Statistical Methods for Immunogenicity Assessment select the least immunogenic lead candidates to advance into clinical testing stage. Equally important is the assessment of other triggers that may cause undesirable immune responses. It is now well understood that there are many other factors related to the product, process, and patient that may cause immunogenicity. It involves understanding the characteristics of the molecules, their intended use, factors impacting immunogenicity, development of sensitive assays for the detection, quantification, characterization of ADA, and careful design of risk-mitigation strategies. However, effective immunogenicity risk management in produc- tion and clinical development of the therapeutic proteins depends heavily on the ability to effectively synthesize and quantitatively evaluate information from multiple sources. The intrinsic complexity of quantitative evaluations argues for advanced statistical methods in both study design and analysis. In this chapter, we provide an overview of immunogenicity issues, with a focus on the development of ADAs, regulatory requirements, and strategies that can be used to mitigate immunogenicity risk. Applications of statistical con- cepts, principles, and methods to address immunogenicity issues are briefly described. In depth coverage of statistical methods used to evaluate immuno- genicity is provided in Chapters 2-5. 1.2 Immunogenicity Immune responses are the natural defense mechanism of vertebrates against disease causing pathogens. Immunogenicity is the ability of an antigen to elicit immune responses. There are two types of immune responses. The innate im- mune response is the first line of host defense against a pathogen. Activation of an innate immune response occurs by an antigen binding to germ-line en- coded receptors on antigen presenting cells (APCs), such as macrophages and dendritic cells, followed by antigen internalization and degradation. Following degradation, peptide fragments are moved to the extracellular cell membrane where they form Major Histocompatibility (MHC)-peptide complexes. The cell-mediated immune system, also known as the adaptive immune system, is the second line of defense against pathogens. In the cell-mediated response, CD4+ helper T cells and CD8+ cytotoxic T cells are activated when they bind to the MHC-peptide complexes on the APCs. Activation of signaling pathways that lead to the development of anti-antigen antibodies is mediated by CD4+ T cells. In brief, cytokines released by CD4+ T cells stimulate B cells to proliferate and differentiate into plasma cells that produce antibodies specific to one of the pathogen’s peptides. Immune responses can be either wanted or unwanted. Wanted immuno- genicity is the immune response against pathogens including viruses and bac- teria, which is typically induced with injection of a vaccine. Unwanted im-
- 32. Introduction 5 munogenicity is the immune response against a therapeutic protein, such as a monoclonal antibody, through production of ADAs. ADA responses to ther- apeutic proteins are a complex phenomenon, owing to myriad factors that can potentially affect immunogenicity. In general, these factors are classified into two categories: product-related and patient-related. Product-related fac- tors that influence the development of ADA responses include species-specific epitopes, levels and types of glycosylation, levels of protein aggregates and impurities, and product formulation; patient-related characteristics encom- pass genetic makeup and immune status of the patient due to disease, route of administration, dosing frequency, and existence of endogenous equivalents (Shankar et al. (2007)). To fulfill the promise that the biologics offer, careful considerations need to be given to both biologics design and production. 1.3 Impact of Immunogenicity As previously discussed, ADA responses to non-vaccine biologics have the potential to negatively affect both patient safety and product efficacy. For ex- ample, neutralizing ADAs (NAbs) can bind to the therapeutic protein, thus reducing the efficacy of the product, and it can be harmful when it cross- reacts with an endogenous counterpart of the therapeutic in patients. Exam- ples of adverse ADA responses include autoimmune thrombocytopenia (ITP) following exposure to recombinant thrombopoietin, and pure red cell aplasia caused by antibodies to recombinant human EPO (rhEPO) that neutralize the product as well as endogenous EPO (Kromminga and Deray (2008), Kirshner (2011)). Additionally, both ADAs and NAbs may affect the pharmacokinetic properties of the drug by either increasing or decreasing product serum half-life which may require dosing modifications. These immunologically based conse- quences may cause drug developers to either terminate development or limit use of otherwise effective therapies. Since the impact of immunogenicity can be quite severe, regulatory agencies have provided guidelines for various as- pects of immunogenicity assessment such as ADA assay development and risk factor identification. Common to all the regulatory documents is the require- ment of managing immunogenicity risk using a risk-based approach, which helps develop risk management plan through a systematic method that links the extent of immunogenicity monitoring to the immunogenicity risk of the therapeutic protein under development.
- 33. 6 Statistical Methods for Immunogenicity Assessment 1.4 Regulatory Environment and Guidelines In recent years, assessment of immunogenicity has emerged as an important regulatory initiative as evidenced by a growing number of white papers on the subject, and publication of regulatory guidelines. Over the past decade, concerted efforts have been made by various working groups, consortiums, and regulatory bodies to gain a deep understanding of immunogenicity, establish best practices for ADA detection and characterization, and develop risk-based approaches to immunogenicity assessment and management. In 2000, the Lig- and Binding Assay Bioanalytical Focus Group was formed under the auspice of the American Association of Pharmaceutical Scientists (AAPS), which was followed by the establishment of the Immunogenicity Working Group. Ever since, the group has produced several important publications on design, op- timization, and validation of immunoassays and bioassays for detection of ADA, regulatory implications of immunogenicity, immunogenicity testing for non-clinical and clinical studies, and risk management. Immunogenicity is a complex phenomenon involving multiple components of the immune system. Because multiple epitopes on the therapeutic may be immunogenic, ADA responses are polyclonal in nature and therefore ADAs and NAbs may both be present in samples. Therefore, it is challenging to attribute ADAs to one particular cause. In addition, it is equally difficult to obtain ADA positive controls, making detection and quantifications of ADAs formidable. Oftentimes, polyclonal or monoclonal ADAs are derived from an- imals and used as surrogate controls. However, they do not provide full rep- resentation of ADA from subjects given the therapeutic protein. Moreover, because ADA response vary from subject to subject, there is not only the lack of similarity between the positive control and test sample, but also lack of similarity among subjects. Therefore quantification of ADAs based on the dose response curve of positive controls is likely to introduce substantial bi- ases. In 2004, the concept of a tiered assay approach to ADA detection and quantification was first introduced by Mire-Sluis et al. (2004). Centered on what is now known as the first tier screening assay, the authors discuss prac- tical considerations of and provide recommendations for screening assay design and optimization. Subsequently the tiered approach concept was expanded by Geng et al. (2005) to explicitly include a tiered process in which all samples are tested using the screening assay, and only samples that are tested positive are further tested using the confirmatory assay to determine specific bind- ing to the therapeutic protein. In 2007, the method was further expanded in Shankar et al. (2007) to include a third tier in which characterization of antibody isotopes and determination of their neutralizing activities is carried out on samples which are deemed positive by the confirmatory assay. Recom- mendations for the development of cell-based assays for the detection of neu- tralizing antibodies are provided in Gupta et al. (2007) for both non-clinical
- 34. Introduction 7 and clinical studies. A seminal paper by Shankar et al. (2008) suggests a more unified approach to ADA assay validation, which utilizes formal statistical experimental design and analysis for data normalization and transformation, outlier removal, cut point analysis, and assay quantification and validation. Simple but sufficiently rigorous statistical methods are discussed, with the understanding that use of more rigorous methods is advisable with the help of statistical professionals. More recently, risk-based strategies for detection and characterization of ADAs were recommended by Koren et al. (2008). The method proposes that the extent of ADA testing and characterization be re- sults from risk-based assessments that determine the likelihood and severity of ADA responses. In general, the greater the risk is to the patient, the more testing and characterization is needed. These collective efforts and important publications along with other re- search outcomes paved the way for several milestone publications of regulatory guidelines on assay development, including the draft guidelines “Guideline on Immunogenicity Assessment of Biotechnology-Derived Therapeutic Pro- teins” by the European Medicines Agency (EMA) Committee for Medicinal Products for Human Use (CHMP) (EMA (2007)), and the U.S. FDA “Draft Guidance for Industry: Assay Development for Immunogenicity Testing of Therapeutic Proteins” (FDA (2009)). In 2014, the FDA issued “Guidance for Industry:Immunogenicity Assessment for Therapeutic Protein Products” (FDA (2014)), recommending a risk-based approach be adopted for the evalu- ation and mitigation of ADA responses to therapeutic proteins. The scientific rationale of the FDA risk assessment strategy was fully expounded in three publications by FDA researchers which included discussions on clinical con- sequences of immune responses to protein therapeutics, impact of process-, product- and patient-related factors on immunogenicity, effects of changes in manufacturing, and the utility of animal models in assessing a products immunogenicity (Rosenberg and Worobec (2004a,b, 2005)). The guidelines released by the FDA and EMA provide the necessary framework for the devel- opment of robust immunogenicity assays and sound risk mitigation strategies for products under clinical development. Although there has been no formal regulatory guideline issued by the Japanese regulatory authorities, papers and presentations given by Japanese regulators suggest their requirements for im- munogenicity testing, characterization, and risk control are consistent with those recommended by the EMA and FDA. In the following section, regula- tory guidelines from various regulatory agencies are reviewed, and key recom- mendations highlighted. 1.4.1 FDA Guidelines 1.4.1.1 Tiered Approach to ADA Assay Development In light of the cumulative knowledge of immunogenicity risk and recommen- dations in white papers published between 2004 and 2009 on ADA assay
- 35. 8 Statistical Methods for Immunogenicity Assessment development, the 2009 FDA guidance adopts a tiered approach to ADA assay development. To start, a screening assay is used to classify samples into either a positive or negative category. Often easy to run, fast and sensitive, the assay provides an efficient way to detect potential ADA positive samples. Because the screening assay detects all antibodies against the therapeutic protein re- gardless of their functional impact, positive samples from the screening assay are subjected to a confirmatory assay to determine specificity of the ADAs against the therapeutic protein. The confirmed samples are further tested by a neutralizing antibody (NAb) assay to assess the neutralizing capability of the ADA against the therapeutic protein. Additionally, other tests aimed at assessing immunoglobulin subclasses or the isotypes of the ADAs and their epitope specificity, cross-reactivity with endogenous proteins, and other char- acteristics may also be carried out. This approach has been widely adopted by analytical laboratories that specialize in immunogenicity detection and quan- tification. A diagram of this tiered approach is presented in Figure 1.1. Serum samples Screening assay Confirmatory assay Characterization assays (titer, Nab, epitopes) ADA - – – + + FIGURE 1.1 Tiered approach to immunogenicity assessment. The FDA guidance points out that results of pre-clinical testing is not pre- dictive of immunogenicity in human subjects, but acknowledges that immuno- genicity from animal models may be useful for pre-clinical safety assessments, and may provide insight for the monitoring of antibody-related toxicities in human studies. The FDA guidance also allows for an evolving ADA assay development and validation paradigm that allows implementation of prelimi- nary validated assays for use in preclinical and early clinical trials. However, results from fully validated assays are required for the licensure application. Key parameters that need to be considered for ADA assays intended for
- 36. Introduction 9 human sample testing include: (1) Sensitivity: the assays should detect clini- cally meaningful levels of ADAs; (2) Interference: effects of therapeutic drug in sample and sample matrix should be evaluated; (3) Functional or physio- logical consequences: the assays should detect neutralizing activity of ADAs; and (4) Risk-based application: for each product, a careful sampling plan and testing strategy should be developed based on probability that product will elicit an immune response and the potential severity of the ADA response. For assays in each of the three tiers, the FDA guidance provides ADA assay design considerations such as assay format, positive and negative controls, and minimum required dilution, cut point, matrix effect, and drug tolerance (FDA (2009)). Also suggested are key validation parameters including sensitivity, specificity, precision, robustness and stability. For all assays, the guidance recommends test samples be obtained at appropriate time points considering the product’s’s half-life and dosing frequency. For example, for IgM detection, it is optimal to collect samples 7-14 days after exposure whereas samples taken at 4-6 weeks post treatment are recommended for determining IgG responses. It is also recommended to collect pre-exposure samples that can be tested, and used as baseline measures for assessing drug-related ADA responses. 1.4.1.2 Immunogenicity Risk Assessment In August 2014, the U.S. FDA published a guidance entitled “Guidance for In- dustry: Immunogenicity Assessment for Therapeutic Protein Products.” Rec- ognizing the immunogenic potential of biologics that may adversely impact patient safety and product efficacy, the guideline recommends a risk-based approach to mitigating immunogenicity risk. Central to this approach is a thorough understanding of clinical consequences of ADA responses, identifi- cation of factors that may affect the immunogenicity of the product under development, and development of risk control strategies. To that end, the guidance describes various product- and patient-specific factors that have the potential to affect the immunogenicity of protein therapeutics and provides detailed recommendations pertaining to each of these factors that may reduce the likelihood of unwanted immune responses. Furthermore, the guidance af- fords a series of risk mitigation strategies which can be employed in the clinical development of protein therapeutics. In addition, supplemental information on the diagnosis and management of particular adverse consequences of immune responses is provided, along with discussions of the uses of animal studies, and the conduct of comparative immunogenicity studies. Clinical Consequences As stated in the FDA guidances, administration of therapeutic protein products in patients often results in unwanted immune responses of varying clinical relevance. Adverse events may range from transient antibody responses with no apparent clinical manifestations to life-threatening and catastrophic reactions. To mitigate such risk, it is imperative to understand the underlying immunologic mechanism, and devise risk control strategies accordingly.
- 37. 10 Statistical Methods for Immunogenicity Assessment Impact on Efficacy Development of both neutralizing and non-neutralizing antibodies as a re- sult of immunogenicity can cause loss of efficacy in product recipients. This is of particular concern when the product is a life-saving therapeutic. For exam- ple, persistent ADAs to the treatments of metabolic diseases may complicate the intervention, resulting in diminished clinical benefits, disease progression, and even death. Less severe consequences include alteration of the pharmacoki- netic profile of the product, which may require dose modifications. In addition, ADAs may have an impact on pharmacodynamics by misdirecting the ther- apeutic protein to target Fc Receptor (FcR) bearing cells, thereby reducing efficacy. Therefore, the FDA guidance recommends determining the clinical relevance of both binding and neutralizing antibody responses by correlating them with clinical manifestations. Consequences for Safety As previously discussed, the safety consequences of immunogenicity are wide-ranged. Although ADA responses often do not cause adverse events, some incidences of ADA-induced adverse events have been observed. The guideline outlines major safety concerns associated with immunogenicity. They include acute allergic reactions such as anaphylaxis. However, it is recognized in the guidance that the presence of ADA alone is not necessarily predictive of ana- phylaxis or other hypersensitivity reactions. The clinical relevance of these ADAs can only be elucidated through correlation with clinical responses. This is a subject that is studied at length in Chapter 4. Other safety concerns comprise cytokine release syndrome(CRS) caused by the rapid release of pro- inflammatory cytokines, infusion reactions ranging from discomfort to severe reactions, non-acute reactions such as delayed hypersensitivity, and finally, in cases where the products have endogenous counterparts critical for certain physiological functions, cross-reactivity to endogenous proteins. Factors Affecting Immunogenicity Immunogenicity is a complex phenomenon, owing to myriad factors po- tentially affecting immunogenicity. The risk factors can be categorized into patient-specific factors such as genetic makeup and immune status of the pa- tient and product-specific characteristics of which some are product-intrinsic such as glycosylation, aggregation, impurities and formulation and others are product-extrinsic factors including route of administration, dosing frequency, and existence of endogenous equivalents. Understanding of these risk factors is the foundation for the development of effective risk mitigation strategies for unwanted immunogenicity. For example, murine mAbs are known to elicit immunogenicity. Recent technology advances make possible replacement of murine amino acid sequences with human sequences. As such, there has been notable decline of immunogenicity incidences due to use of humanized mAbs. However, despite the progress, immunogenicity issues persist even for thera- pies of humanized mAbs. It is critical to approach immunogenicity risk control from a holistic perspective, which includes identification of risk factors and development of risk-mitigating strategies.
- 38. Introduction 11 • Patient-Specific Factors Certain patient’s characteristics may predispose the subjects to the de- velopment of undesirable immunogenic reactions to some products. For example, patients with an activated immune system due to certain infec- tions or autoimmune disease may have high chances to mount immune responses to therapeutic product than those whose immune systems are compromised. Other factors include the patient’s age, genetic makeup. The guidance emphasizes the need for the manufacturer to provide a clear rationale to support the selection of an appropriate study population, es- pecially for first-in-human studies. Additional risk factors include sensiti- zation of patient’s prior exposure to the drug or similar protein, and/or to the excipients, and/or process/product-related impurities; route, dose, and frequency of administration; patient’s genetic makeup and status of tolerance to the homologous endogenous proteins. Table 1.1 summarizes patient-specific factors along with risk control recommendation in the FDA guidance. TABLE 1.1 Patient-specific factors that affect immunogenicity and risk-mitigating strate- gies Factors Recommendation Immunologic Status and Competency Provide a rationale to support selection of study population, especially for first-in- human studies Prior Sensitization Screen for history of relevant allergies. Dos- ing based on individual risk-benefit assess- ment Route of Administration, Dose, and Frequency Carefully select route of administration Genetic Status Evaluate genetic factors to predispose pa- tients to ADA development Tolerance to Endogenous Protein Gain robust understanding of immune tol- erance to endogenous protein • Product-Specific Factors The FDA guidance also lists nine product-intrinsic factors that may affect immunogenicity. They include product origin, structural and functional characteristics of the product such as presence of aggregates, glycosyla- tion and pegylation variants. Detailed discussion of the impact of these factors may have on the product can be found in the guidance. The fac- tors are listed in Table 1.2, with corresponding recommendations of risk management.
- 39. 12 Statistical Methods for Immunogenicity Assessment TABLE 1.2 Product-specific factors that affect immunogenicity and risk-mitigating strate- gies Factors Recommendation Product Origin Naturally sourced products should be evalu- ated for other protein and non-protein com- ponents Prior Structure and Post Translational Modification For fusion molecules, studies to define anti- genic site of antibody response are recom- mended Aggregates Minimize protein aggregation to the maxi- mal extent possible Glycosylation/Pegylation Use proper cell substrate production system that glycosylates the therapeutic protein in a nonimmunogenic manner. Assays for an- tibodies to PEG itself should be developed and utilized Impurities with Adjuvant Activity Use assays of high sensitivity and clinical relevance to detect and quantitate levels of innate immune response modulating impu- rities Immunomodulatory Properties Monitor product potential for autoimmu- nity from the earliest stages of product de- velopment Formulation Evaluate excipients for potential to prevent product denaturation and degradation Container Closure Test for leachables on product under both stress and real-time storage conditions Product Custody Educate patients. Ensure cold chain security A Risk-Based Strategy A risk-based approach to immunogenicity assessment is recommended in the FDA guidances. The method consists of assessments of the overall risk of the therapeutic protein, identification of risk factors, and development of control strategies, which includes a risk management plan for clinical testing, and measures to minimize chances of unwanted immune responses. In general, the greater the immunogenicity potential the protein has, the more stringent the risk management plan should be. The risk-based approach lends drug de- velopers the tools for development of therapeutics, in the presence of potential immunogenicity risk.
- 40. Introduction 13 1.4.2 European Medicines Agency (EMA) Guidance 1.4.2.1 EMA Guidelines on Immunogenicity Assessment In 2007, EMA released a draft guideline entitled “Immunogenicity Assessment of Biotechnology-Derived Therapeutic Proteins” (EMA (2007)). The scope of the guideline covers proteins and polypeptides, their derivatives, and prod- ucts of which they are components, for example, conjugates. These proteins and polypeptides are primarily derived from recombinant or non-recombinant expression systems. General recommendations and principles are provided to guide developers and assessors on immunogenicity evaluation. The guideline discusses factors that may affect immunogenic responses, utilities of non- clinical assessment of immunogenicity, development of analytical methods for detecting and quantifying ADAs in clinical samples, potential clinical con- sequences of immunogenicity, immunogenicity and clinical development, and immunogenicity risk management plans. The EMA 2007 guideline stresses that therapeutic proteins should be seen as individual products, and experience from related proteins can only be con- sidered supportive. Therefore, immunogenicity evaluation needs to be stud- ied for each product and each indication/patient population. The guideline also classifies risk factors into patient- and disease-related, and product- and process-related characteristics which are similar to the lists provided in Ta- bles 1.1 and 1.2. Patient-related factors include genetic factors that are either modulating the immune response or related to a gene defect, age, disease- related factors (severity and stage of a disease), concomitant treatment(s), duration of and route of administration, treatment modalities, and previous exposure to similar or related proteins. Product factors that may influence the immunogenic potential of the therapeutic are the origin and nature of the related/similar protein (structural homology, post-translational modifi- cations), modification of the native protein (e.g. pegylation), product- and process-related impurities (e.g. breakdown products, aggregates and host cell proteins, lipids or DNA), and formulation. Like the FDA guidance, the EMA document indicates that there is lim- ited utility in using results of non-clinical studies to predict immunogenicity in human. However, animal models can be used to compare immunogenicity responses for similar biological products, and for changes to the manufactur- ing process. The EMA guideline also recommends a tiered approach to the development of reliable and robust methods for immunogenicity assessment. The guideline also stresses that careful consideration should be given to sam- pling schedule for immunogenicity assessment in clinical trials, and inclusion of all patients for such assessment. For the market authorization application, an immunogenicity risk management plan is required. Post-marketing studies may be needed to further evaluate the product immunogenic potential. Subsequent to the 2007 EMA guideline, EMA issued a guideline enti- tled “Guideline on Immunogenicity Assessment of Monoclonal Antibodies In- tended for In Vivo Clinical Use,” which came into effect in December, 2012.
- 41. 14 Statistical Methods for Immunogenicity Assessment This guideline addresses immunogenicity issues of mAbs intended for clinical use. Once again, a risk-based approach is recommended. The guideline in- cludes assessments of risk factors impacting immunogenicity of mAbs, clinical consequences, and considerations in ADA assay development, in particular, for neutralizing antibodies. Two additional EMA guidelines, which discuss immunogenicity issues, were released in 2012 and 2013. One is a guideline on similar biological medici- nal products containing mAbs (EMA (2013)) and the other on similar biologi- cal medicinal products containing biotechnology-derived proteins as the active substance (EMA (2012)). Both guidelines recommend comparability immuno- genicity assessments between the innovator drug and biosimilar version of the drug. It is further recommended that the innovator therapeutic protein and the biosimilar product be tested using the same validated ADA assays. In instances when the biosimilar product has a higher level or frequency of im- munogenicity than the innovator biologic, a comparison of risk to benefit may be necessary. In addition, long-term immunogenicity data post-authorization might be required especially in cases where the duration of comparability study included in the licensure application was short. 1.4.2.2 Latest Development of EMA Immunogenicity Guidelines Since the publications of 2007 and 2009 EMA guidelines, a considerable amount of information and knowledge has been gained concerning ADA as- says, risk factors, and clinical consequences of immunogenicity including loss of efficacy, hypersensitivity, and cross-reactivity with endogenous proteins. In consideration of these cumulative knowledge and issues observed in assess- ment of market authorization applications (MAAs), in March 2014, the EMA published a concept paper, announcing its intent to revise the 2007 guideline. The revision will address issues including: (1) More specific guidance for the presentation of immunogenicity data; (2) Requirements of data on antibody assays; (3) Role of in vitro and in vivo non-clinical studies; (4) Risk-based approaches to immunogenicity; (5) Clinical data to study the correlations of the induced antibodies to allergic and anaphylactic/anaphylactoid reactions, delayed immunological reactions, pharmacokinetics, lack of efficacy; (6) Com- parative immunogenicity studies; and (7) Post-licensing immunological stud- ies. The EMA stated that the aim of the revision is not to increase the number of studies on immunogenicity, but to increase the quality of the studies and their clarity to the assessors. 1.4.3 Japanese Regulatory Requirements of Immunogenicity So far, there have been no formal regulatory documents issued by the Japanese regulatory authorities. However, the agency fully recognizes that immuno- genicity is a critical issue in the manufacture, clinical, and commercial use of the therapeutic proteins (Hayakawa and Ishii-Watabe (2011)). Since the
- 42. Introduction 15 middle of 1980s, many biological products have been approved in Japan. Drawing from the experience and knowledge gained from those approvals, the Japanese regulatory authority requires immunogenicity be tested in both non-clinical and clinical studies. They also suggest mitigating immunogenic- ity risk using risk-minimizing approaches, which include reducing product- and process-related risk factors and management of treatments with known risk. For example, in silico studies can be carried out to help select candidate sequences that are less immunogenic, and biomarker studies can be used ex- clude high-risk patients from clinical trials. The Japanese regulatory authority also acknowledges that, in the long run, advances in in silico technology and development of more relevant animal models will doubtlessly help select less immunogenic proteins for clinical development. However, in the short term, pre-approval data including those from both nonclinical and clinical studies might not be sufficient to gain an adequate understanding of product immuno- genic profiles. Firms have to rely on post-approval surveillance programs to gain a solid understanding of the product immunogenicity risk. In principle, similar to the U.S FDA and the EMA, the Japanese regulatory authority advocates a risk-based approach to immunogenicity assessment. 1.5 Statistics in Immunogenicity Risk Assessment All biologics become potentially immunogenic under specific circumstances. The risk varies considerably among products, patient populations, and treat- ment regimens. In general, a risk-based approach to assessing immunogenicity is recommended by regulatory guidelines and has been widely adopted by drug manufacturers and clinical investigators. The method is developed on a case-by-case basis. As pointed out by Koren et al. (2008), the essential components of a risk-based approach to immunogenicity assessment include (1) understanding molecular characteristics of the therapeutic protein; (2) its mechanism of action (MOA) and intended use; (3) target population; (4) risk factors that were discussed in the previous section; (5) associated risk control strategies. The knowledge of the protein’s molecular characteristics, MOA, therapeutic indication, and intended recipients helps classify the molecule into different risk categories such as low, moderate, or high risk. These classifica- tions will impact the ADA testing plan including sampling time and frequency during clinical development. For example, the more immunogenic the molecule is, the more frequent the ADA testing needs to be conducted. It is also essen- tial to identify risk factors through an objective risk assessment, based on prior knowledge and historical data. For this purpose, laboratory, nonclinical data, and clinical experience can be utilized. Once the risk factors are identified and risk level determined, proper control strategies can be devised. Virtually all aspects of immunogenicity risk assessment involve applica- tions of statistical methods and principles. The unique challenges of predicting
- 43. 16 Statistical Methods for Immunogenicity Assessment immunogenicity either in silico, in vitro or in vivo in animal models, develop- ing sensitive and reliable ADA assays in the absence of reference standards, establishing ADA association with clinical sequelae, identifying risk factors, and devising immunogenicity risk-mitigating strategies require the application of advanced statistical methods. For example, since there are usually multiple factors that contribute to risk of immunogenicity, risk assessment needs to be conducted in a holistic manner. Multivariate analyses, which account for inter- dependence among potential risk factors, and time-dependent nature of ADA measurements collected at different sampling points, are likely to be effective tools for exploring association of ADAs with the various risk factors described in the introduction. In addition, understanding that the sample sizes might be small and incidence of ADA might be low for some early studies, mixture models or meta-analysis might be explored to describe the associations. 1.5.1 In Silico Prediction of Immunogenicity A brief synopsis of the roles of MHC molecules in regulating T-cell responses, which are well characterized, are briefly explained here. MHC molecules bind to peptides which are expressed of the surface of APCs and which are recog- nizable by T-cell receptor, thus activating T-cell responses. In recent years, in silico methods have been developed to identify peptides which may have high affinity to the MHC binding grove, which could trigger strong immune responses. Since the binding grove of MHC class I molecules are closed at both ends, the sizes of peptide sequences that bind to this class of molecules are short, and it is relatively easy to predict their affinity to MHC-I molecules. By contrast, the binding groove of MHC Class II molecules are open at both ends. As a consequence, peptides of varying lengths can bind to the MHC Class II groove. While useful in rational design of therapeutic proteins, prediction of MHC Class II binding peptides is much more challenging. Research of binding motifs revealed that a segment of nine amino acids within a peptide is instru- mental in peptide-MHC-binding (Zhang et al. (2008)). Computer algorithms based on various statistical modeling methods such as artificial neural network and Gibbs sampling can be used to predict MHC-II-binding peptides (Brusic et al. (1998), Nielsen et al. (2004)). 1.5.2 ADA Detection and Quantification Key to successful evaluation of immunogenicity is to have well developed, validated, and sensitive assays. Such assays enable objective detection and characterization of ADAs, and render confidence in the data used for im- munogenicity assessment. However, there are many challenges associated with ADA assay development, optimization, and validation. For example, lack of reference standards makes it extremely hard to accurately estimate an ADA assay detection limit (Dodge et al. (2009)). Furthermore, although various assay platforms such as enzyme-linked immunosorbent assay (ELISA) and
- 44. Introduction 17 electrochemiluminescene-based assay (ECLA) are well understood scientifi- cally and several guidelines are available for ADA assay development and validation (FDA (2009), Mire-Sluis et al. (2004), Shankar et al. (2008)), data analysis and interpretation has become increasingly sophisticated and remains challenging. While the published works (e.g., Shankar et al. (2008)) cover a broad array of statistical issues related to ADA assay development, they fall short on development of effective statistical methods to solve the problems (Zhang et al. (2013)). For example, Kubiak et al. (2013) demonstrated that the results from screening assays and confirmatory assays are highly correlated. Yet, there has been no method developed so far that can effectively account for such correlations. As a result, estimated false-positive rates are likely inflated. Likewise, for sample size determination, cut point analysis, outlier removal, comparison between sample populations from validation and in-study experi- ments, utility of tiered approach, and overall reporting of ADA results all need careful scientific and statistical evaluations (Zhang et al. (2013), Zhang et al. (2014), Gorovits (2009)). In addition, because ADA assays are often developed under stringent timeline, utilization of statistical design of experiment (DOE) strategies is advantageous in understanding the effects of multiple factors and their interaction on ADA assay performance (Ray et al. (2009)). 1.5.3 Clinical Characterization of ADA As previously discussed, clinical consequences of immunogenic response com- prise altered exposure to the drug, compromised efficacy, and adverse events ranging from transient appearance of ADAs to severe life-threatening reac- tions. It is critically important to assess the impact of ADA on the efficacy and safety of the therapeutic protein. The ADA responses also need to be further correlated with parameters such as pharmacokinetics and pharmacodynam- ics to gain a deeper understanding of the clinical effects of ADA (Wadhwa et al. (2003)). Efforts have been made to develop a harmonized strategy for the assessment and reporting of data from clinical immunogenicity studies of therapeutic proteins and peptides (Shankar et al. (2014)). This approach has the potential to maximize the utility of immunogenicity data from clinical tri- als. Three aspects of ADAs are of particularly interest, and are recommended to be reported. They include (1) characteristics of ADA reactions; (2) relation- ships of ADAs with clinical pharmacokinetics (PK) and pharmacodynamics (PD); and (3) relationships of ADAs with clinical safety and efficacy. 1.5.3.1 Characteristics of ADA Immune Response The ADA incidence rate is a key measure of immunogenic response. Accurate estimation of ADA responses helps assess the immunogenic potential of the therapeutic protein. ADA can be characterized through many clinical end- points such as number of ADA incidences within a fixed observational time, probability of occurrence, and time to first detection of ADAs. Statistical
- 45. 18 Statistical Methods for Immunogenicity Assessment methods that provide estimation of these endpoints are discussed in Chapter 4. 1.5.3.2 Correlation between ADA and PK/PD, Clinical Safety and Efficacy ADAs induced by administration of protein therapeutics has an impact on their PK/PD characteristics. ADAs with high affinities for the protein thera- peutic have an increased likelihood of modulating and even neutralizing the drugs therapeutic effects. For example, drug-clearing or drug-sustaining ADA responses can cause increased or decreased drug clearance rates, respectively, thus necessitating dose modifications. The characterization of these ADA re- sponses presents experimental challenges because of many technological limi- tations including inability to measure absolute ADA concentration or affinity. For example, ADA responses may not correlate well with the true immune response as they vary when different assay platforms are used. Mathematical models incorporating PK/PD and ADA features can provide significant in- sights which can be used to characterize ADA responses such as severity of the response, and the elimination rate of ADA-drug complexes. Discussions of these models are provided in Chapter 4. In addition, traditional statistical models such as fixed, mixed effects, and/or repeated measurement models can be used to assess the impact of ADA frequency, time to first ADA episode, du- ration of ADA on PK/PD parameters/profiles. Bayesian analysis may also be utilized to differentiate PK/PD profiles between ADA-positive and -negative subjects. 1.5.3.3 Relationship of ADA with Clinical Efficacy and Safety The correlation of ADA characteristics, in terms of magnitude, frequency, and duration, with clinical endpoints can be modeled within a multivariate frame- work. Multivariate approaches which accommodate inter-dependence among response variables may enhance the detectability of statistical significant as- sociations. These approaches are particularly useful when data are limited. 1.5.3.4 Identification of Risk Factors Identification of risk factor starts with the determination of the criticality of each factor. The criticality of a factor is determined based on the impact im- munogenicity has on product safety and efficacy. The criticality of a factor can usually be determined through a risk assessment process. As the first step, the assessment determines both the severity of immunogenicity due to the risk fac- tor and the probability or likelihood for the immunogenic event to occur. The severity ranks the risk factor based on the consequences of the immunogenic reaction; whereas, likelihood of the immunogenic event characterizes the prob- ability for an immunogenic event to occur when the risk factor is outside of its acceptable range. The determination of both severity and likelihood requires
- 46. Introduction 19 knowledge of both the product and process, and depends heavily on labora- tory, non-clinical, and clinical data. Statistical tools ranging from graphical displays to sophisticated modeling are important in the identification of risk factors. 1.5.4 Control of Immunogenicity Risk Although general strategies for immunogenicity risk control have been dis- cussed in the literature and outlined in regulatory guidelines, the control strategies are product-dependent. In addition, as immunogenicity is impacted by so many risk factors which are usually correlated and interdependent, it is important to account for their joint effects in the design of risk mitigation strategies. In this regard, two methods are of particular interest. One is to re- duce immunogenicity risk due to process- and product-related factors through setting appropriate acceptance ranges for the critical factors predicted to be involved in the development of immunogenicity. The other concerns segment- ing the patient population so as to identify a subset of the population that has less potential to have immune response to the treatment. Such a subset can be used as the target population for the clinical development of the therapeutic protein. 1.5.4.1 Control of Process/Product Factors Product- and process- related factors are often involved in the development of immunogenicity. Oftentimes, these risk factors are correlated, and have a joint impact on immunogenicity. For example, aggregation is frequently iden- tified as a risk factor of great concern. However, the immunogenic potential of aggregation is likely dependent on the mechanism of action of the therapeutic protein. It may or may not have a significant impact depending on the nature of the target molecule (immunostimulatory or immunomodulatory). By iden- tifying critical risk factors and controlling them jointly within ranges such that movement within the ranges would not cause immunogenicity concerns. Clin- ical data with measurements of risk factors and immunogenicity incidences are most useful in establishing the acceptance ranges. Statistical models can be used to link the risk factors to the ADA incidences, thus enabling accep- tance criteria to be set in a clinical meaningful fashion. More discussion on this subject is given in Chapter 5. 1.5.4.2 Biomarkers for Immunogenicity As previously discussed, some patient’s characteristics predispose the subject to have an immunogenic response to the biological therapies. One way to mit- igate this immunogenic risk is to segment the patient population so that a subset of the population that is less likely to have immune reaction is identi- fied. This subpopulation will be the target population for the biologic clinical development. In the literature, some markers are used to indicate possible im-
- 47. 20 Statistical Methods for Immunogenicity Assessment munogenicity. For example, usually high and persisting levels of neutralizing antibodies are indicative of immunogenicity (Cornips and Schellekens (2010)), and can be potentially used to exclude high risk patients from enrolling into clinical studies. However, since factors that impact immunogenicity are rarely independent, the use of a univariate marker is inadequate to differentiate one group of subjects from another. In Chapter 5, various statistical methods for patient segmentation and biomarker discovery are introduced. They include cluster analysis, principal component analysis, and predictive modeling based on discriminant analysis. These methods are illustrated using simulated ex- amples. 1.6 Statistical Considerations in Comparative Immuno- genicity Studies Immunogenicity issues can also arise in the development of biosimilar prod- ucts. Although the follow-on biologic might be derived from the same substrate as the reference product, the raw materials might be from different sources and there might be differences in manufacturing process that can potentially increase the risk of immunogenicity. In the EMA guideline on “Similar Biolog- ical Medicinal Products Containing Biotechnology-Derived Proteins as Active Substance: Non-Clinical and Clinical Issues,” (EMA (2013)) it is stated that immunogenicity testing of the biosimilar and the reference products should be conducted within the comparability exercise by using the same assay for- mat and sampling schedule. Assays should be performed with both the ref- erence and biosimilar molecules in parallel (in a blinded fashion) to measure the immune response against the product that was received by each patient. Therefore, a formal immunogenicity study is required to establish compa- rable immunogenicity profiles between the biosimilar and reference product. The study needs to be carefully planned, sample size adequately determined, and sample testing scheme well thought out to ensure collection of sufficient and quality data. Several study design considerations are discussed in Chow (2013). They include patient population, randomization, and washout period for cross-over design, inter- and intra-product variability, sample size, and surrogate endpoints. Various methods for sample size calculations are pro- vided. As comparative immunogenicity study design and analysis is beyond the scope of this book, readers who are interested in the topic should refer to Chow (2013).
- 48. Introduction 21 1.7 Concluding Remarks Immunogenicity is a critical issue in the development of biologics. If not man- aged well, it may cause either early termination or limited use of the products. Therefore immunogenicity is a potential barrier to further development of oth- erwise effective treatments. Assessment and management of immunogenicity risk is a regulatory requirement and demands careful scientific, regulatory, and statistical considerations. When effectively utilized, statistical methods can be advantageous in addressing issues related to ADA assay development, risk factor identification, and development of effective control strategies. It is also worth pointing out that immunogenicity risk assessment requires ex- pertise from various disciplines. Only through collective efforts of individuals from various scientific and regulatory disciplines, including statisticians, phar- macokineticists, toxicologists, clinical assay developers, clinicians, biopharma- ceutical engineers, and regulatory reviewers, can immunogenicity issues be adequately addressed.
- 49. This page intentionally left blank This page intentionally left blank
- 50. 2 ADA Assay Development and Validation CONTENTS 2.1 ADA Assays ...................................................... 23 2.1.1 Multi-Tiered Approach .................................. 24 2.1.1.1 Screening Assay .............................. 25 2.1.1.2 Confirmatory Assay .......................... 25 2.1.1.3 Neutralizing Assay ........................... 26 2.1.2 Assay Platforms ......................................... 26 2.2 Assay Development and Validation .............................. 28 2.2.1 Assay Parameters ........................................ 29 2.2.1.1 Cut Point ..................................... 29 2.2.1.2 Sensitivity .................................... 29 2.2.1.3 Drug Tolerance ............................... 33 2.2.1.4 Precision ...................................... 33 2.2.1.5 Robustness ................................... 35 2.2.1.6 Ruggedness/Reproducibility ................. 35 2.2.2 Life-Cycle Approach ..................................... 35 2.3 Design of Experiment ............................................ 36 2.3.1 Fractional Factorial Design .............................. 38 2.3.2 Response Surface Design ................................ 41 2.3.3 Split-Plot Design ........................................ 43 2.3.4 Design Optimality ....................................... 46 2.3.5 An Example of Neutralizing Antibody Assay Development and Validation ............................ 48 2.4 Method Transfer ................................................. 48 2.1 ADA Assays The assessment of immunogenicity depends on appropriate detection, quantifi- cation, and characterization of ADAs. However, there are multiple challenges for successfully developing ADA assays. Unlike biological assays for measuring drug concentration or protein concentration in human serum, concentrations of ADAs are very low and often in magnitude of ng/mL. Due to the hetero- geneity of patient characteristics and multiple epitopes on biotherapeutics, 23
- 51. 24 Statistical Methods for Immunogenicity Assessment ADA responses are likely to vary from patient to patient and change dur- ing the course of treatment. For instance, immunogenic responses can develop against epitopes previously not immunogenic. In addition, various factors such as the presence of circulating drug and concomitant drug adminstration make accurate detection of ADA even more difficult. 2.1.1 Multi-Tiered Approach Depending on the assay signal readouts, assays typically fall into one of the following four categories: definitive quantitative, relative quantitative, quasi- quantitative, and qualitative (Lee et al. (2003)). Definitive and relative quanti- tative assays are those where the signal readouts having a continuous relation- ship with the concentration of the analyte. These assays typically have a ref- erence standard. Concentrations of testing samples are obtained through cali- bration from the standard curve of the reference standard. In contrast, quasi- quantitative assays do not have reference standards and thus calibration is either not used or should be interpreted cautiously. Unlike quasi-quantitative results which are also expressed in continuous units or counts, qualitative assays generate data which take nominal values such as positive/negative or discrete values such as ordinal scores. The immunoassays for the detection and measurement of ADAs are quasi-quantitative assays, due to the lack of truly ADA positive human samples at the time of assay development and valida- tion. Therefore, it is impossible to interpolate ADA concentration as is done for measuring drug concentration with pharmacokinetic immunoassays. As men- tioned earlier, immune responses are of a rather complex nature. Therefore, a straightforward method is not available for ADA detection and characteriza- tion. Based on the many years of experimentation in the biopharmaceutical industry, it is generally accepted that a multi-tiered approach works well for ADA assays. Since then, several regulatory guidances (FDA (2009), EMA (2007), USP<1106>) and industry white papers (Mire-Sluis et al. (2004), Shankar et al. (2008), Gupta et al. (2007), Gupta et al. (2011)) regarding immunogenicity assay development and validation have been published. The multi-tiered approach is depicted in Figure 1.1 (see page 8). In the first tier, samples are tested in a screening assay. The assay re- sponses of these samples are compared with a threshold (cut point). Samples tested as ADA negative are not subject to further testing. If samples are tested as potentially positive, these samples are further tested in the confir- matory assay. The confirmatory assay, as a second tier is designed to confirm the specificity of potential ADA positive samples and eliminate false positive samples detected in the screening assay. The third tier of testing consists of further characterization of immune response, such as ADA titer, neutralizing activity, isotype, etc.
- 52. ADA Assay Development and Validation 25 2.1.1.1 Screening Assay Screening assays serve as the first step in the immunogenicity testing. They allow rapid testing of clinical trial samples. The screening assays allow a certain number of false positive samples due to the nature of complex antibody-binding. Immunoassays used for ADA detection generally are quasi- quantitative because it is not possible to generate genuine human-specific ADAs as calibrators or positive controls at the time of assay development and validation. Instead, hyperimmune serum from animals immunized with the protein therapeutic often serve as surrogates for positive controls. However, it is well known that these surrogate drug-induced ADAs generally cannot represent the human-specific ADAs. In addition, different hyperimmunized surrogate ADAs usually have different binding affinities and thus any analysis results cannot directly extrapolate to results of human population. During the screening testing, sample immune responses are compared against a screening cut point. The screening cut point of an immunogenicity assay is the level of immune response readout in the screening assay at and above which the sample is deemed to be potentially positive for the presence of ADAs. Samples with responses below the screening cut point are declared negative and excluded from further testing. Samples with immune responses at or above the screening cut point are declared potentially positive and di- rected for additional testing in a confirmatory assay. Since samples below the screening cut point are not tested further, it is important to minimize false negative results. Therefore, selection of an appropriate screening cut point in- volves a tradeoff between false-positive and false-negative classifications. From a risk-based perspective, it is appropriate to have more false positives than false negatives during the initial screening step. Regulatory agencies and white papers recommend setting the screening cut point to allow 5% of false posi- tive classifications. To maintain the desired false positive rate, the screening cut point is usually established using an appropriate statistical method on the data generated from individual samples. Chapter 3 is devoted to detailing various statistical methods for determining cut points. 2.1.1.2 Confirmatory Assay Samples that are potentially positive in the screening assay are confirmed in the second tier of immunogenicity testing. The confirmatory assay is usually the same as that used for the screening assay with the exception that excess labeled soluable drug is added. By competing with a labeled drug in the screening assay, immune responses generated by ADA should be inhibited by addition of the drug, while immune responses from non-specific binding should not be inhibited or inhibited to a lesser degree. False positive samples detected in the screening step are expected to be ruled out in the confirmatory assay. As with the screening assay, a confirmatory cut point is established using statistical methods. The confirmatory cut point value of an assay is the minimum level of inhibition of a positive samples response in the presence
- 53. 26 Statistical Methods for Immunogenicity Assessment of excess drug that determines whether the samples response is specific to the drug. Drug-specific ADA is expected to produce high inhibition and thus should be greater than a confirmatory cut point. Samples with inhibition lower than the confirmed cut point are declared negative. If samples are positive, further characterization may be necessary. 2.1.1.3 Neutralizing Assay As mentioned before, neutralizing antibodies (NAbs) can block the biologi- cal activity of the drug and potentially impact clinical efficacy. One way that NAbs interfere with the binding of the drug to its target is by sequestering the therapeutic from its intended target and thus preventing the drug from elicit- ing the desired pharmacological effect. When the biotherapeutic drug has non- redundant endogenous counterpart in the human body, the NAbs are a great concern to the patient safety. Therefore, depending on the ADA incidence rate as well as the risk of the biotherapeutic drug, it is sometimes necessary to investigate the neutralizing activity of the confirmed ADA-positive sam- ples. The neutralizing effect of NAbs can be detected using cell-based assays which best mimic the in vivo biological activity. Therefore, it is a regulatory expectation that cell-based assays will be used for NAb detection whenever possible. However, cell-based assays can prove extremely challenging due to a lack of appropriate cell lines and difficulties associated with their growth and maintenance. In addition, cell-based assays often suffer from low sensitivity, high variability, poor tolerance to the presence of the drug in the samples and very narrow dynamic ranges. Alternatively, ligand-binding assays can be used to analyze NAbs and are not subject to the shortcomings of cell-based assays. 2.1.2 Assay Platforms An analytical platform utilizes certain unique physicochemical properties of the analyte of interest in order to detect and quantitate it in a matrix. A thor- ough understanding of these properties is indispensable to properly interpret resulting bioanalytical data. Currently, the ligand-binding assays (LBAs) are widely used for immunogenicity testing. LBAs depend on interactions between a receptor and its ligand. These interactions are characterized by high affinity and specificity. These properties make ligand-binding assays ideal for detec- tion and quantification of biological molecules in complex matrices. Generally, in the screening step, the therapeutic protein is immobilized onto a solid sup- port such as microtiter plate or beads. Immobilization of the therapeutic is accomplished by labeling the therapeutic protein with biotin and the solid support with either streptavidin or avidin. Because of the high affinity be- tween biotin and streptavidin/avidin, the therapeutic protein is retained on the support matrix. If ADA molecules exist, the ADA molecules will bind to the immobilized drug proteins. The captured ADAs are then detected using another labeled reagent or drug protein. In bridged enzyme-linked immunosor-
- 54. ADA Assay Development and Validation 27 bent assays (ELISAs), a signal is generated by labeling the drug with enzymes that generate colorimetric readouts. In electrochemiluminescence immunoas- says, the drug is labeled with a ruthenium complex that generates light upon application of an electric current to the solid surface on which the ADA-drug complex is captured (see Figure 2.1 for a bridging assay format). One short- coming of these assays is that the labeling and coating can block the epitopes of the therapeutic drug from binding to the ADAs, leading to potential false negatives. It is therefore important to optimize the assay for drug concen- tration and labeling ratios. In all, a solid understanding of advantages and limitations of the assay format can aid in the design of immunogenicity as- sessment program and interpretation of the resulting data. The most common formats for immunogenicity assays are briefly described below. Reporter Capture Anti-Drug Antibody Drug Drug Solid Support FIGURE 2.1 Schematic of bridging assay format. Bridging assays are arguably the most common immunogenicity assay for- mat. It depends on the ability of ADA molecules to bind more than one molecule of the drug at the same time. Two different forms of the drug are needed for the assay: one that allows ADAs to be captured onto solid surfaces and one that can serve as a reporter for generation of a signal. Upon binding of the ADAs with the two different forms of the drug, the resulting antibody- drug complex can be easily detected. Bridging assays are widely used due to their high sensitivity and capability to detect all immunoglobulin subclasses and most isotypes. Additionally, these assays can be easily applied for de- tection of ADAs from different species which allows use of the same format for non-clinical and clinical samples. The major disadvantage of this format is that it does not detect antibodies directly; rather, multivalent entities are capable of binding more than one molecule of the drug at the same time. For
- 55. Random documents with unrelated content Scribd suggests to you:
- 56. afraid. Some day he would woo his beloved, and I might fare ill with my conditions. No! I will keep my ring!" "Just heaven! can it really be that one sister can so torture another?" Kitty cried, in indignant pain. "And yet at this very moment, seeing as I do your incorrigible egotism, your pitiless nature, your invincible passion for intrigue more clearly than ever before, I am all the more impelled to deliver your former lover at any price from the vampire that thirsts for his life-blood. You must not retain any hold upon him. He shall begin his life anew, in a home where he will find happiness and peace, now that he is no longer condemned to lead a mere life of society by the side of a heartless coquette——" "Many thanks for your flattering description! You show far too much enthusiasm for his happiness to allow of my entrusting my treasure to your keeping." "Give it to me; you may do so without fear." "Even if he should indeed and in truth love you?" The girl's lips quivered in absolute agony, she wrung her hands as in despair, but she was firm. "What if it were so? I should be no irreparable loss. He can easily find a better than I. His past bitter experience is warrant that he will not again deceive himself. Give me the ring, the counterfeit. Although I know that not the least particle of value attaches to it in reality, I promise you to respect it as the one now lying in the river, since it is a sign and pledge of Bruck's enfranchisement." She held out her hand. "I know you to be honourable enough never to use it for your own advantage," Flora said, slowly and with emphasis, drawing off the ring. A tremor shook Kitty's limbs as the gold touched her palm,
- 57. and her fingers closed tight upon the circlet, while a contemptuous smile hovered upon her lips; she was too proud to assert by a single syllable her purity of purpose. "Well?" Flora cried. "I have given you my word; now I am the puppet whom you rule by this wire,"—she raised her closed hand,—"are you satisfied?" And she left the room. As she crossed the threshold, Doctor Bruck was ascending the opposite staircase. He glanced towards the two figures, the one erect and triumphant in the middle of the room, coldly smiling, while the girl, issuing from it flushed and agitated, almost broke down at sight of him. He hurried to her side, and, regardless of all else, put his arm around her to support her. The door closed behind them to the accompaniment of a low, mocking laugh.
- 58. CHAPTER XXVIII. In the afternoon the tempest which flying reports had presaged, as sea-mews announce the coming storm, broke over the house. The legal authorities had been expected since the early morning, and yet when they made their appearance it was like an electric shock. They came too soon for every one. The servants were engaged in moving the Frau President's old-fashioned mahogany furniture, with its dusty and torn coverings, from the garrets down into the hall; Flora's trunks were still awaiting the tardy express-wagon; the cellars were still filled with the wine that there had been no time to remove. The Frau President proudly retired to her bedroom, refusing to see the gentlemen; but, although they were perfectly respectful in demeanour, they could not regard her nerves, but were obliged to ask if the furniture of the room belonged to her, and, when answered in the negative, to request her to remove to an adjoining empty cabinet, since the room must be officially sealed up. In this small apartment the old furniture was placed, the bed aired, and covered with the faded brown silk coverlet which the Frau President had not seen for years, and which caused her a shudder of disgust. Her maid arranged everything as comfortably as possible, putting flowers upon the little mahogany table, and bringing from the bedroom many a trifle that her spoiled mistress had been accustomed to use; but the old lady never noticed the pains she was
- 59. taking: she sat by the window gazing towards the pavilion, the new roof of which was just visible among the trees. This dreaded and detested "dower-house" had grown into a fairy habitation. Rich curtains hung at the windows; everything shone in newness and beauty,—the smooth floors, the elegant furniture, the frescoes, the chandeliers; even the kitchen was thoroughly fitted up, down to the commonest iron spoon. This "bijou" was to have been hers as long as she lived, and she had scorned it for fear lest it might exile her from the society wont to gather at the councillor's. And now—and now! Meanwhile, Flora was contending for her possessions; but all her arguments, even her appeal to the testimony of the servants, were in vain. "Fräulein Mangold," the officials courteously persisted, "might reclaim her own afterwards, but at present everything must be placed under seal." And for hours there was a passing to and fro, up and down stairs. All the plants adorning the house were placed in the conservatories, one key after another was turned in the lock, and every open window was closed. It was dreary to mark the silence and darkness that settled down wherever the officials had finished their work. Amidst it all the servants grumbled openly about the wages due them; but each one made ready to leave the house, where every comfort lay behind lock and key, and where the flesh- pots no longer simmered on the fire. The gardener alone remained, and was lodged in the servants' hall. While this confusion reigned, the soul of the sick girl above- stairs unfolded its wings to leave, calmly and peacefully, after the conflict of years, the worn and weary body.
- 60. Henriette's room was unvisited by the officials; everything about the dying girl was her own. Great pains were taken to avoid even a loud footfall on the third floor, and nothing approached the parting soul that could startle or annoy it. She looked through her window into the rosy heavens; she watched the swallows, their white breasts and wings looking like silver crosses floating among the pink evening clouds. On the previous day, thin wreaths of vapour had still floated above the ruin, and distant noises had troubled the sick girl's mind, causing it to dwell painfully upon the terrible spot where the crashing walls had buried beneath their fragments the "rash man" to whom, with all his weaknesses, she had clung in sisterly affection. But at this solemn evening hour, at the close of the day and of a brief mortal existence, there was nothing to remind one of previous horrors. The doctor sat by Henriette's bedside. He saw how the rapid finger of death emphasized and sharpened each outline of the face, still informed for a brief space of time with consciousness. The ebbing stream of life moved her pulses in faint isolated throbs, like retreating waves returning now and then to plash once more upon a deserted shore. "Flora!" the dying girl whispered, with a speaking glance. "Do you wish to see her?" he asked, making ready to go for her. Henriette faintly shook her head. "You will not be vexed that I wish to be alone with you and Kitty until——" She did not finish the sentence, but plucked at the fading crimson vine-leaves upon the coverlet. "I will spare her, and she will be grateful,"—there was a faint shade of irony in her smile,—"she detests touching scenes. You will take her my farewell, Leo."
- 61. The doctor silently inclined his head. By his side stood Kitty. Her heart beat fast; her dying sister had no suspicion that the relations upon which her mind was dwelling no longer existed. Should she learn the truth? She glanced anxiously at the doctor's face: it was grave and composed; no sudden and unexpected announcement should disturb the peace of the departing soul, and for preparation there was no time. Henriette's eyes wandered to the evening sky. "How exquisitely clear and rosy! It must be a heavenly delight for the freed soul to bathe in such splendour!" she whispered, fervently. "Will it ever be allowed to look back here? I only want to look once, to see"—she turned her head on the pillow with difficulty, and gazed, with eyes glowing for the first time with unutterable love, full at Bruck—"if you are happy, Leo. Then I care not how distant are the starry worlds to which I may be borne." Even in this her last hour the poor girl could not bring herself to say, "I must know you happy, or I shall not be content, for I have loved you intensely with every fibre of my heart." A transfiguring glow seemed to illumine the doctor's bowed head. "All is well with me, Henriette," he said, with emotion. "I dare to hope that I shall not pass a lonely and embittered life; nay, better still, I know that even at the eleventh hour my dream of the true happiness of existence will be fulfilled. Does that content you, my sister?" He pressed his lips upon the small hand that was growing cold in his own. "I thank you from my soul," he added. A blush, faint and rosy as the evening sky, came and went upon the cheek of the dying girl; her timid glance involuntarily sought her sister, who, her hand leaning upon Bruck's chair, was evidently
- 62. struggling to control her grief. At sight of her Henriette's heart melted in pity and sympathy. "Look at my Kitty, Leo!" she said, imploringly, in a failing voice. "Let me tell you of what has so often distressed and pained me. You have always been so cold to her,—once harsh even to cruelty,—and yet there is none to be compared to her. Leo, I have never understood your prejudice against her. Be kind to her—befriend her ——" "To my latest breath! while life lasts!" he interrupted her, scarce able to master his emotion. "Then all is well! I know you will take care of her,—and my strong, brave darling will stand between you and all annoyance——" "Like a faithful sister, which from this moment I am," Kitty completed the sentence, in a choking voice. An ecstatic smile hovered about Henriette's mouth. She closed her eyes, and did not see the shudder that shook her strong sister's frame as the doctor held out his hand to her and she rejected it as if she had no right to its mute pressure. The smile faded, and the dying girl struggled for breath. "Say farewell to grandmamma. Now I would rest,—ah, give me rest, Leo, I entreat!" she gasped. "In ten minutes you will fall asleep, Henriette," he said, in a low, soothing tone. He laid her hand upon the coverlet, and softly put his arm beneath the pillow supporting her head; she lay like a child upon his breast,—a happy death! And before the ten minutes were passed she slept. The fluttering vine-leaves at the window stirred, as if lightly touched, and the rosy light in the sky, in which the parting soul had longed to bathe, suddenly glowed to deepest crimson. The little tame bird
- 63. perched upon the window-sill as usual at sunset,—his soft twitter towards the waxen face upon the pillow was heard for the last time, —and then these windows also were closed, not to be opened until the councillor's house had passed into stranger hands. The Frau President came up to the room, bowed as with a sudden added weight of the age she had so steadily tried to ignore. The white cloud of tulle once more enveloped cheek and chin: no mourning should be worn for a scoundrel, she said. She went to the bedside, and a spasm passed over her features as she gazed upon the calm countenance of the dead. "She is happy," she said, in a broken voice. "She has chosen the better part,—she need not go into exile,—she is spared the bitter, bitter struggle with poverty." But Flora came and went without a word. She took no note of the two faithful guardians at the bedside. She kissed her dead sister upon the brow, and then walked with head erect to the door by which she had entered. She paused, it is true, upon the threshold, but she never turned either her eyes or her head towards where the doctor stood and gravely delivered to her her sister's last message. She bent her head almost imperceptibly in token that she heard what was said, and then rustled down the stairs, to put on her bonnet and go to the nearest hotel, where she had engaged lodgings for herself and her grandmother. No one, not even the dead, was permitted to pass another night beneath the criminal's roof. And when, after nightfall, Henriette's form had been borne away to the hall, where all, clad for the grave and heaped with flowers, await the opening of their latest earthly portal, the last room on the third floor was closed and locked, and the doctor and Kitty
- 64. descended the stairs together. Their steps echoed drearily through the silent, deserted house. The lantern carried before them by the gardener shed abroad a ghostly light over the lonely walls and passages, where so lately the stream of life had flowed in luxurious evidence of what was after all but a false, fleeting show of wealth. The soft night air, as they walked along, was as balm to Kitty's burning eyes. A clear, starry sky canopied the silent park, the single groups of trees could be distinguished, and the mirror of the pond gleamed like dull silver through a misty veil. The gravel crunched beneath their tread, and from afar was heard the water of the weir, but not a leaf or a twig stirred,—it was as quiet as it had been for hours in Henriette's room. And therefore Kitty started in terror when the doctor's full deep voice broke the silence. They had reached the leafy entrance of the avenue, and he paused. "I leave the capital in a few days, and I fear that, until then, you will neither visit my aunt nor allow me to come to the mill," he said, with both sorrow and eagerness in his tone. "I tell myself also that we are walking together for the last time,—that is, for the present ——" "Forever!" she interrupted him, sadly but firmly. "No, Kitty!" he said, as firmly. "It would be a separation forever if your words spoken a few hours since could not be gainsaid. I do not want a sister. Do you think a man can content himself with sisterly letters when he is thirsting for loving words from beloved lips? But no,—I did not mean to speak thus to-day. Only selfishness could betray me into such entreaties while you are suffering as at present. One thing I must say to you, however. This afternoon you had an interview which, when I met you, had agitated you
- 65. profoundly. You had been told what has happened, and of course the whole odium that always attaches to the sudden rupture of an engagement had been thrown upon me,—I saw that in your face; and afterwards, when for love of Henriette you promised to be a sister to me, I heard the power that evil whispers had gained over you,—thank God, not for always! I know—I know that your clear, just insight may be dimmed for a while; but this cannot last. Kitty, on that terrible afternoon I was in my garden, and saw how, on the opposite river-bank, a girl leaned her brow against a tree and wept bitterly." Kitty turned as if to flee down the avenue, but Bruck had taken her hand and held it in a firm grasp. "I saw before me the girl whom I was longing to clasp in my arms. I had just been victorious in the last of those self-conflicts from which I had suffered for months; victorious, because I had liberated myself from false views of life and had admitted that I should be a perjured traitor if I contracted a hated marriage while my whole being was filled with an invincible passion. There stood the one who was dearer to me than all else beside, and my heart leaped, for her streaming eyes did not look towards my aunt's windows, but——" He paused, and pressed the hand which he held to his lips, while she leaned against the trunk of a linden, incapable of uttering a word. "I cannot blame her who was to have been my wife; that matters have been allowed to go so far is my fault,—mine only. I was weak enough, for dread of what the world might say, to continue our engagement after I had discovered, with shame and anguish, that I had been attracted by a beautiful exterior animated by no qualities of mind or heart that did not crumble to
- 66. insignificance if subjected to the slightest test. This discovery I made in the first weeks of our betrothal." He was wrong; the qualities enshrined within that lovely form were not insignificant. Flora's was a nature incredibly malicious. She had known then of Bruck's love for her sister, of course from his own confession. What a contemptible plot! Her victim had the ring in her possession; she had bought it with a price; her word was pledged even though Bruck should woo herself. The young girl's eyes wandered in despair to the starry heavens. She knew that Flora would never release her from her promise although she should implore her on her knees. There would be no need even of Flora's eloquence to convince the world that she was betrayed and deceived, the dupe of her younger sister, who had lured her lover from her. That this was the colour she would give to what had taken place was clear as the stars above. How they sparkled, those shining worlds! To which of those golden orbs had the spirit of her sister been borne upon the rosy evening air? Could she look back to see how the happiness of the man whom she had loved would be wrecked? "You do not speak, Kitty. Your silence rebukes me; I ought not to have spoken to-day," he began again. "I will not press you further. I do not ignore the fact that my desires will arouse a conflict within you: you were not else the strictly just and honourable girl that you are; but I know also that I shall attain the goal I so long for without stormy arguments and entreaties. I will leave you time for consideration and recovery from the grief that now fills your soul and colours every thought and feeling. I go without the assurance that alone can give me peace, but—I shall come again. And now we
- 67. will go on to the mill. Take my arm in full confidence that no brother could care for you with less thought of self than fills my soul at this moment. You might with equal tranquillity put yourself in charge of my aunt and myself when we set out on our way to L——." "I shall not return to Saxony," she said. She had placed her hand within his arm, and they walked slowly along the avenue. The girl's limbs seemed possessed with a mortal torpor that clutched at her throbbing heart and deadened the voice that came so hard and cold from her lips. "I found when I was last in Dresden that in my present state of mind there is no help for me in incessant study or the performance of my trifling household duties. I must have some occupation requiring sustained absorbing labour day after day. Until a few days ago I hesitated to express this need; I knew my first hint at such a thing would arouse a storm of expostulation from my guardian. The heiress's duty was all marked out for her, and consisted in spending her income as brilliantly as possible. All that is past. The dreaded safe is no longer in existence, or rather its paper contents were worthless before it was destroyed. This I have been quite sure of, since Nanni whispered to me this afternoon that everything was being sealed up. I am right, my hundreds of thousands have vanished, have they not?" "I hardly think anything can be saved——" "But I still have my mill, and there I will stay. I shall, perhaps, lay myself open to your serious disapproval when I tell you that from this time I wish to attend to my affairs myself. It savours, perhaps, of 'women's rights' for a young girl to undertake the management of business affairs and represent a firm in her own person."
- 68. "I am not so prejudiced; I advocate warmly such independence upon a woman's part, and I know that you, with your force and energy, would do well; but it is not your vocation, Kitty. Your place is at the head of a happy home, not standing day after day reckoning up columns of figures at a desk in a counting-room. Do not begin it! For at some future day you will be carried off without a question as to the debit and credit in your books, and terrible confusion might be the consequence." If the light of the stars could only have illuminated the dark avenue, the speaker would never have allowed the girl at his side to leave him, so hopeless a despair was painted on her face; he would have taken her in charge then and there, and wrung from her the thoughts that were torturing her. But the darkness covered the terrible struggle that was going on beside him, betrayed by no word or sign, not even a sigh, and he ascribed the depression and discouragement which had made her voice so dull and monotonous to the misery of the parting scene she had gone through with her dead sister. Now and then a pebble rattled from beneath their feet on the gravelled road, and the rushing of the waters of the stream sounded loud and near in the silence that followed the doctor's last words. The lindens of the avenue retreated; the heavens stretched broadly above, and standing clear against their sparkling depths were the two slim poplars that flanked the wooden bridge. At sight of them the doctor involuntarily pressed the girl's arm closer to his side. "There, Kitty," he whispered; "there you used to look for the first violets. I promised you you should do so in future,
- 69. and I can keep my word: I shall always spend my Easter holidays here." Kitty pressed her clenched hand to her breast; she thought the violent throbbing of her heart would suffocate her; and yet she asked, quietly, "Will your aunt accompany you to L——?" "Yes; she will undertake the care of my household so long as I am alone. She sacrifices much to do so, and will be thankful to shake the dust of the large city from her feet and return hither to her green country home. I know that the brave, true heart for which I sue will not delay her release too long," he added, in a tone of tender entreaty. A light appeared twinkling from the mill window. Franz the miller had been buried this afternoon, leaving behind him a widow and three children. The roof that still sheltered them did not belong to them, and the miller's small savings were not sufficient for their support. Susy had been to the villa for a few moments to look after her mistress, and had described to Kitty the despair of the poor wretches, and mourned over "the topsy-turvy state of the business without any master." The bow-window of the room in the lower story looking towards the park was dark. The outline of the mill buildings rose black and shapeless against the sky,—it all seemed lonely and deserted; the bark of the watch-dog, who resented the approaching footsteps, sounded lost as in some endless desert. The wheels were silent, and the huge room was so empty and echoing that one might have fancied that, since the strong human hand so lately working here had stiffened in death, each friendly busy elf had pulled his cap over his peevish face and slipped away.
- 70. The doctor drew the young girl towards him before he opened the gate. "I seem to be leading you into exile," he said, anxiously. "You ought not to give me the pain of knowing you alone after this sad and weary day. Come with me; my aunt will be only too happy to receive and take care of you." "No, no!" she said, hurriedly. "Do not think that I shall resign myself to a passion of useless grief when I am alone. I have no time for it, and I shall not do so. I must," and she pointed to the bow- window, where the dim light of a lamp began to shine behind the chintz curtain, "play the part of comforter there. Those four poor people are dependent upon my energy and assistance." "Dear, dear Kitty!" he said, clasping her right hand in both his own and pressing it to his breast. "Go then in God's name! I should hold it a crime to place one stone in the hard but sure path you have chosen through your present suffering. Only remember that you are not yet quite recovered. Do not make too great a demand upon your strength; and wear the bandage upon your forehead for a few days longer. And now farewell: at Easter, when the last wintry mist has flown, when the ice and snow are thawed, when human hearts throb joyously,—at Easter I shall return. Until then, think of one whose every thought is yours, and do not let slander or mistrust come between us!" "Never!" This one word came almost like a groan from her lips. She withdrew the hand he pressed to his lips, and the gate in the wall clanged to behind her. She took no step forward; leaning against the cold damp wall, her face buried in her hands, she listened breathlessly to his departing footsteps. What was death in comparison with the tortures of this wildly-beating heart condemned
- 71. to live? She listened until the soft night air, brushing her cheek, brought no sound upon its wings, and then, with tearless, weary eyes, she passed on into the house, to enter upon her mission of comforter and protector. Three days later, immediately after Henriette's burial, Doctor Bruck and his aunt left the capital. Kitty had not seen the doctor again, but his aunt had repeatedly passed an hour with her. The same day Flora left also, accompanied by the Frau President. The old lady was to visit the baths; and Flora went to Zürich, where, report said, she was to devote herself for a time to the study of medicine.
- 72. CHAPTER XXIX. More than a year had passed since the day in March when Kitty Mangold, grandchild and sole heir of the wealthy castle miller, had been walking upon the high-road from the town on her way to present herself at her guardian the councillor's in her new character of heiress. Those who now turned aside into the by-road leading to the mill found upon their right a row of pretty little cottages, that belonged to the workmen in the factory, and had been erected upon the waste portion of the mill-garden,—the strip of land that Kitty had begged of her guardian for the convenience of these men. And the townspeople liked much to walk in this direction. Formerly the high massive wall enclosing the mill-grounds had cast its shade so far that the footpath beneath it was almost always damp and had long been avoided. Now the wall had gone, and the pretty path was planted with acacias. The cottages looked neat and trim, with their air of Dutch cleanliness, the pretty porch in front of each, and the small gardens which had been planted the previous autumn with all kinds of flowering shrubs. Behind them loomed the castle mill, hoary with age, its windows looking in the opposite direction, as if angry that its ancient mantle of green had been thus bordered with gay embroidery. It had undergone no alteration, save that the shabby old dial had been
- 73. brightened, and the little gate leading through the wall into the adjacent park had been walled up. There was no longer any connection between the mill and the former estate of the vanished Von Baumgartens from whom the old structure had derived its high- sounding title. But the deafening noise, the throbbing heart of the old pile went on with rejuvenated vigour, and the road to the mill- yard was more frequented than ever,—the masterless business was directed by a firm cool hand and a prudent head. Kitty's undertaking had been attended with success. She had found an experienced foreman, and poor Lenz, the merchant who had lost his all, was her assistant book-keeper. She set herself to work in the office she had fitted up in the mill, to learn the mysteries of business, and her thorough education and excellent capacity soon enabled her to acquire all that Lenz could teach. She did actually work like a man, "day by day;" the business increased, and produced such results as would have astonished the old castle miller himself. And the sight of the contented faces about her smoothed the rough path she had chosen to tread. She had taken charge of poor Franz's widow and orphans, giving them rooms for life in a small out-building of the mill, which she had fitted up for their occupation. The woman continued, as heretofore, to assist Susy in her housekeeping, while the children received such an education as their father, whose mind had been occupied entirely with material considerations, had never dreamed of giving them. It was true that of all the vast wealth left behind him by the castle miller nothing remained for Kitty but the mill and a few thousand thalers which she had induced her guardian to allow her to lend to the workmen to enable them to build their cottages upon the
- 74. mill-land. Her hundreds of thousands had vanished in the flames, and the small amount of gold and silver recovered in a melted condition from beneath the ruins was far more likely to be the remains of tankards and platters than of coin. In the disastrous confusion that followed the explosion there were many creditors whose claims even the real estate and valuable collections were not sufficient to satisfy; the failure proved to be one of the worst and most hopeless that occurred in that time of ruin and uncertainty. Villa and park passed again into the hands of an old and noble family, and the new owner had the ruins of the ancient tower cleared away, the ditch filled up, and even the artificial mound levelled, that there might be nothing upon the aristocratic soil to bring to mind the miserable parvenu who had there met his wretched and disgraceful death. And the ancient wooden arched bridge leading across the stream to the house by the river was also destroyed. The doctor's house was now reached by a stone bridge, crossing the river near the factory, and a pretty footpath along the opposite shore. The house, which had been completely restored late in the autumn, was still unoccupied; the Frau Dean's old friend had passed the winter in the doctor's former town-house, and was to move out only with the return of fine spring weather. Kitty used to stroll hither almost every day. Although the autumn mists hung dank and chill, although snow-flakes filled the air, and the wind blew keen from the north, at the approach of twilight she would lay aside her pen, put on her wraps, and sally forth into the open air. Then for half an hour she would throw away all thought of the columns of figures, the dry business details in which she sought all
- 75. day to bury her warm, longing heart. She was no longer the strict mistress, whose watchful eye never overlooked the smallest irregularity, who exacted a rigid performance of duty from herself as well as from her people, inducing it in the latter case by such a judicious mixture of praise and blame that no harsh word was ever needed from her lips. At this twilight hour she was only the young ardent girl, who, hard and stern as she might be to the passion that possessed her soul, still permitted herself some moments of dreaming melancholy, of unrestrained suffering. Then she would pass through the narrow, creaking wicket-gate leading out into the fields; the gate to which, after the attack in the forest, she, with Henriette in her arms, had bent her weary steps. As she reached the moss-grown fragment of a pedestal in the centre of the grassy lawn, beside which she had stood with Bruck, she would pass her hand lightly over it, as if in a caress, and then seek the spot where the pardon-table had stood, where the doctor, as she now knew, had so suffered for her sake. She walked around the lonely house, with its closed shutters, its new unblackened chimneys, and its creaking weather-cock, to mount the damp, slippery steps and listen at the house-door. Through the key-hole came the soft, low sigh caused by the draught of air sweeping through the wide hall, the withered vines about the doorway rustled, and now and then a belated sparrow would dart in beneath the eaves. This was the only sign of life stirring in the loneliness, but the girl looked for it eagerly; at least the silence was not that of the grave. The right to open this door belonged to beloved hands, and some day footsteps would resound within and dear faces look from the windows; this was sure, although Kitty, at the thought of it, told herself that then she should
- 76. leave her home and wander afar, until—Bruck should conduct hither some bride to whose hand she might confide the ring. His career in L—— was a brilliant one. His reputation spread from day to day. Large and distinguished audiences attended his lectures, and several fortunate cures, of which the objects were individuals of high rank, were everywhere talked of. His aunt's letters to Kitty—she wrote frequently—breathed peace and content; they were a source of immense enjoyment to the young girl, but also of terrible mental conflict, for which reason she replied but seldom and briefly. The doctor himself never wrote,—he adhered strictly to his promise not to assail her with entreaties, and contented himself by sending some message of remembrance, which she kindly and punctually reciprocated. In this solitude her young life passed, day after day. She never dreamed that she was a subject of great interest in the town, that her bold assertion of her independence, her resolute and energetic assumption of authority at the head of her affairs, excited far more attention and respect than had ever been awarded to the heiress. The distinction thus falling to her lot was the cause of a series of visits to the castle mill, of which the first when paid was received with no little astonishment. The Frau President Urach when walking with her faithful maid no longer disdained to make the mill a resting- place, in order, "as her duty to her poor dear lost Mangold required, to look after his youngest child." The old lady had returned to the capital a few weeks after her departure from the villa. She occupied a couple of rooms very high up in a narrow little street, living in a pinched way, in accordance with her very small means, and half forgotten by the world. The
- 77. councillor of medicine, Von Bär, had purchased a country-seat, and grumbling turned his back upon the capital; for her he had vanished entirely, and of all her former acquaintance her only visitors were some few of the friends of her youth and the pensioned Colonel von Giese, who sometimes came to play cards with her. She suddenly found it very comfortable "in this fine old room in the castle mill, where there is really space to breathe in," and, weary with her walk, she would seat herself contentedly in the old- fashioned chintz-covered sofa, that had once sustained the castle miller's burly form, and enjoy the delicious coffee which Kitty always prepared for her, making no sort of remonstrance when Susy, at a nod from her young mistress, hung upon the maid's arm a basket filled with fresh butter and eggs. It was best not to speak to her of Flora, who of course had not lost one penny of her fortune, and who now indeed paid the rent of her grandmother's rooms and the wages of her maid, but could do nothing more, since, as she wrote, she needed all the rest of her income for herself, and could hardly manage to live upon it. She had soon quitted Zürich, where the study of "that disgusting medicine irritated the nerves almost to madness." She was one of those intellectual coquettes who pose for a certain part, greedy for notoriety and a reputation for profound and thorough attainment, while in reality they recoil from the slightest amount of genuine serious study. Easter was at hand. For several weeks improvements had been going on in the garden of the house by the river. The doctor had sent a gardener from L——, who laid out new paths, or rather tried
- 78. to restore the pretty old garden to its original plan. Many men were busy digging and planting, and places were arranged for some statues which had arrived from L—— and were still unpacked in the hall. The shutters of the house had been thrown open for two weeks; the rooms had been freshly painted and papered, and a flag- pole had been erected upon the roof. Then the Frau Dean's friend moved out from town, bringing with her a host of charwomen, who made the house a shining mirror of neatness and cleanliness from garret to cellar. Kitty had not discontinued her walks. On the very day before Easter she came hither once more, at noon. The men were still at work in the garden, but the evergreens that had overgrown the land belonging to the house, forming here and there an impenetrable thicket, had been thinned and left only within the boundaries first assigned them, while from among their dark foliage gleamed the new statues. The winding paths were freshly gravelled, the old creaking wooden gate had been replaced by one of wrought iron; the Frau Dean's arbour had been freshly painted, and behind the house a high picket-fence enclosed a new poultry-yard. Upon the familiar stone pedestal before the door stood a Terpsichore with arms gracefully extended, just as Kitty had imagined her from the remains of the little marble foot. "The statue is very pretty," the strange gardener said to her with a shrug, "but it ought to be more elegantly placed. This lawn," and he looked around upon the old bleaching-ground, "is quite wild, by no means in proper order, but the Herr Professor strictly forbade my touching it." Kitty stooped with crimson cheeks and plucked the first violet, winch had opened fully in all its fragrance at the base of
- 79. the pedestal. "Yes, the grass is full of weeds," the man said over his shoulder, as he walked on. And the house, now really a little castle, actually shone with freshness and beauty—"fitted up as if for a bride," the Frau Dean's old friend remarked to Kitty with an unsuspecting smile. The snow- white kitten came softly to the door over the new tiles of the hall. In the Frau Dean's sitting-room, behind the crocheted curtains, in the midst of the laurels and large-leaved plants that had been moved out from town, the canary-bird piped his clear shrill song. The former life was beginning here anew, and the Frau Dean herself was to arrive by the afternoon train. She was to bring a guest with her, her old friend had remarked with a mysterious twinkle of the eye; who it was she did not know, but she had been commissioned to provide the guest-chamber with new furniture. And as she spoke she threw open the folding-doors leading into it from the hall, and tears filled Kitty's eyes as she thought of Henriette, who had lain here in such pain, and yet peaceful and happy as never before in her sad life. But even while her thoughts were thus occupied she was conscious of a sharp, unfamiliar pang of jealousy. Who was this guest who had become so dear to the Frau Dean's heart that she had been invited to stay with her? The gay rose-covered curtains and the hanging-baskets filled their old places, but the rickety furniture had made way for what was new and pretty, although very simple, and instead of the faded illustrations of Vosz's "Luise" some fine landscapes hung upon the freshly-papered walls. The well-remembered room had been converted into a pretty sitting-room, and an adjoining cabinet that
- 80. had formerly stood empty had been arranged for a sleeping- apartment. All this Kitty looked at once more, with tear-dimmed eyes, and then walked home to place herself at her desk and answer several business letters. Lenz was to return in the evening from a business trip he had undertaken, and his young mistress was anxious to have all in readiness to be entrusted to his hands while she spent the next fortnight with her foster-parents in Dresden. Ah, how difficult it was to fix her attention! Her pulses throbbed, and the handwriting, usually so clear and firm, looked scrawled and careless. She was interrupted too by the Frau President's maid, who came with a large empty market-basket on her arm, on her way to make her Easter purchases of provisions, and the Frau President had told her, since it was only a little out of her road, to stop at the mill and give Fräulein Kitty Fräulein Flora's letter to read. It had just come. Susy was immediately instructed to fill the basket with all sorts of delicacies from her pantry, but the letter lay untouched upon Fräulein Kitty's writing-table long after the maid had returned to her mistress. The Frau President had several times previously sent the young girl her step-sister's letters. The sheets had seemed to burn beneath her touch, but she had dutifully read them through that she might not seem ill-natured. And now a flickering flame seemed creeping towards her from the perfumed envelope lying near her elbow. Impatiently she moved her arm and pushed it beneath a pile of bill- headings. She could not see why, to-day, she should give herself the
- 81. Welcome to our website – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookbell.com