DNA Arrays in Neurobiology (International Review of Neurobiology, Vol 60)

International REVIEW OF Neurobiology Volume 60 International REVIEW OF Neurobiology Volume 60 SERIES EDITORS RONALD...

Author: Michael Miles

78 downloads 952 Views 8MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

International REVIEW OF

Neurobiology Volume 60

International REVIEW OF

Neurobiology Volume 60 SERIES EDITORS RONALD J. BRADLEY Department of Psychiatry, School of Medicine Louisiana State University Medical Center Shreveport, Louisiana, USA

R. ADRON HARRIS Waggoner Center for Alcohol and Drug Addiction Research The University of Texas at Austin Austin, Texas, USA

PETER JENNER Division of Pharmacology and Therapeutics GKT School of Biomedical Sciences King’s College, London, UK EDITORIAL BOARD PHILIPPE ASCHER TAMAS BARTFAI FLOYD E. BLOOM MATTHEW J. DURING PAUL GREENGARD KINYA KURIYAMA HERBERT Y. MELTZER SALVADOR MONCADA SOLOMON H. SNYDER CHEN-PING WU

ROSS BALDESSARINI COLIN BLAKEMORE DAVID A. BROWN KJELL FUXE SUSAN D. IVERSEN BRUCE S. MCEWEN NOBORU MIZUNO TREVOR W. ROBBINS STEPHEN G. WAXMAN RICHARD J. WYATT

DNA Arrays IN

Neurobiology EDITED BY

MICHAEL F. MILES Department of Pharmacology and Toxicology Virginia Commonwealth University Richmond, Virginia

Front Cover Photograph: Photo courtesy of Dr. Sebastiano Cavallaro at the Italian National Research council. (see Chapter 4, Figure 5)

Elsevier Academic Press 525 B Street, Suite 1900, San Diego, California 92101-4495, USA 84 Theobald’s Road, London WC1X 8RR, UK

This book is printed on acid-free paper. Copyright ß 2004, Elsevier Inc. All Rights Reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the Publisher. The appearance of the code at the bottom of the first page of a chapter in this book indicates the Publisher’s consent that copies of the chapter may be made for personal or internal use of specific clients. This consent is given on the condition, however, that the copier pay the stated per copy fee through the Copyright Clearance Center, Inc. (www.copyright.com), for copying beyond that permitted by Sections 107 or 108 of the U.S. Copyright Law. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works, or for resale. Copy fees for pre-2004 chapters are as shown on the title pages. If no fee code appears on the title page, the copy fee is the same as for current chapters. 0074-7742/2004 $35.00 Permissions may be sought directly from Elsevier’s Science & Technology Right Department in Oxford, UK: phone: (þ44) 1865 843830, fax: (þ44) 1865 853333, E-mail: [email protected]. You may also complete your request on-line via the Elsevier homepage (http://elsevier.com), by selecting ‘‘Customer Support’’ and then ‘‘Obtaining Permissions.’’ For all information on all Academic Press publications visit our Web site at www.books.elsevier.com ISBN: 0-12-366861-1 PRINTED IN THE UNITED STATES OF AMERICA 04 05 06 07 08 9 8 7 6 5 4 3 2 1

CONTENTS

Contributors............................................................................ Introduction ............................................................................

ix xi

Microarray Platforms: Introduction and Application to Neurobiology Stanislav L. Karsten, Lili C. Kudo, and Daniel H. Geschwind I. II. III. IV. V. VI. VII. VIII. IX. X.

Introduction . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Experimental Flow . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Microarray Design . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Slides and Probes . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Sample Preparation and Labeling . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Data Analysis . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Data Interpretation. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Confirmation of Microarray Results . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Conclusion. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Online Resources . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. References . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . ..

1 3 6 7 11 12 13 14 15 16 18

Experimental Design and Low-Level Analysis of Microarray Data B. M. Bolstad, F. Collin, K. M. Simpson, R. A. Irizarry, and T. P. Speed I. II. III. IV. V. VI. VII.

Introduction . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Design of Experiments . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Sample Size Considerations . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Normalization . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Expression Summaries for GeneChip Data. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Quality Assessment . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. Detection of Absolute Gene Expression . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. References . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . ..

v

25 26 32 38 45 47 54 57

vi

CONTENTS

Brain Gene Expression: Genomics and Genetics Elissa J. Chesler and Robert W. Williams I. II. III. IV. V. VI. VII. VIII.

Introduction . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Transcript Abundance as a Complex Trait . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Genetic Dissection of Transcription Regulation .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . How Many QTLs Are There? Statistical Control in Two Dimensions. .. . . . Probe Versus Probe Set–Level Phenotypes. . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Transcription Regulatory Networks . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Functional Correlates of Brain Transcriptional Activity. . . . . . . . . . . . . . . . . .. . . . Conclusions: Building a Model of Brain Function from Multipoint, Multi-tissue Gene Expression, and Phenomic Observation. . . . . . . . . . . . . .. . . . References. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . .

60 62 66 77 79 83 85 90 92

DNA Microarrays and Animal Models of Learning and Memory Sebastiano Cavallaro I. II. III. IV.

Introduction . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . DNA Microarray Technology . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Use of DNA Microarrays for Studying Learning and Memory . . . . . . . . . .. . . . Conclusion . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . References. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . .

97 98 104 126 126

Microarray Analysis of Human Nervous System Gene Expression in Neurological Disease Steven A. Greenberg I. II. III. IV. V.

Introduction . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Disease Pathophysiology . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Microarray-Based Disease Classification. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . Human Neurological Diseases Studied by Microarray Technology . . . .. . . . Conclusion . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . References. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . .

135 136 137 141 145 146

DNA Microarray Analysis of Postmortem Brain Tissue K A´ roly Mirnics, Pat Levitt, and David A. Lewis I. Introduction . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . II. Challenges in Studies of the Postmortem Human Brain . . . . . . . . . . . . . . . .. . . . III. Microarray Analysis of Human Brain Disorders . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . .

153 154 160

CONTENTS

vii

IV. Where Do We Go from Here?. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. References . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . ..

171 173

Index........................................................................................ Contents of Recent Volumes ....................................................

183 191

This Page Intentionally Left Blank

CONTRIBUTORS

Numbers in parentheses indicate the pages on which the authors’ contributions begin.

B. M. Bolstad (25), Division of Biostatistics, University of California, Berkeley, California 94720-3860 Sebastiano Cavallaro (97), Institute of Neurological Sciences, Italian National Research Council, 95123 Catania, Italy Elissa J. Chesler (59), Department of Anatomy and Neurobiology, Center for Genomics and Bioinformatics, University of Tennessee Health Science Center, Memphis, Tennessee 38163 F. Collin (25), Department of Statistics, University of California, Berkeley, California 94720 Daniel H. Geschwind (1), Department of Neurology, Program in Neurogenetics, University of California, Los Angeles, California 90095 Steven A. Greenberg (135), Department of Neurology, Division of Neuromuscular Disease, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts 02115 R. A. Irizarry (25), Department of Biostatistics, John Hopkins University, Baltimore, Maryland 21205 Stanislav L. Karsten (1), Department of Neurology, Program in Neurogenetics Lili C. Kudo (1), Neuroscience IDP David GeVen School of Medicine, University of California, Los Angeles, California 90095 Pat Levitt (153), Department of Psychiatry, School of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania 15261; and John F. Kennedy Center for Human Development and Department of Pharmacology, Vanderbilt University, Nashville, Tennessee David A. Lewis (153), Department of Psychiatry and Department of Neuroscience, School of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania 15261 Ka´roly Mirnics (153), Department of Psychiatry and Department of Neurobiology, School of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania 15261 K. M. Simpson (25), Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute, Melbourne 3050, Australia

ix

x

CONTRIBUTORS

T. P. Speed (25), Department of Statistics, University of California, Berkeley, California 94720; and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute, Melbourne 3050, Australia Robert W. Williams (59), Department of Anatomy and Neurobiology, Center for Genomics and Bioinformatics, University of Tennessee Health Science Center, Memphis, Tennessee 38163

INTRODUCTION

The advent of whole genome sequence information has ushered in a new generation of biology. No longer do scientists have to conduct complex searches to discover the structure of genes and proteins associated with molecular traits or human disease. The structures of the genes are now available – warehoused and awaiting annotation for their role in biology and disease. The new challenge for biology is to discover how to unravel the complexity of the function of the genome, now that the DNA structure is known. This daunting task requires progress on a genomic scale in several diVerent areas. Functional genomics studies are documenting expression patterns for messenger RNA molecules (mRNA) transcribed from the genome. High-throughput proteomics eVorts are cataloguing the structure of proteins and their functional modifications by post-translational processes. Finally, studies on signal transduction are starting to unravel the complex interactions between proteins or their metabolic products, which ultimately produce phenotypic expression of the genome at the cellular and organism level. The steps mentioned above for unraveling the function of the genome are largely in their infancy. Functional genomics is the area making most rapid progress, at least in terms of sheer numbers of data points. This largely refers to studies on mRNA abundance being conducted on a genome scale. Such studies were nearly beyond our wildest imagination a decade ago. The advent of DNA microarrays has provided the technical ability to simultaneously derive information on the abundance of every mRNA transcribed from the genome. This candy store of information on the workings of the genome has produced over 9000 publications in less than ten years. However, large-scale data derivation does not directly correspond to large-scale increases in knowledge. The complexities of obtaining quality data from microarray studies and translating this information into testable biological hypotheses have proven most diYcult. Deriving functional knowledge from microarray data is perhaps nowhere more complex than with studies on brain gene expression. The brain is thought to express a greater percentage of the genome than most other organs and the expression patterns can be distributed over a variety of cell types and intermingled brain functional domains. Furthermore, the enormous functional diversity of neurons and networks of neurons adds additional biological variance to the regulation of brain gene expression. These diYculties perhaps contribute to the fact that less than 10% of all microarray studies published to date involve the central nervous system (using ‘‘brain’’ and ‘‘microarray’’ as keywords). Advances in microarray technology (allowing smaller sample sizes), analysis xi

xii

INTRODUCTION

methods (reducing/identifying biological and technical variance) and bioinformatics (identifying gene functions and relations between genes) promise to improve the utility of performing microarray studies on the nervous system. This volume strives to summarize important aspects of microarray studies in neurobiology. We will give an overview of important basic concepts in microarray studies, portray important new advances combining genetics and microarrays for study of brain gene expression and provide several examples for application of microarray studies to complex questions of brain function and disease. An introduction to microarray platforms, applications in neurobiology and fundamentals of analysis are provided by Geschwind and colleagues. This chapter also introduces the multi-level nature of microarray analysis, with a critical dependence on a growing number of bioinformatics resources. Anyone conducting microarray studies without a strong working knowledge of bioinformatics resources for analysis of expression patterns will simply further complicate the nature of the problem. Since experimental design and low-level analysis of microarray data is crucial to all higher-order analysis algorithms, we have included a rigorous discussion of these issues. Terry Speed and co-workers discuss recent concepts in optimal design of microarray experiments and provide an in-depth discussion of newer methods for normalization, quality assessment and summarization of microarray data. Their discussion also touches upon the vital topic of power analysis in microarray studies. This is an important area for consideration since microarray studies are generally considered to be terribly ‘‘under powered’’ by traditional statistical standards. A discussion of statistical techniques for identifying lists of significant changes, such as permutation methods, and higher-level analyses such as multivariate studies are not provided since these have been covered extensively elsewhere. Identifying clusters of genes with correlated expression patterns is a traditional first step toward identifying possible functional patterns within microarray data. However, such clustering techniques provide only a limited insight into functional associations between individual gene expression profiles since there are generally only a small number of diVerent experimental conditions included in the analysis. Superimposing other types of biological information upon the expression patterns is crucial to providing improved resolution of functionally important groups of genes. Chesler and Williams provide a thorough discussion of a very recent advance in studying brain gene expression, namely, the combining of genetics and genomics. This so-called genetical genomics approach allows the study of gene expression patterns in relationship to genetic variance across the entire genome. This involves treating gene expression as another quantitative trait for the identification of expression quantitative trait loci (eQTLs). This exciting application of microarray studies provides additional means for identifying relationships between expression patterns of individual genes. Furthermore,

INTRODUCTION

xiii

by performing such studies on a well-characterized source of renewable genetic material, such as recombinant inbred (RI) mouse lines, these authors show how genetics, expression patterns and complex phenotypes such as behaviors can be superimposed. Further development of this area is likely to provide powerful tools for analysis of brain microarray data. The last three chapters of this volume involve applying microarray studies to complex brain functions or disease. Studies on learning and memory in animal models have provided a wealth of candidate molecules for the molecular substrates of learning and memory. However, studying single genes in isolation has not provided a thorough understanding of the processes involved in memory. D’Agata and Cavallaro describe their work using DNA microarrays to study learning and memory in animal models. They provide a detailed example of how microarray studies and multivariate analyses such as clustering, together with powerful animal models and experimental design, can start to provide a complex understanding of brain mechanisms in learning and memory. Greenberg broadens this discussion by providing an overview of application of brain microarray studies for a variety of neurological diseases and discusses application of such work to classification of clinical samples. Using microarray data to provide novel classification schemes is already finding widespread application in cancer biology. The application of similar strategies to subdivide supposedly homogeneous neurological disorders such as multiple sclerosis or Alzheimer’s disease may provide new insights with both diagnostic and therapeutic implications. In the final chapter, Mirnics et al. provide a detailed description of microarray studies performed on postmortem human brain tissue. This very important avenue of work serves as a reference point for the various animal model data derived for certain diseases. The diYculties in obtaining quality microarray data, issues in interpretation of expression diVerences, and impact of co-variables such as treatment history are discussed. Importantly, Mirnics and colleagues provide examples of how comparing similar microarray datasets across multiple laboratories can focus attention on common and unique genes in the various data lists. Such early eVorts at meta-analysis of microarray data are particularly crucial for studies on clinical material where confounds such as genotype, clinical history, sample preparation variance, and endophenotypes may obscure important results in a single dataset. Taken together, the chapters of this volume provide a broad but detailed and up-to-date description of the methods and application of microarray studies in neurobiology. The volume is not meant to be exhaustive in its treatment of the subject since the scope of the material is expanding at a phenomenal rate. Rather, the authors provide fundamental information and an overall framework for performing functional genomics studies on the nervous system. We hope that this discussion might lead others to future innovative functional genomics studies in neurobiology. This is a unique time for biomedical research, particularly in

xiv

INTRODUCTION

neuroscience, where the complexity of biology and disease can finally be studied with tools of near equal complexity. The burden is now upon the investigator to exploit these opportunities. Michael F. Miles

MICROARRAY PLATFORMS: INTRODUCTION AND APPLICATION TO NEUROBIOLOGY

Stanislav L. Karsten,* Lili C. Kudo,y and Daniel H. Geschwind* *

Department of Neurology, Program in Neurogenetics, and y Neuroscience IDP David Geffen School of Medicine, University of California Los Angeles, California 90095

I. Introduction II. Experimental Flow A. Replication and Confirmation III. Microarray Design IV. Slides and Probes V. Sample Preparation and Labeling VI. Data Analysis VII. Data Interpretation VIII. Confirmation of Microarray Results IX. Conclusion X. Online Resources A. General Microarray Sites B. Data Collection, Annotation, and Interpretation Tools C. Data Analysis Tools D. Metabolic and Regulatory Pathway Databases References

I. Introduction

Over the last decade, huge advances in genome sequencing promoted the development of various techniques aimed at studying gene expression on a genome-wide level. With the introduction of complementary DNA (cDNA) and oligonucleotide microarrays in 1995 (Lockhart et al., 1996; Schena et al., 1995), high-throughput simultaneous monitoring of gene expression became possible. DNA microarrays allow the simultaneous study of the expression patterns of thousands of genes in the same tissue or cell (Karsten and Geschwind, 2002). cDNA and oligonucleotide arrays have been successfully used in studying the nervous system in healthy (Geschwind et al., 2001; Karsten et al., 2003; Sandberg et al., 2000) and diseased tissues (Brown et al., 2001; Mirnics et al., 2000, 2001; Pomeroy et al., 2002; Tang et al., 2001; Whitney et al., 1999). Attempts to identify

INTERNATIONAL REVIEW OF NEUROBIOLOGY, VOL. 60

1

Copyright 2004, Elsevier Inc. All rights reserved. 0074-7742/04 $35.00

2

KARSTEN et al.

disease-specific genes in human tissues have been made for such neurological conditions as multiple sclerosis (Mycko et al., 2003), Alzheimer’s disease (Colangelo et al., 2002; Ginsberg et al., 2000; Marvanova et al., 2003), psychiatric disorders (Bunney et al., 2003; Middleton et al., 2002; Mirnics et al., 2000; Pongrac et al., 2002), and epilepsy (Elliott et al., 2003). Despite these preliminary successes and the power of this technology, neurobiologists have been hesitant to readily adopt this method, especially relative to other fields, such as oncology. Some of this hesitance may be due to a general conservative attitude toward new methods among neurobiologists or the belief that the nervous system is somehow diVerent at a molecular level from other tissues. The non–hypothesis-driven, more discovery-based nature of genetic methods, typified by microarrays, also contributes to this relative bias against microarrays. One real stumbling block facing microarray users in neuroscience is that nervous system tissue is more complex than most other tissues, especially cancer cell lines. This cellular diversity and the complexity of the nervous system do pose unique obstacles for neuroscientists in applying microarray technology to their work (Evans et al., 2002; Geschwind, 2000). This issue must be carefully considered in the experimental design of any microarray project. Most cDNA or oligonucleotide array systems are able to reliably detect diVerences in species with relative abundance as low as 1 per 100,000 in a sample. Such low abundance messages in a single cell may very well be undetected in a tissue sample containing various cell types (Geschwind, 2000). For isolation of samples from specific populations of neurons or glia, techniques such as microdissection by laser capture (Kamme et al., 2003; Luo et al., 1999), microscopic dissection (Zirlinger et al., 2001), and single cell polymerase chain reaction (PCR) approaches (Ginsberg and Che, 2002; Kamme et al., 2003) need to be considered. In addition to the analysis of gene expression, microarrays provide a highthroughput genotyping platform through the use of hybridization of genomic DNA to matching and mismatching oligonucleotides present on the array surface (Hacia et al., 1999; Hirschhorn et al., 2000; Lindblad-Toh et al., 2000). Although few studies have been published using single nucleotide polymorphism (SNP) genotyping arrays, they are likely to replace microsatellite-based genotyping. A complete genome scan using 10,000 SNPs, with average heterozygosity between 0.3 and 0.4, can be completed for less than 10 cents a genotype. Because the information attained from a typical microsatellite marker is worth two to three SNPs, this SNP scan is equivalent to 3000–4000 microsatellites, or 5–10 times the current density of the typical whole genome scan. All SNPs are assayed in a single hybridization step, making this a very eYcient method of genotyping; 10,000 SNP genotypes in 20 aVected individuals can be assayed in 2 days using the AVymetrix genotyping chip (www.aVymetrix.com/products/arrays/specific/ 10k.aVx). An increase in oligonucleotide density by 10–30-fold to 300,000 would

MICROARRAY PLATFORMS

3

allow the performance of a whole genome association study with reasonable power in a single hybridization step (Cutler et al., 2001; Fan et al., 2000). The same principle of microarray-based detection of DNA variations was extended to investigate the changes in DNA methylation using a specifically designed methylation-specific oligonucleotide microarray (Gitan et al., 2002). DNA microarray platforms were also successfully used for comparative genomic hybridization (Cheung et al., 1998), identification of splice variants (Clark et al., 2002; Hu et al., 2001), and promoter regulatory sites (Pilpel et al., 2001; Ren et al., 2000; Roth et al., 1998; Spellman et al., 1998). This chapter focuses on DNA microarray-based large-scale gene expression studies in the nervous system concentrating on the choice of microarray platform, data analysis, and interpretation. We use the term platform to describe the combination of sample preparation (RNA isolation and labeling), hybridization, type of microarray used, imaging, data extraction, and data analysis. We describe the use of various DNA microarray platforms in neuroscience and suggest possible avenues for microarray data interpretation, as well as follow-up experimentation. Because the problems unique to the interpretation of microarray data sets are just beginning to be addressed, we also discuss some of the critical issues that are frequently encountered in the course of microarray experiments.

II. Experimental Flow

Figure 1 outlines the typical procedures involved in a microarray experiment. Messenger or total RNA is extracted from the sample, such as tissue or cell cultures, to serve as a template for either cRNA or cDNA targets. Targets are usually labeled with fluorophores (e.g., Cy3-dCTP or Cy5-dCTP) in conjunction with or after cDNA synthesis. In the case of comparative hybridization, two samples are labeled with diVerent fluorophores that emit diVerent wavelengths that can be independently quantified—a two-dye system. Therefore, the signals from the two samples can be compared. Both samples are hybridized onto the same slide. In this design, diVerences in gene expression can be expressed as a ratio rather than as an absolute value (Fig. 1). Such a system is routinely used with most custom and commercial cDNA microarrays, as well as with some commercial oligonucleotide platforms (e.g., Agilent). In a one-dye experiment (Cy3 or Cy5), all samples are labeled with one dye and only one sample is hybridized onto the slide. The one-dye approach depends on stringent quality control that limits the array-to-array variability. The signal detected for a probe upon laser excitation is proportional to the amount of target bound to it, thereby allowing for quantitative analysis of the abundance of that target in a given sample.

4

KARSTEN et al.

Fig. 1. Flowchart of a typical microarray experiment. Messenger or total RNA samples to be compared are extracted from tissue or cells (e.g., normal tissue vs disease tissue) and cRNA or cDNA target is synthesized from each RNA template. Each sample is labeled with a diVerent fluorophore, typically Cy3-dCTP or Cy5-dCTP, either during or after cDNA synthesis. Because the fluorophores emit light at diVerent wavelengths, the signal from each sample can be independently quantified, and the two samples can be compared. The amount of signal detected after laser excitation is proportional to the amount of target bound to each probe, permitting quantitative analysis of the relative abundance of each target transcript in a tissue or cell line. Radioactively labeled cDNA samples have also been successfully applied in the microarray format (Whitney and Becker, 2001). DiVerentially


5

The samples (slides) are compared between each other and the diVerences in the expression are identified. Examples of such platforms are CodeLink Expression Bioarrays (Amersham Biosciences) and AVymetrix GeneChips (AVymetrix, California). Though not expressly designed for that purpose, Agilent arrays can also be used for one-sample, one-dye-per-slide experiments because of the high level of quality control in these and most commercial arrays. The entire microarray experiment from labeling through hybridization takes an average of 2 days without the use of amplification techniques, which can add a day (e.g., one round of T7 amplification, DNA dendrimers) or several days (e.g., two rounds of T7 amplification, TSA) depending on the method (Karsten and Geschwind, 2002). Most protocols (see Online Resources: The Institute for Genomic Research, Stanford University, De Risi Laboratory, University of California at San Francisco.) use incorporation during the reverse transcriptase (RT) reaction rather than Klenow. We use Klenow labeling for direct labeling and find high levels of incorporation and reproducibility. Methods of amplification add more steps to the protocol, thus more time spent. RNA isolation and quality analysis are carried out separately, because they take several hours. In most labeling protocols, target labeling to initiation of the hybridization spans 10–12 hours or 2 divided days, depending on the amplification technique used.

A. Replication and Confirmation Because of the large scale of a microarray experiment, there are a number of procedural considerations in almost every step of the experimental flow. One of the restricting factors in applying microarray technology in a laboratory is the cost of replicating experiments. Unfortunately, to produce reliable results, replication of experiments is currently the only way to deal with biological and experimental noise (Karsten and Geschwind, 2002; Lee et al., 2000; Sabatti et al., 2002). The number of replicates for an experiment will vary depending on the amount of experimental noise and lenience for false-positive results; however, replicates introduce greater reliability to the expression data and should not be neglected. When using most labeling techniques and high-quality arrays, one should run three independent replicates, each duplicated with switched dyes, to obtain a low enough number of false-positive signals. Noise measures can be empirically derived and screening thresholds set appropriately, based on the technical and biological noise in a particular system (Lee et al., 2000; Sabatti expressed genes are identified using one of the statistical programs designed for microarray data analysis (GeneSight, GeneSpring, CyberT, etc.). Data interpretation involves identification of specific metabolic and signaling pathways altered in the experimental conditions studied. Various data mining tools are available (see the section Online Resources). (See Color Insert.)

6

KARSTEN et al.

et al., 2002; Sandberg et al., 2000). Increasing the number of independent replicates will permit the detection of smaller changes (e.g., 1.4-fold) with higher confidence. Statistical methods that estimate or model variance to increase statistical power are very useful when small numbers of replicates are available compared with the number of measurements being made (Baldi and Long, 2001; Nadon and Shoemaker, 2002; http://visitor.ics.uci.edu/genex/cybert/) and several excellent reviews of statistical methods are available (Nadon and Shoemaker, 2002; Quackenbush, 2002; Yang and Speed, 2002; Yang et al., 2002). In addition, tools are available online, as exemplified by the website maintained by Terry Speed’s group (www.stat.berkeley.edu/users/terry/zarray/Html/index.html). One question often posed is, do we need to confirm our results using an independent technique such as RT-PCR or Northern blotting? Even with good statistical methods, confirmation of some small cross section of the results using an alternative method is necessary. Whether to pool samples is another question that is often raised. We view pooling as an eVective way to diminish the eVects of individual variability within biological samples. But pooling does not obviate the need for independent replication, and the power of this approach depends on the integrity of the samples being pooled. Pooling RNA samples can also reduce the cost of microarray experiments instead of testing several individual replicates. But again, one sample with significant deviation from the rest of the pool may spoil an experiment consisting of a comparison of two pools.

III. Microarray Design

Microarray experiments start with array design and synthesis. A typical DNA microarray consists of tens of thousands of elements, called probes, densely deposited onto a solid surface such as glass (or on a membrane) in a grid format. The probes comprise either cDNA sequences of varying lengths for cDNA microarrays (Schena et al., 1995) or short synthetic oligonucleotides of up to 70 nucleotides for oligonucleotide microarrays (AVymetrix, Santa Clara, California; Barczak et al., 2003; Fodor et al., 1991; Lipshutz et al., 1999). Microarry platforms can, therefore, be divided into two major formats: oligonucleotide arrays and cDNA arrays (Brown and Botstein, 1999; Lockhart and Winzeler, 2000). Some arrays cover nearly the entire genome, containing 30,000–45,000 genes on one slide (AVymetrix; Incyte, Palo Alto, California; Hillier et al., 1996), whereas others oVer specific arrays with cDNA clones enriched in a particular tissue-specific library (Chiang et al., 2001), bioinformatically or logically derived neuroarray (Becker, 2001; Geschwind and Nelson, unpublished data; www.brainmapping.com), and arrays of clones from a subtractive hybridization (Geschwind et al., 2001; Welford et al., 1998; Yang et al., 1999). Because one of


7

the main goals of a microarray study is generation of hypotheses using information about specific biological processes and pathways altered in the experimental conditions studied, usually, microarrays consist of 9,000–15,000 representative genes (e.g., Genome Systems, Riken and National Institute of Aging (NIA) mouse cDNA clone sets; http://lgsun.grc.nia.nih.gov/) and contain, at least partially, most signaling and metabolic pathways. It is usually not necessary to have the whole genome, as long as cross sections of most pathways are present. Generation of hypotheses based on altered pathways allows for testing genes not present on the array, thereby providing more complete coverage of a particular transcriptome. Such arrays may be appropriate for general screening of genes; however, those who study specific systems, such as a neuroscientist, may require arrays containing genes yet to be discovered or not present on available arrays. This calls for the use of subtracted or specific libraries, or Serial Analysis of Gene Expression (SAGE) (Luo and Geschwind, 2001; Velculescu et al., 1995). Even large whole genome arrays are lacking genes that many neuroscientists want to probe, further supporting the development of large-scale neuroarrays.

IV. Slides and Probes

If one chooses a commercial platform, there is no need to consider slide choice. However, if one prints arrays in-house, choosing appropriate array slides is essential. A typical cDNA microarray consists of densely deposited PCR-amplified cDNA clones or expressed sequence tags (ESTs) onto a solid surface, such as specially coated silicon or glass slides. These amplified elements (probes) are noncovalently deposited on the slide surfaces that are optimized for DNA attachment and minimal background noise. Slide treatment must be performed with care to avoid the subsequent loss of the target from the glass surface during hybridization, resulting in loss of specific hybridization signal. Glass surfaces are typically treated with poly-llysine or amino-propyl-ethoxy-silane (DeRisi et al., 1997), which both provide low background and adequate attachment of cDNA for most purposes. Other new surfaces that may perform better and allow for covalent attachment are also available. However, good-quality silane- or lysine-coated slides are of reasonable cost and perform well. Commercial slides may not always perform better than homemade slides; however, reliable slide coating is a crucial factor that should not be overlooked. Depending on the cost for quality control to produce in-house slides, purchasing commercial slides may be more desirable. Commercial vendors oVer a variety of quality slides. Some of the major vendor web sites are listed in Karsten and Geschwind (2002) and Geschwind and Gregg (2002). In preparing custom DNA arrays, one must consider positioning of probe elements. To control for position eVects, duplicate or control spots are

8

KARSTEN et al.

placed in various array quadrants. A number of commercial oligonucleotide arrays have many diVerent strategically located controls for qualities, such as RNA synthesis and labeling (Lockhart et al., 1996). This is a significant design advantage. To construct custom cDNA arrays, an array facility usually acquires verified cDNA clones from Unigene clusters or IMAGE consortium clones from commercial sources, such as Genome Systems (St. Louis) or Invitrogen (Carlsbad, California), or noncommercial resources, such as 7,400 and 15,000 mouse cDNA clone sets from the NIA (http://lgsun.grc.nia.nih.gov) and FANTOM clones from RIKEN (http://fantom2.gsc.riken.go.jp/; the RIKEN Genome Exploration Research Group Phase II Team and the FANTOM Consortium, 2001; the FANTOM Consortium and the RIKEN Genome Exploration Research Group Phase I & II Team, 2002). The cDNA clones are available either as bacterial glycerol stocks or PCR-amplified clones of full- or partial-length cDNAs that are ready to be arrayed. After amplification and purification, cDNAs are resuspended in arraying buVer and loaded into a high-throughput DNA arraying robot and amplified probes are printed onto preferred type of coated slides. Various printing methods exist, along with arraying robots and a range of buVers and slides (Gregg and Baldwin, 2001; Schena et al., 1998). The lengths of the cDNA target sequences may not allow for the discrimination of sequences with high homology, such as those belonging to channel gene families, or splice variants. When direct comparison of two species is hindered because of similar sequences, secondary structures, or repetitive elements, specially designed oligonucleotide arrays may be more suitable in detecting particular isoforms or family members. Oligonucleotide microarrays can be manufactured either using in situ synthesis by photolithography (e.g., AVymetrix; Fodor et al., 1993) or deposition of already synthesized oligonucleotides (e.g., inkjet technology, Agilent; Hughes et al., 2001). Some of the strategies for probe selection are common to all oligonucleotide arrays. Melting temperature of an oligonucleotide probe is calculated based on experimentally derived computer models calculating hybridization behavior of target sequences in complex mixtures under particular conditions (pH, salt concentration, temperature). The GeneChip (AVymetrix) arrays are manufactured using the combination of photolithography and combinatorial chemistry (Lipshutz et al., 1999). This allows the synthesis of hundreds of thousands of diVerent oligonucleotides on the same surface at an extremely high density. Because the resulting surface area is very small, it enables researchers to use small sample volumes, thereby reducing the amounts of starting RNA, and the samples may be reused. AVymetrix oVers a range of preprinted arrays covering up to 54,000 genes in a set of five slides (Table I). Each transcript is represented by 11–16 short 25-mer oligonucleotides selected according to their specificity to the desired transcript and low cross-hybridization with similar but unrelated sequences. Because probes are

9


TABLE I Commercial Microarray Platforms Vendor/ genes per slide

AVymetrixa

cDNA microarrays Human

NA

Mouse

NA

Rat

NA

Oligonucleotide microarrays Human Human Genome Focus Array: 8500

Agilentb

Human 1 cDNA Microarray: 13,675 Human 2 cDNA Microarray: 14,355 Mouse cDNA Microarray: 9600 Rat cDNA Microarray: 14,815 Agilent Human Genome Set (2): 36,000

Human Genome U133 Set (2): 33,000 Human Genome U95 Set (5): 54,000

Mouse

Rat

GeneChip Mouse Expression Set 430 (2): 34,000 Murine Genome U74v2 Set (3): 36,000

GeneChip Rat Expression Set 230 (2): 29,000 Rat Genome U34 Set (3): 24,000 Rat Neurobiology U34 Array: 1200 Rat Toxicology U34 Array: 850

Mouse Oligo Microarray: 20,000 Mouse (Development) Oligo Microarray: 20,000

Rat Oligo Microarray: 20,000

Amersham biosciences CodeLinkc

BD biosciences clontechd

NA

NA

NA

NA

NA

NA

CodeLink Atlas Plastic Uniset Human 12 K Human 1 Microarray: Bioarray: 12,000 10,000 Atlas Plastic Human 8 K Microarray: 8300 Atlas Glass Human 1.0 Microarray: 1081 Atlas Glass Human 3.8 I/II Microarray: 3757 Atlas Glass Human 7.6 Microarray (2): 7532 Uniset Rat I Atlas Plastic Mouse Bioarray: 5 K Microarray: 10,000 5002 oligos Atlas Glass Mouse 1.0 Microarray: 1081

Atlas Glass Mouse 3.8 I Microarray: 3748 Mouse I Atlas Plastic Rat 4 K Bioarray: Microarray: 10,000 3897 oligos Atlas Glass Rat 1.0 Microarray: 1081 Atlas Glass Rat 3.8 I Microarray: 3757

(Continued )

10

KARSTEN et al.

designed for significantly unique regions of genes even among gene family members, GeneChip arrays can distinguish transcripts that are up to 90% identical. In addition, some probes are designed to distinguish multiple splice or polyadenylation variants. An industrial noncontact inkjet printing process is used for the manufacturing of Agilent microarrays. Both oligonucleotide and cDNA are deposited onto specially treated glass slides. The reproducible deposition of oligonucleotide or cDNA molecules is achieved without actual contact with a glass surface, thereby reducing the risk of potential anomalies caused by the physical contact of slide and printer surfaces (www.chem.agilent.com). The technology requires only picoliters of DNA per spot. The 60-mer oligonucleotides are synthesized using standard phosphoramidite chemistry. Microarrays covering up to 20,000 genes per slide have been in use for several years (Table I). More recently the Human Whole Genome Bioarray representing 57,000 transcripts from CodeLink and the Whole Human Genome Oligo Microarray representing 41,000 transcripts on a single slide from Agilent became available. For further updates on commercial microarrays, one should visit the web sites from Table 1. CodeLink Activated Slides (Amersham Biosciences) are specially treated to covalently immobilize amine-modified DNA. The combination of cross-linked polymer and end-point attachment allows the oligonucleotides to be more accessible to the labeled targets hybridized onto the slides (www1.amershambiosciences. com). Nucleotide probe sets are derived from UniGene, Incyte LifeSeq database (for human), NCBI (for rat), and fourth quarter 2001 GenBank RefSeq (for mouse). A CodeLink Bioarray contains probes for approximately 10,000 genes, each of which has one specific prescreened 30-mer probe, functionally validated, along with housekeeping and control probes. As is the case for the other cDNA and oligonucleotide arrays described here, publications support the sensitivity, reproducibility, and validity of the data obtained with this platform (El Atifi et al., 2002; Hughes et al., 2001; Ramakrishnan et al., 2002; Taniguchi et al., 2001; Yuen et al., 2002).

Note: The specific microarray set and number of genes per slide or set of slides (indicated in brackets) for rodents and humans provided by each vendor is presented. AVymetrix also oVers D. melanogaster and C. elegans genome arrays that represent >13,500 transcripts and >22,150 transcripts, respectively. The number of genes represented for each slide/set is indicated after the microarray title. This information was obtained from the vendor web sites in September 2003: a www.aVymetrix.com; b www.chem.agilent.com; c www1.amershambiosciences.com; d www.bdbiosciences.com.


11

V. Sample Preparation and Labeling

The purity and quality of the starting RNA has a significant eVect on the results of microarray experiments, so it is essential that all steps of RNA isolation be carried out with maximum care and speed. RNase-free reagents and materials should be used at all times (Karsten and Geschwind, 2002). Total RNA is isolated from cultured cells and frozen tissues by acid phenol extraction (e.g., Trizol LS or Trizol, GIBCO/BRL) or with bead and column-based methods (e.g., RNeasy Total RNA System and Oligotex mRNA Purification System, QIAGEN; S.N.A.P. Total RNA Isolation Kit, Invitrogen; PolyATtract Systems for mRNA Purification). Both total RNA and polyA RNA give reliable array results and most investigators prefer to use total RNA (Karsten et al., 2002). RNA quality must meet high standards so the OD 260/280 is between 1.8 and 2.1 and the total RNA on a denaturing agarose gel is characterized by two clear bands at about 4.5 Kb and about 1.9 Kb for mammalian total RNA, representing the 28S and 18S ribosomal subunits respectively, with a 2:1 or greater (28S:18S) ratio. The major limiting factor in gene expression experiments of human disease is the availability of quality disease specimens. Formalin- and ethanol-fixed as well as paraYn-embedded tissues may be used as sources of RNA; however, archived frozen tissue seems most suited for microarray experiments (Karsten et al., 2002). Ethanol-fixed tissues provide marginal RNA, and heavily formalin-fixed tissues give weak overall signals and less reproducibility with oligodT priming for cDNA synthesis (Karsten et al., 2002). Using random priming rather than oligodT primers for cDNA synthesis and rapid light fixation is likely to improve studies using archived or fixed specimens (Godfrey et al., 2000; Specht et al., 2001). The signal intensity from hybridization depends on the target concentration, the amount of immobilized probe molecules, and the method of labeling. Ideally, cDNA targets generated by direct labeling upon hybridization with probe immobilized on the microarray surface should give signal intensity proportional to the abundance of the specific target cDNA molecule. Signal amplification methods yield signal intensity with nonlinear relation to the transcript abundance and require special attention when concluding the amount of transcript in a sample. However, comparisons can still be made between similarly prepared samples, even those made from PCR amplicons (Geschwind et al., 2001; Tietjen et al., 2003). Amplification techniques, such as PCR amplification, amino-allyl labeling, DNA dendrimers, tyramide signal amplification (TSA), and T7-directed in vitro transcription and amplification (Eberwine et al., 1992), are reliable in generating labeled products from small quantities of RNA on a consistent basis, in some cases from a single cell or a few laser-captured cells (Tietjen et al., 2003). In general, amplification techniques use two approaches to increase signal to background ratios: signal amplification and sample amplification (Karsten and Geschwind, 2002; Stears et al., 2000; www.tigr.org/tdb/microarray/).

12

KARSTEN et al.

Many new methods of labeling and detection are available, for example, a system using spherical gold and silver Resonance Light Scattering (RLS) particles to enhance the signal, which seems to be very sensitive (Genicon Sciences, California). Direct labeling of RNA uses platinum-linked cyanine dyes to directly chemically label mRNA from as little as 2 mg of total RNA (PerkinElmer; Gupta et al., 2003). This method has very high precision, low error, and no labeling bias. It allows using low amounts of starting material, yet avoiding enzyme-introduced labeling and sequence bias. Although these newer techniques oVer potential advantages, the most common reliable methods used now include T7-amplification followed by dye incorporation, direct labeling, and amino-allyl coupled labeling.

VI. Data Analysis

Before recommending the strategies of microarray data analysis and interpretation, one should be aware that there is no standard or consensus on the best way to represent or analyze microarray data. This is a rapidly evolving field and our suggestions are likely to change. Nevertheless, we propose several general data analysis and interpretation steps found prerequisite for most microarray experiments. Additionally, although statistical analysis varies, consensus has been reached about how microarray data should be presented, shared, and annotated: The Minimal Information About a Microarray Experiment (Brazma et al., 2001; MIAME, www.mged.org/you). Let us discuss two basic kinds of experiments. The first is designed to perform a rapid screen of the genome to identify a group of genes for detailed follow-up experiments (Geschwind et al., 2001; Zirlinger et al., 2001). Statistical validation or extensive replication may not be necessary in this case. In the second, more typical experiment, one is attempting to use microarrays to identify a relatively certain list of genes characterizing a particular system. This requires independent replicates and appropriate statistical testing. Microarray-based expression analysis implies a complex experimental procedure with a large number of parameters aVecting the final result. Statistical analysis must consider that nearly every parameter represents an individual source of variance, whose relative contribution will vary between platforms and laboratories (Sabatti et al., 2002; Yue et al., 2001). The two major sources of variability are methodological and biological variance. Both sources necessitate independent replication to generate reliable results (Geschwind, 2000; Lee et al., 2000; Sabatti et al., 2002). The group directed by Terry Speed and colleagues has done much to provide sound guidance to the field for statistical analysis of microarray data (Irizarry et al., 2003; Smyth et al., 2003; Yang and Speed, 2002).


13

Significant sources of methodological variance, such as dye eVects, position eVects, and gene eVects, can be controlled. Labeling two aliquots of the same sample with diVerent fluors and hybridizing both samples on the same array— homotypical hybridization—can test incorporation rates of Cy3 and Cy5 into cDNA during enzymatic reaction. Ideally the correlation slope should be equal to 1.0, but experimentally, around 10% deviations might be observed. Generally, labeling and hybridization will yield correlations between 0.95 and 0.99 for the same sample that are independently labeled, if both fluors have equal incorporation rates (Karsten et al., 2002; Lee et al., 2000; Sabatti et al., 2002; Sandberg et al., 2000). Biological variability will likely cause independent replicates from diVerent samples to produce fewer correlated signals than with the co-hybridization of the exact same sample (Lee et al., 2000). The variability also depends on the methods used to normalize channels and correct for background (Kerr et al., 2000; Tseng et al., 2001). Signal intensity– based non linear normalization is often preferred over global linear normalization, especially in the case of signal amplification techniques (Sabatti et al., 2002; Smyth et al., 2003; Speed, 2002). Some local background correction methods can increase variance compared to quadrant-based or modeled background correction (Sabatti et al., 2002; Tseng et al., 2001). Various shareware (e.g., Bioconductor; www.bioconductor.org/) and commercial software that perform nonlinear normalization and adaptable background correction are available (e.g., GeneSight, BioDiscovery; GeneSpring, SiliconGenetics).

VII. Data Interpretation

Advances in cDNA microarray technologies have enabled the definition of global changes in gene expression across thousands of analyzed genes, but elucidation of functional relations among identified genes remains the most important and challenging task. Often, time-consuming and elaborate literature searches might be the only option to construct a putative pathway of regulated genes and their products’ interrelations. Nevertheless, several bioinformatic approaches are available, which significantly facilitate the data interpretation and further hypothesis building. Because this is a young field, novel approaches and online tools for data mining are constantly developed. Several general microarray web sites might provide important updates in the field (see the section Online Resources later in this chapter). To more clearly visualize the biological significance of the overall changes in gene expression, we should categorize identified genes into functional groups. If one does not use a relatively neutral classification scheme, a biased interpretation of the microarray data may result. Several automatic online tools for the identification of functional relationships among large sets of genes are available

14

KARSTEN et al.

and updated to provide practical, automated annotation (see the section Online Resources). The Gene Ontology Consortium oVers classification of genes based on three major groups: molecular function, cellular localization, and involvement in specific biological process (www.geneontology.org/). Involvement of genes in specific metabolic pathways can be evaluated using the KEGG database (www.genome.ad.jp/kegg/). The most altered groups can be determined by comparison of the percentage of genes altered in expression in a particular functional group to the overall percentage of change during the diVerentiation program, using shareware such as the Expression Analysis Systematic Explorer (EASE; http://apps1.niaid.nih.gov/david/upload.asp), MAPPFinder (Dahlquist et al., 2002; www.genmapp.org/mappfinder.html), or a local algorithm (Karsten et al., 2003). Functional groups that show highly significant alterations of expression indicate pathways most dramatically altered between the experimental conditions studied. PubGene developed at the University of Oslo (www.pubgene.uio.no) oVers an automatic approach of extracting information about desired genes from published literature. The program allows building a network of genes mentioned in the same abstracts. GenMapp (Doniger et al., 2003; Gene MicroArray Pathway Profiler; www.genmapp.org) is another developing new tool designed to visualize gene expression data on maps representing biological pathways and functional groupings of genes.

VIII. Confirmation of Microarray Results

The potential presence of false positives and false negatives in a microarray experiment requires validating a subset of the information (genes) identified by microarray analysis. Because additional microarray experiments result in significant cost increase for end users, alternative methods might be used for confirming relative expression levels of a gene product. The most widely accepted and reliable techniques used for validation of microarray data are quantitative RT-PCR, Northern blot analysis, and in situ hybridization. Quantitative and real-time RT-PCR are both especially suitable in validating a large number of gene expression, however, in situ hybridization experiments can also be performed to confirm the expression of more than a dozen genes within a short period (Geschwind et al., 2001; Karsten et al., 2003). Northern blots also oVer a way to confirm the expression of transcripts and have demonstrated the consistency and validity of cDNA microarray data (Taniguchi et al., 2001). Western blots can provide protein expression levels, which are not necessarily correlated with the transcript levels but could give further biological relevance to the data. Real-time quantitative RT-PCR allows monitoring the progress of a PCR in real time as each amplification cycle takes place. We recommend the SYBR


15

Green PCR detection system based on direct fluorescence detection of the PCR product using an intercalating dye (SYBR Green; Molecular Probes). The specific advantage of this approach is that it is only necessary to provide standard PCR reagents that are used in combination with a low-cost intercalating dye. The fluorescence response can be monitored by a variety of commercially available quantitative RT-PCR platforms, such as Stratagene’s Mx4000, ABI Prism 7900, 7700, 5700, and GeneAmp 7000, and the BioRad iCycler. The drawback of using SYBR Green is that if the PCR is not clean, nonspecific products will obscure the data. In this case, more specific probes for RT-PCR, such as ABI Prism Real Time Quantitative PCR instrument TaqMan (Applied Biosystems), can be used. Although a number of commercial microarray platforms claim that their DNA oligonucleotide probes give specific hybridizations for a particular gene, there can be a number of splice isoforms that may not be distinguished on the arrays. Northern blots will provide both quantitative expression data and information on the presence of alternative transcripts. Further, in situ hybridizations using antisense probes reveal temporal and spatial expression patterns for genes identified on the array at the cellular level. This may be especially important when whole tissue, rather than specific cell types is used to make RNA for the array experiment. Digoxygenin (DIG) or radio-labeled probes (mainly [35S]) when coupled with immunohistochemistry, can both provide the resolution to confirm the presence of particular mRNA in a particular region of the CNS, in a specific cell type, as well as the cellular localization. While radiolabeled probes hybridized onto tissue slides can be quantified, DIG-labeled probes can be used on slides and for whole-mount in situ hybridization. Because brain tissues consist of high density of heterogeneous cell types, it is a desirable way to confirm the microarray results.

IX. Conclusion

The field has changed rapidly over the last eight years but is settling down. Although each experiment must be carefully scrutinized by peer review and specific pitfalls remain, no longer should the general validity of the technology be broadly questioned. Commercial oligonucleotide platforms such as GeneChip/AVymetrix, Agilent, and Codelink/Amersham are becoming standards in the field. cDNA arrays, most of which are produced in universities in core facilities or individual laboratories, have also proven to be reliable and valid and have the advantage of low cost. Chip sets with 80% of the genome are available, but in some cases, genes of interest to neuroscience are often missing, supporting the need for focused neuroarray chips. Single or small cell pool PCR and other amplification techniques can be used reproducibly for microarray experiments and have already yielded valuable data. Analytical tools are becoming more standardized and statistical thresholds are

16

KARSTEN et al.

more accepted. Bioinformatic mining of the data and functional confirmation remains an important challenge and is one of the exciting areas of progress. Resources are rapidly changing, so here we provide some general array web sites that should keep you up to date. X. Online Resources

A. General Microarray Sites MGED: Microarray Gene Expression Database group www.mged.org/ TIGR: The Institute for Genomic Research www.tigr.org/tdb/microarray Stanford University cmgm.stanford.edu/pbrown/protocols/index.html De Risi Laboratory, University of California at San Francisco www.microarrays.org Y. F. Leung’s Functional Genomics http://ihome.cuhk.edu.hk/%7eb400559/array.html B. Data Collection, Annotation, and Interpretation Tools PubMed www.ncbi.nlm.nih.gov/pubmed DAVID, Database for Annotation, Visualization and Integrated Discovery http://appsl.niaid.nih.gov/david/upload.asp PubGene, University of Oslo www.pubgene.org/ SOURCE, Stanford University http://source.stanford.edu GenMapp, Gene Microarray Pathway Profiler www.genmapp.org/ 2HAPI, High-density Array Pattern Interpreter, version 2, UCSD http://array.sdsc.edu/

C. Data Analysis Tools ARRAY-VIEWER, The Institute for Genomic Research (TIGR) www.tigr.org/softlab


17

Cyber-T, Institute for Genomics and Bioinformatics, University of California http://visitor.ics.uci.edu/genex/cybert/ EMBL, European Bioinformatics Institute http://ep.ebi.ac.uk PaGE (Patterns from Gene Expression), University of Pennsylvania www.cbil.upenn.edu/page SCAN-ALYZE, Lawrence Berkeley National Lab www.microarrays.org/software.html Rosetta Resolver System www.rosettabio.com/products/resolver/default.htm ImaGene and GeneSight, BioDiscovery, Inc. www.biodiscovery.com GeneSpring: Silicon Genetics, Inc. www.sigenetics.com Spotfire DecisionSite for Functional Genomics: Spotfire, Inc. www.spotfire.com

D. Metabolic and Regulatory Pathway Databases KEGG: Kyoto Encyclopedia of Genes and Genomes www.genome.ad.jp/kegg/ CSNDB: Cell Signaling Networks Database describes signaling pathways of human cells. http://geo.nihs.go.jp/csndb/ TRANSPATH database describes the signal transduction from the ligand at the surface of a cell up to the transcription factor. http://transpath.gbf.de/ SPAD: The Signaling PAthway Database www.grt.kyushu-u.ac.jp/spad/ Wnt signaling pathway database www.stanford.edu/rnusse/wntwindow.html Boehringer Mannheim Biochemical Pathways www.expasy.org/cgi-bin/search-biochem-index EMP, Enzymes and Metabolic Pathways database http://emp.mcs.anl.gov PathDB, Biochemical Pathways www.ncgr.org/pathdb BioCarta www.biocarta.com

18

KARSTEN et al.

References

Baldi, P., and Long, A. D. (2001). A Bayesian framework for the analysis of microarray expression data: Regularized t test and statistical inferences of gene changes. Bioinformatics 17, 509–519. Barczak, A., Rodriguez, M. W., Hanspers, K., Koth, L. L., Tai, Y. C., Bolstad, B. M., Speed, T. P., and Erle, D. J. (2003). Spotted long oligonucleotide arrays for human gene expression analysis. Genome Res. 13, 1775–1785. Becker, K. G. (2001). The sharing of cDNA microarray data. Nat. Rev. Neurosci. 2, 438–440. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeckert, C., Aach, J., Ansorge, W., Ball, C. A., Causton, H. C., Gaasterland, T., Glenisson, P., Holstege, F. C., Kim, I. F., Markowitz, V., Matese, J. C., Parkinson, H., Robinson, A., Sarkans, U., SchulzeKremer, S., Stewart, J., Taylor, R., Vilo, J., and Vingron, M. (2001). Minimum information about a microarray experiment (MIAME)—toward standards for microarray data. Nat. Genet. 29, 365–371. Brown, P. O., and Botstein, D. (1999). Exploring the new world of the genome with DNA microarrays. Nat. Genet. 21, 33–37. Brown, V., Jin, P., Ceman, S., Darnell, J. C., O’Donnell, W. T., Tenenbaum, S. A., Jin, X., Feng, Y., Wilkinson, K. D., Keene, J. D., Darnell, R. B., and Warren, S. T. (2001). Microarray identification of FMRP-associated brain mRNAs and altered mRNA translational profiles in fragile X syndrome. Cell 107, 477–487. Bunney, W. E., Bunney, B. G., Vawter, M. P., Tomita, H., Li, J., Evans, S. J., Choudary, P. V., Myers, R. M., Jones, E. G., Watson, S. J., and Akil, H. (2003). Microarray technology: A review of new strategies to discover candidate vulnerability genes in psychiatric disorders. Am. J. Psychiatry 160, 657–666. Cheung, V. G., Gregg, J. P., Gogolin-Ewens, K. J., Bandong, J., Stanley, C. A., Baker, L., Higgins, M. J., Nowak, N. J., Shows, T. B., Ewens, W. J., Nelson, S. F., and Spielman, R. S. (1998). Linkage-disequilibrium mapping without genotyping. Nat. Genet. 18, 225–230. Chiang, L. W., Grenier, J. M., Ettwiller, L., Jenkins, L. P., Ficenec, D., Martin, J., Jin, F., DiStefano, P. S., and Wood, A. (2001). An orchestrated gene expression component of neuronal programmed cell death revealed by cDNA array analysis. Proc. Natl. Acad. Sci. USA 98, 2814–2819. Clark, T. A., Sugnet, C. W., and Ares, M., Jr. (2002). Genomewide analysis of mRNA processing in yeast using splicing-specific microarrays. Science 296, 907–910. Colangelo, V., Schurr, J., Ball, M. J., Pelaez, R. P., Bazan, N. G., and Lukiw, W. J. (2002). Gene expression profiling of 12633 genes in Alzheimer hippocampal CA1: Transcription and neurotrophic factor down-regulation and up-regulation of apoptotic and pro-inflammatory signaling. J. Neurosci. Res. 70, 462–473. Cutler, D. J., Zwick, M. E., Carrasquillo, M. M., Yohn, C. T., Tobin, K. P., Kashuk, C., Mathews, D. J., Shah, N. A., Eichler, E. E., Warrington, J. A., and Chakravarti, A. (2001). High-throughput variation detection and genotyping using microarrays. Genome Res. 11, 1913–1925. Dahlquist, K. D., Salomonis, N., Vranizan, K., Lawlor, S. C., and Conklin, B. R. (2002). GenMAPP, a new tool for viewing and analyzing microarray data on biological pathways. Nat. Genet. 31, 19–20. DeRisi, J. L., Iyer, V. R., and Brown, P. O. (1997). Exploring the metabolic and genetic control of gene expression on a genomic scale. Science 278, 680–686. Doniger, S. W., Salomonis, N., Dahlquist, K. D., Vranizan, K., Lawlor, S. C., and Conklin, B. R. (2003). MAPPFinder: Using Gene Ontology and GenMAPP to create a global gene-expression profile from microarray data. Genome Biol. 4, R7.


19

Eberwine, J., Yeh, H., Miyashiro, K., Cao, Y., Nair, S., Finnell, R., Zettel, M., and Coleman, P. (1992). Analysis of gene expression in single live neurons. Proc. Natl. Acad. Sci. USA 89, 3010–3014. El Atifi, M., Dupre, I., Rostaing, B., Chambaz, E. M., Benabid, A. L., and Berger, F. (2002). Long oligonucleotide arrays on nylon for large-scale gene expression analysis. Biotechniques 33, 612–616, 618. Elliott, R. C., Miles, M. F., and Lowenstein, D. H. (2003). Overlapping microarray profiles of dentate gyrus gene expression during development- and epilepsy-associated neurogenesis and axon outgrowth. J. Neurosci. 23, 2218–2227. Evans, S. J., Datson, N. A., Kabbaj, M., Thompson, R. C., Vreugdenhil, E., De Kloet, E. R., Watson, S. J., and Akil, H. (2002). Evaluation of AVymetrix GeneChip sensitivity in rat hippo campal tissue using SAGE analysis. Serial Analysis of Gene Expression. Eur. J. Neurosci. 16, 409–419. Fan, J. B., Chen, X., Halushka, M. K., Berno, A., Huang, X., Ryder, T., Lipshutz, R. J., Lockhart, D. J., and Chakravarti, A. (2000). Parallel genotyping of human SNPs using generic high-density oligonucleotide tag arrays. Genome Res. 10, 853–860. The FANTOM Consortium and the RIKEN Genome Exploration Research Group Phase I & II Team. (2002). Analysis of the mouse transcriptome based on functional annotation of 60,770 full-length cDNAs. Nature 420, 563–573. Fodor, S. P. A., Rava, R., Huang, X. C., Pease, A. C., Holmes, C. P., and Adams, C. L. (1991). Light-directed, spatially addressable parallel chemical synthesis. Science 251, 767–773. Fodor, S. P., Rava, R. P., Huang, X. C., Pease, A. C., Holmes, C. P., and Adams, C. L. (1993). Multiplexed biochemical assays with biological chips. Nature 364, 555–556. Geschwind, D. H. (2000). Mice, microarrays, and the genetic diversity of the brain. Proc. Natl. Acad. Sci. USA 97, 10676–10678. Geschwind, D. H., Ou, J., Easterday, M. C., Dougherty, J. D., Jackson, R. L., Chen, Z., Antoine, H., Terskikh, A., Weissman, I. L., Nelson, S. F., and Kornblum, H. I. (2001). A genetic analysis of neural progenitor diVerentiation. Neuron 29, 325–339. Geschwind, D. H., and Gregg, J. (2002). Microarrays for the neurosciences: An essential guide. Cambridge, Mass, MIT Press. Ginsberg, S. D., Hemby, S. E., Lee, V. M., Eberwine, J. H., and Trojanowski, J. Q. (2000). Expression profile of transcripts in Alzheimer’s disease tangle-bearing CA1 neurons. Ann. Neurol. 48, 77–87. Ginsberg, S. D., and Che, S. (2002). RNA amplification in brain tissues. Neurochem. Res. 27, 981–992. Gitan, R. S., Shi, H., Chen, C. M., Yan, P. S., and Huang, T. H. (2002). Methylation-specific oligonucleotide microarray: A new potential for high-throughput methylation analysis. Genome Res. 12, 158–164. Godfrey, T. E., Kim, S. H., Chavira, M., RuV, D. W., Warren, R. S., Gray, J. W., and Jensen, R. H. (2000). Quantitative mRNA expression analysis from formalin-fixed, paraYn-embedded tissues using 50 nuclease quantitative reverse transcription-polymerase chain reaction. J. Mol. Diagn. 2, 84–91. Gregg, J., and Baldwin, D. (2001). ‘‘Microarrays: an introduction. Microarrays: The New Frontier in Gene Discovery and Gene Expression Analysis.’’ Society for Neuroscience Short Course Syllabus. Society for Neuroscience, Washington D. C. Gupta, V., Cherkassky, A., Chatis, P., Joseph, R., Johnson, A. L., Broadbent, J., Erickson, T., and DiMeo, J. (2003). Directly labeled mRNA produces highly precise and unbiased diVerential gene expression data. Nucl. Acids Res. 31, e13. Hacia, J. G., Fan, J. B., Ryder, O., Jin, L., Edgemon, K., Ghandour, G., Mayer, R. A., Sun, B., Hsie, L., Robbins, C. M., Brody, L. C., Wang, D., Lander, E. S., Lipshutz, R., Fodor, S. P., and Collins, F. S. (1999). Determination of ancestral alleles for human single-nucleotide polymorphisms using high-density oligonucleotide arrays. Nat. Genet. 22, 164–167.

20

KARSTEN et al.

Hillier, L. D., Lennon, G., Becker, M., Bonaldo, M. F., Chiapelli, B., Chissoe, S., Dietrich, N., DuBuque, T., Favello, A., Gish, W., Hawkins, M., Hultman, M., Kucaba, T., Lacy, M., Le, M., Le, N., Mardis, E., Moore, B., Morris, M., Parsons, J., Prange, C., Rifkin, L., Rohlfing, T., Schellenberg, K., and Marra, M. (1996). Generation and analysis of 280,000 human expressed sequence tags. Genome Res. 6, 807–828. Hirschhorn, J. N., Sklar, P., Lindblad-Toh, K., Lim, Y. M., Ruiz-Gutierrez, M., Bolk, S., Langhorst, B., SchaVner, S., Winchester, E., and Lander, E. S. (2000). SBE-TAGS: An array-based method for eYcient single-nucleotide polymorphism genotyping. Proc. Natl. Acad. Sci. USA 97, 12164–12169. Hu, G. K., Madore, S. J., Moldover, B., Jatkoe, T., Balaban, D., Thomas, J., and Wang, Y. (2001). Predicting splice variant from DNA chip expression data. Genome Res. 11, 1237–1245. Hughes, T. R., Mao, M., Jones, A. R., Burchard, J., Marton, M. J., Shannon, K. W., Lefkowitz, S. M., Ziman, M., Schelter, J. M., Meyer, M. R., Kobayashi, S., Davis, C., Dai, H., He, Y. D., Stephaniants, S. B., Cavet, G., Walker, W. L., West, A., CoVey, E., Shoemaker, D. D., Stoughton, R., Blanchard, A. P., Friend, S. H., and Linsley, P. S. (2001). Expression profiling using microarrays fabricated by an ink-jet oligonucleotide synthesizer. Nat. Biotechnol. 19, 342–347. Irizarry, R. A., Hobbs, B., Collin, F., Beazer-Barclay, Y. D., Antonellis, K. J., Scherf, U., and Speed, T. P. (2003). Exploration, normalization, and summaries of high density oligonucleotide arrayprobe level data. Biostatistics 4, 249–264. Kamme, F., Salunga, R., Yu, J., Tran, D. T., Zhu, J., Luo, L., Bittner, A., Guo, H. Q., Miller, N., Wan, J., and Erlander, M. (2003). Single-cell microarray analysis in hippocampus CA1: demonstration and validation of cellular heterogeneity. J. Neurosci. 23, 3607–3615. Karsten, S., and Geschwind, D. H. (2002). Gene expression analysis using cDNA microarrays. In ‘‘Current protocols in neuroscience’’ ( J. N. Crawley, Ed.), Vol. 1, John Wiley & Sons, New York. Karsten, S. L., Van Deerlin, V. M. D., Sabatti, C., Gill, L., and Geschwind, D. H. (2002). An evaluation of tyramide signal amplification and archived fixed and frozen tissue in microarray gene expression analysis. Nucl. Acids Res. 30, e4. Karsten, S. L., Kudo, L., Sabatti, C., Jackson, R., Kornblum, H., and Geschwind, D. (2003). Global analysis of gene expression in neural progenitors reveals specific cell-cycle, signaling and metabolic networks. Dev. Bio. 261, 165–182. Kerr, M. K., Martin, M., and Churchill, G. A. (2000). Analysis of variance for gene expression microarray data. J. Comput. Biol. 7, 819–837. Lee, M. L., Kuo, F. C., Whitmore, C. A., and Sklar, J. (2000). Importance of replication in microarray gene expression studies: Statistical methods and evidence from repetitive cDNA hybridizations. Proc. Natl. Acad. Sci. USA 97, 9834–9839. Lindblad-Toh, K., Winchester, E., Daly, M. J., Wang, D. G., Hirschhorn, J. N., Laviolette, J. P., Ardlie, K., Reich, D. E., Robinson, E., Sklar, P., Shah, N., Thomas, D., Fan, J. B., Gingeras, T., Warrington, J., Patil, N., Hudson, T. J., and Lander, E. S. (2000). Large-scale discovery and genotyping of single-nucleotide polymorphisms in the mouse. Nat. Genet. 24, 381–386. Lipshutz, R. J., Fodor, S. P., Gingeras, T. R., and Lockhart, D. J. (1999). High density synthetic oligonucleotide arrays. Nat. Genet. 21, 20–24. Lockhart, D. J., Dong, H., Byrne, M. C., Follettie, M. T., Gallo, M. V., Chee, M. S., Mittmann, M., Wang, C., Kobayashi, M., Horton, H., and Brown, E. L. (1996). Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat. Biotechnol. 14, 1675–1680. Lockhart, D. J., and Winzeler, E. A. (2000). Genomics, gene expression and DNA arrays. Nature 405, 827–836. Luo, L., Salunga, R. C., Guo, H., Bittner, A., Joy, K. C., Galindo, J. E., Xiao, H., Rogers, K. E., Wan, J. S., Jackson, M. R., and Erlander, M. G. (1999). Gene expression profiles of laser-captured adjacent neuronal subtypes. Nat. Med. 5, 117–122.


21

Luo, Z., and Geschwind, D. H. (2001). Microarray applications in neuroscience. Neurobiol. Dis. 8, 183–193. Marvanova, M., Menager, J., Bezard, E., Bontrop, R. E., Pradier, L., and Wong, G. (2003). Microarray analysis of nonhuman primates: Validation of experimental models in neurological disorders. FASEB J. 17, 929–931. Middleton, F. A., Mirnics, K., Pierri, J. N., Lewis, D. A., and Levitt, P. (2002). Gene expression profiling reveals alterations of specific metabolic pathways in schizophrenia. J. Neurosci. 22, 2718–2729. Mirnics, K., Middleton, F. A., Marquez, A., Lewis, D. A., and Levitt, P. (2000). Molecular characterization of schizophrenia viewed by microarray analysis of gene expression in prefrontal cortex. Neuron 28, 53–67. Mirnics, K., Middleton, F. A., Lewis, D. A., and Levitt, P. (2001). Analysis of complex brain disorders with gene expression microarrays: Schizophrenia as a disease of the synapse. Trends Neurosci. 24, 479–486. Mycko, M. P., Papoian, R., Boschert, U., Raine, C. S., and Selmaj, K. W. (2003). cDNA microarray analysis in multiple sclerosis lesions: Detection of genes associated with disease activity. Brain 126, 1048–1057. Nadon, R., and Shoemaker, J. (2002). Statistical issues with microarrays: Processing and analysis. Trends Genet. 18, 265–271. Pilpel, Y., Sudarsanam, P., and Church, G. M. (2001). Identifying regulatory networks by combinatorial analysis of promoter elements. Nat. Genet. 29, 153–159. Pomeroy, S. L., Tamayo, P., Gaasenbeek, M., Sturla, L. M., Angelo, M., McLaughlin, M. E., Kim, J. Y., Goumnerova, L. C., Black, P. M., Lau, C., Allen, J. C., Zagzag, D., Olson, J. M., Curran, T., Wetmore, C., Biegel, J. A., Poggio, T., Mukherjee, S., Rifkin, R., Califano, A., Stolovitzky, G., Louis, D. N., Mesirov, J. P., Lander, E. S., and Golub, T. R. (2002). Prediction of central nervous system embryonal tumour outcome based on gene expression. Nature 415, 436–442. Pongrac, J., Middleton, F. A., Lewis, D. A., Levitt, P., and Mirnics, K. (2002). Gene expression profiling with DNA microarrays: Advancing our understanding of psychiatric disorders. Neurochem. Res. 27, 1049–1063. Quackenbush, J. (2002). Microarray data normalization and transformation. Nat. Genet. 32, 496–501. Ramakrishnan, R., Dorris, D., Lublinsky, A., Nguyen, A., Domanus, M., Prokhorova, A., Gieser, L., Touma, E., Lockner, R., Tata, M., Zhu, X., Patterson, M., Shippy, R., Sendera, T. J., and Mazumder, A. (2002). An assessment of Motorola CodeLink microarray performance for gene expression profiling applications. Nucl. Acids Res. 30, e30. Ren, B., Robert, F., Wyrick, J. J., Aparicio, O., Jennings, E. G., Simon, I., Zeitlinger, J., Schreiber, J., Hannett, N., Kanin, E., Volkert, T. L., Wilson, C. J., Bell, S. P., and Young, R. A. (2000). Genome-wide location and function of DNA binding proteins. Science 290, 2306–2309. The RIKEN Genome Exploration Research Group Phase II Team and the FANTOM Consortium. (2001). Functional annotation of a full-length mouse cDNA collection. Nature 409, 685–690. Roth, F. P., Hughes, J. D., Estep, P. W., and Church, G. M. (1998). Finding DNA regulatory motifs within unaligned noncoding sequences clustered by whole-genome mRNA quantitation. Nat. Biotechnol. 16, 939–945. Sabatti, C., Karsten, S., and Geschwind, D. (2002). Thresholding rules to recover a sparse signal from microarray experiments. Math. Biosci. 176, 17–34. Sandberg, R., Yasuda, R., Pankratz, D. G., Carter, T. A., Del Rio, J. A., Wodicka, L., Mayford, M., Lockhart, D. J., and Barlow, C. (2000). Regional and strain-specific gene expression mapping in the adult mouse brain. Proc. Natl. Acad. Sci. USA 97, 11038–11043. Schena, M., Shalon, D., Davis, R. W., and Brown, P. (1995). Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270, 467–470.

22

KARSTEN et al.

Schena, M., Heller, R., Theriault, T., Konrad, K., Lachenmeier, E., and Davis, R. (1998). Microarrays: Biotechnology’s discovery platform for functional genomics. Trends Biotechnol. 16, 301–306. Smyth, G. K., Yang, Y. H., and Speed, T. (2003). Statistical issues in cDNA microarray data analysis. Methods Mol. Biol. 224, 111–136. Specht, K., Richter, T., Muller, U., Walch, A., Werner, M., and Hofler, H. (2001). Quantitative gene expression analysis in microdissected archival formalin-fixed and paraYn-embedded tumor tissue. Am. J. Pathol. 158, 419–429. Speed, T. P. (2002). Statistical analysis of gene expression microarray data. Boca Raton, F1, CRC Press LLC. Spellman, P. T., Sherlock, G., Zhang, M. Q., Iyer, V. R., Anders, K., Eisen, M. B., Brown, P. O., Botstein, D., and Futcher, B. (1998). Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization. Mol. Biol. Cell. 9, 3273–3297. Stears, R. L., Getts, R. C., and Gullans, S. R. (2000). A novel, sensitive detection system for high-density microarrays using dendrimer technology. Physiol. Genomics 3, 93–99. Tang, Y., Lu, A., Aronow, B., and Sharp, F. R. (2001). Blood genomic responses diVer after stroke, seizures, hypoglycemia, and hypoxia: Blood genomic fingerprints of disease. Ann. Neurol. 50, 699–707. Taniguchi, M., Miura, K., Iwao, H., and Yamanaka, S. (2001). Quantitative assessment of DNA microarrays—comparison with Northern blot analyses. Genomics 71, 34–39. Tietjen, I., Rihel, J. M., Cao, Y., Koentges, G., Zakhary, L., and Dulac, C. (2003). Single-cell transcriptional analysis of neuronal progenitors. Neuron 38, 161–175. Tseng, G., Oh, M. K., Rohlin, L., Liao, J., and Wong, W. H. (2001). Issues in cDNA microarray analysis: Quality filtering, channel normalization, models of variation, and assessment of gene eVects. Nucl. Acids Res. 29, 2549–2557. Velculescu, V. E., Zhang, L., Vogelstein, B., and Kinzler, K. W. (1995). Serial analysis of gene expression. Science 270, 484–487. Welford, S. M., Gregg, J., Chen, E., Garrison, D., Sorensen, P. H., Denny, C. T., and Nelson, S. F. (1998). Detection of diVerentially expressed genes in primary tumor tissues using representational diVerences analysis coupled to microarray hybridization. Nucl. Acids Res. 26, 3059–3065. Whitney, L. W., Becker, K. G., Tresser, N. J., Caballero-Ramos, C. I., Munson, P. J., Prabhu, V. V., Trent, J. M., McFarland, H. F., and Biddison, W. E. (1999). Analysis of gene expression in multiple sclerosis lesions using cDNA microarrays. Ann. Neurol. 46, 425–428. Whitney, L. W., and Becker, K. G. (2001). Radioactive 33-P probes in hybridization to glass cDNA microarrays using neural tissues. J. Neurosci. Methods 106, 9–13. Yang, G. P., Ross, D. T., Kuang, W. W., Brown, P. O., and Weigel, R. J. (1999). Combining SSH and cDNA microarrays for rapid identification of diVerentially expressed genes. Nucl. Acids Res. 27, 1517–1523. Yang, Y. H., Dudoit, S., Luu, P., Lin, D. M., Peng, V., Ngai, J., and Speed, T. P. (2002). Normalization for cDNA microarray data: A robust composite method addressing single and multiple slide systematic variation. Nucl. Acids Res. 30, e15. Yang, Y. H., and Speed, T. (2002). Design issues for cDNA microarray experiments. Nat. Rev. Genet. 3, 579–588. Yue, H., Eastman, P. S., Wang, B. B., Minor, J., Doctolero, M. H., Nuttall, R. L., Stack, R., Becker, J. W., Montgomery, J. R., Vainer, M., and Johnston, R. (2001). An evaluation of the performance of cDNA microarrays for detecting changes in global mRNA expression. Nucl. Acids Res. 29, E41–1.


23

Yuen, T., Wurmbach, E., PfeVer, R. L., Ebersole, B. J., and Sealfon, S. C. (2002). Accuracy and calibration of commercial oligonucleotide and custom cDNA microarrays. Nucl. Acids Res. 30, e48. Zirlinger, M., Kreiman, G., and Anderson, D. J. (2001). Amygdala-enriched genes identified by microarray technology are restricted to specific amygdaloid subnuclei. Proc. Natl. Acad. Sci. USA 98, 5270–5275.


EXPERIMENTAL DESIGN AND LOW-LEVEL ANALYSIS OF MICROARRAY DATA

B. M. Bolstad,* F. Collin,y K. M. Simpson,z R. A. Irizarry,x and T. P. Speedy;z *Division of Biostatistics and Department of Statistics, University of California, Berkeley, California 94720 z Division of Genetics and Bioinformatics, WEHI, Melbourne, Australia x Department of Biostatistics, John Hopkins University, Baltimore, Maryland 21205 y

I. Introduction II. Design of Experiments A. Design of Experiments Using AVymetrix Arrays B. Design of Experiments Using cDNA Arrays III. Sample Size Considerations A. A Classic Power Calculation B. Some Observations about Power Calculations in the Microarray Context C. What Can Be Done? Part I D. Is There a Multiple Testing Power Analysis? E. What Can Be Done? Part II IV. Normalization A. Normalization for cDNA Arrays B. Normalization for AVymetrix Arrays V. Expression Summaries for GeneChip Data VI. Quality Assessment A. Quality Assessment for AVymetrix GeneChip Expression Data B. Quality Assessment for cDNA Microarray Experiments VII. Detection of Absolute Gene Expression A. The AVymetrix Presence=Absence Algorithm B. Alternative Methods References

I. Introduction

In this chapter, we review the design and low-level analysis of microarray experiments. Microarray experiments are widely used to quantify and compare gene expression on a large scale. Such experiments can be costly in terms of equipment, consumables, and time. For this reason, careful design is particularly important if the resulting experiment is to be maximally informative, given the eVort and the resources. A number of issues must be addressed when designing a microarray experiment: what will have the most impact on the accuracy and precision of the resulting measurements? How should the diVerent components


25


26

BOLSTAD et al.

of the experimental design be balanced to reach a decision? For example, should we replicate, and if so, how? Which samples should be hybridized to which slides? Should samples be pooled? If the design is inadequate, the experimenter will be left with a less than eVective use of resources and the resulting conclusions might be biased. The scientific question of interest may even be left unanswerable with the collected data. Low-level analysis is carried out between the image analysis phase and interrogation of gene expression data. The goal of low-level analysis is to take raw data from the scanner, without any biological interpretation, and process it to produce cleaner and ultimately more meaningful gene expression measures. This is in contrast to higher level analysis in which questions of a more biological nature are addressed. Such high-level questions include detecting diVerential expression in treatment and control tissues, gene function, pathway analysis, and changes in gene expression over time, among others. We shall not further consider such high-level analysis here. However, improved low-level analysis ultimately aids the downstream data investigations and for this reason is very important to consider. In this chapter, we consider the following low-level topics: normalization, in which the goal is to reduce or remove sources of nonbiological variability (for both complementary DNA [cDNA] and AVymetrix arrays), summarization, in which one combines multiple probes to produce a gene expression measure (a topic of great importance for AVymetrix-like arrays), data quality, as may be assessed by image analysis for cDNA arrays and from probe-level modeling of AVymetrix arrays, and the detection of absolute expression.

II. Design of Experiments

A. DESIGN OF EXPERIMENTS USING AFFYMETRIX ARRAYS Comparative experiments with an AVymetrix chip share many similarities with comparative experiments more generally. In contrast with two-color cDNA microarray experiments, which we discuss next, we can make immediate use of the extensive statistical literature on experimental design (Box et al., 1978; Cobb, 1998; Cox, 1992; Montgomery, 2000). These books present excellent discussions of the general principles of randomization, replication, and local control within the context of agricultural, industrial, and scientific experimentation. In this section, we briefly summarize the particular aspects of these issues relevant to comparative experiments involving AVymetrix and other high-density short oligonucleotide chips and refer the reader to the books for more general considerations. Randomization is not widely used and not popular in biomedical laboratory experiments. In our view, it would be a step forward if this attitude changed. One

DESIGN AND ANALYSIS OF MICROARRAY DATA

27

of the most striking aspects of AVymetrix chip experiments is the extent to which the conditions of an experiment—the messenger RNA (mRNA) extraction and processing, the reagents, the operator, the hybridization conditions, the scanner, and so on—can leave a ‘‘global signature’’ in the resulting expression data. If an experiment involving a number of chips is carried out over a number of days, more often than not, chips processed on the same day are readily identifiable as such, for example, by cluster analysis or some other similar global analysis. If diVerent parts of an experiment are carried out on diVerent equipment, at very diVerent times, or with diVerent personnel, these global diVerences can be striking, raising the possibility that they could obscure or confound the very diVerences the experiment seeks to measure. The normalization methods we discuss can reduce the impact of some of these diVerences, but if they cannot be controlled by good design, randomization is the best way to deal with them. We illustrate the idea next. Replication is the foundation stone of statistical inference. No one should expect to be able to reach conclusions about variable populations based on a sample size of one. Appropriate replication is the key to generalizing from a sample to a population, but the term appropriate is important here. Taking an RNA sample from, say, the liver of one mouse from an inbred strain, treating it correctly, and hybridizing portions of the result on to two AVymetrix chips leads to what is known as technical replicate data: two sets of measurements on one mouse. This is to be contrasted with taking RNA from the livers of two mice from the same strain and hybridizing RNA from the diVerent mice onto diVerent chips. These are called biological replicates: one set of measurements on each of two mice. Clearly biological replication leads to data that are better for reaching conclusions that might apply more generally, and using more than two mice is better still. Equally clearly, technical replication leads to data better for reaching conclusions about that particular mouse, under the conditions of that particular experiment. The term appropriate depends on the aim of the experiment, where this term includes the level of generalization sought for any conclusions. In most experiments involving animals or humans, biological replicates are more relevant to the aims of an experimenter than technical replicates. Indeed, it is arguable that in an experiment in which variable material such as animal or human tissue is involved, technical replicates are never preferable to biological replicates. Chips are expensive, and a technical replicate will (in a well-conducted experiment) usually give results very similar to the original hybridization, whereas even inbred animals can lead to quite diVerent results. Biological replication can also be expensive, but in general it will be better to spend limited resources getting more data at the level exhibiting higher variation, across animals, say, than at the level exhibiting lower variation, within animals. Averaging results can reduce the impact of chance variation on summary statistics, but if there are chance features common to all results being averaged, as in the case of technical replicates, averaging will not reduce their impact. If the ability to obtain

28

BOLSTAD et al.

technical replicates is limited, but resources for experimentation are not, then carrying technical replicates is a good way to improve the quality of the data you have, but only about the samples measured. In a way, you get better information about a more limited universe, which is reasonable if it is not possible to expand that universe, but in general the expanded universe leads to more robust scientific conclusions. There are many intermediate levels of replication between what one might call pure technical and pure biological replicates, and in general a more eYcient use of resources results from carrying out replication at the level at which the greatest variability is to be found. Local control is the general term statisticians use for arranging experimental material, in the present case, RNA samples to be treated and hybridized to chips, in relation to extraneous sources of variability. By extraneous we mean variability in the measurement process, not variability in the experimental material, which is what is relevant to replication. For example, if we propose to compare gene expression between wild-type and particular mutant animals and plan to use six animals in each group, it would be most unwise to carry out the processing of all six animals in one group on one day and then process the six animals from the other group on a second day. If this was done, day-to-day diVerences, which are sometimes diYcult to avoid, could be wrongly seen as genetic diVerences. Equally, it would be unwise to carry out the processing of the six wild-type animals first, spread out over days, and then follow with the six mutant animals. A more appropriate approach in this example would be to proceed as follows. First decide how many samples can be processed in a single day, and hence the number of days over which the processing will be spread. Suppose that six per day for 2 days is deemed feasible. Then an appropriate design would be to process three wildtype and three mutant animals on each day, each set of three chosen at random from the available six, and within a day, process the mix of three wild-type and three mutants in a random order. A more complex design might seek to avoid processing all three mutants or all three wild-type animals on any given half-day, although here it is clearly impossible to arrange for equal numbers in each group to be processed in each half-day. The principles being illustrated should be clear here: group the material so that any unique chance features of a day are shared equally by the two groups and thus are not potentially confounded with group diVerences, and to the extent that this is not possible, randomize over the remaining features, such as time within day. Similar considerations apply whenever diVerences will occur across other factors known to contribute extraneous variability, such as chip batch, operator, reagents, scanners, and so on. Arrange those varying factors that can be controlled in a manner similar to that just described, and randomize across the remainder. This is the general idea, and more details of its implementation can be found from the books listed. For the design of a complex experiment, it is advisable to seek statistical advice.


29

We are frequently asked about the advantages or disadvantages of pooling RNA samples before hybridization. For example, in the small wild-type versus mutant experiment just discussed, RNA from the six wild-type animals could be pooled, and the resulting mix hybridized to six (or fewer) chips, with the mutant animals being dealt with likewise. Many people feel that pooling provides a form of ‘‘biological averaging’’ and should make it possible to get more precise results with fewer chips than hybridizing RNA from individual animals to separate chips. Our view is that although this may well be true, compelling evidence is not yet present in the literature. On the other hand, we have seen evidence that pooling could mislead, but perhaps more importantly, could mislead without this being apparent. Suppose that one animal contributing to the pool is very diVerent from the remainder, for example, perhaps one mouse has been injured by fighting with other mice, and that its immune system is in a quite diVerent state. Depending on the tissue under study, pooling this animal’s RNA with that of the other mice gives it the ability to exert a large influence on the measured expression values for many genes, whereas running it on a separate chip leaves open the possibility of identifying this animal as an outlier. If the level of a particular mRNA transcript is 50 times the average of the other animals, biological averaging is not taking place, rather this outlier animal is biasing conclusions reached for the group. At this point it is probably best to say that if pooling is envisaged in an experiment that could be carried out at the same cost without pooling, it is better not to pool. If pooling is carried out to save cost, the possible drawback just mentioned should be borne in mind. Frequently, pooling is seen to be necessary to get suYcient mRNA from the tissue in question, in which case the possible drawbacks must inevitably be accepted or at least weighed against the possible drawbacks of the alternative, which is usually amplification.

B. DESIGN OF EXPERIMENTS USING CDNA ARRAYS Because the two-color microarray system is inherently comparative, the major design issue for cDNA microarrays is which samples should be cohybridized. For any proposed design to be desirable, it needs to satisfy two types of constraints: physical and scientific. Physical constraints include the quantity of RNA available, the number of slides, and other cost considerations. Scientific constraints should motivate the design in that more important questions should be able to be answered more precisely than less important ones. The primary decision with cDNA experiments is which samples to hybridize together. This often becomes a question of whether to use a direct or an indirect comparison. Direct comparisons are between two samples hybridized to the same slide. Indirect comparisons are those between samples that can be compared only via multiple slides. In other words, should we be comparing within or between

30

BOLSTAD et al.

FIG. 1. Common designs for complementary DNA experiments. The target RNA labeled with Cy3 is at the tail of the arrow, and the RNA labeled with Cy5 is at the head. (A) Direct comparison between A and B. (B) An indirect comparison between A and B using reference R. (C) A dye swap experiment. (D) A simple loop design. Each sample is directly compared with the others.

slides? Figure 1 demonstrates some of the basic components of cDNA experiments using a simple graphical representation. The squares represent target mRNA samples and the arrows represent hybridization between the two samples. By convention, the sample at the head of the arrow will be labeled with Cy5 (red) and the tail with Cy3 (green). Sometimes there is an obvious design choice, given the available materials and the goal of the experiment. For example, suppose we have a series of cells each receiving treatment from a diVerent drug and the aim is to compare them with untreated cells. An appropriate design in this case would be one in which the untreated cells become a de facto reference and each one of the treated sets of cells is hybridized with the untreated cells (as in Fig. 1B). The key diVerence between direct and indirect comparisons can be illustrated by examining the variance of the estimated log fold change in each comparison. Consider a direct comparison of the form shown in Fig. 1A. We will assume that the variance of the log ratio log(B=A) is given by 2. With two direct comparisons, we would take the average of the two independent observed ratios, yielding an estimate with variance 2=2. If we instead make an indirect comparison, as in Fig. 1B, then an estimate of the log ratio is given by logðB=AÞ ¼ logðB=RÞ logðR=AÞ. Assuming independence, this would have a variance of 22. In other words, two direct comparisons have one fourth the variance of an indirect comparison. Note that in practice this might not be observed, because the independence assumption is not always valid.


31

A dye-swap experiment (Fig. 1C) is one in which each hybridization is done twice, with the dye assignments reversed in the second hybridization. This is useful for removing systematic bias. Systematic dye biases are commonly observed with cDNA microarrays, and although normalization (as will be discussed later) can partially correct for this eVect, it is still possible that individual spots have a residual color bias. By swapping the dyes and averaging across the two hybridizations, it is hoped that this bias will be reduced. Figure 1D represents a simple loop design. In this example, a comparison between any two pairs can be made directly (e.g., A–B) and indirectly (through C). Maintaining our previous assumptions, the variance of such a comparison would be 22=3. With more than three sources of mRNA, the number of hybridizations needed to allow direct comparisons between every pair of sources increases rapidly. In practice, the comparisons of greatest interest would be done directly and those of lesser interest done more indirectly, with those of least interest the most distant. In a time-course experiment, the design choice depends on the comparison of interest. For instance, if comparisons to the initial time are of interest, then each of the subsequent time points should be hybridized with a sample from the initial time. If changes between time points are of particular interest instead, then a more sequential design would be desirable, with samples from consecutive time points hybridized together. Often, a combination of the two might be desirable, for example, a direct comparison between the first and final time points and direct comparisons between each of the intermediate time points. The second important decision in cDNA microarray experiments is how replication should be carried out. Replication is important because it reduces variability in summary statistics and allows data to be analyzed using more formal statistical methods. With cDNA microarrays, individual genes can be quite variable between hybridizations. By replicating, then averaging, less variable estimates are obtained. A common form of replication is to place replicates of the same spot (cDNA probe) on each slide. However, because such spots will typically share systematic eVects from printing, general hybridization, and scanning, the lack of independence between these measurements reduces their value for more sophisticated statistical inference. If duplicate spots are to be used, it is recommended that they are distributed across the slide because this will give a better reflection of intraslide variability. There are two methods by which between-slide replicates can be created: technical replication and biological replication. Technical replication is when mRNA from the same extraction is hybridized to multiple slides. Because RNA extraction typically has characteristic repeatable elements, technical replicates usually have smaller variability than biological replicates. In addition, these shared systematic features will remain even after averaging. Biological replication

32

BOLSTAD et al.

refers to hybridizations that use mRNA from diVerent extractions from, for example, the same cell line or tissue. Ideally, sample labeling is also carried out separately for each extraction. In the context of a microarray experiment, this leads more closely to independent experimental results. Biological replication should therefore, be favored as the primary method of replication. Biological replication also refers to the situation in which target mRNA is extracted from diVerent individuals or perhaps from diVerent versions of a cell line. This form will be more variable than the biological variation discussed earlier. Although this might make it more diYcult to detect real expression diVerences, the conclusions made from such an experiment might be more generalizable. For example, multiple extractions from a single individual cannot be treated as being representative of all individuals with the same condition, so any conclusions made using the former might be flawed. A mixture of biological and technical replication is often desirable. Biological replication will allow more generalization of conclusions. Technical replication will reduce variability. Just how much and which types of replication can be done are determined by the physical constraints of the experiment.

III. Sample Size Considerations

A. A CLASSIC POWER CALCULATION Suppose that we use a microarray to measure the expression of one gene in a class of cells on unmatched samples from cases and controls. (The approach in what follows applies equally to matched samples, but the actual figures will change.) We suppose that the measurements are given in the log2 scale. They will be subject to measurement error, which we view on the same log2 scale. Our aim in this section is to explore power issues in a context in which the aim of the study is to identify diVerentially expressed genes based on gene expression microarray data. The test we discuss here is the standard two-sided, two-sample t test, with type I error (i.e., false-positive) rate of 5%. Such a test seeks to identify diVerential expression between the case and control groups on the log2 scale, in other words, relative expression, because a diVerence of logs of means equals a log ratio of means. For example, a mean diVerence of 1 between the two groups on this log2 scale is equivalent to a twofold diVerence, on average, on the original (concentration) scale. Microarray gene expression measurements are subject to measurement error, and in our analysis, we make a range of realistic assumptions concerning the standard deviation (SD) of this error.

33


Of primary interest in power studies is the probability of detecting a mean diVerence of a given magnitude between the two groups, given a sample size n and value SD for the error standard deviation, and our test procedure (with given type I error). Put another way, of interest is the power of our test to reject the null hypothesis of no mean diVerence in the gene’s expression between the two groups, given that the true mean diVerence has a magnitude , the sample size is n, and the measurement error has standard deviation SD. Realistic values of in this context are 0.5, 1.0, and 2.0, corresponding to fold changes of 1.4, 2.0, and 4.0. Similarly, realistic values of SD (on the log2 scale) are 0.5, 1.0, and 2.0, while we consider sample sizes (for each of the case and control groups) of 10, 20, and 30. We will see shortly that the main points we make are not dependent on the precise values of these numbers. As foreshadowed previously, we calculate and display later the power to reject the null hypothesis for all 27 combinations of these input variables, using the abbreviations LO and HI for values of the power close to 0 or 1, respectively. Each of the three tables later in this chapter correspond to one value for the SD, the first to the value 0.5, the second to 1.0, and the third to 2.0. Rows of the tables are labeled by the value for , and columns by the value of n. All the entries of the tables are power values, that is the probability of (correctly) rejecting the null of no mean group diVerence, given the values for the parameters and the fact that we are using a two-sided two-sample t test, with type I error 5%. The power values are taken from Table 10 of Pearson and Hartley (1962).

¼ 0.5 ¼1 ¼2

n ¼ 10

n ¼ 20

n ¼ 30

.60 .98 HI

.8 HI HI

.95 HI HI

SD ¼ .5

n ¼ 10 n ¼ 20 n ¼ 30 LO .6 .98

LO .8 HI SD ¼ 1

.5 .95 HI

n ¼ 10 n ¼ 20 n ¼ 30 LO LO .6

LO LO .8

LO .5 .95

SD ¼ 2

What can we learn from these values? The most striking thing is that the actual power varies from LO to HI through the values 0.5, 0.6, 0.8, and 0.95 as we vary the postulated mean group diVerence , the SD, and the sample size through plausible values in the microarray context. Put another way, depending on the group mean diVerence and the single gene expression measurement SD, we can have a power from HI to middling to LO, as the sample size varies. For large and small SD, we can get by with small samples and feel sure that group mean diVerences of this magnitude will be detected, whereas for smaller values of , larger SDs, or both, detection of such group mean diVerences may be diYcult with even the highest sample size we currently contemplate.

34

BOLSTAD et al.

B. SOME OBSERVATIONS ABOUT POWER CALCULATIONS IN THE MICROARRAY CONTEXT We now make some important observations about power calculations in this microarray context. Together, they suggest that the previous power calculations are of little direct value in the microarray context. First, in general we do not necessarily know in advance the genes whose between case control mean diVerential expression we wish to ascertain. Some of these will be probes corresponding to unknown ESTs, and our finding them to be diVerentially expressed between cases and controls might be the first interesting fact found out about them. Often the main reason for doing a microarray experiment is to find genes that are diVerentially expressed between cases and controls. Second, even if we do know the names of the genes of interest, and there are often many likely candidates that could be specified in advance, we will not necessarily be able to state in advance the magnitude of the relative mean expression that is of interest. Of course, we might nominate a fold change such as 1.4 and claim interest in detecting all genes with fold changes of 1.4 or greater. However, there are undoubtedly genes for which smaller fold changes are still biologically significant and that might be detected with greater replication. But we will not know in advance which these are. Third, even if we did know the gene or genes of interest and we were able to specify to the magnitude of the diVerence of interest, we are extremely unlikely to know in advance the SD of a single microarray expression measurement on that gene. Again, we might nominate a ( perhaps conservative) value for the SD and hope that in so doing, we avoid this objection. But again, we will not know in advance which genes have SDs smaller than this conservative value, and which do not. And SDs of genes can vary enormously. Fourth, the analysis underlying the power calculation just given assumes that all cases have the same normally distributed gene expression measurements (on the log2 scale) and similarly for the controls. It ignores the well-known tendency for cases in particular to be heterogeneous in gene expression, and to a lesser extent, this also applies to controls. Heterogeneity compromises the power calculation, and unknown heterogeneity does so to an unknown extent. Summarizing these four points, a conventional power analysis is of limited value in the context of analyzing microarray gene expression values, because we will only rarely be able to nominate in advance the gene, the magnitude of interest, or the magnitude of the measurement error SD, or be sure about homogeneity of response. Both the SD and the biologically meaningful change are notorious variables across genes in expression microarray studies. Nevertheless, it might still be argued that we should be able to determine the power we have with a given sample to detect all genes whose log fold change is some value


35

or greater, and whose measurement SD is no more than a given (conservative) value, even if we will not know in advance which genes these are. However, there is a much greater obstacle to making use of conventional power analyses in this context than these four points. In microarray studies, or other similar ‘‘genome-wide’’ studies, we are not measuring just one gene’s expression; in a typical experiment, we measure the gene expression in 20,000 genes. When we do this, another important issue comes to the fore: the issue of multiple testing. Put simply, the multiple testing issue is that even when no genes are diVerentially expressed, on average, between cases and controls, many will appear to be so, by chance, according to conventional analyses. In other words, even though the power analysis presented earlier is appropriate for examining whether a single (named) gene is diVerentially expressed between cases and controls, that is not the context in which we operate with gene expression microarrays. In this context, we are screening tens of thousands of genes for diVerential expression, and the conventional power analysis just presented is quite inappropriate. To identify diVerentially expressed genes in the context of tens of thousands of genes, we would never use the conventional two-sided twosample t tests with a conventional cutoV appropriate to a (single test) type I error rate 5%, or we would risk obtaining hundreds or thousands of type I errors. Indeed, we might not even use the t statistic at all.

C. WHAT CAN BE DONE? PART I The next question is naturally, which power analyses are appropriate in this multiple testing context? Sadly, but almost inevitably, the answer is that with our present understanding of the statistical analysis of microarray gene expression data (Speed, 2003), there are no analyses strictly similar to conventional power analyses. The reason is not diYcult to see. As already noted, with tens of thousands of genes, the observed log2 fold changes vary greatly, and the variability of single gene expression measurements vary greatly. If the analysis is to be global, that is, if it is to use all genes with probes on the array, then many genes will be genuinely diVerentially expressed at levels that would be of interest to an experimenter if he or she knew, but these will not be identified, because they will be masked by the thousands of genes quite probably not diVerentially expressed, but simply varying greatly. These statements can be (and have been) supported by theoretical and simulation analyses. Put crudely, for a gene to be noticed against a background of tens of thousands of other genes, it must stand out, that is, it must have a level of diVerential expression substantially greater than the others, and in particular, substantially greater than we would require in conventional one-gene-at-a-time testing.

36

BOLSTAD et al.

Statisticians have developed a number of ways of dealing with the generic testing issues in this context, with names like family-wise (type I) error rate (FWER), false-discovery rate (FDR), and positive FDR (pFDR). These are all multiple testing analogs of the simple type I error for a single test, with diVerent assumptions and diVerent properties (Ge et al., 2003). Their natures diVer greatly: FWER refers to single genes, and FDR and pFDR only to sets of genes. The ability to use them diVers greatly, with FDR techniques more broadly applicable and FWER approaches less so. And their values can diVer greatly. The story is not so simple, but the main point is we cannot and should not speak of the power to reject the null hypothesis of no diVerence in the mean expression for single genes, with microarray experiments. If we want to use the advantage of the microarray experiment in permitting us to screen tens of thousands of genes simultaneously, we have to appreciate that decisions on individual genes in a microarray experiment are always made in the context of thousands of similar decisions on other genes. This is both a boon for biologists and a drawback for them, because their thinking concerning power (and a few other issues) needs to be modified.

D. IS THERE A MULTIPLE TESTING POWER ANALYSIS? The nearest analog to a power analysis in the large-scale multiple testing context is embodied in the following idea. We could conduct a computational study in which expression levels for cases and controls for all genes in a large set are simulated, where we include a specified subset of genes as having predetermined levels of diVerential expression between the groups, and we give all genes predetermined standard deviations for their measurement errors. Then for a specified test statistic and multiple testing procedure, we could examine how many of the genes ‘‘known’’ to be diVerentially expressed from the design of the study are correctly identified as diVerentially expressed. This could be done for many procedures at the same time, and the performances compared (Lo¨nnstedt and Speed, 2002). The catch is that we need to simulate data that we hope look like our actual data. All power studies assume that the actual data are very close in their statistical properties to the underlying theoretical model, and frequently this is not such a bad assumption. In other words, power studies frequently turn out to be useful. However, with gene expression measurements on 20,000 genes, nobody knows how to describe a model (and hence simulate data) that leads to data closely approximating the data we observe. The microarray process is simply too complex for this to be achievable right now, and perhaps for the foreseeable future. We can hope that our simulations capture features of the data relevant to our analysis, and there is some evidence—it is early days yet for this—that our hopes are fulfilled.


37

Such simulation studies have been carried out, and one of the clear conclusions is that, just as one might expect, when there are a lot of genes diVerentially expressed between two conditions and these diVerences in average level vary from high to low, there is no chance of identifying all of them. False-positive rates are controllable, but at the price of what might appear at first glance to be disturbingly high false-negative rates, that is, very low power, in the traditional (one-gene-at-a-time) thinking. When there are just a few genes diVerentially expressed and the diVerences in average level are not small, then with a large enough number of replicates, it is reasonable to expect them all to be identified. As is so often the case with microarrays, the catch is that we do not generally know in advance which case we are in, although we frequently suspect that it is the first, and more diYcult, rather than the second, relatively easier case.

E. WHAT CAN BE DONE? PART II Where are we now? If we think that our average gene expression diVerences between cases and controls are not likely to be large, then we are simultaneously faced with the prospect of possibly missing many genes whose diVerential expression might be of interest to us and having little that we can do in advance to influence this. It remains true that more independent replicates always helps our aims, as long as the heterogeneity does not increase at the same rate and the analyses are appropriate. What we cannot do is say how many replicates are enough. Even answering the question ‘‘enough for what?’’ is not easy. Two simple conclusions follow from this discussion. First, we should aim to get as many case and control samples as we can, bearing in mind the important requirement of homogeneity. And secondly, we need to improve the quality of our statistical analysis, knowing that we cannot increase our sample size beyond a certain point. In brief we need a ‘‘smarter analysis’’ to try and overcome the limitations of modest sample size. We need to do better than using the standard t test (or its analogs) and instead use appropriate calibrated moderated t statistics. We need to depart from the conventional ‘‘context-free,’’ search for diVerentially expressed genes that get embodied in power calculations and their multiple testing analogs, and more fully integrate biological knowledge with statistical analysis. Recent eVorts for connecting pathways, the Gene Ontology classification and related tools are along these lines (Mootha et al., 2003). Even when we do these things, multiple testing issues will remain, so we have to extend that theory to apply to our stronger analysis. In brief, we are still in the early days of the statistical analysis of microarray data. Much traditional thinking must be extended and strengthened.

38

BOLSTAD et al.

IV. Normalization

Normalization is a process performed to compensate for systematic technical diVerences both between and within arrays. The process of normalization should reduce or remove this variation while leaving the more scientifically interesting biological diVerences that may exist. Systematic nonbiological diVerences between samples become apparent in several common ways. For instance, it is often observed that one array is brighter overall than another. With cDNA arrays, a systematic diVerence in the intensity of signals from diVerent dyes is a frequently observed source of variation. There are many possible causes of systematic nonbiological variation. DiVering amounts of RNA, scanner settings, and diVering hybridization or experimental conditions are all commonly observed contributors to this sort of variation. In two-color arrays, dye biases are commonly observed. These biases could be due to various factors such as physical properties of the dyes (light and heat sensitivity), how eYciently dye is incorporated, or experimental variation in the labeling process. It should be remembered that all normalization methods require some level of assumption about the underlying data. The most common assumption is that most genes are not changing across conditions. A second common assumption is that the number of upregulated genes is roughly equal to the number of downregulated genes across conditions.

A. NORMALIZATION FOR cDNA ARRAYS The MA-plot (Dudoit et al., 2002a) is a very useful tool for normalizing twocolor microarray data. For each spot on the array, we have a (R, G) fluorescence intensity pair (where R ¼ red, for Cy5, and G ¼ green, for Cy3). An MA-plot is used to represent these (R, G) data pairs, where we define M ¼ log2 R=G and pffiffiffiffiffiffiffi A ¼ log2 RG . MA-plots help to identify spot artifacts and to detect intensitydependent patterns in the log ratios. Note that the MA-plot is a rotation and rescaling of a plot of R versus G. Because equal amounts of RNA are generally hybridized to both channels, a diVerence in brightness between channels must be due to diVerent uptakes of the labeling dye. The simplest adjustment one can make to two-color array data is to scale the data so that both channels have equal total intensity. We can think of this as relating the two channels by a constant so that R ¼ kG. Note that this is equivalent to subtracting a constant from the log ratio. Thus, the transformation is


39

log2 ðR=GÞ ! log2 ðR=GÞ c ! log2 ðR=kGÞ: We would usually choose a constant that centers the distribution of M ¼ log2 R=G around 0, by setting c ¼ log2 k to be the mean (or median) of the M values. However, this method does not adequately deal with nonlinear intensity-dependent diVerences in the dye bias. A commonly used intensity-based adjustment (Yang et al., 2002b) is a loess scatterplot smoother (Cleveland and Devlin, 1988) fitted to the MA-plot. A loess smoother is a locally robust linear fit. It will not typically be aVected by the small fraction of diVerential genes, which would appear as outliers on an MA-plot. The loess adjustment is given by log2 ðR=GÞ ! log2 ðR=GÞ cðAÞ ! log2 ðR=kðAÞGÞ; where c(A) is the loess fit to the MA-plot. A span of f ¼ 0.4 is typically used for the loess curve. Although the global loess adjustment deals eVectively with intensity dependent diVerences in the dye bias, we sometimes observe diVerences resulting from spatial location on the array. This is most clearly illustrated in Fig. 2 where printtip–specific eVects are still evident after global loess normalization has been applied. An improvement to the global loess method is to use print-tip–specific loess smoothers (Yang et al., 2002b). The adjustment is now given by log2 ðR=GÞ ! log2 ðR=GÞ ci ðAÞ ! log2 ðR=ki ðAÞGÞ; where the ci(A) are loess smoothers fitted individually to data from each print tip. This method is recommended for routine use because it deals with both intensity-dependent eVects and subarray variation. There are many other features of the data that could be used in the normalization process. However, further normalization should be applied only when there is clear evidence from diagnostic plots indicating the need for such normalization. Unnecessary estimation of such eVects and trend removal may add noise to the data. One variation, after using print-tip loess normalization, is to further standardize the M values from each print tip to have the same scale. In particular, it is assumed that the variance of the M from print-tip group is given by ai2. pQ ffiffi each I We can robustly estimate âi ¼ MADi = i¼1 MADi where MAD is the median absolute deviation. The scale-normalized values for grid i are then given by Mi ¼ Mi =âi . This normalization is typically not required except in cases in which the arrays are extremely noisy. An example in which such normalization might be required is shown in Fig. 3, where we see that the variability of M from the fourth row of grids is significantly larger than that for the other grids. Applying the scale normalization removes this diVerence. Usually all the spots on the array are used in the normalization methods described previously, because this provides the most stability in terms of the

40

BOLSTAD et al.

FIG. 2. An MA-plot after global loess normalization (top) shows print tip eVects, which are eliminated using the print-tip specific normalization (bottom). (See Color Insert.)

number of spots and the flexibility to operate in a print-tip–specific manner. However, sometimes the expression profiles in the biological samples are more divergent than has been assumed in the cases mentioned. The previous strategies


41

FIG. 3. Box plots by print-tip group of M after print-tip normalization indicate a need for further scale normalization. (See Color Insert.)

can be employed if a suitable set of control spots that are known to be not diVerentially expressed are printed on the array. Ideally these would span the range of possible concentrations. One such method is to use a microarray sample pool (MSP) titration series in which the entire clone library is pooled and then titrated at diVerent concentrations. Because, in theory, all labeled cDNA sequences should hybridize to this series, it should not be subject to sample-specific biases. DiVerential genes should not bias a loess curve through the control spots. Sometimes it is useful to combine the MSP normalization and the printtip–specific normalization. It is suggested (Yang et al., 2002) that one take a weighted average of the print-tip–specific adjustment and the MSP normalization, in which the weights are dependent on the intensity. Define ci ðAÞ ¼ wðAÞ^g ðAÞ þ ð1 wðAÞÞ ^fi ðAÞ; where ^g ðAÞ is a loess curve fitted to spots from the MSP series, ^fi ðAÞ is a loess for print-tip group i, and w(A) is usually defined as the proportion of spots less than intensity A. The adjustment is then done as before. The idea is to increasingly use the MSP curve at higher intensities, where there are fewer spots and the print-tip–specific curves may be more unreliable.

42

BOLSTAD et al.

We have discussed normalization within slides, but sometimes there are large diVerences in scale when comparing data between slides. The advised procedure is to first normalize within slides using the methods previously discussed, and then consider scaling of M between slides, as described previously. This adjustment is needed so that the relative expression levels from one slide do not dominate the expression levels from others when averaging across replicate slides. It should be noted that there is a tradeoV between the gains achieved by scale normalization and any variability that may be introduced. Often this normalization will not be required. Software implementing these normalization methods for cDNA data may be found in the SMA package (Dudoit et al., 2002b) and downloaded from CRAN (http:==cran.r-project.org=).

B. NORMALIZATION FOR AFFYMETRIX ARRAYS There are two main approaches to normalization of AVymetrix GeneChip data. A recent paper (Bolstad et al., 2003) categorizes these into methods that use a baseline array and methods that are complete data methods. A complete data method does not use a baseline array, instead using data from all the chips to form the normalization. Examining box plots of raw probe intensities by array can often show the need for normalization. Such a plot is shown in Fig. 4A, for five arrays from part of a dilution series dataset (Gene Logic, 2001). The only diVerence between the arrays is the scanner that was used, yet the box plot shows quite diVerent levels of expression for each array. A number of normalization methods have been proposed. The simplest approach, scaling, is to scale each array so that all arrays in a dataset have the same mean intensity. Trimmed means are often used instead of means, and this is the method used by AVymetrix in the MAS 5.0 software (AVymetrix, 2001). i is the mean (trimmed) intensity for array i and K is the target mean intensity, If X i. The target intensity is often then array i is normalized by multiplying by K =X chosen to be the mean of one of the arrays. We would, thus, classify this method as a baseline method. Figure 4B shows the five arrays after scaling normalization. The scaling approach can be applied in a time-eYcient manner, but it does not adequately deal with possible nonlinear trends between arrays, as shown in Fig. 5. The AVymetrix HG-U113A chip has 100 normalization control probe sets that may be used for normalization in this context. These probe sets have been chosen because of their stability of expression across a wide range of tissues. Another approach is to choose a baseline array, then fit nonlinear relationships between the baseline array and each of the other arrays (in this context, we call these the treatment arrays). Such an approach fitting splines was suggested by


43

FIG. 4. Box plots of log-scale PM intensities for five arrays from diVerent scanners when (A) unnormalized, (B ) normalized by scaling, and (C ) quantile normalized.

Schadt et al. (2001) and used with a running median line (Li and Wong, 2001a,b). A rank invariant set of probes is chosen between the baseline and the treatment array. These probes are then used to fit the nonlinear relation. The curve is then used to map from the treatment array to the baseline array and defines the normalization. Several complete data adaptations of the MA-plot loess method for cDNA arrays have been proposed for normalizing AVymetrix arrays. The first is the cyclic loess method, in which arrays are normalized against each other in a pairwise fashion using a loess fit to an MA-plot. Unfortunately this requires O(N 2) MA-plot normalizations and so it is quite time consuming. A second adaptation of the MA-plot loess method is to transform the data using an orthonormal basis to give a set of contrasts (A˚strand, 2003). The normalization is applied to the transformed data. The data are then transformed back to the original basis. This method requires only O(N ) MA-plot normalizations and is, therefore, faster than the cyclic loess method. However, loess normalizations are slow for probe-intensity data. Typical implementations use only a subset of the probes to improve the processing time. Another complete data method is the quantile normalization method, in which the goal is to normalize arrays so that each array has a common intensity

44

BOLSTAD et al.

distribution. This method uses a simple non-parametric algorithm to quickly normalize a batch of arrays. In particular, averaging the quantiles of all the arrays in the set forms the reference distribution. Each array is then assigned the reference intensity distribution. The quantile normalization method is a specific case of the transformation xi0 ¼ F 1Gðxi Þ, where we estimate G by the empirical distribution of each array and F using the empirical distribution of the averaged sample quantiles. Extensions of the method could be implemented where F 1 and G are more smoothly estimated. However, we have found the current method to perform satisfactorily in practice. Figure 4C shows the arrays after quantile normalization. Figure 5 demonstrates that the quantile normalization deals adequately with nonlinear relationships in the data. In practice, this normalization can be carried out in a very time-eYcient manner. These methods were compared in a recent paper (Bolstad et al., 2003), where it was demonstrated that the scaling method was least eVective at reducing variability. Figure 6 illustrates this result using the RMA expression measure. These graphs show the ratios of the variance of the expression measure across five arrays plotted against mean expression for two diVerent normalization

FIG. 5. MA plots comparing PM probes from array 1 with array 5 when (A) unnormalized, (B ) normalized by scaling, and (C ) quantile normalized.


45

FIG. 6. Comparing the variances of the RMA expression measure across five arrays: (A) unnormalized against scaling, (B) unnormalized against quantile normalization. The curves are loess smoothers. Smaller values indicate greater reductions in variance.

methods. We see that scaling and quantile normalization both reduce the variability, with the greater reduction achieved by the quantile normalization method. The complete data methods were favored in Bolstad et al. (2003) because baseline methods can introduce peculiarities of the baseline array into the data for the treatment arrays. It was found that the quantile method was the fastest, with acceptable reductions in variance and little change in bias. Software implementing these normalization methods may be found in the aVy package (Gautier et al., 2003), which is part of the Bioconductor project (see www.bioconductor.org).

V. Expression Summaries for GeneChip Data

Figure 7 shows that background-corrected probe intensities follow an additive model on the log scale. For each probe set, we can write the following model: log2 ðBðPMij ÞÞ ¼ ai þ mj þ "ij

ð1Þ

for i ¼ 1, , I and j ¼ 1, , J. The quantity B(PMij) is the backgroundadjusted, normalized PM intensity, ai is a probe aYnity eVect, mj is a quantity

46

BOLSTAD et al.

FIG. 7. A plot of background-adjusted and normalized log PM intensities against concentration for a spike-in probe_set suggests an additive model. (See Color Insert.)

proportional to the amount of transcript on array j, and "ij is an independent identically distributed error term with mean 0. For identifiability of the parameters, we assume that the sum of the ai is 0 for each gene. Notice that this assumption translates to assuming that the AVymetrix technology has probes with expected intensities that, on average, are representative of the associated gene expression. Under this model an unbiased estimate of mj for each array j could be obtained using the average of the log2(B(PMij)) across the i ¼ 1, , I probes. This average can be used to estimate a simple expression measure. We can demonstrate empirically that this expression measure works well. If the errors are normally distributed, this estimate is according to various statistical criteria. However, many researchers (Li and Wong, 2001b) have observed that outliers (observations too extreme to occur under the normality assumption) are relatively common. For some arrays, the proportion of outliers is as high as 15%. This suggests that the aforementioned model should be fit using robust procedures. Median polish is a simple ad hoc procedure for fitting such a model robustly (Holder et al., 2002; Tukey, 1977). Irizarry et al. (2003a) demonstrate that the expression measure obtained using median polish provides estimates with comparable accuracy to and much better precision than the two leading expression measures, namely those obtained from MAS 5.0 and from dChip MBEI (Li and Wong, 2001a,b). Irizarry et al. (2003a,b) call this procedure the Robust Multi-Array Analysis (RMA).


47

The additive model lends itself to various practical extensions. For example, if we are comparing two populations of RNA species for which we have many technical replicates that we assume have the same expected RNA expression, we can write log2 ðBðPMijk ÞÞ ¼ ai þ mj þ "ijk for i ¼ 1, , I, j ¼ 1, , J, and k ¼ 1, , K. The estimate of mj would then be based on K times more data than RMA. If we had technical replicates instead of biological replicates, we could add a Zij term to the model, representing a random eVect (Chu et al., 2002).

VI. Quality Assessment

A. QUALITY ASSESSMENT FOR AFFYMETRIX GENECHIP EXPRESSION DATA Producing gene expression data using microarray technology is an elaborate process with many potential sources of variability. To maximize the scientific value of gene expression information derived from microarrays, we must make rigorous quality assessments throughout the process. Standard sample preparation protocols include a number of qualitative assessments meant to ensure that good quality RNA is used in the hybridization experiments. After hybridization and image processing, each microarray provides a wealth of information that can be used to assess the quality of the data. Recommended post-hybridization quality assessments include general image quality assessment and analysis of intensity measures of specialized probes (AVymetrix, 2001). In this section, we suggest some methods to assess data quality based on the analysis of residuals from the models fitted to estimate gene expression. Departures from quality standards may be attributable to various sources: RNA preparation, hybridization, chip scan, wash, image processing, or faulty chips. The eVects of departures from quality may be localized to a small area on a chip or may be uniformly distributed over an entire array, possibly aVecting numerous arrays. In most cases, departures from quality standards attributable to processing failures will be reflected by inflated residuals from fits to models such as Eq. 1. Residuals are, therefore, expected to provide useful information for data quality assessment. Quality assessment can be focused at diVerent levels: at the level of individual probes, of probe set summaries, of probe sets, or of chips. Fitting the probe level models robustly will automatically reduce the eVect of malfunctioning probes (cross-hybridizing or non-responding probes) on the estimated expression values, so diagnosis of dysfunctional probes is not required to obtain good expression

48

BOLSTAD et al.

summaries in this context. It may still be useful to identify dysfunctional probes (by means of residual analysis) for other purposes, for example, when seeking cross hybridizing probes or genes with alternative splicing. At the probe set summary level, residuals can be combined to produce estimated standard errors of probe set summaries. These can be used to derive weights for individual probe set summaries for downstream analysis. Careful analysis is required to ensure that these weights are beneficial to the downstream analysis. At the probe set level, residuals can be used to estimate the scale of the residual variance for each probe set or to produce a goodness-of-fit measure for the models fitted to each probe set. These goodness-of-fit measures can be used to derive appropriate weights for combining expression measures for diVerent probe sets. Our focus in this section is on obtaining an overall chip data quality index, which can be used to distinguish among chips of varying quality. We also suggest a way to visualize the distribution of residuals on a chip to help diagnose the source of departures from quality. Finally, we suggest some chip data quality assessment based on analysis of relative log expression. To illustrate the methodology we use a set of 19 cel files from the AVymetrix HG-U95A Spike-In Experiment, the 2353 series. The cel files and corresponding chips are identified by the letters A through T (note that the C experiment is missing from this series). DiVerential concentrations of 14 human transcripts were spiked in a common pool of pancreas mRNA. The behavior of the 14 spike-in probe sets does not play a role in overall chip data quality assessment. For the remainder of the probe sets, the arrays in this experiment constitute a set of technical replicates. The data are available from www.aVymetrix.com=analysis= download_center2.aVx. 1. Summarizing Residuals from Fits A simple way to summarize residuals for an entire chip is by means of their empirical distribution. Box plots provide a useful way to compare distributions for a large data set. The top panel of Fig. 8 shows box plots of residuals for each chip. In these, we note a slightly inflated variability in residuals for experiments A and P of the series. Note that the box plots of residuals will be centered close to zero (exactly zero for a least-squares fit), and that their distribution is approximately symmetrical about zero, so the diVerences between chips could eVectively be summarized by the 75th percentile of the chip residual distribution. Because our biggest concern is the eVect of low-quality probe data on expression summaries, it makes sense to combine residuals into estimated standard errors of expression estimates and summarize these at the chip level. To derive the standard errors, we assume that the models were fitted robustly by iteratively re-weighted least squares (IRLS). This fitting procedure can be used to obtain the various M estimators (Holland and Welsch, 1977) as well as the


49

FIG. 8. Box plots of individual probe residuals by chip in the AVymetrix HG-U95A Latin Square experiment, series 2353 (top panel). Box plots of normalized unscaled probe set standard errors (bottom panel).

maximum likelihood fit assuming t error distributions (Lange et al., 1989). IRLS estimates of parameters are obtained as weighted least-squares estimates. The weights are updated at each step by applying a transformation to the residuals from the previous fit. The choice of weight function depends on the particular M estimator desired (Huber, 1972, 1981). Applying the M-estimation techniques to the model specified in Eq. 1 we get m ^j ¼

I X

wij log2 ðBðPMij ÞÞ

i¼1

vffiffiffiffiffiffiffiffiffiffiffiffiffi u I uX wij SEð^ mj Þ ¼ ^=t i¼1

For each chip, indexed by j, we thus get both an expression value and a standard error estimate. These vectors can be summarized at the chip level to obtain an index of quality for each chip.

50

BOLSTAD et al.

The standard errors of estimated expression within a chip form a heterogeneous set by virtue of the fact that the value of ^ varies from probe set to probe set. We can remove this source of heterogeneity by using unscaled standard errors to assess the precision of the estimated expressions. Removing the ^ factor from the standard error does not aVect the assessed relative precision of estimated expressions across chips, which is our main interest. There still remains some heterogeneity in the unscaled standard errors across probe sets, because the eVective number of probes used in estimating the expression for chip j may vary from probe set to probe set. To remove this source of heterogeneity, we can normalize the unscaled standard errors by dividing by the average or median standard error across a set of chips. The bottom panel of Fig. 8 shows box plots of normalized unscaled standard errors (NUSE) of probe set summaries for each chip. In these, we see that the NUSE of probe set summaries are quite sensitive to deviations in assessed expression variability, with experiments A and P clearly standing out from the rest. We can summarize the batches of NUSEs for each chip by the median, for example, and this value can be used as a chip data quality index. NUSE values fluctuate around 1.0. A median value of 1.05 for a chip may be interpreted as a 5% average loss in precision. The question that naturally arises is what is a good range for this quality index? This is a diYcult question and may not have a single answer. The answer depends on the specific application and the various costs involved. For a specific application, one could judge at what level of quality including a chip in an analysis becomes disruptive, by a ‘‘leave-one-out’’ comparison, for example. For a carefully performed analysis, such as one that combines expression measures robustly, the answer may be that including a small number of lower quality chips in an analysis will be harmless in most cases. This does not mean that detecting departures from quality standards that have small eVects on downstream analysis is not useful. For example, in a large-scale production environment, having a sensitive tool to monitor quality may help detect and correct problems before they have an impact on expression measures and critical results of downstream analysis. AVymetrix recommends a number of quality checks to be performed after the analysis of the raw data by the AVymetrix MAS 5.0 software (AVymetrix, 2001). Some are qualitative and involve judging the overall quality of the chip image by visual inspection, whereas others are quantitative. Of the quantitative assessments, some involve examining the expression level of special-purpose probe sets—the hybridization controls, poly(A) controls, and housekeeping genes. Other quantitative assessments are based on a more comprehensive summary of expression and signal level on a chip. Figure 9 examines the relationship between the chip data quality index derived from the residuals (the median NUSE) and three of the quantitative quality assessment measures recommended by AVymetrix: Scaling Factor (target ¼ 500), RawQ, and Percent Present calls.


51

FIG. 9. Comparing the median NUSE with AVymetrix quality standards.

Other recommended measures that summarize the probe intensities are highly correlated with these and do not provide much additional information. Figure 9 demonstrates that the median NUSE is highly sensitive to departures from quality standards. In sets of chips varying over a wide range of quality levels, we find that the index of quality based on assessed variability of expression, the median NUSE, is highly correlated with some of the recommended quality assessment measures. We believe that assessed variability provides a better basis for making decisions to rerun an experiment or exclude a chip from an analysis set, whereas other measures are potentially more useful at identifying the source of a problem. Analyzing expression levels of specialized spike-ins or housekeeping genes for quality assessment purposes poses a special challenge. Because there are only a few spike-in probe sets, measures derived from them tend to be noisy, requiring substantial departures from quality standards for a problem to be detectible. These measures may nonetheless be useful for tracking the source of departures from quality standards that are more easily detectible by other means. 2. Spatial Analysis of Residuals Residuals can be imaged in a manner similar to the way probe cell intensities are typically imaged. It is common practice to assess chip quality by visually inspecting probe-intensity images. Artifacts like bright or dim spots, scratches, or uneven brightness can be identified this way. Because cell intensities within a chip vary over a wide range and most of this variation comes from the fixed part of the model (Eq. 1), the imaged residuals are expected to provide increased resolution for visually detecting image artifacts.

52

BOLSTAD et al.

Spatial patterns of residuals can be profitably examined when seeking an explanation for elevated standard errors of expression estimates on a chip. Spatial patterns may provide evidence of SAPE residue caused by poor wash, uneven hybridization, bubbles, or other local artifacts. A uniform distribution of elevated residuals is another possibility, indicating a diVerent kind of problem with the assay. Note that spatial patterns of residuals may sometimes detect artifacts that are not detectible at the level of gene expression variability. Such artifacts would probably not play a role in accepting or rejecting a chip for analysis but may be valuable in monitoring a chip production process. Spatial patterns of residuals themselves have proven diYcult to visualize. The challenge is to capture spatial patterns of a dense scatter of numbers having both sign and amplitude. Each of these features, sign and amplitude, are readily visualized separately. The weights used in the IRLS fit can be imaged to capture the magnitude of the residuals, highlighting residuals that deviate substantially from an overall estimated scale. The sign of residuals can also be imaged and such images add to the pseudo-images of the weights by telling us whether a region of outlying residuals corresponds to a bright or a dim region on the chip. In addition, the image of the sign of residuals will capture small eVects that are not detectible in the weights, which are insensitive to small deviations of the residuals from their expected value of zero. In Fig. 10 the log intensities (top row), probe weights (middle row), and the residuals (bottom row) are imaged for three chips: two with elevated assessed variability of expression, A and P, and one with average assessed variability of expression, H. Low probe weights, corresponding to residuals with high absolute values, appear as the intense green spots on the chip pseudo-image of the weights (middle row). Clusters of probes with high absolute residuals are clearly visible for chips A and P. The patterns are also discernible in the pseudo-images of log intensities, but not nearly as clearly. Clusters of positive residuals corresponding to bright areas on the chip are clearly visible in the images of the sign of the residuals (bottom row). Determining the source of variability accounting for specific patterns of high absolute residuals—local versus global, and possible trends—is an open question. Software for producing these images is available as part of the aVyPLM library from www.bioconductor.org. 3. Quality Assessment Based on Relative Expression The standard error estimates provide a measure of expression summary variability that is independent of expression level. We can also gauge variability of expression measures by summarizing the distribution of relative log expressions. To compute relative log expression values, we use a virtual median chip constructed by taking, for each probe set, the median log expression from a reference set of chips. We can summarize a vector of relative expression by a measure of bias: median(RE), a measure of variability: IQR(RE), or total error:


53

FIG. 10. (Top row) Image of log probe intensities. (Middle row) Image of probe weights. Intense green areas denote high concentration of large absolute residuals. (Bottom row) Image of residuals. Red represents areas of highly positive residuals, and blue represents areas of highly negative residuals. (See Color Insert.)

IQR þ |Bias|. These summaries are sensitive to technical sources of variability that are large compared to biological variation. This assessment will be highly correlated with an assessment based on estimated standard errors of probe set expressions, but it has the advantage of being derived from the expression estimates alone (as opposed to probe-level residuals). Figure 11 shows box plots of relative log expressions for the 2353 series. We can readily see the elevated variability in chips A and P, as was assessed by the residual analysis. In addition, we note a downward bias in the expressions for chip P. As the chips being compared here were hybridized with a common source of RNA, the relative log expression should be zero for all non spike-in probe sets, and the diVerences in variability between chips can therefore be attributed to technical or processing variability. When comparing chips with diVerent sources of RNA, the variability in relative log expression will be inflated by real biological variability. This is not seen as a serious handicap in the use of relative log expression to assess data quality, because the technical variability that we are interested in is typically greater than the biological variability.

54

BOLSTAD et al.

FIG. 11. Box plots of each chip’s expression levels relative to a virtual reference chip. The higher variability of chips A and P represents departures from quality.

B. QUALITY ASSESSMENT FOR CDNA MICROARRAY EXPERIMENTS The quality of the expression data derived from cDNA microarray experiments depends on experimental and production factors similar to those aVecting oligonucleotide microarrays. The extraction of gene expression information from a scanned array requires a complicated image analysis process. This process is an additional source of potential variability. Yang et al. (2002a) discuss image analysis for spotted arrays in detail. As a by-product of the image analysis step, a number of spot characteristics are generated: spot size and shape, spot intensity, and background intensity. These can be used as quality indicators (Wang et al., 2001). When some clones are spotted at several locations on the array, the repeated measurements for clones can be combined to obtain some assessment of the reproducibility of the measurements, just as probes within probe sets are used to measure reproducibility with the oligonucleotide microarrays. Jenssen et al. (2002) and Tseng et al. (2001) discuss the use of multiply spotted clones in quality assessment. Ritchie et al. (2003) demonstrate that spot-quality measures are correlated to spot reproducibility for the multiply spotted clones and suggest that this relationship could be exploited to derive spot weights to be used in gene-wise regressions.

VII. Detection of Absolute Gene Expression

The problem of classifying genes as present or absent in a given sample has been largely overlooked in the literature. The only widely used detection call for oligonucleotide microarrays is the one implemented in the MAS software developed by AVymetrix (2001). Although the detection of absolute expression is


55

not generally regarded as important as that of diVerential expression, it has definite biological relevance in some circumstances. For example, a biologist studying gene expression in neural stem cells may want to know which genes go from being absent to being present at a particular time, and vice versa. A. THE AFFYMETRIX PRESENCE=ABSENCE ALGORITHM The AVymetrix MAS 5.0 software makes a detection call for each probe set by defining a discrimination score Ri ¼

PMi MMi ; PMi þ MMi

where PMi is the perfect match intensity of the i’th probe in the probe set, and MMi is the corresponding mismatch intensity. This is done for the nonsaturated probe pairs. A one-sided Wilcoxon signed-rank test is then used to test H0 : medianðRi Þ ¼ ; H1 : medianðRi Þ > where H0 is the null hypothesis and H1 is the alternate. is a small positive number, tunable by the user, and set to a default of 0.015. AVymetrix has determined this value as being one that minimizes the number of incorrect calls without sacrificing sensitivity. The p value from the signed-rank test is used as a determinant of gene presence or absence. MAS 5.0 actually uses two user-configurable significance levels 1 and 2, such that 0 < 1 < 2 < 0.5. Probe sets are called present if p < 1, absent if p 2, and marginal (no call) if 1 p < 2. The defaults in MAS 5.0 are 1 ¼ 0.04, 2 ¼ 0.06. These are found to be optimal (based on analyses of spikein data) for the default value of . More details about the AVymetrix presence= absence methodology can be found in Liu et al. (2001) and Liu et al. (2002). B. ALTERNATIVE METHODS Zhou and Abagyan (2002) have developed an algorithm to calculate expression summaries that use only the PM intensities. As a side eVect of their procedure, they perform a detection call. The 5% lowest intensity probe sets are designated as background, and the empirical cumulative distribution of the background intensities on the linear scale, B(I ), is then calculated. For each probe set, they calculate the empirical cumulative distribution of the probe signals, Sk(I ), and compare this to B(I ). The authors’ claim is that genes that are absent will tend to have integral distributions that are close to the background distribution. They, therefore, compare each Sk to B using a Kolmogorov-Smirnov test. There is no recommendation

56

BOLSTAD et al.

for an appropriate threshold on the p value from the K-S test for calling presence or absence. Instead, the authors state that ‘‘those signal sets that can be easily explained by noise are assigned a log10p value closer to zero.’’ Rubinstein and Speed (2003) have approached the problem of transcript detection using several novel methods. They define three broad classes of detection algorithms: thresholding rank sums of probe-pair summaries, thresholding robust averages of probe-pair summaries, and thresholding expressionlevel estimates. Possible probe-pair summaries include log(PMi =MMi ) and (PMi MMi Þ=ðPMi þ MMi Þ. The latter is the summary used by the MAS software. The PM and MM values may or may not be background corrected and normalized across chips. The authors have developed a framework for evaluating diVerent detection algorithms, using the ROC (Receiver Operating Characteristic) Convex Hull method. Under this scheme, the cost of misclassification is defined as follows: Pð pÞ ð1 TPRÞ CðN ; pÞ þ PðnÞ FPR CðP; nÞ;

FIG. 12. Receiver Operating Characteristic (ROC) curves for some of the detection algorithms discussed in the text. NRALR is the normalized robust average of log ratios, the NRSLR is the normalized rank sum of log ratios, and the RMA is the expression-level estimate obtained using RMA. The gray polygon is the convex hull of the ROC curves and represents the best possible classifier.


57

where P( p) is the prior probability of an example being positive and P (n) of being negative, TPR and FPR are the true and false positive rates, and C(P, n) and C(N, p) are the costs of false negatives and false positives, respectively. Requiring that this cost be minimized allows one to evaluate the optimality of detection algorithms over a particular range of false-negative and false-positive costs, given a set of ROC curves for those algorithms. Rubinstein and Speed find that their normalized robust average of log ratios (NRALR) and normalized rank sum of log ratios (NRSLR) outperform the MAS 5.0 algorithm for a wide range of costs, while thresholding on either expression level derived from RMA (Irizarry et al., 2003a) or the MAS 5.0 signal estimate do not perform as well. ROC curves for these five representative algorithms are shown in Figure 12. References

AVymetrix (2001). AVymetrix Microarray Suite Users Guide, Version 5. Calif, Santa Clara. A˚strand, M. (2003). Contrast normalization of oligonucleotide arrays. J. Comput. Biol. 10, 95–102. Bolstad, B. M., Irizarry, R. A., Astrand, M., and Speed, T. P. (2003). A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19, 185–193. Box, G. E. P., Hunter, W. G., and Hunter, J. S. (1978). Statistics for experimenters: An introduction to design, data analysis, and model building. Wiley, New York. Chu, T. M., Weir, B., and Wolfinger, R. (2002). A systematic statistical linear modeling approach to oligonucleotide array experiments. Mathematical Biosci. 176, 35–51. Cleveland, W. S., and Devlin, S. J. (1988). Locally-weighted regression: An approach to regression analysis by local fitting. J. Am. Statistical Assoc. 83, 596–610. Cobb, G. W. (1998). Introduction to design and analysis of experiments. Springer, New York. Cox, D. R. (1992). Planning of experiments. Wiley, New York. Dudoit, S., Yang, Y. H., Speed, T. P., and Callow, M. J. (2002a). Statistical methods for identifying diVerentially expressed genes in replicated cDNA microarray experiments. Statistica Sinica 12, 111–139. Dudoit, S., Yang, Y. H., and Bolstad, B. M. (2002b). Using R for the analysis of DNA microarray data. R. News 2, 24–32. Gautier, L., Cope, L. M., Bolstad, B. M., and Irizarry, R. A. (2003). Analysis of AVymetrix GeneChip data at the probe level. Bioinformatics 20(3), 307–315. Ge, Y., Dudoit, S., and Speed, T. P. (2003). Resampling-based multiple testing for microarray data analysis [with Discussion]. Test 12, 1–77. Gene Logic (2001). Dilution series data available at: www.genelogic.com=media=studies=dilution.cfm. Holder, D., Pikounis, V., Raubertas, R., Svetnik, V., and Soper, K. (2002). Statistical analysis of high density oligonucleotide arrays: A SAFER approach: Proceedings of the American Statistical Association. Atlanta, Georgia. Holland, P. W., and Welsch, R. E. (1977). Robust regression using iteratively reweighted least-squares. Comm. Stat. Theory Methods A6(9), 813–827. Huber, P. J. (1972). Robust statistics: A review. Ann. Mathematical Stat. 43, 1041–1067. Huber, P. J. (1981). Robust statistics. John Wiley & Sons, New York. Irizarry, R. A., Bolstad, B. M., Collin, F., Cope, L. M., Hobbs, B, and Speed, T. P. (2003a). Summaries of AVymetrix GeneChip probe level data. Nucl. Acids Res. 31, e15.

58

BOLSTAD et al.

Irizarry, R. A., Hobbs, B., Collin, F., Beazer-Barclay, Y. D., Antonellis, K. J., Scherf, U., and Speed, T. P. (2003b). Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4, 249–264. Jenssen, T., Langaas, M., Kuo, W. P., Smith-Sorensen, B., Myklebost, O., and Hovig, E. (2002). Analysis of repeatability in spotted cDNA microarrays. Nucl. Acids Res. 30, 3235–3244. Lange, K. L., Little, R. J. A., and Taylor, J. M. G. (1989). Robust statistical modeling using the t distribution. J. Am. Stat. Assoc. 84, 881–896. Li, C., and Wong, W. H. (2001a). Model-based analysis of oligonucleotide arrays: Model validation, design issues and standard error application. Genome Biol. 2, 0032.1–0032.11. Li, C., and Wong, W. H. (2001b). Model-based analysis of oligonucleotide arrays: Expression index computation and outlier detection. Proc. Natl. Acad. Sci. USA 98, 31–36. Liu, W. M., Mei, R., Bartell, D. M., Di, X., Webster, T. A., and Ryder, T. (2001). Rank-based algorithms for analysis of microarrays. Proc. SPIE 4266, 56–67. Liu, W. M., Mei, R., Di, X., Ryder, T. B., Hubbell, E., Dee, S., Webster, T. A., Harrington, C. A., Ho, M. H., Baid, J., and Smeekens, S. P. (2002). Analysis of high density expression microarrays with signed-rank call algorithms. Bioinformatics 18, 1593–1599. Lo¨nnstedt, I., and Speed, T. P. (2002). Replicated microarray data. Statistica Sinica 12, 31–46. Montgomery, D. C. (2000). Design and analysis of experiments, 5th ed. Wiley, New York. Mootha, V. K., Lindgren, C. M., Eriksson, K. F., Subramanian, A., Sihag, S., Lehar, J., Puigserver, P., Carlsson, E., Ridderstrale, M., Laurila, E., Houstis, N., Daly, M. J., Patterson, N., Mesirov, J. P., Golub, T. R., Tamayo, P., Spiegelman, B., Lander, E. S., Hirschhorn, J. N., Altshuler, D., and Groop, L. C. (2003). PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet. 34, 267–273. Pearson, E. S., and Hartley, H. O. (1962). Biometrika Tables for Statisticians. Cambridge University Press, Cambridge. Ritchie, M., Smyth, G. K., Diyagama, D., Val Laar, R., Holloway, A., and Speed, T. P. (2003). Quality measures for cDNA microarray experiments. Royal Statistical Society posten presentation. Rubinstein, B. I. P., and Speed, T. P. (2005). Detecting gene expression with oligonucleotide microarrays. (Unpublished manuscript). Schadt, E. E., Li, C., Ellis, B., and Wong, W. H. (2001). Feature extraction and normalization algorithms for high-density oligonucleotide gene expression array data. J. Cell. Biochem. Suppl. 37, 120–125. Speed, T., Ed. (2003). Statistical analysis of gene expression microarray data. Boca Raton, FL. Chapman and Hall CRC Press. Tseng, G. C., Oh, M., Rohlin, L., Liao, J. C., and Wong, W. H. (2001). Issues in cDNA microarray analysis: Quality filtering, channel normalization, models of variations and assessment of gene eVects. Nucl. Acids Res. 29, 2549–2557. Tukey, J. W. (1977). Exploratory data analysis. Addison-Wesley, Reading, Mass. Wang, X., Soumitra, G., and Guo, S. (2001). Quantitative quality control in microarray image processing and data acquisition. Nucl. Acids Res. 29, 2549–2557. Yang, Y. H., Buckley, M. J., Dudoit, S., and Speed, T. P. (2002a). Comparison of methods for image analysis on cDNA microarray data. J. Comput. Graph. Stat. 11, 108–136. Yang, Y. H., Dudoit, S., Luu, P., Lin, D. M., Peng, V., Ngai, J., and Speed, T. P. (2002b). Normalization for cDNA microarray data: A robust composite method addressing single and multiple slide systematic variation. Nucl. Acids Res. 30, e1. Yang, Y. H., and Speed, T. P. (2002). Design issues for cDNA microarray experiments. Nat. Rev. Genet. 3, 579–588. Zhou, Y., and Abagyan, R. (2002). Match-only Integral Distribution (MOID) algorithm for high-density oligonucleotide array analysis. BMC Bioinform. 3, 3.

BRAIN GENE EXPRESSION: GENOMICS AND GENETICS

Elissa J. Chesler and Robert W. Williams Department of Anatomy and Neurobiology Center for Genomics and Bioinformatics University of Tennessee Health Science Center Memphis, Tennessee 38163

I. Introduction A. Involvement of Transcription Regulation in Diverse CNS Processes B. Various Processes are Involved in the Modification of Gene Expression C. Genetic and Genomic Approach to Understanding Transcription Regulation II. Transcript Abundance as a Complex Trait A. Strain DiVerences and Genetic Variation in Transcript Abundance B. Other Endogenous Sources of Variance in Brain Gene Expression III. Genetic Dissection of Transcription Regulation A. Distribution of Transcript Abundance and Heritability Estimation of Microarray Measures B. Experimental Crosses for Genetic Analysis C. Detecting Major QTLs D. Gene–Gene Interactions and Multiple QTL Models E. Gene–Environment Interactions F. Identification of QTL Polymorphisms IV. How Many QTLs Are There? Statistical Control in Two Dimensions A. QTL Mapping and Empirical Significance Tests B. Multiple Testing and Microarray Analysis C. Permuting the Transcriptome Analysis V. Probe Versus Probe Set–Level Phenotypes A. Probe Variation B. Genetic Variation and Probe Level EVects C. Probe Redundancy and Overlap D. Implications of Probe Level Variation and Normalization Techniques for QTL Analysis VI. Transcription Regulatory Networks VII. Functional Correlates of Brain Transcriptional Activity A. Genetic Correlations B. Cliques, Clusters, and Genetic Association Networks C. Incorporation of Other Traits into the Transcriptional Network VIII. Conclusions: Building a Model of Brain Function from Multipoint, Multi-tissue Gene Expressions, and Phenomic Observation A. An Expanding Notion of the Neurochemical Pathway B. A Discovery-Based Approach to Dissection of Gene Regulatory Relations C. A Relational Model for Data Integration References


59


60

CHESLER AND WILLIAMS

I. Introduction

Gene expression in the central nervous system (CNS) is a dynamic and complex process that modulates synaptic eYcacy, signal transduction, and numerous other cell functions involved in information processing and cellular homeostasis. As in other tissues, CNS abnormalities and disease also involve compensatory and sometimes maladaptive changes in gene expression in diVerent CNS cells and regions. Little is known about gene regulatory networks in the brain or any other tissue, but the prospects of making rapid progress are now excellent in large part because of high-throughput assays of messenger RNA (mRNA) species and quantities. From massive transcriptome data sets, we can now begin to assemble networks of co-regulated genes. In some cases it is even practical to step up from the level of gene expression to higher order phenotypes such as neuropharmacological responses and even behavior variation. Some of the most interesting diVerences in gene expression are those among individuals. From a genetics perspective, individual diVerences within a species are particularly useful because they can be traced to sequence variants in gene loci using conventional gene mapping methods. By exploiting individual genetic diVerences, a complex gene regulatory network can be defined, dissected, and perturbed experimentally. It is also possible to locate and identify key genes that generate downstream diVerences in mRNA abundance. This is a key topic of this chapter. Some of these key modulators of mRNA abundance may prove useful targets for disease prevention and treatment.

A. Involvement of Transcription Regulation in Diverse CNS Processes Most major nervous system processes are modulated directly or indirectly by variation in gene transcription regulation. Developmental processes are regulated by combinatoric expression of sets of transcription factors. These determine the rostral and caudal axes, cell fates, cell migration, and connections among neurons and among glial cells. Transcription factors, including the well-known immediate early genes, are also involved in the signal transduction cascade that follows neural activation (Morgan and Curran, 1989). These genes play a vital role in inducing long-term synaptic changes that occur during plasticity-dependent processes of learning (Clayton, 2000), including enhancement of neural transmission and structural modification of the dendritic processes themselves (Rose, 1991). Cascades of transcriptional activation also follow injury, cell death, and oxidative stress.

GENETICS OF CNS GENE EXPRESSION

61

B. Various Processes Are Involved in the Modification of Gene Expression Although the most obvious regulators of transcription are the classical transcription factors, the activity of many other proteins and transcripts can also regulate expression directly or indirectly. For example, genetic variation in channel conductance can influence the subsequent activation of transcription in dependent processes, and in this context the ion channel can be identified as a distal genetic modifier of transcription. Variation in receptor-associated proteins and other factors not commonly viewed as transcription regulatory genes also indirectly aVects mRNA levels. Exogenous factors such as stress, reproductive cues, pathogens, heat, and other environmental factors can also have rapid and dramatic eVects on gene expression through many genetically variable transduction processes. It is, therefore, not too surprising that in a study of the major transcription regulatory loci in the yeast transcriptome, few of the loci that regulated target transcripts contained transcription factors (Yvert et al., 2003). Single-gene approaches to characterize transcription control often start with a prior hypothesis and attempt to confirm the identity of a small handful of downstream genes using knockout mice or other gene perturbation experiments. In light of the many modifiers of gene expression, a complementary, broad, and assumption-free view of transcriptional control is also highly rewarding. An approach that considers the full diversity of transcription modifiers can elucidate numerous unexpected regulators of gene expression.

C. Genetic and Genomic Approach to Understanding Transcription Regulation Microarray technology has greatly simplified the genomic analysis of expression patterns in the CNS, making it possible to compare and analyze both global and local patterns of gene expression (Geschwind, 2000; Pavlidis and Nobel, 2001; Sandberg et al., 2000; Zhao et al., 2001; Zirlinger et al., 2001). For the most part, these studies have generated lists of genes that are expressed, upregulated, or downregulated under particular conditions. The regulatory mechanisms behind patterns of variation remain unspecified. Analysis of transcriptional diVerences can begin to expose the common regulatory mechanisms. This is almost always done by comparing experimental and control samples or more rarely, a time series, but it is also entirely feasible to exploit the very substantial natural variation in gene expression among a population of genetically heterogeneous individuals. This approach gives us an opportunity to identify the actual regulatory

62


genes and gene loci that control gene expression in a normal setting in which all of the ‘‘experimental’’ variation is actually attributable to a multitude of common polymorphisms. Gene polymorphisms that influence the level of expression of one of more gene products trigger a cascade of downstream eVects and will alter expression of several pathways. Multiple polymorphic loci can interact in complex and nonlinear ways. Locating and identifying the set of critical polymorphisms that influence a gene regulatory network is still diYcult but can be achieved in principle and in practice by associating variation in gene expression with genotype markers of known location, an approach that has been successfully deployed in yeast (Brem et al., 2002), mouse (Chesler et al., 2003; Schadt et al., 2003; Wang et al., 2003), and even in humans (Schadt et al., 2003). The genetic approach can also be used to identify associations between gene expression changes and behavior and other brain-related phenotypes. This approach facilitates discovery of the causal interactions among gene expression traits and the higher order functional traits (Chesler et al., 2003, 2004).

II. Transcript Abundance as a Complex Trait

In contrast to mendelian traits, which are regulated by single major mutations or variants, variation in gene expression typically has a complex basis and is influenced by multiple factors, including polymorphic upstream modifier genes, environmental factors, and interactions between genetic and environmental factors. These factors cause gene expression to vary along a continuum (Fig. 1). Dissection of complex phenotypes of this type involves estimating and subdividing the numerous sources of variance (e.g., the Complex Trait Consortium, 2003). The analysis of complex traits typically begins by estimating heritable variation. This is now often performed by comparing inbred lines in the mouse, or by comparing monozygotic and dizygotic human twins. Of course, gene expression is a comparatively simple trait, and one that can now be measured with reasonable accuracy simultaneously across the entire genome. For example, using AVymetrix arrays, we can compute the heritability of each of 45,000 transcripts. It is important to point out that typical microarray-based assays of expression generally measure the steady-state relative abundance of mRNAs; this quantity results from the net eVect of production, processing, and degradation of mRNA in the tissue sample. Intracellular distribution of message and rapid fluctuations at the cell level are still relatively inaccessible.


63

Fig. 1. Frequency histograms showing the distribution of strain mean transcript abundance (log base 2) for several representative transcripts. In most cases transcript abundance is approximately normal. All arrays are log transformed and standardized to a mean of 8 and standard deviation of 2. Thus, each unit change corresponds to approximately a twofold diVerence in expression as assayed using the MAS 5 data transform. Gene names corresponding to the gene symbols (followed by the last two digits of the probe set in the event of multiple probe sets) are as follows: Cacna1e (voltage-dependent calcium channel alpha 1E, probe set 100337_at and 100338_s_at), Mc1r (melanocortin 1 receptor, probe set 101161_at), Grin2b (NMDA 2B ionotropic glutamate receptor subunit, probe set 101312_at), Htr1a (5-hydroxytryptamine receptor 1a, probe set 101140_at), Veli3 (vertebrate homolog of C. elegans Lin7 type 3, probe sets 103370_at and 161086_at), Apoe (apolipoprotein E, probe sets 161321_i_at and 95356_at), App (amyloid beta (A4) precursor protein, 93063_at), Risc (retinoid-inducible serine carboxypeptidase, probe set 98042_at), and Rpo2tc1 (RNA polymerase II transcriptional coactivator, probe set 01980_at).

A. Strain Differences and Genetic Variation in Transcript Abundance Although early microarray analysis of gene expression in the brain was primarily aimed at determining the presence or absence of genes in particular regions, these early studies also highlighted the influence of genetic variation (Sandberg et al., 2000). However, because of limitations in sample size and the

64


statistical approach, variation in gene expression resulting from genetic factors was underestimated in these early studies. Fewer than 1% of transcripts were reported to be diVerentially expressed between strains C57BL/6J and 129/SvTac (Geschwind, 2000). The magnitude of this underestimate was noted in a reanalysis by Pavlidis and Nobel (2001) using analysis of variance (ANOVA) methods to simultaneously examine strain- and region-specific eVects on gene expression, but they still found diVerences in only 2% of transcripts. This estimate was based on data collected in a two-strain comparison with modest numbers of biological replicates. In a more genetically diverse 50-array subset of our large data set containing males and females from several strains (C57BL/6J, DBA/2J, CXB5, CXB10, and CXB12) and five tissue types, we found that there are at least 314 diVerentially expressed genes between strains using Benjamini and Hochberg’s (1995) false discovery rate—also equivalent to roughly 2.5%. However, other approaches to multiple testing (Storey and Tibshirani, 2003), which take into account the actual distribution of results, provide estimates that there are truepositive strain diVerences for roughly 50% of all transcripts, and 700 strain diVerences in mRNA abundance can be detected with false discovery rates of less than 5%. In a preliminary analysis of strain diVerences, we find major transcript variation between strains in the hippocampus, forebrain, olfactory bulb, and cerebellum. For forebrain gene transcription, the comparison of the C57BL/6J and DBA/2J strains (Fig. 2) reveals marked diVerences between Gnb1, Pam, and Kcnj9 (all upregulated in C57BL/6J), and Rpl26, Rps28, Comt, and Viaat (all upregulated in DBA/2J). From this brief summary, we can conclude that the number of diVerentially regulated transcripts is simply a function of statistical power and the asymptotic limit will approach 100%. The key issue is the functional significance of variation. Do we consider 1%, 10%, 50%, or 100% diVerences in expression level interesting? This will clearly depend on the type of gene product, its position in molecular networks, and the potential of other related transcripts to assume joint functional responsibility.

B. Other Endogenous Sources of Variance in Brain Gene Expression Regional variation, the main subject of the studies cited earlier, is a major source of variance in brain gene expression (Pavlidis and Nobel, 2001; Sandberg et al., 2000). Although the early analyses were performed in small sample sizes with limited genotype diversity, as previously mentioned in the discussion of strain diVerences, these studies reported that 1–5% of genes studied were diVerentially expressed between the brain regions. In our analysis of 50 microarrays, at a false discovery rate of 5%, 25% of transcripts were diVerentially expressed between brain regions, although some of these eVects varied by sex and genotype


65

Fig. 2. Comparison plot of gene expression between C57BL/6J and DBA/2J strains in the forebrain. Transcripts that have higher expression in the C57BL/6J mice are shown in gray, and transcripts with higher expression in DBA/2J mice are shown in black (MAS 5 transform).

(Shou et al., 2002). The tremendous regional variation of gene expression in the brain implies that there are region-specific controllers of expression. Indeed, diVerent regions of the brain including the cerebellum and forebrain, and diVerent tissues including hematopoietic stem cells do have diVerent modulatory loci regulating expression of the same transcript (see www.webqtl.org). Sex diVerences in gene expression in the brain tend to be much more subtle than variation between tissues. However, diVerentially expressed transcripts have also been identified between the sexes. Xist and Dby are two truly dimorphic transcripts, with non-overlapping distributions of gene expression between males and females (Chesler et al., 2002). Expression diVerences between these genes are so reliably detectable in the brain that they can be used to retrospectively classify microarrays by sex. Age-related changes in gene expression are also apparent. In our analysis, we have sampled a limited range of ages within typical adulthood spanning 8 weeks to 1 year of age. Among females of this age range, only a small handful of transcripts vary significantly by age (E. J. Chesler, L. Lu, and R. W. Williams, 2002, unpublished observations). These include Prm3, Mlp, Pawr, Gbp2, Ckap2, Car4, Maprel, F9, Serpin f 2, Kcnk1, Sfxn1, Traf5, Gpr97, Apoh, a list rich in cell death, injury, and disease-related genes. During more dynamic periods of development and aging, dramatic gene expression diVerences are likely to be detected.

66


III. Genetic Dissection of Transcription Regulation

Because transcript abundance is a polygenic trait determined by the action of many other genes and both internal and external environmental factors, complex trait analysis methods can be profitably applied to study the genetic control of gene expression (Cheung, 2002; Jansen, 2001). Gene transcription regulatory loci have been mapped for the control of individual transcripts in the brain. For example, Janowsky et al. (2001) were able to identify regulatory loci that influence dopamine transporter density using standard radioligand methods of protein binding and were able to associate this locus with several behaviors. In an early mapping study, a single-gene mRNA-based approach was used to map transcription of endogenous mouse mammary tumor viruses (Traina-Dorge et al., 1985). Massively parallel analysis technology has allowed simultaneous mapping of many thousands of gene expression or proteomic traits. In the first mouse study of this kind, Klose et al. (2002) mapped genetic loci regulating variation in proteins separated via two-dimensional gel electrophoresis. The combination of microarray and complex trait analysis was first performed in yeast by Brem et al. (2002) and later extended to the mouse liver (Schadt et al., 2003), brain (Chesler et al., 2003, 2004; Wang et al., 2003), and hematopoietic stem cells (DeHaan et al., 2003). Complex trait analysis or quantitative trait locus (QTL) mapping involves associations of known genetic markers with phenotypical values in the progeny of a segregating cross. Several QTL detection methods have been developed, all essentially estimating the position and eVect size of upstream polymorphic loci on mRNA expression. The estimator is computed across the genome, allowing identification of loci without prior hypotheses of their identity. Mathematical approaches to quantitative trait locus mapping are becoming increasingly sophisticated, from the earliest single marker to trait association analyses to slightly more complex methods that allow detection of multiple QTLs for multiple traits (Doerge, 2002). The method has been widely used in neuroscience, particularly through application to behavioral traits (reviewed by Flint, 2003).

A. Distribution of Transcript Abundance and Heritability Estimation of Microarray Measures Phenotypical variation is typically caused by either a single major eVect gene (mendelian) or several smaller eVect genes. Examination of the distribution of gene expression means for inbred strains reveals the existence of both types of gene expression traits. Transcript abundance may be continuously distributed, in which estimates within a genotype take on a wide range of values, often overlapping with those of other strains. Such a distribution indicates that the


67

transcript is under the control of polygenic factors and/or a multitude of environmental eVects. Other transcript appear to be under the control of a single genetic locus. For these transcripts the distributions within each marker genotype at the regulatory locus do not overlap (Fig. 3). A single QTL is observed in these cases, with high LOD scores because of the near-perfect ability of genotype to predict phenotype. Often, but not always, the polymorphism regulating these transcripts is located in proximity to the transcript itself. Many phenotypes are transgressive. Sets of diVerent strains may have polymorphisms modulating transcription, some of which increase and others that decrease abundance of the transcript. Because these are fixed in the inbred lines and may have opposing eVects on transcription, many heritable traits do not obviously vary between a pair of strains. However, in the progeny which contain a diVerent assortment of these loci, an accumulation of increasor or decreasor alleles in the particular strains may exist. Thus, when the inbred lines are crossed, the recombinant progeny often have more extreme phenotypes than either parental strain (Fig. 4). Estimates of the genetic contribution to variation in transcript abundance are useful indicators of the potential for complex trait analysis. Typically, mapping is fruitful for traits with approximately 30% or more of the phenotypic variance attributable to genetic sources relative to total phenotypic (genetic plus technical) sources. The cost of the phenotyping assay severely limits the precision one can obtain in gene expression estimates by collecting large numbers of independent replicates. Complex trait analysis ordinarily involves the use of hundreds of F2 progeny. Analysis of recombinant inbred (RI) strains is a bit more eYcient. For highly variable behavioral traits, RI strain surveys typically use 10–15 mice per strain, although as few as three samples per strain can be used to map traits with high heritabilities (Belknap, 1998). With the small number of arrays typically used in a mapping study, one can increase the signal/noise ratio by pooling tissue mRNA extracts within arrays. This approach reduces the environmental variance, thereby increasing the ability to detect genetic variance, improving power for mapping. Although the relative variance within (environmental) and between (genetic) strains cannot be estimated using the intraclass correlation method, heritability can be estimated by comparing the parental strains to the isogenic and recombinant lines. The relative amount of genetic variation in transcript abundance is estimated by determining the percentage of variance in the phenotype accounted for by genetic factors. Depending on the level of replication present in the samples, this can be a standard heritability estimate (genetic variance/total phenotypic variance where total phenotypical variance is the sum of technical and environmental variation). In the case of pooled samples created to reduce the impact of environmental variability in a more cost-eVective design, this estimate is the amount of genetic variance over an estimate of phenotypic variance, which is

68


Fig. 3. Single marker eVect plots for Kcnj9 (top) and Drd2 (bottom) at their respective quantitative trait loci (QTLs). EVect of allelic variation at distal chromosome 1 on Kcnj9 transcript abundance and at chromosome 9 on Drd2 transcript abundance (from www.webqtl.org). A value of 1 indicates the B6 allele at the marker, 0 indicates heterozygosity, and 1 indicates D2 allele. The positions of the B6, D2, and heterozygous mice are shown. There is virtually no overlap between the strains carrying


69

Fig. 4. A quantile-quantile (Q-Q) plot reveals transgression of the Drd2 phenotype. Strain expression means are plotted against z-score. This type of plot is useful for examining normality of the strain distribution pattern. Expression levels in several of the RI lines are more extreme than those of either parental line. For complete strain information, see www.webqtl.org.

largely technical. Strain diVerences account for as much as 89% of transcript expression variance by intraclass correlation, and for approximately 1100 transcripts, more than 30% of the variance in the trait is accounted for by genetic factors. These observations indicate that the identity of transcription modulators and their complex interactions can be vastly facilitated by transcriptome-QTL analysis, which allows identification of the genetic loci that generates expression variation. B. Experimental Crosses for Genetic Analysis Heritable variation can often be harnessed for detection of regulatory polymorphisms. The genetic variability can be reduced to a small set of quantitative trait loci and candidate QTLs. After determination of the heritability of the phenotype, mapping studies proceed with the generation of recombinant progeny C57BL/6J alleles and the strains carrying DBA/2J alleles. Mice that are heterozygous at this locus have intermediate levels of this transcript. The Kcnj9 expression distribution is almost bimodal and mendelian. An advanced intercross line (A12), which is eight or more generations away from being fully inbred, has an intermediate phenotype, as does the F1 line. Drd2 appears to have polygenic regulation.

70


(Fig. 5). The first generation of the cross produces an F1 generation of mice heterozygous at all loci. These mice are typically crossed to form an F2 generation. Each mouse in this generation carries the alleles of the parental lines, but because of meiotic recombinations, they may be homozygous for one parental allele and heterozygous or homozygous for the other parental allele in a 1:2:1 ratio. Another common mapping strategy is to backcross the F1 lines back to one of the progenitor lines, creating genetically unique mice that are either homozygous for the parental strain or heterozygous at each locus. 1. Mapping Transcription Regulation with F2 Crosses Schadt et al. (2003) have mapped gene expression–related QTLs in the mouse liver using an F2 cross design and demonstrated the relationship of gene expression in these mice to a gross morphological phenotype, fat pad mass. As illustrated in Fig. 5, this approach requires the assay of genotypes and phenotypes in more than 110 genetically unique mice. When performing this type of work in a

Fig. 5. Crosses for genetic mapping. In the F2 cross, each individual progeny is unique and both phenotypical and genotypical data must be obtained for each mouse. If these mice are sibling-pair mated for 20 or more generations, recombinant inbred lines will be generated. Genotypical data need only be required once. These lines can be maintained indefinitely, and multiple phenotypes can be assayed in many experiments and environments.


71

reference population that has been characterized on many measures, such as the human CEPH pedigrees, we may combine results with other previously obtained phenotypes. A small number of phenotypes collected in the specific individuals can be related to gene expression in a single experiment. For phenotypes of interest to most neuroscientists, repeated testing of the same individuals may not be possible, and it is not economically feasible to perform microarray analysis for many small sets of traits being investigated. Furthermore, for most tissue samples, the assessment of both pretreatment and posttreatment levels of gene expression is not possible. 2. Recombinant Inbred Lines and Their Advantages Recombinant inbred strains, such as the BXD strain set, are significantly advantageous for the complex trait analysis of brain and behavior (Plomin et al., 1991). These strains were developed by 20 or more generations of sib mating of the progeny of an F2 cross of C57BL/6J and DBA/2J mice; since then, more strains have been added (Taylor et al., 1999; Peirce et al., 2004). The resulting mice are fully inbred, each with novel recombinant chromosomes homozygous for either parental allele at each locus (Fig. 5). The genetic map is expanded fourfold in these strains, because of the accumulation of meiotic recombinations that occur throughout their generation. Because these strains can be repeatedly tested, significant resources can be devoted to the development of high-resolution genetic maps and genotypical data, and within-strain phenotypical variability can be reduced by obtaining numerous biological replicates. The major advantage of these lines is that data collected in them can be aggregated across studies, allowing one to compare diverse phenotypes in genetically identical mice. For example, expression-related traits can be correlated directly with other phenotypes that have been characterized in these strains throughout their history (Chesler et al., 2003). Furthermore, sequence data are known for several of the progenitor lines, with nearly complete sequence data available for the two progenitor strains of BXD set, C57BL/6J and DBA/2J. This allows one to directly go from QTL to potential sequence polymorphisms underlying the phenotypical variability. EVorts are underway to generate a much larger set of recombinant inbred strains, with increased resolution, genetic diversity, and the statistical power to reliably detect many multilocus interactions (Vogel, 2003).

C. Detecting Major QTLs Simple QTL mapping methods fit models of genotype and phenotype association at each point in the genome. This is done for each of the thousands of transcripts on the microarray. These models assume a single QTL and can be used to detect a small number of major loci regulating trait variation. Many

72


neurologically relevant transcripts have strong QTLs regulating forebrain expression including Glyt, Kcnj9, Pam, Gpi1, Viaat, Chrng, Htr1a, and numerous others (Fig. 6). 1. Cis-Acting Modulators of Transcription Transcript QTLs may be classified as ‘‘cis,’’ here intended to designate that they map to the location of the transcript itself (Fig. 6, top panel). This is most likely because of sequence variation in the promotor or enhancer. This region can span from a few kilobases (Kb) to several megabases (Mb) and as regulatory regions are better defined, the gene-specific cis region can be included in the definition. As a rough estimate of cis status, we define these QTLs to have a 1-LOD confidence interval that includes the transcript. A human cis-regulatory SNP located 1 Kb upstream of Htr1a and implicated in major depression and suicide was found to produce a twofold change in expression by impairing activity of transcription factors (Lemonde et al., 2003). Regulatory regions are not the only sites of polymorphisms that influence transcript abundance. The production of some transcript variants may lead to increased levels of transcription. For example, a receptor variant with low aYnity may be produced at higher copy numbers to compensate for the poor activity, in essence because of a lack of negative feedback. A second intragenic regulatory mechanism could be the production of elevated levels of a gene product causing positive feedback, thereby triggering increased copy production. The cis-QTLs typically have high LOD scores. For these, transcripts abundance is bimodally distributed, often with no overlap between strain values (Fig. 3). 2. Transacting Modulators of Transcription Transcript QTLs may also be classified as ‘‘trans,’’ here intended to refer to polymorphism location far from the actual transcript location. Transacting QTLs might be transcription factors, perhaps binding to the promoters or enhancers at the location of the transcript. Transacting QTLs can also be signaling molecules, receptors, or any other polymorphic genes that influence the expression of other genes. Even polymorphisms in transducers of environmental eVects can influence gene expression. Thus, through the study of transcription regulatory QTLs, we are not merely examining the transcriptional network and apparatus itself, but also the genes that influence the activation of this transcription machinery. The pattern of transcription regulation and the modulators of transcription will vary in diVerent tissues and under diVering environmental conditions. Activation of the particular transcriptional pathway is required to reveal the polymorphic regulators of transcription. Mapping these same transcripts in a diVerent brain tissue, the cerebellum, reveals diVerent regulatory loci (Fig. 7).

Fig. 6. Interval maps for single quantitative trait locus (QTL) scans. A cis-regulatory QTL was found for Viaat and a transregulatory QTL was found for Chrng. Likelihood ratio statistic (LRS) is plotted in blue. The green horizontal line is the suggestive permutation threshold and the blue horizontal line is the significant permutation threshold for the LRS. Transcript location is indicated by the orange triangle below the plot. Yellow bars indicate the frequency of peak LRS location in 1000 bootstrap tests. The red line is the additive eVect of the DBA/2J allele. (See Color Insert.)

74


Fig. 7. Tissue specificity of Htr1a regulation. The location of forebrain and cerebellum quantitative trait loci diVer for this transcript.

3. Transcriptome Map: Global Properties of Transcription Regulation Transcript abundance is a unique phenotype. Each transcript has a genomic location and genomic loci responsible for its modulation. Plotting the location of these transcripts by their regulatory loci generates a ‘‘transcriptome map,’’ which reveals some of the global properties of transcription regulation. Such a map contains an intense diagonal band of transcripts regulated by their own location and several vertical bands of transcripts regulated in concert by similar genetic loci. The vertical bands are of tremendous interest because they most likely represent the location of major controllers of transcription. Identifying the polymorphisms underlying these bands and determining the functional impact of polymorphisms at these loci will reveal the source of variation underlying a vast array of molecular events. EVorts are underway to do so, and this has been successfully done in a yeast transcriptome analysis (Yvert et al., 2003).

D. Gene–Gene Interactions and Multiple QTL Models Although there are some traits for which a single strong regulatory locus has been identified, gene expression regulation is more commonly due to the actions and interactions of several genes. For many traits a few significant loci are detected. These loci can interact with one another, so the eVects of one polymorphism are conditional on those of another, a phenomenon referred to as epistasis. A simple test for detecting pairs of loci that together have significant eVects on transcript abundance is used to scan for epistatic interactions. Once these genetic loci are detected, modeling methods can be employed to identify the best fitting multiple QTL model, a polygenic regulatory model for gene expression level. Genetic dissection of truly polygenic traits is best achieved by developing a model for gene expression that takes into account the simultaneous eVects of these loci (Broman, 2002). For example, the transcript Risc appears to have two regulatory QTL, both of which are statistically significant by permutation. Pair-wise scanning of genetic loci followed by multiple QTL modeling reveals that a joint model


75

with both of the QTLs fit simultaneously is best (Fig. 8). Together, these two QTLs account for 66% of the trait variance. Occasionally, these scans reveal pairs of loci that are undetectable as single QTL eVects. As the sample sizes in the existing RI strain sets increase, routine detection of interacting loci will become possible. For several transcript phenotypes, statistical power is suYcient to make a beginning.

E. Gene–Environment Interactions Many environmental eVects may alter gene expression levels. These may interact with genotypes. Because heritability of neurological traits is often less than 50%, a substantial amount of trait regulation is due to environmental

Fig. 8. Pairwise quantitative trait locus (QTL) scan for Risc expression. The lower triangle and right color scale displays joint LOD scores for the full model fitting each pair of loci and their interaction. The upper triangle and right color scale indicates model fit for the interaction term alone. In this plot, the improved fit of the joint model can be seen at the intersection of chromosome 5 and chromosome 11 in the lower triangle of the figure. An interaction of loci on chromosomes 8 and 3, not significant in 1000 permutations of the pairwise scan, can be seen in the upper triangle. (See Color Insert.)

76


factors. The environment can be loosely interpreted to include the distinct hormonal and genetic milieu of the sexes, the context of diVerent tissue types, and thus tissue specific ‘‘microenvironments’’ the eVects of drug administration, disease states, and other experimental contexts. These cofactors can be included in the QTL mapping analysis to study the interaction of genetic and environmental factors on gene expression. Among the strongest sex by genotype interactions in brain gene expression are Klf5 (Kruppel-like factor 5), Osbpl5, Dlgh4 (discs large homolog 4), Heg f l (heparin-binding epidermal growth factor–like growth factor), Klk8 (kallikrein8), DNAjc3 (Hsp40 homolog, subfamily C, member 3), and Htr7 (5-hydroxytryptamine receptor-7). Because of these interactions of sex and genetic factors, sex-specific regulatory loci may be detected or the magnitude of genetic control of gene expression may vary between the sexes. For example, no main eVect QTLs are found on chromosome 1 for the control of Htr7 abundance, but in considering data from the two sexes separately, a mid-chromosome 1 locus is found for females and a distal chromosome 1 regulatory locus is found for males.

F. Identification of QTL Polymorphisms Identification of the specific genetic basis for variation in gene expression is a long process, but successes have been attained (Korstanje and Paigen, 2002). Progress in this endeavor has been speeded by technological advances in high-resolution mapping, the availability of complete genome sequences, converging evidence from knockout mice, analysis of gene expression variation, and numerous other methods (Glazier et al., 2002; the Complex Trait Consortium, 2003). Accumulation of suYcient evidence for causal relations between genetic polymorphisms and expression phenotypes requires both sequence analysis and further study of the relationship between genotype and phenotype. Validation of the phenotypical assay through quantitative reverse transcriptase polymerase chain reaction (RT-PCR), TOGA (SutcliVe et al., 2000), or other transcript abundance assays will become critical, particularly for low-abundance transcripts. High-abundance transcripts are more reliably assayed using the oligonucleotide array methods, so validation through other means may be less essential. Currently there is much discussion regarding the notion of ‘‘proof’’ of genetic eVect in complex traits (Page et al., 2003). Using a combination of bioinformatics methods and experimentation, we can obtain evidence for the role of specific genes and polymorphisms, ultimately resulting in the positive identification of the precise source of genetic variation responsible for transcript expression and other phenotypical variation (Fig. 9).


77

Fig. 9. A combination of bioinformatics and high-precision genetic methods are the shortest and most economical route to identification of the exact genetic polymorphism responsible for the quantitative trail locus eVect.

IV. How Many QTLs Are There? Statistical Control in Two Dimensions

Transcriptome-QTL analysis creates new challenges in multiple testing corrections. As more tests are run, the probability of at least one false-positive result increases, and the number of false-positive results observed in the entire family of tests increases. In transcriptome-QTL analysis, this problem occurs in two dimensions—one across the large number of markers that are tested for linkage to the phenotype across the genome, and the other across the large number of phenotypes. For 753 markers and 12,422 transcripts, this amounts to a minimum of 9,353,766 point-wise statistical tests! However, because of linkage of adjacent loci, this amounts to approximately 100 independent genetic tests per transcript, a ‘‘mere’’ 1,242,200 tests (Belknap et al., 1996).

78


A. QTL Mapping and Empirical Significance Tests Conservative corrections designed to maintain a constant probability of at least one false-positive result across this huge family of tests assume that the tests are uncorrelated and that the data satisfy distributional assumptions that often do not hold for transcript abundance traits. Simple Bonferroni corrections, therefore, are not practical for this type of analysis. Standards for statistical significance based on the experimental design and genome length proposed by Lander and Kruglyak (1995) also do not adequately address the lack of independence and distributional characteristics of phenotypes. Thus, empirical significance thresholds, obtained by permuting the observations, are preferable to determine how extreme a result is given the actual data (Churchill and Doerge, 1996). B. Multiple Testing and Microarray Analysis A permutation test run for each transcript does not address the multiple testing issues inherent to microarray analysis. The test is designed to determine the probability of detecting at least one QTL for a particular phenotype. With the large number of transcripts present on a microarray, many QTLs will be detected by chance. With a genome-wise error rate of 5%, approximately 621 chance QTLs will be detected under statistically idealized conditions when testing 12,422 phenotypes. The determination of the false discovery rate (Benjamini and Hochberg, 1995) is a straightforward method for determining the error rate given the actual test results, assuming a uniform distribution of test statistics across the experiment by chance. Strict deployment of significance thresholds does not take into account the strength of evidence for QTLs and can result in many false-negative results. Variant approaches to interpreting the false discovery rate, including point-wise estimation (Storey, 2003), make it possible to estimate the probability that a given result is a false positive. This information can be used in combination with prior biological knowledge to interpret the utility of a test statistic. The preponderance of evidence can be used to evaluate the validity of a test result. It is essential that all test results be considered because true positives can be lurking among even weakly significant test results. Attaching the false discovery rate to a mapping result, examining the results of related transcripts, and carefully interpreting with respect to prior knowledge in the area is far more informative than employing an arbitrary significance threshold. C. Permuting the Transcriptome Analysis Analyses that consider the architecture of the entire set of transcriptome-QTL mapping have aided in the interpretation of the transcriptome-QTL map. Permutation of the individual transcripts and plotting the peak LODs from each transcript


79

by transcript location reveals a complete absence of the trans-QTL bands typically observed. This permutation analysis establishes that the presence of the transbands is due to the relations between transcripts (co-regulation or correlation), and that transbands are not an artifact of the genotype distributions. Permutation of the entire array mean vector for each strain results in trans-QTL bands at diVerent locations for each permutation. This permutation analysis reveals that there are large groups of correlated transcripts that share regulatory loci. A consequence of this finding is that a set of 12,422 transcripts may be resolved as a set of far fewer than 1000 transcriptional networks. Even a conventional FDR approach may prove excessively conservative in these circumstances.

V. Probe Versus Probe Set–Level Phenotypes

A. Probe Variation Common to all AVymetrix microarray experiments are issues of probe versus probe-level phenotypes. Treating the individual probes as phenotypes is a daunting task both because of the vast magnification of the number of statistical tests that must be performed and because of the interpretational issues that occur when two adjacent probes produce radically diVerent estimates. The probes together form a multidimensional assay of a single construct—the transcript targeted by a probe set. However, these probes are not independent replicates. Some may perform better than others, some may not be valid or reliable in that they hybridize to multiple gene products, and some overlap their neighbors to varying degrees (see next section). Probe level variation occurs at many levels. Individual probes may hybridize to diVerent exons, and numerous probes span exon boundaries (R. Williams and Y. Qu, 2002, unpublished observations). Furthermore, the hybridization free energies of individual probes and their targets aVect the degree to which they bind. These properties also aVect the hybridization kinetics of match and mismatch probes (Urakawa et al., 2003; Zhang et al., 2003). The base composition of the diVerent probe sequences can be used to determine stacking energies, and this has been used to devise methods of normalizing the array data (Zhang et al., 2003).

B. Genetic Variation and Probe Level Effects Genetic variation also aVects probe level hybridization. Most of the probes were developed using public sequence data from the C57BL/6J strain. As a result, the probes are exact matches for some strains transcripts but may be


81

mismatches for the sequence of other strains. In rare cases, the mismatch for one strain may be a match for another strain, reversing the polarity of a probe level QTL eVect. Probe design is a complex process that has multiple constraints. Identification of SNPs in the probes will facilitate the development of probe level filters to enhance the estimation of transcript abundance. In our comparison of probe sequences with SNPs reported in the Celera Genomics database (Celera Discovery System, Celera Genomics, Rockville, Maryland [7/01/03]), we verified fewer than 1500 probes out of about 390,000 containing SNPs on the U74Av2 array.

C. Probe Redundancy and Overlap Another factor complicating the use of individual oligonucleotide probes on the AVymetrix platform is that a small subset are redundant and recognize multiple gene products. Of 196,670 perfect match probes sequences on the murine U74Av2 array, 762 probe sequences are duplicated at least once; some as many as eight times. In addition, 1659 probes are completely redundant with at least one other probe. In the case of the gamma crystallins D, E, and F, there are six probes that are common to the three probe sets, 12 probes that occur in two of the three probe sets, and 12 probes (4 each) that are unique to each of the three transcripts. The amount of variability observed for each replicate probe on the array is proportional to the intensity of probe hybridization. For the most highly expressed genes, a high degree of consistency to the QTL maps is observed, whereas for low-aYnity probes, QTL maps for identical probes can vary somewhat (Fig. 10). For some related gene products, virtually the entire probe set overlaps, for example, vomeronasal receptors 8 and 9, which share 15 or 16 probes, rendering discrimination of the two transcript expression levels virtually impossible. The presence of multiple identical probes can help reduce noisy estimates and improve precision of array regional normalization methods. In other cases, the probes overlap extensively, so as many as 24 of 25 nucleotides overlap with the adjacent probe (Fig. 11). Naturally, these probes often have highly correlated expression levels. If an SNP is contained in a subset of the probes, or other physical property, or if exon variability is manifest in the probe hybridization, the overlap can lead to a biased assay of expression levels.

Fig. 10. Interval mapping for probes that are shared among several probe sets. These can be thought of as analogous to spot replicates on a complementary DNA array. For abundant (strongly hybridizing) probes consistency is high. As abundance/hybridization decreases, variance is higher and probe level mapping becomes more inconsistent. (See Color Insert.)

82


D. Implications of Probe Level Variation and Normalization Techniques for QTL Analysis Various normalization procedures have been devised to account for diVerential probe level hybridization eVects. This is an actively developing area of analysis, and diVerent analytical methods may be required for diVerent purposes. Some of these algorithms adjust for array-related technical variance simultaneously, whereas others do so in a separate stage of analysis. The various algorithms exhibit diVerent performance for diVerent RNA abundances. The AVymetrix MAS 5.0 algorithm uses a Tukey bi-weight average of the diVerences between perfect match (PM) and those mismatch (MM) probes that do not exceed the perfect match (AVymetrix, 2001). This approach is problematic, particularly in the event that genetic variation results in diVerences in probe level hybridization and in the event that the PM and MM probes have very similar hybridization to probes. Robust multichip average (RMA) uses a background correction, quantile normalization, and linear model fit using robust methods, taking advantage strictly of the PM data (Irizarry et al., 2003). Li and Wong’s (2001) dChip uses an invariant gene set to compute a model-based expression index. Positional-dependent nearest neighbor (PDNN) takes into account the stacking energy of probe level hybridization reactions (Zhang et al., 2003). Each of these algorithms can produce very diVerent gene-specific QTL maps. An unpublished comparison of the overall mapping results for each of the methods by Dr. Kenneth Manly and colleagues reveals that there are more than 600 transcripts for which the peak of linkage is the same for all methods, although these peaks are of varying statistical significance. Additionally, for the significant QTLs by any method, the peak location of linkage does not vary by normalization method for a subset of approximately 100 loci (Fig. 12). This indicates that although these normalization methods can be quite consistent with one another, some may be conservatively biased. This is because normalization must be performed at each locus in the mapping analysis. Adjusting data before analysis by a cofactor will in most cases reduce the magnitude of the estimate of the cofactor eVect (Darlington and Smulders, 2001). Normalizing the arrays before mapping by either using strain as a grouping variable, or not using any grouping variable in the mapping analysis can either increase the noise or reduce the signal, respectively, in expression estimates by genotype. Both of these approaches introduce conservative bias, evidenced in the p value distributions for mapping, which demonstrate greater than expected numbers of high p values. This phenomenon is most extreme in RMA, followed by PDNN and MAS 5.0. The use of simultaneous normalization and mapping is computationally intensive (an entire microarray group comparison analysis must be performed at each locus) but is not impossible.


83

Fig. 11. Substantial overlap of probes occurs in many probe sets. This screen shot from www.webqtl.org illustrates the variety of probe level information available at the site, including sequence, melting temperature, stacking energies, and mean expression levels. VI. Transcription Regulatory Networks

The production and maintenance of a cellular pool of mRNA involves a complex network of transcriptional processes, RNA processing elements, mRNA export from the cell nucleus, and processes of degradation (Maniatis and Reed, 2002). In our analysis of brain transcription in RI strains, we found genetic correlations of variation in RNA polymerase 2 transcription with many other transcription factors and ribosomal proteins, hinting at the large number of genes

Fig. 12. An example of colocalization of quantitative trait loci (QTLs) detected using diVerent normalization methods. Chromosome 1 interval mapping for Pam (peptidyl-glycine- amidating monooxygenase) using positional-dependent nearest neighbor dChip (perfect match [PM] only method), robust multichip average (RMA), and MAS 5.0 methods of normalization. For each method, the peak likelihood ratio statistic (LRS) location is consistent, although the magnitude of the peak varies greatly.


85

involved in the transcriptional machine and the complexity of regulation we may observe. Beyond this gene expression ‘‘machinery,’’ all of which is subject to genetic polymorphism, lies control by various other cellular processes that signal the transcription of genes in response to stimuli, including signal transduction cascades, cellular signaling, and global phenomenon such as transduction of the internal and external environment of the organism. The identification of single QTLs is just a beginning in the eVort to tease apart transcription regulatory networks and study the eVects of experimental perturbations of these networks. A variety of approaches to reverse engineering of gene networks through modeling of microarray data have been employed, including time-series analysis (Cho et al., 1998), bayesian network analysis (Friedman et al., 2000), and coupling of sequence analysis to bayesian networks (Tamada et al., 2003). These approaches require relatively small input sets, so the vast amount of data from microarray analysis must be filtered. QTL analysis can be used to reduce the set of network nodes and candidate genes. Newer experimental approaches involve large-scale reverse engineering of gene networks, in which a small circuit is systematically perturbed to analyze large-scale transcriptional eVects (Tegne´r et al., 2003). Known regulatory relationships can be confirmed using QTL mapping of transcript abundance. Variation in transcript abundance of one gene often maps back to the location of a known regulator of that genes expression. QTLs can provide a causal structure to networks. Genetic analysis provides evidence that the regulators of transcript expression are likely to include members of a set of candidates within a particular location. Typical gene expression networks do not have directionality between the nodes. Network modelers can use QTLs to establish the direction of causality between nodes when connected nodes of a gene expression network contain a transcript and a gene in one of its regulatory loci.

VII. Functional Correlates of Brain Transcriptional Activity

A. Genetic Correlations Recombinant inbred strains are uniquely advantageous for analysis of gene expression data. Over their 20-year history, investigators have already characterized these strain sets on a large number of anatomical, physiological, behavioral, and other functional or disease-related phenotypes. The BXD strain set, in which our transcriptome-QTL mapping study has been performed, has been characterized on several hundred traits. By correlating strain means of these traits with strain means for gene expression, genes associated with the traits can be identified.

86


Because the gene expression traits are studied in completely diVerent individuals than the other phenotypes, the family means correlations can be attributed to shared genetic mediation. When traits are studied simultaneously in the same individuals, some of the correlation may be due to shared environment. The converse question could also be addressed with this analysis. Starting from a gene, one can identify traits in which that gene product may play a role. These correlations are indicative of shared genetic regulation of the traits, or pleiotropy. However, it is also possible that the traits in question are each regulated by diVerent genes that are tightly linked to one another on the chromosome. Genetic correlations of the gene expression levels themselves may also be performed. The vesicle amino acid transporter Viaat gene has expression levels that associate with many transcripts expected from its role in transporting aminobutyric acid (GABA) and glycine (an N-methyl-d-aspartate [NMDA] receptor ligand) into neurotransmitter vesicles (Fig. 13). These include Tln (talin), Narg1 (NMDA receptor–regulated gene), and Nsap1 (NS1-associated protein, a synaptotagmin-binding protein). Evaluating the statistical significance of the many billions of correlations possible in the data is a challenging multiple testing issue. Application of pointwise estimation of the false discovery rate (Storey and Tibshirani, 2003) is a promising approach, but the definition of the family of statistical tests is dependent on the precise question being asked. In this case, the family of tests includes all genes on the array and their association with Viaat expression. At a 10% false

Fig. 13. Correlation of Viaat expression with other forebrain gene expression on the mouse U74Av2 array. Several synaptic proteins are among the correlates.


87

discovery rate, more than 1500 genes are significantly correlated with Viaat expression (Fig. 14). B. Cliques, Clusters, and Genetic Association Networks The associations of genes with each other enable the detection of networks of co-regulated genes. Associative networks can be drawn with edges determined by the genetic correlation of the nodes (Fig. 15). Many very large groups of correlated transcripts can be identified within these networks, particularly those that are regulated by genes residing in the trans QTL bands. The amyloid precursor protein App is part of a larger cluster of genetically correlated transcripts. More than 500 transcripts have absolute correlations with this transcript exceeding 0.66. Clique analysis is a means of visualizing related correlation networks on a massive scale using graphs that has been successfully used in the analysis of plant metabolic activity (Kose et al., 2001). In our transcriptome profiling data, some large cliques of genes have genetic correlations between all members exceeding 0.85 and contain as many as 17 members (Dr. Michael Langston and colleagues, unpublished observations). For tightly correlated sets of genes, multiple common regulatory loci can be found. For less tightly associated groups some shared and some unique genetic modifier loci can be identified. C. Incorporation of Other Traits into the Transcriptional Network The placement of other phenotypes within gene transcriptional networks is possible. For example, adult neurogenesis-related phenotypes (Kempermann and Gage, 2002) assayed in the hippocampus correlate highly with hippocampal regional volumes (Peirce et al., 2003) and neuron numbers in adult mice (Lu et al., 2001). These traits share correlations with gene expression, and common sources of genetic variation are, therefore, likely to be identified. These results indicate that genetic variation in processes leading to increased adult neurogenesis is likely the cause of variation in adult neuron number. For example, 64 genes having high correlations to new neurons (Kempermann and Gage, 2002) are among the top correlates of granule cell layer volume (Peirce et al., 2003). A role for other cellular processes such as early proliferation and cell death may also be identified as determinants of this trait. Examination of these traits and the shared and unique associations of gene expression with the original published phenotypes and regulatory loci can elucidate the network of genes involved in common and separate mediation of the traits. A single principal component explains 72% of the variance in adult neurogenesis and hippocampal size phenotypes. Genetic correlates of gene expression to this component include many neural development– related genes, including Hoxd8, which has an expression correlation of –0.97 with

88


Fig. 14. Estimation of q values for genome-wise genetic correlation of Viaat expression and other transcript abundance traits. Even when considering the multiplicity of tests, a seemingly lenient p value threshold of approximately 0.01 has a 5% false discovery rate. There are approximately 750 transcripts at this level.

this composite trait (Fig. 16). The use of the BXD RI strain panel as a genetic reference population allows this type of relational analysis to be performed. Rather than perturbing individual network players, naturally occurring genetic variation can be exploited to simultaneously assess the relationships of the many thousands of transcriptomic, proteomic, and phenomic characters that have been or will be evaluated in these strains.

Fig. 15. A genetic association network for behavior and brain gene expression. Each trait is a node in a genetic correlation network. Correlations between traits are shown along the edges. Locations of local likelihood ratio statistic (LRS) maxima are indicated for each transcript node. Large filled ovals illustrate shared regulatory loci. The probe set number is listed on the right of the node, and the gene location is given under the gene name. The expression range among the recombinant inbred strains is indicated beside each gene name.

90


Fig. 16. Correlation of the first principal component for several hippocampal anatomy and adult neurogenesis phenotypes with Hoxd8 expression in BXD RI strains.

VIII. Conclusions: Building a Model of Brain Function from Multipoint, Multi-tissue Gene Expression, and Phenomic Observation

A. An Expanding Notion of the Neurochemical Pathway Genomic technology has increased our awareness that the notion of ‘‘one gene–one phenotype’’ is an oversimplification. Even our consideration of the simple classical neurotransmitter pathways must now be expanded to include all proteins involved in the synthesis, posttranslational modification, traYcking, anchoring, and coupling of the enzymes, receptors, and vesicle proteins that were former targets of attention. As annotation of the genome improves, we will better understand the various roles of these proteins and their interactions, and the hypothesis-generating functions of genetic correlation analysis can be a useful vehicle in driving improved knowledge of the complete functional proteome associated with these neuronal activities. By treating gene expression as a complex phenotype, we can detect these large networks due to the presence of individual diVerences that occur in many of their molecular players.


91

B. A Discovery-Based Approach to Dissection of Gene Regulatory Relations Hypothesis-driven science, focused on single-gene analysis has left us with insights into the associations between genes and phenotypes for a small subset of the 30,000 known genes. Genetic genomics is an approach that takes advantage of naturally occurring variation and genome-wide screening of both the candidate transcripts and their regulators. This screening of both dependent and independent variables across the genome casts a wide net for discovery of gene regulatory relations, unconstrained by prior hypotheses. Discovery-based genomic science is gradually increasing our depth of understanding of the broad set of genes and polymorphisms, in a manner that allows rapid assessment of relationships. In time, the precision of these relationships will increase. Tools for integrating the wealth of transcriptome data with our existing gene and phenotypical knowledge base are rapidly becoming available. Accumulating both depth and breadth of data and devising intelligent methods to mine these relations will be the key to the success. Already, confirmation of major known regulatory relations is evident in transcription QTL analysis, and many highly plausible novel hypotheses about gene–gene and gene-phenotype relations have been developed.

C. A Relational Model for Data Integration Using a reference panel of inbred strains and deep phenotype data acquisition and analysis, we can study the complexities of genetic, environmental, and trait interactions between and among many tissue types. It is essential to relate data through a common genetic reference panel to assemble and assess this information eYciently. Naturally occurring genetic variation, viable in the laboratory mouse, is likely to be conserved in other species, including humans. Although the sites of polymorphisms will diVer, viability of a variation and viability of alleles in one species indicates potential nonlethality of similar alleles in other species. Thus, the examination of the genetic basis of individual diVerences of these traits in mice will also translate to human gene regulatory pathways. As new strains are developed, precision and accuracy for detection of regulatory loci increases. All that is needed to tap into or expand the resource is collection of phenotypic data, ranging from additional transcriptomic and proteomic assays in multiple brain regions, to the numerous developmental, functional, and behavioral phenotypes.

92


Acknowledgments

We would like to thank Dr. Lu Lu, Dr. Kenneth F. Manly, Dr. Jintao Wang, Dr. Michael Langston, Dr. Siming Shou, Dr. Li Zhang, Dr. Michael Miles, Dr. Cheng Li, Dr. Bing Zhang, Dr. Jing Gu, Dr. Yanhua Qu, Dr. David Threadgill, Dr. Hui Chen Hsu, Dr. Gerd Kempermann, Dr. John K. Belknap, Dr. John C. Crabbe, Dr. Thomas Sutter, Dr. Divyan Patel, and Mary-Kathleen Sullivan. Our thanks for financial support from (1) the Informatics Center for Mouse Neurogenetics; P20-MH62009 from NIMH, NIDA, and NSF. (2) INIA grants U01AA13499 and U24AA135B from NIAAA and the William and Dorothy Dunavant Chair of Excellence.

References

AVymetrix (2001). ‘‘Statistical algorithms reference guide,’’ Technical report, AVymetrix. Belknap, J. K. (1998). EVect of within-strain sample size on QTL detection and mapping using recombinant inbred mouse strains. Behav. Genet. 28, 29–38. Belknap, J. K., Mitchell, S. R., O’Toole, L. A., Helms, M. L., and Crabbe, J. C. (1996). Type I and type II error rates for quantitative trait loci (QTL) mapping studies using recombinant inbred mouse strains. Behav. Genet. 26, 149–160. Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing. J. R. Statist. Soc. B 57, 289–300. Brem, R. B., Yvert, G., Clinton, R., and Kruglyak, L. (2002). Genetic dissection of transcriptional regulation in budding yeast. Science 296, 752–755. Broman, K. W. (2002). A model selection approach for the identification of quantitative trait loci in experimental crosses. J. R. Statist. Soc. B 64, 1–16. Bystrykh, L., Weersing, E., Sutton, S., Dontje, B., Vellenga, E., Wang, J., Manly, K. F., Williams, R. W., Cooke, M., and Deltaan, G. (2003). Genetical genomics to identify gene pathways regulating hematopoietic stem cells. Complex Trait Consortium Abstracts. Chesler, E. J., Lu, L., Wang, J., Williams, R. W., and Manly, K. F. (2004). Web QTL: Rapid exploratory analysis of gene expression and genetic networks for brain and behavior. Nature Neuroscience 7:4, 85–486. Chesler, E. J., Shou, S., Qu, Y., Yang, X., Lu, L., and Williams, R. W. (2002). Microarray analysis of sex diVerences in the mouse CNS transcriptome. Program No. 6236 ‘‘Abstact Viewer/Itinerary Planner.’’ Washington, D.C.: Society for Neuroscience. Online. Chesler, E. J., Wang, J., Lu, L., Qu, Y., Manly, K. F., and Williams, R. W. (2003). Genetic correlates of gene expression in recombinant inbred strains: A relational model system to explore neurobehavioral phenotypes. Neuroinformatics 3, 342–358. Cheung, V. G., and Spielman, R. S. (2002). The genetics of variation in gene expression. Nat. Genet. 32, 522–525. Churchill, G. A., and Doerge, R. W. (1996). Empirical threshold values for quantitative trait mapping. Genetics 138, 963–971. Cho, R. J., Campbell, M. J., Winzeler, E. A., Steinmetz, L., Conway, A., Wodicka, L., Wolfsberg, T. G., Gabrielian, A. E., Landsman, D., Lockhart, D. J., and Davis, R. W. (1998). A genomewide transcriptional analysis of the mitotic cell cycle. Mol. Cell 2, 65–73. Clayton, D. F. (2000). The genomic action potential. Neurobiol. Learning Memory 74, 185–216. Complex Trait Consortium: Flaherty, L., Abiola, O., Angel, J. M., Avner, P., Bachmanov, A. A., Belknap, J. K., Bennett, B., Blankenhorn, E. P., Blizard, D. A., Bolivar, V., Brockmann, G. A.,


93

Buck, K. J., Bureau, J-F., Casley, W. L., Chesler, E. J., Cheverud, J. M., Churchill, G. A., Cook, M., Crabbe, J. C., Crusio, W. E., Darvasi, A., de Haan, G., Demant, P., Doerge, R. W., Elliott, R. W., Farber, C. R., Flint, J., Gershenfeld, H., Gibson, J. P., Gu, W., Himmelbauer, H., Hitzemann, R., Hsu, H.-C., Hunter, K., Iraqi, F., Jansen, R. C., Johnson, T. E., Jones, B. C., Kempermann, G., Lammert, F., Lu, L., Manly, K. F., Matthews, D. B., Medrano, J. F., Mehrabian, M., Mittleman, G., Mock, B. A., Mogil, J. S., Montagutelli, X., Morahan, G., Mountz, J. D., Nagase, H., Nowakowski, R. S., O’Hara, B. F., Osadchuk, A. V., Paigen, B., Palmer, A. A., Peirce, J. L., Pomp, D., Rosemann, M., Rosen, G. D., Schalkwyk, L. C., Seltzer, Z., Settle, S., Shimomura, K., Shou, S., Sikela, J. M., Siracusa, L. D., Spearow, J. L., Teuscher, C., Threadgill, D. W., Toth, L. A., Toye, A. A., Vadasz, C., Van Zant, G., Wakeland, E., Williams, R. W., Zhang, H-G., and Zou, F. (2003). The nature and identification of quantitative trait loci: A community’s view. Nat. Genet. Rev. Darlington, R. B., and Smulders, T. V. (2001). Problems with residual analysis. Anim. Behav. 62, 599–602. Doerge, R. W. (2002). Mapping and analysis of quantitative trait loci in experimental populations. Nat. Rev. Genet. 3, 43–52. Flint, J. (2003). Analysis of quantitative trait loci that influence animal behavior. J. Neurobiol. 54, 46–77. Friedman, N., Linial, M., Nachman, I., and Pee´r, D. (2000). Using Bayesian networks to analyze expression data. J. Comput. Biol. 7, 601–620. Geschwind, D. H. (2000). Mice, microarrays and the genetic diversity of the brain. Proc. Natl. Acad. Sci. 97, 10676–10678. Glazier, A. M., Nadeau, J. H., and Aitman, T. J. (2002). Finding genes that underlie complex traits. Science 298, 2345–2349. Irizarry, R. A., Hobbs, B., Collin, F., Beazer-Barclay, Y. D., Antonellis, K. J., Scherf, U., and Speed, T. P. (2003). Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4, 249–264. Janowsky, A., Mah, C., Johnson, R. A., Cunningham, C. L., Phillips, T. J., Crabbe, J. C., Eshleman, A. J., and Belknap, J. K. (2001). Mapping genes that regulate density of dopamine transporters and correlated behaviors in recombinant inbred mice. J. Pharmacol. Exp. Ther. 298, 634–643. Jansen, R. C., and Nap, J. P. (2001). Genetical genomics: The added value from segregation. Trends Genet. 17, 388–391. Klose, J., Nock, C., Herrmann, M., Stuhler, K., Marcus, K., Bluggel, M., Krause, E., Schalkwyk, L. C., Rastan, S., Brown, S. D., Bussow, K., Himmelbauer, H., and Lehrach, H. (2002). Genetic analysis of the mouse brain proteome. Nat. Genet. 30, 385–393. Kose, F., Weckwerth, W., Linke, T., and Fiehn, O. (2001). Visualizing plant metabolomic correlation networks using clique-metabolite matrices. Bioinformatics 17, 1198–1208. Korstanje, R., and Paigen, B. (2002). From QTL to gene: The harvest begins. Nat. Genet. 31, 235. Kempermann, G., and Gage, F. H. (2002). Genetic determinants of adult hippocampal neurogenesis correlate with acquisition, but not probe trial performance, in the water maze task. Eur. J. Neurosci. 16, 129–136. Lander, E. S., and Kruglyak, L. (1995). Genetic dissection of complex traits: Guidelines for interpreting and reporting linkage results. Nat. Genet. 11, 241–247. Lemonde, S., Turecki, G., Bakish, D., Du, L., Hrdina, P. D., Brown, C. D., Sequiera, A., Kushwaha, N., Morris, S. J., Basak, A., Ou, X-M., and Albert, P. R. (2003). Impaired repression at a 5-hydroxytryptamine 1A receptor gene polymorphism associated with major depression and suicide. J. Neurosci. 23, 8788–8799. Li, C., and Wong, W. H. (2001). Model-based analysis of oligonucleotide arrays: Model validation, design issues and standard error applications. Genome Biol. 2, 1–11. Lu, L., Airey, D. C., and Williams, R. W. (2001). Complex trait analysis of the hippocampus: Mapping and biometric analysis of two novel gene loci with specific eVects on hippocampal structure in mice. J. Neurosci. 21, 3503–3514.

94


Maniatis, T., and Reed, R. (2002). An extensive network of coupling among gene expression machines. Nature 416, 499–506. Morgan, J. I., and Curran, T. (1989). Stimulus-transcription coupling in neurons: Role of cellular immediate-early genes. Trends Neuroci. 12, 459–462. Page, G. P., George, V., Go, R. C., Page, P. Z., and Allison, D. B. (2003). ‘‘Are we there yet?’’: Deciding when one has demonstrated specific genetic causation in complex diseases and quantitative traits Am. J. Hum. Genet. 73, 711–719. Pavlidis, P., and Noble, W. S. (2001). Analysis of strain and regional variation in gene expression in the mouse brain. Genome Biol. 2, 0042.1–0042.15. Peirce, J. L., Chesler, E. J., Williams, R. W., and Lu, L. (2003). Genetic architecture of the mouse hippocampus: Identification of gene loci with selective regional eVects. Genes Brain Behav. 2, 238–252. Peirce, J. L., Lu, L., Gu, J., Silver, L. M., and Williams, R. W. (2004). A new set of BXD recombinant inbred lines from advanced intercross populations in mice. BMC Genet. 5, 7. Plomin, R., McClearn, G. E., Gora-Maslak, G., and Neiderhiser, J. M. (1991). Use of recombinant inbred strains to detect quantitative trait loci associated with behavior. Behav. Genet. 21, 99–116. Rose, S. P. (1991). How chicks make memories: The cellular cascade from c-fos to dendritic remodeling. Trends Neurosci. 14, 390–397. Sandberg, R., Yasuda, R., Pankratz, D. G., Carter, T. A., Del Rio, J. A., Wodicka, L., Mayford, M., Lockhart, D. J., and Barlow, C. (2000). Regional and strain-specific gene expression mapping in the adult mouse brain. Proc. Natl. Acad. Sci. 97, 11038–11043. Schadt, E. E., Monks, S. A., Drake, T. A., Lusis, A. J., Che, N., Colinay, V., RuV, T. G., Milligan, S. B., Lamb, J. R., Cavet, G., Linsley, P. S., Mao, M., Stoughton, R. B., and Friend, S. H. (2003). Genetics of gene expression surveyed in maize, mouse, and man. Nature 422, 297–302. Shou, S., Lu, L., Qu, Y., Jensen, P., and Williams, R. W. (2002). High transcriptional diversity and complexity among mouse brain regions. 903.8 ‘‘Abstract Viewer/Itinerary Planner’’ Washington, D.C. Society for Neuroscience. Online. Storey, J. D., and Tibshirani, R. (2003). Statistical significance for genomewide experiments. Proc. Natl. Acad. Sci. USA 100, 9440–9445. SutcliVe, J. G., Foye, P. E., Erlander, M. G., Hilbush, B. S., Bodzin, L. J., Durham, J. T., and Hasel, K. W. (2000). TOGA: An automated parsing technology for analyzing expression of nearly all genes. Proc. Natl. Acad. Sci. USA 97, 1976–1981. Tamada, Y., Kim, S., Bannai, H., Imoto, S., Tashiro, K., Kuhara, S., and Miyano, S. (2003). Estimating gene networks from gene expression data by combining Bayesian network model with promoter element detection. Bioinformatics 19(Suppl 2), II227–II236. Taylor, B. A., Wnek, C., Kotlus, B. S., Roemer, N., MacTaggart, T., and Phillips, S. J. (1999). Genotyping new BXD recombinant inbred mouse strains and comparison of BXD and consensus maps. Mamm. Genome 10, 335–348. Tegne´r, J., Yeung, M. K. S., Hasty, J., and Collins, J. C. (2003). Reverse engineering gene networks: Integrating genetic perturbations with dynamical modeling. Proc. Natl. Acad. Sci. USA 100, 5944–5949. Traina-Dorge, V. L., Carr, J. K., Bailey-Wilson, J. E., Elston, R. C., Taylor, B. A., and Cohen, J. C. (1985). Cellular genes in the mouse regulate in trans the expression of endogenous Moure mammary tumor viruses. Genetics 111, 597–615. Urakawa, H., El Fantroussi, S., Smidt, H., Smoot, J. C., Tribout, E. H., Kelly, J. J., Noble, P. A., and Stahl, D. A. (2003). Optimization of single-base-pair mismatch discrimination in oligonucleotide microarrays. Appl. Environ. Microbiol. 69, 2848–2856. Vogel, G. (2003). Scientists dream of 1001 complex mice. Science 301, 456–457. Wang, J., Williams, R. W., and Manly, K. F. (2003). WebQTL: Web-based complex trait analysis. Neuroinformatics 1, 299–308.


95

Yvert, G., Brem, R. B., Whittle, J., Akey, J. M., Foss, E., Smith, E. N., Mackelprang, R., and Kruglyak, L. (2003). Trans-acting regulatory variation in Saccharomyces cerevisiae and the role of transcription factors. Nat. Genet. 35, 57–64. Zhang, L., Miles, M. F., and Aldape, K. D. (2003). A model of molecular interactions on short oligonucleotide microarrays. Nat. Biotechnol. 21, 818–821. Zhao, X., Lein, E. S., He, A., Smith, S. C., Aston, C., and Gage, F. H. (2001). Transcriptional profiling reveals strict boundaries between hippocampal subregions. J. Compar. Neurol. 441, 187–196. Zirlinger, M., Krieman, G., and Anderson, D. J. (2001). Amygdala-enriched genes identified by microarray technology are restricted to specific amygdaloid subnuclei. Proc. Natl. Acad. Sci. 98, 5270–5275.


DNA MICROARRAYS AND ANIMAL MODELS OF LEARNING AND MEMORY

Sebastiano Cavallaro Institute of Neurological Sciences Italian National Research Council, 95123 Catania, Italy

I. Introduction II. DNA Microarray Technology A. Basic Principles B. Microarray Data Analysis III. Use of DNA Microarrays for Studying Learning and Memory A. Physiology of Learning and Memory B. Pathology of Learning and Memory IV. Conclusion References

I. Introduction

Identifying the mechanisms responsible for learning and memory consolidation remains a critical goal of behavioral neuroscience. Many experiments over the past few decades have demonstrated that inhibitors of transcription or translation interfere with long-term memory (LTM) formation, indicating the requirement of de novo gene expression (Davis and Squire, 1984; Stork and Welzl, 1999). Proteins newly synthesized during memory consolidation may contribute to restructuring processes at the synapse and thereby alter the eYciency of synaptic transmission beyond the duration of short-term memory. Revealing the dependence of LTM on protein synthesis, however, provides no information about the identity and specificity of the required proteins. Because the quantity of a particular protein is often reflected by the abundance of its messenger RNA (mRNA), a variety of methods have been used to describe a limited number of diVerentially expressed mRNAs during LTM. Increased or, less often, decreased expression of genes has been demonstrated during specific time windows following learning (Stork and Welzl, 1999). In the past we have used RNA fingerprinting to identify genes that were upregulated in the hippocampus of water maze–trained rats (Cavallaro et al., 1997). Spatial learning–induced changes in expression of some of these genes occur at selective times and in specific hippocampal subfields (Cavallaro et al., 1997; Zhao et al., 2000), indicating distinct contributions to learning and INTERNATIONAL REVIEW OF NEUROBIOLOGY, VOL. 60

97


98

CAVALLARO

memory. Increased expression of one of these genes, the ryanodine receptor type2, could result in increased mobilization of [Ca2þ] that may participate in the synaptic changes underlying associative memory storage (Alkon et al., 1998). In these past studies, however, we screened only a small fraction of the genes that may have been diVerentially expressed during LTM. Thus, the questions remain how many genes are involved in memory and how do they interact functionally to eVect memory storage. In addition, each of the identified genes may not act in a linear sequence but in complex networks. Successive screenings at diVerent times, therefore, were needed to uncover the networks of genes involved in distinct steps of memory storage. Knowledge of the structure and organization of the human genome and high-throughput gene expression technologies are now opening the door to a new dynamic and functional dimension to the exploration of learning and memory. In this chapter, we highlight the use of one of these technologies, DNA microarray, and illustrate how this can be applied to dissect and analyze the pathophysiology of learning and memory in the mammalian brain. For a more general description of microarray technology, see other reviews (Heller, 2002; Hess et al., 2001; Noordewier and Warren, 2001).

II. DNA Microarray Technology

A. Basic Principles A DNA microarray is a grid of DNA spots, called probes, each containing a unique DNA sequence (Fig. 1). Spots contain either DNA oligomers or a longer DNA sequence designed to be complementary to a particular mRNA of interest. When a microarray is hybridized to fluorescence-tagged complementary DNAs (cDNAs) or RNAs (cRNAs) derived from mRNA or total RNA, each spot is a target for the mRNA encoded by a gene. A laser can then excite the bound cDNAs or cRNAs, and fluorescence intensities from each spot on the slides are collected by a scanner. The intensity of the fluorescence at each array element is proportional to the expression level of that gene in the sample. The choice of having oligomers or longer cDNA sequences yields two diVerent microarray technologies: oligonucleotide and cDNA microarrays, respectively. With cDNA microarrays, two fluor-labeled cDNA samples can be combined and simultaneously hybridized to the same microarray, where they competitively react with thousands of arrayed cDNA molecules. Oligonucleotide microarrays, instead, require that each sample be hybridized onto separate arrays. The thing that makes microarrays the most promising technology for genome-wide expression analysis is the number of DNA probes that it is possible to place on a microarray.

MICROARRAYS AND ANIMAL MODELS OF LEARNING AND MEMORY

99

Already there are microarrays with probes for every gene in yeast, and others with more than 30,000 human genes. This allows researchers to observe the response of whole genomes to various stimuli instead of one gene at a time.

B. Microarray Data Analysis Microarray analysis results in large amounts of data that are diYcult to interpret without computational methods. The simplest analysis involves two samples, representing a test condition and a control condition, and yields a list of paired expression values, one pair for each gene. As illustrated in Fig. 1, these pairs can be represented graphically by a scatter plot, with the values of sample one plotted on the x-axis and the values of sample two plotted on the y-axis. The resulting correlation plot provides a visual image of the relationship between the two expression profiles. In this plot, genes with similar expression levels in the two samples should have points on the identity line (y ¼ x) and genes that are expressed diVerentially lie at some distance from this line. However, the problem is that microarrays do not measure expression levels directly, but intensity levels, as represented by the amount of phosphorescent dye that was recorded by a scanner. Many other factors, such as the overall mRNA concentration of the two samples, the saturation eVects in the hybridization, or the quenching eVect of the phosphorescent dyes, can aVect these intensity values. To correct these diVerences in intensity levels, the raw data can be ‘‘normalized,’’ for example, by using a normalization constant derived from ‘‘housekeeping’’ or ‘‘spiked’’ control genes. Once normalized, a series of restrictions (or filters) can be applied to the data obtained. These restrictions include factors such as quality control, expression-level constraints, sample-to-sample fold comparison, and statistical group comparisons. The simplest way to identify interesting genes in DNA microarray experiments is to search for those that are consistently either upregulated or downregulated. To this end, fold-diVerence thresholds and/or statistical analysis of gene expression levels can be applied. Relative diVerences in expression levels (fold changes) are typically employed in group comparisons of gene expression and have much intuitive appeal for biologists. The choice of thresholds, however, is somewhat arbitrary and inherently subject to high error rates because information on sample variance is not exploited. If array experiments are replicated to an extent that permits direct estimates of the variance of each transcript, parametric or non-parametric statistics can be applied. In these cases, however, many false-positive results are expected by chance when one relies on the nominal p value. For instance, when testing 10,000 transcripts we would expect to misidentify about 500 genes as significant ( p < .05), even when there is no real diVerence in gene expression. Multiple testing corrections, therefore, are needed to adjust the individual p value to account for this eVect.

100

CAVALLARO

Fig. 1. Schematic representation of DNA microarray methodology. Total or messenger RNA is extracted, reverse transcribed, labeled, and hybridized to oligonucleotide microarrays. In complementary DNA microarray platforms, two RNA samples can be reverse transcribed, labeled with diVerent fluorochromes (e.g., Cy3 and Cy5), and simultaneously cohybridized to arrays. At the end of the hybridization, the image produced by the dye is collected by a laser scanner. Intensity values from each


101

More complex computational methods are needed to monitor several gene expression profiles, such as those arising from time course studies, and various clustering techniques have been applied to the identification of patterns in gene expression data. Cluster analysis is a commonly used method to investigate and interpret gene expression data sets. By grouping together genes that have similar expression profiles, cluster analysis can be used for extraction of regulatory motifs, inference of functional annotation, and classification of cell types or tissue samples. 1. Cluster Analysis The term clustering stands for a method that makes it possible to partition a set of objects (genes) into subgroups with similar features called clusters. These partitions have to satisfy the following features: homogeneity in the cluster (the objects that belong to the same cluster have to be as similar as possible) and heterogeneity among clusters (the objects that belong to diVerent clusters have to be as diVerent as possible). Briefly, a clustering method generally consists of two distinct components: a distance measure (or similarity coeYcient) that indicates how similar two gene expression patterns are (or more generally, two clusters) and a clustering algorithm, which uses some heuristics to identify clusters of similar gene expression patterns, based on the distance measure. a. Measure of Distance or Similarity Coefficient. Many of the advanced analysis techniques are based on measures of gene similarity. Similarity or ‘‘nearness’’ between genes is usually based on the correlation between the expression profiles of the genes. For expression data, we can solve the problem of ‘‘similarity’’ mathematically by defining an ‘‘expression vector’’ for each gene that represents its location in ‘‘expression space.’’ In this way, expression data can be represented in n-dimensional expression space, where n is the number of experiments and where each gene expression vector is represented as a single point in that data space. b. Clustering Algorithms. After providing a means of measuring distance between genes, clustering algorithms sort the data and group genes together on the basis of their separation in expression space. Various clustering techniques have been applied to the identification of patterns in gene expression data. Most cluster analysis techniques are hierarchical, the resultant classification has an increasing number of nested classes, and the result resembles a phylogenetic classification. Nonhierarchical clustering techniques also exist, such as k-means clustering, which simply partitions objects into diVerent clusters without trying to specify

spot are calculated and then analyzed by specific software. Data can be represented graphically by a scatter plot, with the values of sample one plotted on the x-axis and the values of sample two plotted on the y-axis. Data obtained under diVerent conditions (e.g., diVerent time points) can be analyzed with diVerent algorithms such as hierarchical or k-mean clustering. (See Color Insert.)

102

CAVALLARO

the relationship between individual elements. Clustering techniques can further be classified as divisive or agglomerative. A divisive method begins with all elements in one cluster that is gradually broken down into smaller and smaller clusters. Agglomerative techniques start with (usually) single-member clusters and gradually fuse them together. Finally, clustering can be either supervised or unsupervised. Supervised methods use existing biological information about specific genes that are functionally related to guide the clustering algorithm. However, most methods are unsupervised. Although cluster analysis techniques are extremely powerful, great care must be taken in applying this family of techniques. Even though the methods used are objective in the sense that the algorithms are well defined and reproducible, they are still subjective in the sense that selecting diVerent algorithms, diVerent normalizations, or diVerent distance metrics will place diVerent objects into diVerent clusters. Furthermore, clustering unrelated data would still produce clusters, although they might not be biologically meaningful. c. Hierarchical Clustering. Hierarchical clustering is advantageous because it is simple and the result can be easily visualized. It has become one of the most widely used techniques for the analysis of gene expression data. Hierarchical clustering is an agglomerative approach in which single expression profiles are joined to form groups, which are further joined until the process has been carried to completion, forming a single hierarchical tree. Relationships among objects (genes) are represented by a tree, called dendrogram, whose branch lengths reflect the degree of similarity between the objects. An example is reproduced in Fig. 1. There are several variations on hierarchical clustering that diVer in the rules governing the way distances are measured between clusters as they are constructed. Each of these will produce slightly diVerent results, as will any of the algorithms if the metric distance is changed. One potential problem with many hierarchical clustering methods is that as clusters grow, the expression vector that represents the cluster might no longer represent any of the genes in the cluster. Consequently, as clustering progresses, the actual expression patterns of the genes themselves become less relevant. Furthermore, if a bad assignment is made early in the process, it cannot be corrected. An alternative, which can avoid these artifacts, is to use a divisive clustering approach, such as k-means or selforganizing maps, to partition data (either genes or experiments) into groups that have similar expression patterns. d. k-Means Clustering. Having a priori knowledge about the number of clusters that should be represented in the data, k-means clustering is a good alternative to hierarchical methods. In k-means clustering, objects are partitioned into a fixed number (k) of clusters, so the clusters are internally similar but externally diVerent; no dendrograms are produced. An example is reproduced in Fig. 1. Some implementations of k-means clustering allow not only the number of clusters, but also seed cases (or genes) for each cluster, to be specified. This has the


103

potential to allow, for example, use of previous knowledge of the system to help define the cluster output. For example, an attempt to classify patients with two morphologically similar but clinically distinct diseases using microarray expression patterns can be imagined. By using k-means clustering on experiments with k ¼ 2, the data will be partitioned into two groups. The challenge then faced is to determine whether there are really only two distinct groups represented in the data or not. The main disadvantage of the k-means algorithm is that the number of clusters, k, must be supplied as a parameter. A simple validity measure based on the intracluster and intercluster distance measures can be used to determine automatically the number of clusters. e. Semantic Clustering. Cluster analysis is a methodology to identify groups of genes that share expression characteristics and behaviors. It has been frequently exploited in the analysis of genome-wide expression data as the experimental observation that a set of genes that is coexpressed implies that the genes share a biological function and are under common regulatory control. Frequently, the clustering is used to group genes considering only similar expression profiles, but it does not consider other well-known features of the gene properties. Actually, genes with a diVerent profile expression could have similar functions as well and the classic clustering methodologies do not put it in evidence. To extract knowledge from gene expression information, cluster analysis can be organized in two approaches: numerical and semantic clustering. The numerical clustering method is applied to the levels of gene expression. It tends to group genes with a similar expression profile in the same clusters and makes sure that genes having diVerent profiles with similar semantic features fall in diVerent clusters. These considerations suggest that simple numerical clustering algorithms are inadequate to infer the genes and proteins role. To discover more complex relationships among gene sequences, semantic clustering is used. It allows to group genes showing common biological characteristics. The term semantic clustering indicates methods of clustering based on semantic characteristics such as gene ontologies. Before performing the semantic clustering, the features have to be turned in to numerical values. At the beginning, the semantic clustering turns the features in numerical values to transform features that are similar functionally in near values. After that, methods of classic numerical clustering can be applied to that data set. In this way, the cluster analysis could make groups with similar semantic features but diVerent profiles. f. Visual Representation of Clustering. To interpret the results from any analysis of multiple experiments, it is helpful to have an intuitive visual representation. A commonly used approach relies on the creation of an expression matrix in which each row of the matrix represents the expression vector for a particular gene and each column represents a single experiment. Coloring each of the matrix elements on the basis of its expression value creates a visual representation of gene

104

CAVALLARO

expression patterns across the collection of experiments. The most commonly used method colors genes on the basis of their relative expression level in each experiment (Fig. 1). For each element in the matrix, the relative intensity represents the relative expression, with brighter elements being more highly diVerentially expressed. III. Use of DNA Microarrays for Studying Learning and Memory

This section focuses on the use of DNA microarray technology to dissect and analyze the pathophysiology of learning and memory in the mammalian brain. We start with experiments performed in diVerent behavioral paradigms to study the physiology of learning and memory, and then we move to an animal model of cognition disorders. A. Physiology of Learning and Memory 1. Eyelid Conditioning To begin a comprehensive survey of the molecular mechanisms that underlie LTM, we have used cDNA microarray technology to perform genomewide expression analysis after classical conditioning of the rabbit’s nictitating membrane response (NMR), a uniquely well-controlled associative learning paradigm (Fig. 1) (Cavallaro et al., 2001). Classical conditioning of the rabbit NMR involves the presentation of an innocuous stimulus such as a tone followed by a noxious stimulus such as air puV to or electrical stimulation around the eye (Gormezano et al., 1962). Extensive lesion and recording data have implicated the cortex of the cerebellum and in particular, lobule HVI, in classical conditioning of the rabbit NMR (Berthier and Moore, 1986; Gould and Steinmetz, 1996; Gruart and Yeo, 1995; Schreurs et al., 1991; Yeo et al., 1985). Although the hippocampus may not be necessary for NMR conditioning, recording data do show consistent eye blink conditioning-specific hippocampal changes (Coulter et al., 1989; Sanchez-Andres and Alkon, 1991). In addition, imaging studies have implicated both structures in human eye blink conditioning (Blaxton et al., 1996; Logan and Grafton, 1995; Molchan et al., 1994; Schreurs et al., 1997). Messenger RNA levels from cerebellar lobule HVI and hippocampus of unpaired and paired rabbits were simultaneously analyzed with high-density cDNA microarrays containing more than 8700 cDNAs (Cavallaro et al., 2001). When gene expression patterns were compared, mRNA levels of 79 (0.9%) and 17 (0.2%) genes diVered more than twofold in lobule HVI and hippocampus, respectively (Fig. 2B and C). Approximately 50% (eight) of the genes diVerentially


105

expressed in the hippocampus were also diVerentially expressed in the HVI lobule, suggesting common mechanisms of memory storage in the two areas. A majority of diVerentially expressed genes were downregulated, whereas only two genes that diVered by a factor greater than 2 were upregulated in lobule HVI of paired animals (Fig. 2B). Because LTM can be blocked by transcription and protein synthesis inhibitors, most previous reports have focused on the identification of proteins whose expression is upregulated (Davis and Squire, 1984). The preponderant reduction of gene expression during LTM, therefore, would not have been predicted and provides new and unexpected insights into the molecular mechanisms that underlie it. The specific role of the downregulation of these genes following learning remains a matter of speculation. Downregulation of a gene may be the end-point in a dynamic gene expression process that begins with upregulation during acquisition of the learned response. Alternatively, memory storage may require a balance of upregulation of some genes and downregulation of genes that exert inhibitory constraints on memory formation (Alberini et al., 1994). These latter genes might be termed memory suppressor genes (Abel and Kandel, 1998). Although our data represented the average gene expression from separate microarray analyses of cerebellar and hippocampal tissue obtained from a group of paired and a group of unpaired rabbits, there could be diVerences in gene expression between individual rabbit-derived tissues or between trained and sit control animals. To address these questions and confirm the microarray results, we selected eight (10%) of the genes that were diVerently expressed and performed in situ hybridization in cerebellum and forebrain tissue sections from individual paired, unpaired, and sit rabbits. In addition to corroborating the microarray data, the in situ hybridization analysis revealed distinct spatial distribution patterns of the genes. Figure 2D shows the regional mRNA expression of EST W18585.1, insulin-like growth factor-I (IGF-I), and Bach 2. All these mRNAs were abundantly expressed in the cerebellar cortex and were reduced in the lobule HVI of paired rabbits. In addition to this lobule, downregulation of EST W18585.1 was also found in other cerebellar lobules. In paired animals, a marked downregulation of Bach 2 was also revealed in the dentate gyrus, CA1, and CA3 areas of the hippocampus. A majority of the diVerential expressed genes implicated have no recognized function and are not yet named. Complete nucleotide sequence determination, conceptual translation, expression monitoring, and biochemical analysis are underway and should provide a detailed functional understanding of these genes. Seventeen genes have significant similarity to known genes and can be grouped into three classes (Fig. 2E): (1) signal transduction, (2) protein modification, and (3) DNA transcription regulation. It is important to note that some of these genes have been previously related to synaptic plasticity, memory, or cognitive disorders.

106

CAVALLARO

Fig. 2. Microarray analyses of eye-blink–conditioned rabbits. (A) Mean percent conditioned responses in paired, unpaired, and sit control rabbits as a function of three training sessions. To relate changes in gene expression to a learning task, we used pairings of a tone and periorbital electrical


107

a. Signal Transduction. The first group of genes encodes proteins involved in signal transduction and includes growth factors and proteins engaged in phosphorylation. One of the identified growth factors is IGF-I, a peptide with trophic and neuromodulatory actions. In the cerebellum, IGF-I is locally synthesized by Purkinje cells but also originates from climbing fibers, which are thought to convey information to the cerebellum about the reinforcing properties of the unconditioned stimulus. IGF-I modulates the size of dendritic spines on Purkinje cells (Nieto-Bona et al., 1997) and inhibits glutamate-induced -aminobutyric acid (GABA) release by Purkinje cells (Castro-Alamancos and Torres-Aleman, 1993). Interestingly, IGF-I levels have been correlated with cognitive test performance in aging humans (Aleman et al., 2000) and administration of IGF-I has been shown to ameliorate age-related behavioral deficits in rats (Markowska et al., 1998). Two diVerentially expressed growth factors whose functions in the central nervous system (CNS) are not known were growth diVerentiation factor-9 (Fitzpatrick et al., 1998), a member of the transforming growth factor- (TGF-) family, and a fibrinogen/angiopoietin-related protein (Kim et al., 2000). In lobule HVI, we observed the combined downregulation of a leukocyte common antigen-related (LAR) protein-tyrosine phosphatase and liprin-beta 2, a LARinteracting protein-like gene. The LAR gene is a transmembrane protein tyrosine phosphatase (PTPase) with sequence similarity in the extracellular region to cell adhesion molecules such as the neural cell adhesion molecule NCAM (Zhang et al., 1994). Liprins function to localize the LAR tyrosine PTPase at specific sites stimulation in a standard delay conditioning procedure, training rabbits to asymptotic levels of conditioning over 3 consecutive days. Paired rabbits (n ¼ 12) acquired conditioned responses to the tone and reached a mean terminal level of 94.7% conditioned responses, whereas the unpaired control rabbits (n ¼ 12) responded to the tone at mean levels of less than 1.3% across the 3 days of stimulus presentations and sit control rabbits (n ¼ 5) had spontaneous blink rates of less than 1% ( p < .001). Without further training or testing, rabbits show a level of 80% conditioned responses as long as 1 month after the 3 days of the stimulus pairings used in the present experiments. Consequently, harvesting cerebellar and hippocampal tissue 24 hours after 3 days of pairings ensured that rabbits were still at an asymptotic level of conditioning. Scatter plot of gene expression levels for paired and unpaired animals in (B) cerebellar lobule HVI and (C) hippocampus. Messenger RNA (mRNA) levels from cerebellar lobule HVI and hippocampus of unpaired and paired rabbits (n ¼ 7 per group) were simultaneously analyzed with high-density complementary DNA (cDNA) microarrays containing more than 8700 cDNA mouse clones with a length of 500–5000 bp and with averages in the 1-kb region. The estimated percentage of homology between mouse clones and rabbit genes is 88.98 3.7 (mean SD). The cross-species similarity and a complete list of the diVerentially expressed genes are available online at www.ct.isn.cnr.it/genomic-center/microarraydata/eye-blink.htm as supplementary information. (D) In situ hybridization validation of microarray results. Specific riboprobes labelled with [-35S] for insulin-like growth factor-I, Bach-2, and EST W18585.1 were hybridized with brain sections of sit paired and unpaired rabbits. Labeled mRNA signals were revealed with autoradiography. The color spectrum on the right side of each panel represents the pixel value of gray levels. (E) DiVerentially expressed genes with a known function are ordered into functional groups. (See Color Insert.)

108

CAVALLARO

on the plasma membrane, possibly regulating their interaction with the extracellular environment and their association with substrates (Serra-Pages et al., 1998). Although the extracellular ligands and physiological substrates of LAR-PTPase are not known, it may be part of specific signal transduction cascades that have eVects on neuronal plasticity by functioning as signal transducers of cell contact phenomena. The final identified gene that may play a role in signal transduction is phocein, a protein that binds striatin, a Ca2þ/calmodulin-binding protein mostly found in dendritic spines where it is essential for the maintenance and growth of dendrites (Bartoli et al., 1999). b. Protein Modification. The group of proteins involved in protein degradation includes a protein similar to CD156, a transmembrane glycoprotein with metalloprotease activity (Kataoka et al., 1997); the F-box protein FBX8, a specificity-conferring component of the ubiquitin protein ligase SCFs complex, which functions in phosphorylation-dependent ubiquitination of a wide array of regulatory molecules (Winston et al., 1999); and hippostasin, a brain-related serine protease of unknown function. Although the substrates of these proteindegrading enzymes are unknown, their diVerential expression may play a critical role in synaptic plasticity and axonal remodeling. c. Transcription Regulation. Among the group of diVerentially expressed genes involved in transcription regulation, 7SK is a small nuclear RNA involved in the control on transcription (Krause, 1996), TR2 is an orphan receptor belonging to the family of steroid/thyroid hormone receptors (Young et al., 1998), whereas the rest have functions related to transcription factors. One of these is similar to the CCAAT enhancer-binding protein (C/EBP) family of transcription factors, which have been implicated in LTM consolidation after inhibitory avoidance learning (Taubenfeld et al., 2001) and longterm facilitation, a synaptic mechanism that in Aplysia is thought to contribute to LTM (Alberini et al., 1994). Interestingly, selectively enhanced contextual fear conditioning (24 hours after training) has been shown in mice lacking the transcriptional regulator C/EBP-delta, implicating some isoforms of this family of proteins in specific types of learning and memory as memory suppressor genes (Sterneck et al., 1998). WBSCR11 is a putative transcription factor gene that is commonly deleted in Williams-Beuren syndrome and may contribute to the spectrum of developmental symptoms that includes mental retardation and profound impairment of visuospatial cognition (Osborne et al., 1999). The bifunctional protein dimerization cofactor of transcription factor HNF1/pterin-4-alpha-carbinolamine dehydratase (DCoH/PCD) is both a dimerization cofactor of transcription factor HNF1 and a cytoplasmatic enzyme PCD involved in the regeneration of tetrahydrobiopterin, the cofactor for aromatic amino acid hydroxylases (Strandmann et al., 1998). Bach 2 is a transcription factor almost exclusively expressed in neurons that forms heterodimers with MafK and may play important roles in coordinating transcription activation and repression (Oyake et al., 1996). Pirin is a putative nuclear factor


109

I–interacting protein (Wendler et al., 1997). DRG11 is a paired homeodomain protein specifically expressed in sensory neurons and a subset of their CNS targets (Saito et al., 1995). The data reported previously were the first reported in the literature to demonstrate the feasibility and utility of a cDNA microarray system as a means of dissecting the molecular mechanisms of associative memory. Further studies, however, were required at diVerent times and behavioral conditions to better understand the role of the implicated genes. To perform such studies, we changed animal species and moved to rat, which is better suitable for genomic studies than rabbit in terms of sequenced genes and available microarrays. In the following two sections, we review studies obtained in rats following water-maze and passive-avoidance training using the same microarray platform. 2. Water-Maze Learning We measured hippocampal gene expression profiles in naive, swimming control and water-maze–trained animals using microarrays containing more than 1200 genes relevant to neurobiology (Fig. 3) (Cavallaro et al., 2002). When gene expression profiles in naive and swimming control animals 1, 6, and 24 hours after swimming sessions were compared, 345 genes (27.3%) were found diVerentially expressed more than twofold in at least two of the four conditions (Fig. 3C). These genes, operationally defined as ‘‘physical activity–related genes’’ (PARGs), indicate that physical activity and mild stress associated with behavioral training has a significant impact on hippocampal gene expression. When gene expression levels in swimming control animals were compared to water-maze–trained animals 1, 6, or 24 hours after training, 140 genes (11%) were found diVerentially expressed and operationally defined as ‘‘memoryrelated genes’’ (MRGs) (Fig. 3C). Most of these MRGs (110 out of 140) were also PARGs, that is, influenced by physical activity. Among MRGs, 55 genes were upregulated in the hippocampus of water-maze–trained animals (Fig. 1D, G, I, and M), whereas 91 genes were downregulated (Fig. 1E, F, H, and L). Most of the MRGs, those diVerentially expressed between the swimming and spatial learning animal groups, were also aVected during swimming alone but with entirely diVerent temporal patterns of expression (Fig. 3F–M). Although learning and physical activity involves common groups of genes, the behavior of learning and memory can be distinguished from unique patterns of gene expression across time. Genes implicated by gene expression profiling participate in various stages of learning and memory, and further studies are required to fully characterize their exact role. Their encoded proteins, however, may represent potential drugs or molecular targets whose activity and modulation may improve successive stages of memory (e.g., learning, consolidation, and long-term retention), under normal conditions and in disorders that aVect cognitive functioning, such as

110

CAVALLARO

Alzheimer’s disease. A promising example is represented by fibroblast growth factor-18 (FGF-18), one peptide whose sustained increase during memory retention was implicated by microarray analysis (Fig. 1G, I, and M). To explore the eVect of FGF-18 in spatial learning, we tested the eVects of a single exogenous dose of FGF-18. As shown in Table I, animals treated intracerebroventricularly with 0.94 pmol of FGF-18 displayed significantly improved spatial learning behavior ( p < .05) compared with vehicle-injected control animals. FGF-18 treatment induced a 49% reduction in the escape latency but no significant changes in motor activity. All of the ‘‘LRGs’’ identified have a recognized function and can be classified into six major groups based on their translated product (a complete list of the diVerentially expressed genes and their function is available online at www.ct.isn.cnr.it/genomic-center/microarray-data/water-maze.htm as supplementary information): (1) cell signaling, (2) synaptic proteins, (3) cell–cell interaction and cytoskeletal proteins, (4) apoptosis, (5) enzymes, and (6) transcription or translation regulation. Some of these genes have been previously related to synaptic plasticity, memory, or cognitive disorders, whereas others provide a significant number of unique and novel entry points. The exact role and functional relationships of the genes and proteins implicated, however, are


Fig. 3. Continued.

111

112

CAVALLARO

113


TABLE I Effects of a Single Exogenous Administration of Fibroblast Growth Factor-18 on Water-Maze Learning Latency (sec) Day 1 Control Drug

48.2 16.1 46.6 16.7

Distance (m) Day 2

37.1 11.1 19.2 6.3a,b

Day 1

Day 2

13.6 4.3 15.2 5.7

10.4 3.1 6.1 2.0a

Note: Thirteen male Wistar rats (250–300 g) were implanted stereotaxically with stainless-steel guide cannulas in the right and left lateral ventricles. On day 1, 1 week after surgery, animals were subjected to a 2-min swimming training session. Then, a water-maze training session was performed on days 2 and 3, and consisted of finding a submerged platform to escape from the water. Two trials were given to the animal for each session. The escape latency and distance to find the platform were monitored as described above. Ten minutes after the second trial on day 2, an intracerebroventricular administration of drug or vehicle was performed in both lateral ventricles. Six animals received 0.94 pmol of FGF-18 and the other seven received a control injection of vehicle (saline). a Day 1 vs day 2, p < .05. b Control vs drug, p < .05.

Fig. 3. Water-maze learning. (A) Escape latencies of rats swimming to a submerged platform in the water maze during four consecutive trials. To reduce stress in the experimental day, the first day was dedicated to swimming training in the absence of an island. Each rat was placed in the pool for 2 minutes and was returned to its home cage. On the next day, half of the rats were placed again in the pool for a 2.5-minute swimming session and were used as swimming controls. The other half were given four consecutive trials to locate the platform, each trial lasting up to 2 minutes. Rats were required to spend 30 seconds of an intertrial interval on the platform. The rats’ escape latency was measured using a HVS2020 video tracking system (HVS Image Ltd, U.K.). (B) Probe trial. To verify that the trained rats in fact learned the spatial location of the island, six rats were trained to find the island and tested 24 hours later on a quadrant analysis test. The trained rats swam significantly longer in the quadrant (red) where the island was located. (C–E) Venn diagrams of diVerentially expressed hippocampal genes. Hippocampal gene expression profiles in naive swimming–control, and water-maze–trained rats were measured using microarrays containing 1263 genes relevant to neurobiology (AVymetrix GeneChip Rat Neurobiology U34 array). Genes diVerentially expressed in naive and swimming control animals 1, 6, and 24 hours after training were operationally defined as ‘‘physical activity–related genes’’ (PARGs), whereas genes diVerentially expressed in water-maze– trained animals compared to swimming controls were operationally defined as ‘‘memory-related genes’’ (MRGs) (C). Among these, 55 genes were upregulated (D), whereas 91 genes were downregulated (E) in at least one of three time points examined. (F–M) DiVerentially expressed genes in swimming control versus water-maze–trained animals at 1, 6, and 24 hours after training. A complete list of the diVerentially expressed genes is available online at www.ct.isn.cnr.it/genomiccenter/microarray-data/water-maze.htm as supplementary information. (See Color Insert.)

114

CAVALLARO

presumably those we cannot yet recognize. For this reason, in the following section, we discuss only some of the MRGs implicated by microarray analysis during spatial learning. a. Cell Signaling. The group of genes involved in cell signaling is the largest and includes a subgroup of neuropeptides, growth factors, and their receptors. Among them is FGF-18, a novel member of the FGF family, which was shown to stimulateneurite outgrowth (Ohbayashi et al., 1998). Although the function of this peptide is still unknown, the other members of its family are important signaling molecules in several inductive and patterning processes and act as brain organizer– derived signals during formation of the early vertebrate nervous system. Watermaze training but not physical activity induced the expression of FGF-18. This, together with the ability of FGF-18 to enhance spatial memory when exogenously administered, is strong and novel evidence in favor of its involvement in learning and memory. DiVerential expression of interleukin-1 (IL-1), interleukin-15, and interleukin-2 receptor chain suggest a physiological role of brain cytokines in memory consolidation processes. Indeed, the reduction of IL-1 mRNA in water-maze– trained animals is consistent with previous studies showing that central IL-1 administration and agents that induce central IL-1 activity impair the consolidation of memories that depend on the hippocampal formation (Rachal et al., 2001). Enhanced expression of corticotropin-releasing hormone in water-maze– trained animals is also in line with other evidence obtained in another learning paradigm (Lee et al., 1996). The subgroup of G-protein–coupled receptors includes two GABA B-type receptor splice variants, GABA-B1d and GABA-B2a. Functional GABA-B receptors, whose function depends on dimerization of GABA-B1 and GABA-B2, are known to activate second messenger systems and modulate potassium and calcium channel activity, thereby controlling presynaptic transmitter release and postsynaptic silencing of excitatory neurotransmission (Dutar and Nicoll, 1988). GABA-B receptor agonists or antagonists are known to impair or facilitate, respectively, cognitive performance in the Morris water-maze task and other kinds of learning (Mott and Lewis, 1994). By reducing GABA-B receptor signaling, the downregulation of GABA-B1d and GABA-B2a 1 hour after water-maze training may exert a mnemonic eVect similar to that produced by GABA-B receptor antagonists. Dopamine 1A and D4 receptors are downregulated and upregulated, respectively, 1 hour after water-maze training. These receptors are coupled to diVerent G proteins and their change in expression may allow modulation of neuronal dopamine-mediated signal. The opioid receptor–like is decreased 1 hour after WM training. This receptor is a G protein-coupled receptor structurally related to the opioid receptors,


115

whose endogenous ligand is the heptadecapeptide nociceptin that has been implicated in sensory perception, memory process, and emotional behavior (Calo’ et al., 2000). The adenosine receptor A1, which is negatively coupled to adenylate cyclase, decreases 1 hour after WM training. Adenosine is thought to exert a tonic inhibitory role on synaptic plasticity in the hippocampus (de Mendonca and Ribeiro, 1994). Its decrease, therefore, may exert a facilitatory role during learning and memory. The insulin receptor was increased in swimming control and decreased in water-maze–trained rats. Instead, the precursor of its endogenous ligand, insulin, was detectable only 24 hours after WM training. The fine balance of brain insulin and its receptor may regulate cognitive functions (Park, 2001). The subgroup of ligand-gated ion channels includes five GABA-A receptor subunits, which were all diVerentially expressed 1 hour after water-maze training. Four of them, 4, 5, 2, and 2 were downregulated, whereas one, the pi subunit, was upregulated. Changes in the expression of specific GABA-A receptor subunits may aVect the composition and pharmacology of GABA-A receptor assemblies. These changes may also be relevant in consideration of vast number of drugs such as anxiolytics, anticonvulsants, general anesthetics, barbiturates, ethanol, and neurosteroids, which are known to elicit at least some of their pharmacological eVects via GABA-A receptor subunits (Smith, 2001). The expression of glutamate ionotropic receptors is dynamically regulated during spatial learning. N-methyl-d-aspartic acid receptor (NMDA-R) 1, which possesses all properties characteristic of the NMDA receptor–channel complex, is downregulated 1 hour after water-maze training, whereas NMDA-R2A, which has regulatory activities, is upregulated after 24 hours. One L-alpha-amino-3hydroxy-5-methylisoxazole-4-propionate (AMPA) receptor 3 subunit is downregulated 1 hour after training. Two kainate receptors, GluR6 and GluR5-2, are upregulated 6 and 24 hours after training, respectively. Plastic changes of diVerent combinations of glutamate receptors might have profound eVects on glutamate responsiveness (Madden, 2002). The subgroup of ion channels includes several proteins that play a role in the maintenance of ionic homeostasis. Among these are 10 potassium (Kþ) channel subunits: two Shaker (Kcna5 and Kcna6), two Shab (Kcnb1 and Kcnb2), one Shal (Kcnd2), and one EAG-related (Kcnh5) voltage-dependent Kþ channel subunits; one Ca2þ-activated (Kcnn2) and three inwardly rectifying (Kcjn4, Kcjn11, and Kcjn16). Expression changes of diVerent Kþ channel subunits may alter composition of the channel complexes and aVect cellular excitability (Choe, 2002). Although the exact contribution of each of these subunits during spatial memory is unknown, 7 of the 10 are downregulated after water-maze training and may produce increased excitability.

116

CAVALLARO

The subgroup of proteins involved in intracellular signaling includes several proteins involved in the intracellular homeostasis of calcium, sodium, and potassium ions. Among these is the frequenin homolog, also known as neuronal calcium sensor-1, which has been shown to regulate associative learning (Gomez et al., 2001). The subgroups of proteins involved in neurotransmitter transport include GABA, glutamate, and serotonin transporters. The GABA and glutamate transporters are downregulated 1, 6, or 24 hours after water-maze training, whereas the serotonin transporter is upregulated after 1 hour. Neurotransmitter uptake by nerve terminals and glial cells is crucial for providing a reservoir of transmitter or transmitter precursors and the termination of synaptic events (Masson et al., 1999). Changes in the expression of these transporters, therefore, may have profound eVects on neurotransmission by controlling neurotransmitter levels at the synaptic cleft. The subgroup of signaling enzymes includes a number of proteins previously implicated in learning and memory. After water-maze training, a strong induction of the inducible form of nitric oxide synthase (iNOS) was observed. This enzyme produces nitric oxide (NO), a molecule involved in neurosynaptic transmission, and is induced in many pathological conditions. Although the role of NO in learning and memory is still unclear, some studies have reported that systemic NO inhibition had deleterious eVects in water maze learning (Chapman et al., 1992; Estall et al., 1993; Yamada et al., 1995). The role of iNOS in the hippocampus, therefore, may go beyond its well-established detrimental function in neurological disorders and could contribute to the mechanisms underlying learning and memory. Two genes encoding enzymes involved in the mitogen-activated protein kinase (MAPK) signaling cascade, p38 MAPK and MAPK phosphatase, were found diVerentially expressed after water maze training. This signaling cascade has been previously implicated in the development of synaptic plasticity underlying learning and memory (Impey et al., 1999; Kornhauser and Greenberg, 1997; Zhen et al., 2001). However, there are three subfamilies of MAPKs that are activated by diVerent upstream cascades and are involved in the regulation of distinct nuclear transcriptional factors (Davis, 1993). As suggested by the present observations and previous studies (Berman et al., 1998), LTM may involve diVerent MAPKs and/or their MAPK phosphatases. DiVerential expression of two Ca2þ/calmodulin-dependent protein kinases, belonging to a class of signaling enzymes extensively implicated in memory formation and consolidation (Mayford et al., 1996), was observed after water-maze training. Other proteins involved in signal transduction include Ania-3, a short form of the Homer family of proteins, which bind to group I metabotropic glutamate receptors, inositol trisphosphate receptors, ryanodine receptors, and NMDA


117

receptor–associated Shank proteins and have been implicated in synaptogenesis, signal transduction, receptor traYcking, and axon pathfinding (Xiao et al., 2000). The long Homer forms are constitutively expressed and self-associate to function as adaptors to couple membrane receptors to intracellular pools of releasable Ca2þ. The short Homer forms compete with the long Homer proteins for binding to signaling components, thus functioning as endogenous dominantnegative regulators of receptor-induced Ca2þ release from intracellular stores. Downregulation of Ania-3 in water-maze–trained animals may modulate the properties of the long Homer forms and be involved in activity-dependent alterations of synaptic structure and function. Upregulation of another signaling molecule, citron, was found 24 hours after water-maze training. Citron is a neuronal Rho-target molecule associated to the postsynaptic scaVold protein PSD-95, which plays an important role in the anchoring and clustering of neurotransmitter receptors at synapses (Zhang et al., 1999). The expression of citron may provide a cross-talk between the Rho signaling pathway, which has been implicated in mechanisms of neuronal plasticity, and neurotransmitter receptors like the NMDA receptor. b. Cell–Cell Interactions and Cytoskeletal Protein. The group of cell–cell interactions and cytoskeletal proteins includes a vast number of proteins whose change in expression may reflect morphological adaptation of brain cells during formation of memory. Among them, for example, is delta-catenin, a component of the cell– cell adherens junctions expressed specifically in the nervous system. Delta-catenin is downregulated during neuronal migration and expressed in the apical dentrites of postmitotic neurons (Ho et al., 2000). Changes in delta-catenin expression, therefore, are considered fundamental for the establishment and maintenance of dendrites and synaptogenesis. Delta-catenin was originally discovered as an interactor with presenilin-1 (Zhou et al., 1997) whose mutation causes early onset familial Alzheimer’s disease. In addition, hemizygosity of delta-catenin is associated with severe mental retardation in the cri-du-chat syndrome, which is associated with severe mental retardation (Medina et al., 2000). The hippocampal expression of several proteins involved in microtubule formation was reduced 1 hour after water-maze training. Among these are beta-tubulin, neuraxin, and microtubule-associated protein-2 (MAP2) and -5. Reduced expression of MAP2, in particular, was confirmed in three redundant probe sets. Altered expression of MAP2, which is critical for dendritic stability (Kosik et al., 1984), has been shown with contextual memory, long-term potentiation, aging, epilepsy, Alzheimer’s disease, and Rett syndrome (Fukunaga et al., 1996; Kaufmann et al., 1995; Kosik et al., 1984; Leterrier and Eyer, 1992; Woolf et al., 1999; Yamanouchi et al., 1998). We have also found altered expression of MAP2 in a transgenic animal model of fragile X syndrome (see below) (D’Agata et al., 2002), which shows behavioral deficits in the Morris water maze (the Dutch-Belgian Fragile X Consortium, 1994). Expression of

118

CAVALLARO

several others proteins involved in cell–cell and cell–matrix interactions was found increased (intercellular adhesion molecule-1, C-CAM2a isoform) or more often decreased (neurexin-1, connexin-43, contactin-1, chondroitin sulfate proteoglycan-3, myelin-associated glycoprotein, and axonal glycoprotein). CAMs have already been implicated in synaptic plasticity, learning, and memory (Wright et al., 2002). Together, their changes may be critical in regulating cell–cell recognition and establishing mature dendritic relationships in the neuropil. c. Apoptosis. The group of proteins involved in apoptosis includes Bcl-2 relateddeath gene product BOD-L, caspase-1, caspase-6, and DP5 that are all upregulated after water-maze training. In agreement with other studies (Mattson and Duan, 1999), our data suggest that beyond their roles in cell death, apoptotic and antiapoptotic cascades may play roles in synaptic plasticity. d. Enzymes. The group of enzymes includes two proteins involved in free radical metabolism, heme oxygenase-1, and superoxide dismutase-3, whose expression was reduced in the hippocampus of water-maze–trained animals. Besides their role in oxidative stress, these enzymes may be implicated in other physiological roles such as learning and memory. Indeed, impaired spatial memory is found in mice overexpressing these two proteins (Gahtan et al., 1998; Morgan et al., 1998). e. Transcription or Translation Regulation. Among the group of diVerentially expressed genes involved in transcription or translation regulation is the upregulated gene encoding for cyclin Ania-6a, whose splicing is dynamically controlled by diVerent forms of neuronal stimulation (Berke et al., 2001), and Jun-B, which is induced after diVerent memory tasks (Tischmeyer and Grimm, 1999). f. Synaptic Proteins. The group of synaptic proteins includes a number of proteins that regulate membrane traYcking and fusion. They include synaptojanin1, four members of the syntaxin family of proteins (syntaxin-2, -5, -8, and -12), five synaptotagmins (2, 4, 5, 7, and 8), and synaptosomal-associated protein-25. DiVerential expression of these proteins, which are involved in diVerent steps of membrane traYcking and fusion ( Jahn and Sudhof, 1999), may regulate synaptic plasticity by aVecting cellular functions, such as secretion, endocytosis, and axonal growth. The data obtained in hippocampus of water-maze–trained rats presented previously represent the first temporal gene expression comparison reported in the long-term retention of learning and memory and further demonstrate the utility of a cDNA microarray system as a means of dissecting the molecular basis of associative memory. This approach provides information on the gene expression changes that occur during physical activity, stress, learning, and memory and allows the identification of molecular targets and pathways whose modulation may allow new therapeutic approaches for improving cognition. As shown in previous studies or in the present section for FGF-18, pharmacological or genetic


119

modulation of some of these pathways can indeed be eVective in facilitating learning and memory. 3. Passive Avoidance Learning We have extended the genome-wide screenings described previously to an additional behavioral animal model, a step-through passive avoidance test, known to require hippocampus-dependent learning and dependence on transcription (Stubley-Weatherly et al., 1996). In these experiments, conditioned animals (CAs) were trained to avoid moving from the lighted to the darkened section of a conditioning chamber by delivering a foot shock when they entered the darkened section. Control rats included untrained (naive) animals and animals exposed to the unconditioned (USTA) or the conditioned (CSTA) stimulus. To verify that the trained rats in fact learned the passive avoidance task, learning was assessed in a comparable group of animals by evaluating the latency of stepthrough in a retention test. Twenty-four hours after the one-trial training period, only CAs learned to associate stepping through the darkened chamber with the foot shock (Fig. 4A). Hippocampal gene expression profiles in CAs, USTA, CSTA, and naive animals were measured 6 hours after training using microarrays containing 1263 genes relevant to neurobiology. Gene expression data in each of the four experimental conditions represented the average from four separate microarray analyses performed on hippocampal RNA samples from individual animals. When gene expression profiles of naive animals were compared to those of CSTA or USTA, 46 and 60 genes, respectively, were found diVerentially expressed (Fig. 4B). These genes indicate that physical activity and mild stress associated with behavioral training has a significant impact on hippocampal gene expression. When gene expression levels in naive animals were compared to those of CAs, 38 genes (3%) were found diVerentially expressed and operationally defined as MRGs (Fig. 4B). Among MRGs, 21 genes were downregulated and 17 genes were upregulated. Some of these MRGs (21 of 38) were also diVerentially expressed in CSTA (16) and USTA (16) (Fig. 4B). A hierarchical clustering method was used to group memory related genes on the basis of similarity in their expression patterns (Fig. 4C). The most evident traits of the clustered data was that MRGs showed entirely diVerent patterns of expression in CA versus CSTA or USTA. Genes segregating into nine major branches of the dendrogram were assigned to nine clusters (Fig. 4C). Clusters 1 through 4 represent those genes that were downregulated, whereas clusters 5 through 9 include those that were upregulated in CAs. Some of the MRGs, those diVerentially expressed between naive animals and CAs, were also aVected by exposing the rats to the conditioned or the unconditioned stimulus alone, whereas others were uniquely induced when the two were associated and the animals were conditioned (Fig. 4C, clusters 2 and 8). Expression changes of MRGs in

120

CAVALLARO

Fig. 4. Passive avoidance learning. (A) Passive avoidance retention test. Conditioned animals (CAs) were trained to avoid moving from the lighted to the darkened section of a conditioning chamber by delivering a foot shock when they entered the darkened section. Control rats included untrained (naive) animals, and animals exposed to the conditioned (CSTA) or the unconditioned (USTA) stimulus. Twenty-four hours after the training trial, half of the animals (n ¼ 4 per group)


121

CSTA or USTA had diVerent magnitudes or more often opposite trends than CA (Fig. 4C, clusters 1, 2, 3, 5, 8, and 9). As we have previously observed in watermaze–trained animals, learning, physical activity, and mild stress associated with behavioral training involve common groups of genes. Their behavior in learning and memory, however, could be distinguished from unique patterns of gene expression, as shown in the clustered data. All of the MRGs identified have a recognized function and can be classified into four major groups based on their translated product (Fig. 4C): (1) cell signaling, (2) synaptic and cytoskeletal proteins, (3) apoptosis, and (4) transcription regulation. Some of these genes have been previously related to synaptic plasticity, memory, or cognitive disorders. Six of thirty-eight MRGs found in the hippocampus of rats after passive avoidance training (Fig. 4C, shown in bold) were also diVerentially expressed after water-maze learning (Fig. 3), suggesting common mechanisms of memory storage in diVerent behavioral paradigms.

performed the retention test to verify that the trained rats in fact learned the passive avoidance task. The animals were placed in the safe compartment with the door closed. After 2 minutes of acclimation, the light turned on, the door opened, and the animal was allowed to enter the dark compartment. The latency to enter the dark compartment was recorded and used as the measure of retention. The rats avoiding the dark compartment for more than 300 seconds were considered to have a memory of the training experience. During the retention trial, CAs had a longer mean stepthrough latency than naive, CSTA and USTA (*p < .001). (B) Venn diagrams of diVerentially expressed hippocampal genes. Hippocampal gene expression profiles in CA, USTA, CSTA, and naive animals were measured 6 hours after training using microarrays containing 1263 genes relevant to neurobiology (AVymetrix GeneChip Rat Neurobiology U34 array). Genes diVerentially expressed between naive and CSTA were defined as ‘‘conditioned stimulus–related genes’’ (CSRGs); genes diVerentially expressed between naive and USTA were defined as ‘‘unconditioned stimulus–related genes’’ (USRGs); genes diVerentially expressed between naive and CA were defined as ‘‘memoryrelated genes’’ (MRGs). (C) Hierarchical clustering of MRGs. A hierarchical clustering algorithm (Pearson correlation, separation ratio 0.2, minimum distance 0.001) was used to order MRGs in a dendrogram in which the pattern and length of the branches reflects the relatedness of the samples. Data are presented in a matrix format: Each row represents a single gene and each column an experimental condition. The averaged normalized intensity from four replicates is represented by the color of the corresponding cell in the matrix. Green, black, and red cells, respectively, represent transcript levels below, equal to, or above the median abundance across all conditions. Color intensity reflects the magnitude of the deviation from the median (see scale at the bottom). The graphs on the left of the dendrogram represent the averaged natural log of normalized data SEM of the genes in nine major clusters. The gene expression ratio between naive and CA is shown on the right of the matrix. Functional classification of MRGs is represented in a column on the right of the figure where each functional classes or subclasses are color-coded. The name and GenBank accession number of MRGs uniquely regulated in CAs are indicated in italic, whereas MRGs previously found to be diVerentially expressed in the hippocampus of water-maze–trained rats (Cavallaro et al., 2002) are indicated in bold. A complete list of the diVerentially expressed genes is available online at www.ct.isn.cnr.it/genomic-center/microarray-data/passive-avoidance.htm as supplementary information. (See Color Insert.)

122

CAVALLARO

In the following paragraphs, we discuss only some of the MRGs implicated by microarray analysis emphasizing those whose regulation in CA was diVerent than CSTA or USTA, and those we previously implicated in water-maze learning (Cavallaro et al., 2002). a. Cell Signaling. The group of genes involved in cell signaling is the largest and includes a subgroup of neuropeptides, growth factors, and their receptors. Among these is transforming growth factor (TGF)- receptor 3 whose downregulation in CA is in line with previous observations demonstrating impaired learning after administration of TGF- (Nakazato et al., 2002). DiVerential expression of interleukin8 (IL-8) and interleukin-12 suggests a physiological role of brain cytokines in memory consolidation processes. Recent evidence indicates that interleukins and their receptors are present in the CNS, where they can be involved in various eVects including neuroinflammatory processes, modulation of the synaptic transmission, and regulation of the synaptic connections in the brain (Asensio and Campbell, 1999; Hesselgesser and Horuk, 1999). Although the eVects of IL-8 and IL-12 on learning and memory are unknown, other chemokines are known to modulate cognition or are diVerentially expressed during memory consolidation (Cavallaro et al., 2002; Gibertini et al., 1995; Lynch, 2002; Ma and Zhu, 1997). Among the subgroup of G-protein–coupled receptors and their effectors are three serotonin (5-HT) receptors, the adrenergic 1a and dopamine-1A receptors. After training, we observed the coordinated reduction of 5-HT1B and 5-HT4, and the selective increase of 5-HT3A receptor mRNA in CAs. These data are in agreement with the involvement of the serotonin system through the interplay of diVerent receptor subtypes in learning and memory processes (Meneses, 1999) in addition to a variety of other behaviors, such as emotional states and impulse control (Bouwknecht et al., 2001). Administration of 5-HT1B receptor antagonists, for example, is known to prevent memory impairment and facilitate learning, whereas agonists for 5-HT1B generally have opposite eVects (Meneses, 1999). Furthermore, 5-HT1B receptor knockout mice exhibit enhanced memory performance (Malleret et al., 1999). The subgroup of ligand-gated ion channels includes the 1 and 3 subunits of the strychnine-sensitive glycine-gated chloride channels and the GABA-C receptor rho-3 subunit. Both ion channels have been involved in a variety of other behaviors ( Jentsch et al., 2002) and mediate fast postsynaptic inhibition in the brain, and altered expression of their subunits has been shown to modulate hippocampal excitability (Bormann, 2000; Chattipakorn and McMahon, 2002). The subgroup of ion channels includes three Ca2þ-activated potassium channels (Kcnmal, Kcnmb1, Kcnn3) and one voltage-gated sodium channel type IV . Although the exact contribution of these subunits during learning and memory is unknown, regulation of their expression can produce a flexible tuning of electrical excitability of hippocampal neurons in response to neurotransmitters. Indeed, downregulation of potassium channels (Kcnma1, Kcnn3) and upregulation of


123

the sodium channel after passive avoidance training may produce increased excitability. The subgroup of intracellular signaling proteins includes two kinases, phosphoinositide 3-kinase and p38 MAPK, whose expression was increased in CAs. These two signaling enzymes are implicated in a variety of receptor-stimulated cell responses and have been involved in neuronal synaptic plasticity and memory formation (Barros et al., 2001; Blum et al., 1999; Cavallaro et al., 2002; Ming et al., 1999; Sweatt, 2001; Zhen et al., 2001). b. Synaptic and Cytoskeletal Proteins. This group includes synaptogyrin-I, an abundant synaptic vesicle protein involved in short-term and long-term synaptic plasticity ( Janz et al., 1999), and microtubule-associated protein 1B (MAP1B), which regulates neuronal cytoskeleton during neurite outgrowth, plasticity, and regeneration (Edelmann et al., 1996). Decreased expression of these proteins may reflect morphological adaptation of brain cells during formation of memory. c. Apoptosis. The group of proteins involved in apoptosis includes Bcl-2 and death eVector domain-containing protein. The former is increased, whereas the latter is decreased in passive avoidance CAs. In agreement with other studies (Yamada et al., 1995), our data suggest that beyond their roles in cell death, apoptotic, and antiapoptotic cascades may play roles in synaptic plasticity. d. Transcription Regulation. Among the group of diVerentially expressed genes involved in transcription is the downregulated gene encoding for estrogen receptor-2, which is thought to mediate at least part of the complex and time-dependent eVects of estrogens on memory (Rissman et al., 2002).

B. Pathology of Learning and Memory As described previously, several of these genes implicated by microarray have been related to synaptic plasticity, memory, or cognitive disorders. The exact role and functional relationships of the genes and proteins implicated, however, are presumably those we cannot yet recognize. To obtain a more complete interpretive framework, we are extending our genome-wide expression analysis not only to diVerent behavioral paradigms but also to pathophysiological conditions. In this part of this chapter, we describe one of these analyses performed in FMR1 knockout mice, an animal model showing a phenotype that mimics the human syndrome, such as macro-orchidism and behavioral abnormalities (the Dutch-Belgian Fragile X Consortium, 1994). Fragile X syndrome is the most common inherited form of mental retardation and originates from the loss of FMR1 expression due to trinucleotide repeat expansion (Fu et al., 1991; Verkerk et al., 1991). In addition to global cognitive deficits, the disorder can be manifest as specific impairments in visuospatial learning and auditory and visual short-term memory (Fisch et al., 1996, 1999;

124

CAVALLARO

Freund and Reiss, 1991). Although the function of the FMR1 gene product, FMRP, is still unknown, the presence of three RNA binding regions (two KH domains and an RGB box) suggests that FMRP is an RNA binding protein (Ashley et al., 1993; Siomi et al., 1993). Indeed, in vitro–translated FMRP has been demonstrated to preferentially bind certain RNA homopolymers and to selectively bind a subset of brain transcripts including its own message (Siomi et al., 1993). The observation that FMRP is an RNA binding protein and may be implicated in RNA metabolism suggests that other genes, whose products may vary in the absence of FMRP, could play a significant role in the cognitive deficits associated with fragile X syndrome. To test this hypothesis and gain insights into the molecular mechanisms leading to mental retardation in fragile X syndrome, we have performed genome-wide expression analysis in brains of control wildtype littermates and FMR1 knockout mice. Among the gene expression changes that result from a deficiency of FMRP, our analysis involved a number of genes previously involved in other memory or cognitive disorders (D’Agata et al., 2002) (Fig. 5). Altered expression of MAP2, for example, has been shown with contextual memory, long-term potentiation, aging, epilepsy, Alzheimer’s disease, and Rett syndrome (Fukunaga et al., 1996; Johnson and Jope, 1992; Kaufmann et al., 1995; Kosik et al., 1984; Leterrier and Eyer, 1992; Woolf et al., 1999; Yamanouchi et al., 1998). MAP2 is heavily concentrated in mature dendrites and may be critical for dendritic stability ( Johnson and Jope, 1992). The Ser/Thr kinase KKIAMRE, whose expression is decreased in FMR1 knockout mice, is a cell division cycle 2–related protein kinase that has been shown to be induced after eyeblink conditioning (Gomi et al., 1999). Decreased expression of RAB, a member of the Rab small G-protein family, was also observed in FMR1 knockout mice. Rab proteins are key regulators of vesicular transport and play critical roles in synaptic plasticity, and their dysfunction has been linked to mental retardation phenotypes (Seabra et al., 2002). The Werner gene encodes a DNA helicase, which is mutated in the Werner syndrome, an autosomal recessive genetic disorder that is manifested by accelerated aging, as also reflected by extensive deposition of amyloid- peptide in the CNS (Leverenz et al., 1998). In FMR1 knockout mice, we also observed increased mRNA expression of the APP, whose altered expression has been extensively linked to Alzheimer’s disease and Down syndrome ( Jiang et al., 1999). Finally, altered expression of two proteins involved in the ubiquitin-proteosome protein degradation pathway, the ubiquitin-specific protease-7 and the ubiquitin-binding protein homolog, was found in the brains of FMR1 knockout mice. Abnormality in the ubiquitin system has been demonstrated in other cognitive disorders, such as Alzheimer’s disease and the Angelman syndrome (Lennon et al., 1996; Zhao et al., 2000).

125 Fig. 5. Microarray analysis of brains from wild-type and FMR1 knockout mice. (A) Scatter plot of 6789 genes showing measurable levels of expression in wild-type versus FMR1 knockout mice. Red lines indicate the n-fold (1, 2) change. A complete list of the diVerentially expressed genes is available online at www.ct.isn.cnr.it/genomic-center/microarray-data/FMR1.htm as supplementary information. (B–C). In situ hybridization validation of microarray results. Specific riboprobes labeled with [-35S] for MAP2 and APP mRNAs were hybridized with brain sections of individual FMR1 knockout mice and wild-type littermates. Labeled mRNA signals were revealed with autoradiography. The color spectrum on the right side of each panel represents the pixel value of gray levels. (See Color Insert.)

126

CAVALLARO

IV. Conclusion

The exact role and functional relationships of the genes and proteins implicated are presumably those we cannot yet recognize. Gene expression profiles described here unlock virtually unexplored frontiers and we will learn as we explore them. To facilitate this exploration, the data generated in diVerent behavioral and pathophysiological conditions are available online (www.ct.isn.cnr.it/genomiccenter/microarray-data). Systematic characterization of expression patterns associated with cognition and cognition-related disorders has just started (Blalock et al., 2003; Cotman and Berchtold, 2002; Cotman and Engesser-Cesar, 2002; Dubnau et al., 2003; Hata et al., 2001; Ho et al., 2001; Hsieh et al., 2003; Leil et al., 2003; Loring et al., 2001; Pasinetti, 2001; Pasinetti and Ho, 2001; Rampon et al., 2000; Tudor et al., 2002; Yao et al., 2003) and will provide a framework for interpreting the biological significance of the expression patterns observed in the long-term retention of learning and memory. In addition, the value of these experiments will progressively increase as more is learned about the function of each gene and microarray databases will be established to enable cross-platform comparisons. Although sure to be just the tip of the iceberg, the results described point toward genes or sets of genes that may play critical roles in learning and memory. This could point researchers toward therapies to improve learning and memory, under normal conditions and in disorders that aVect cognitive functioning, such as Alzheimer’s disease.

Acknowledgments

We gratefully acknowledge Alfia Corsino, Maria Patrizia D’Angelo, and Francesco Marino for their administrative and technical support.

References

Abel, T., and Kandel, E. (1998). Positive and negative regulatory mechanisms that mediate long-term memory storage. Brain Res. Rev. 26, 360–378. Alberini, C. M., Ghirardi, M., Metz, R., and Kandel, E. R. (1994). C/EBP is an immediate-early gene required for the consolidation of long-term facilitation in Aplysia. Cell 76, 1099–1114. Aleman, A., de Vries, W. R., de Haan, E. H., Verhaar, H. J., Samson, M. M., and Koppeschaar, H. P. (2000). Age-sensitive cognitive function, growth hormone and insulin-like growth factor 1 plasma levels in healthy older men. Neuropsychobiology 41, 73–78.


127

Alkon, D. L., Nelson, T. J., Zhao, W., and Cavallaro, S. (1998). Time domains of neuronal Ca2þ signaling and associative memory: steps through a calexcitin, ryanodine receptor, Kþ channel cascade. Trends Neurosci. 21, 529–537. Asensio, V. C., and Campbell, I. L. (1999). Chemokines in the CNS: Plurifunctional mediators in diverse states. Trends Neurosci. 22, 504–512. Ashley, C. T., Jr., Wilkinson, K. D., Reines, D., and Warren, S. T. (1993). FMR1 protein: Conserved RNP family domains and selective RNA binding. Science 262, 563–566. Barros, D. M., Mello e Souza, de Souza, M. M., Choi, H., DeDavid e Silva, Lenz, G., Medina, J. H., and Izquierdo, I. (2001). LY294002, an inhibitor of phosphoinositide 3-kinase given into rat hippocampus impairs acquisition, consolidation and retrieval of memory for one-trial step-down inhibitory avoidance. Behav. Pharmacol. 12, 629–634. Bartoli, M., Ternaux, J. P., Forni, C., Portalier, P., Salin, P., Amalric, M., and Monneron, A. (1999). Down-regulation of striatin, a neuronal calmodulin-binding protein, impairs rat locomotor activity. J. Neurobiol. 40, 234–243. Berke, J. D., Sgambato, V., Zhu, P. P., Lavoie, B., Vincent, M., Krause, M., and Hyman, S. E. (2001). Dopamine and glutamate induce distinct striatal splice forms of Ania-6, an RNA polymerase II–associated cyclin. Neuron 32, 277–287. Berman, D. E., Hazvi, S., Rosenblum, K., Seger, R., and Dudai, Y. (1998). Specific and diVerential activation of mitogen-activated protein kinase cascades by unfamiliar taste in the insular cortex of the behaving rat. J. Neurosci. 18, 10037–10044. Berthier, N. E., and Moore, J. W. (1986). Cerebellar Purkinje cell activity related to the classically conditioned nictitating membrane response. Exp. Brain Res. 63, 341–350. Blalock, E. M., Chen, K. C., Sharrow, K., Herman, J. P., Porter, N. M., Foster, T. C., and Landfield, P. W. (2003). Gene microarrays in hippocampal aging: Statistical profiling identifies novel processes correlated with cognitive impairment. J. Neurosci. 23, 3807–3819. Blaxton, T. A., Bookheimer, S. Y., ZeYro, T. A., Figlozzi, C. M., Gaillard, W. D., and Theodore, W. H. (1996). Functional mapping of human memory using PET: Comparisons of conceptual and perceptual tasks. Can. J. Exp. Psychol. 50, 42–56. Blum, S., Moore, A. N., Adams, F., and Dash, P. K. (1999). A mitogen-activated protein kinase cascade in the CA1/CA2 subfield of the dorsal hippocampus is essential for long-term spatial memory. J. Neurosci. 19, 3535–3544. Bormann, J. (2000). The ‘ABC’ of GABA receptors. Trends Pharmacol. Sci. 21, 16–19. Bouwknecht, J. A., Hijzen, T. H., Van der, G. J., Maes, R. A., Hen, R., and Olivier, B. (2001). Absence of 5-HT(1B) receptors is associated with impaired impulse control in male 5-HT(1B) knockout mice. Biol. Psychiatry 49, 557–568. Calo’, G., Guerrini, R., Rizzi, A., Salvadori, S., and Regoli, D. (2000). Pharmacology of nociceptin and its receptor: A novel therapeutic target. Br. J. Pharmacol. 129, 1261–1283. Castro-Alamancos, M. A., and Torres-Aleman, I. (1993). Long-term depression of glutamate-induced gamma-aminobutyric acid release in cerebellum by insulin-like growth factor I. Proc. Natl. Acad. Sci. USA 90, 7386–7390. Cavallaro, S., D’Agata, V., Manickam, P., Dufour, F., and Alkon, D. L. (2002). Memory specific temporal profiles of gene expression in the hippocampus. Proc. Natl. Acad. Sci. USA 99, 16279–16284. Cavallaro, S., Meiri, N., Yi, C. L., Musco, S., Ma, W., Goldberg, J., and Alkon, D. L. (1997). Late memory-related genes in the hippocampus revealed by RNA fingerprinting. Proc. Natl. Acad. Sci. USA 94, 9669–9673. Cavallaro, S., Schreurs, B. G., Zhao, W., D’Agata, V., and Alkon, D. L. (2001). Gene expression profiles during long-term memory consolidation. Eur. J. Neurosci. 13, 1809–1815. Chapman, P. F., Atkins, C. M., Allen, M. T., Haley, J. E., and Steinmetz, J. E. (1992). Inhibition of nitric oxide synthesis impairs two diVerent forms of learning. Neuroreport 3, 567–570.

128

CAVALLARO

Chattipakorn, S. C., and McMahon, L. L. (2002). Pharmacological characterization of glycine-gated chloride currents recorded in rat hippocampal slices. J. Neurophysiol. 87, 1515–1525. Choe, S. (2002). Potassium channel structures. Nat. Rev. Neurosci. 3, 115–121. Cotman, C. W., and Berchtold, N. C. (2002). Exercise: A behavioral intervention to enhance brain health and plasticity. Trends Neurosci. 25, 295–301. Cotman, C. W., and Engesser-Cesar, C. (2002). Exercise enhances and protects brain function. Exerc. Sport Sci. Rev. 30, 75–79. Coulter, D. A., Lo Turco, J. J., Kubota, M., Disterhoft, J. F., Moore, J. W., and Alkon, D. L. (1989). Classical conditioning reduces amplitude and duration of calcium-dependent after hyperpolarization in rabbit hippocampal pyramidal cells. J. Neurophysiol. 61, 971–981. D’Agata, V., Warren, S., Zhao, W., Torre, E., Alkon, D., and Cavallaro, S. (2002). Gene expression profiles in a transgenic animal model of fragile x syndrome. Neurobiol. Dis. 10, 211. Davis, H. P., and Squire, L. R. (1984). Protein synthesis and memory: A review. Psychol. Bull. 96, 518–559. Davis, R. J. (1993). The mitogen-activated protein kinase signal transduction pathway. J. Biol. Chem. 268, 14553–14556. de Mendonca, A., and Ribeiro, J. A. (1994). Endogenous adenosine modulates long-term potentiation in the hippocampus. Neuroscience 62, 385–390. Dubnau, J., Chiang, A. S., Grady, L., Barditch, J., Gossweiler, S., McNeil, J., Smith, P., Buldoc, F., Scott, R., Certa, U., Broger, C., and Tully, T. (2003). The staufen/pumilio pathway is involved in Drosophila long-term memory. Curr. Biol. 13, 286–296. Dutar, P., and Nicoll, R. A. (1988). A physiological role for GABAB receptors in the central nervous system. Nature 332, 156–158. Edelmann, W., Zervas, M., Costello, P., Roback, L., Fischer, I., Hammarback, J. A., Cowan, N., Davies, P., Wainer, B., and Kucherlapati, R. (1996). Neuronal abnormalities in microtubuleassociated protein 1B mutant mice. Proc. Natl. Acad. Sci. USA 93, 1270–1275. Estall, L. B., Grant, S. J., and Cicala, G. A. (1993). Inhibition of nitric oxide (NO) production selectively impairs learning and memory in the rat. Pharmacol. Biochem. Behav. 46, 959–962. Fisch, G. S., Carpenter, N., Holden, J. J., Howard-Peebles, P. N., Maddalena, A., Borghgraef, M., Steyaert, J., and Fryns, J. P. (1999). Longitudinal changes in cognitive and adaptive behavior in fragile X females: A prospective multicenter analysis. Am. J. Med. Genet. 83, 308–312. Fisch, G. S., Simensen, R., Tarleton, J., Chalifoux, M., Holden, J. J., Carpenter, N., Howard-Peebles, P. N., and Maddalena, A. (1996). Longitudinal study of cognitive abilities and adaptive behavior levels in fragile X males: A prospective multicenter analysis. Am. J. Med. Genet. 64, 356–361. Fitzpatrick, S. L., Sindoni, D. M., Shughrue, P. J., Lane, M. V., Merchenthaler, I. J., and Frail, D. E. (1998). Expression of growth diVerentiation factor-9 messenger ribonucleic acid in ovarian and nonovarian rodent and human tissues. Endocrinology 139, 2571–2578. Freund, L. S., and Reiss, A. L. (1991). Cognitive profiles associated with the fra(X) syndrome in males and females. Am. J. Med. Genet. 38, 542–547. Fu, Y. H., Kuhl, D. P., Pizzuti, A., Pieretti, M., SutcliVe, J. S., Richards, S., Verkerk, A. J., Holden, J. J., Fenwick, R. G., Jr., and Warren, S. T. (1991). Variation of the CGG repeat at the fragile X site results in genetic instability: Resolution of the Sherman paradox. Cell 67, 1047–1058. Fukunaga, K., Muller, D., and Miyamoto, E. (1996). CaM kinase II in long-term potentiation. Neurochem. Int. 28, 343–358. Gahtan, E., Auerbach, J. M., Groner, Y., and Segal, M. (1998). Reversible impairment of long-term potentiation in transgenic Cu/Zn-SOD mice. Eur. J. Neurosci. 10, 538–544. Gibertini, M., Newton, C., Friedman, H., and Klein, T. W. (1995). Spatial learning impairment in mice infected with Legionella pneumophila or administered exogenous interleukin-1-beta. Brain Behav. Immun. 9, 113–128.


129

Gomez, M., De Castro, E., Guarin, E., Sasakura, H., Kuhara, A., Mori, I., Bartfai, T., Bargmann, C. I., and Nef, P. (2001). Ca2þ signaling via the neuronal calcium sensor-1 regulates associative learning and memory in C. elegans. Neuron 30, 241–248. Gomi, H., Sun, W., Finch, C. E., Itohara, S., Yoshimi, K., and Thompson, R. F. (1999). Learning induces a CDC2-related protein kinase, KKIAMRE. J. Neurosci. 19, 9530–9537. Gormezano, I., Schneiderman, N., Deaux, E. G., and Fuentes, I. (1962). Nictitating membrane: Classical conditioning and extinction in the albino rabbit. Science 138, 33–34. Gould, T. J., and Steinmetz, J. E. (1996). Changes in rabbit cerebellar cortical and interpositus nucleus activity during acquisition, extinction, and backward classical eyelid conditioning. Neurobiol. Learn. Mem. 65, 17–34. Gruart, A., and Yeo, C. H. (1995). Cerebellar cortex and eyeblink conditioning: bilateral regulation of conditioned responses. Exp. Brain Res. 104, 431–448. Hata, R., Masumura, M., Akatsu, H., Li, F., Fujita, H., Nagai, Y., Yamamoto, T., Okada, H., Kosaka, K., Sakanaka, M., and Sawada, T. (2001). Up-regulation of calcineurin Abeta mRNA in the Alzheimer’s disease brain: Assessment by cDNA microarray. Biochem. Biophys. Res. Commun. 284, 310–316. Heller, M. J. (2002). DNA Microarray Technology: Devices, systems, and applications. Annu. Rev. Biomed. Eng. 4, 129–153. Hess, K. R., Zhang, W., Baggerly, K. A., Stivers, D. N., and Coombes, K. R. (2001). Microarrays: Handling the deluge of data and extracting reliable information. Trends Biotechnol. 19, 463–468. Hesselgesser, J., and Horuk, R. (1999). Chemokine and chemokine receptor expression in the central nervous system. J. Neurovirol. 5, 13–26. Ho, C., Zhou, J., Medina, M., Goto, T., Jacobson, M., Bhide, P. G., and Kosik, K. S. (2000). Deltacatenin is a nervous system–specific adherens junction protein which undergoes dynamic relocalization during development. J. Comp Neurol. 420, 261–276. Ho, L., Guo, Y., Spielman, L., Petrescu, O., Haroutunian, V., Purohit, D., Czernik, A., Yemul, S., Aisen, P. S., Mohs, R., and Pasinetti, G. M. (2001). Altered expression of a-type but not b-type synapsin isoform in the brain of patients at high risk for Alzheimer’s disease assessed by DNA microarray technique. Neurosci. Lett. 298, 191–194. Hsieh, M. T., Hsieh, C. L., Lin, L. W., Wu, C. R., and Huang, G. S. (2003). DiVerential gene expression of scopolamine-treated rat hippocampus-application of cDNA microarray technology. Life Sci. 73, 1007–1016. Impey, S., Obrietan, K., and Storm, D. R. (1999). Making new connections: Role of ERK/MAP kinase signaling in neuronal plasticity. Neuron 23, 11–14. Jahn, R., and Sudhof, T. C. (1999). Membrane fusion and exocytosis. Annu. Rev. Biochem. 68, 863–911. Janz, R., Sudhof, T. C., Hammer, R. E., Unni, V., Siegelbaum, S. A., and Bolshakov, V. Y. (1999). Essential roles in synaptic plasticity for synaptogyrin I and synaptophysin I. Neuron 24, 687–700. Jentsch, T. J., Stein, V., Weinreich, F., and Zdebik, A. A. (2002). Molecular structure and physiological function of chloride channels. Physiol. Rev. 82, 503–568. Jiang, Y., Lev-Lehman, E., Bressler, J., Tsai, T. F., and Beaudet, A. L. (1999). Genetics of Angelman syndrome. Am. J. Hum. Genet. 65, 1–6. Johnson, G. V., and Jope, R. S. (1992). The role of microtubule-associated protein 2 (MAP-2) in neuronal growth, plasticity, and degeneration. J. Neurosci. Res. 33, 505–512. Kataoka, M., Yoshiyama, K., Matsuura, K., Hijiya, N., Higuchi, Y., and Yamamoto, S. (1997). Structure of the murine CD156 gene, characterization of its promoter, and chromosomal location. J. Biol. Chem. 272, 18209–18215. Kaufmann, W. E., Naidu, S., and Budden, S. (1995). Abnormal expression of microtubule-associated protein 2 (MAP-2) in neocortex in Rett syndrome. Neuropediatrics 26, 109–113.

130

CAVALLARO

Kim, I., Kim, H. G., Kim, H., Kim, H. H., Park, S. K., Uhm, C. S., Lee, Z. H., and Koh, G. Y. (2000). Hepatic expression, synthesis and secretion of a novel fibrinogen/angiopoietin-related protein that prevents endothelial-cell apoptosis. Biochem. J. 346(pt. 3), 603–610. Kornhauser, J. M., and Greenberg, M. E. (1997). A kinase to remember: Dual roles for MAP kinase in long-term memory. Neuron 18, 839–842. Kosik, K. S., DuVy, L. K., Dowling, M. M., Abraham, C., McCluskey, A., and Selkoe, D. J. (1984). Microtubule-associated protein 2: Monoclonal antibodies demonstrate the selective incorporation of certain epitopes into Alzheimer neurofibrillary tangles. Proc. Natl. Acad. Sci. USA 81, 7941–7945. Krause, M. O. (1996). Chromatin structure and function: The heretical path to an RNA transcription factor. Biochem. Cell Biol. 74, 623–632. Lee, E. H., Huang, A. M., Tsuei, K. S., and Lee, W. Y. (1996). Enhanced hippocampal corticotropinreleasing factor gene expression associated with memory consolidation and memory storage in rats. Chin. J. Physiol. 39, 197–203. Leil, T. A., Ossadtchi, A., Nichols, T. E., Leahy, R. M., and Smith, D. J. (2003). Genes regulated by learning in the hippocampus. J. Neurosci. Res. 71, 763–768. Lennon, G., AuVray, C., Polymeropoulos, M., and Soares, M. B. (1996). The I.M.A.G.E. Consortium: An integrated molecular analysis of genomes and their expression. Genomics 33, 151–152. Leterrier, J. F., and Eyer, J. (1992). Age-dependent changes in the ultrastructure and in the molecular composition of rat brain microtubules. J. Neurochem. 59, 1126–1137. Leverenz, J. B., Yu, C. E., and Schellenberg, G. D. (1998). Aging-associated neuropathology in Werner syndrome. Acta Neuropathol. (Berl) 96, 421–424. Logan, C. G., and Grafton, S. T. (1995). Functional anatomy of human eyeblink conditioning determined with regional cerebral glucose metabolism and positron-emission tomography. Proc. Natl. Acad. Sci. USA 92, 7500–7504. Loring, J. F., Wen, X., Lee, J. M., Seilhamer, J., and Somogyi, R. (2001). A gene expression profile of Alzheimer’s disease. DNA Cell Biol. 20, 683–695. Lynch, M. A. (2002). Interleukin-1 beta exerts a myriad of eVects in the brain and in particular in the hippocampus: Analysis of some of these actions. Vitam. Horm. 64, 185–219. Ma, T. C., and Zhu, X. Z. (1997). Intrahippocampal infusion of interleukin-6 impairs avoidance learning in rats. Zhongguo Yao Li Xue. Bao. 18, 121–123. Madden, D. R. (2002). The structure and function of glutamate receptor ion channels. Nat. Rev. Neurosci. 3, 91–101. Malleret, G., Hen, R., Guillou, J. L., Segu, L., and Buhot, M. C. (1999). 5-HT1B receptor knock-out mice exhibit increased exploratory activity and enhanced spatial memory performance in the Morris water maze. J. Neurosci. 19, 6157–6168. Markowska, A. L., Mooney, M., and Sonntag, W. E. (1998). Insulin-like growth factor-1 ameliorates age-related behavioral deficits. Neuroscience 87, 559–569. Masson, J., Sagne, C., Hamon, M., and El Mestikawy, S. (1999). Neurotransmitter transporters in the central nervous system. Pharmacol. Rev. 51, 439–464. Mattson, M. P., and Duan, W. (1999). ‘‘Apoptotic’’ biochemical cascades in synaptic compartments: Roles in adaptive plasticity and neurodegenerative disorders J. Neurosci. Res. 58, 152–166. Mayford, M., Bach, M. E., Huang, Y. Y., Wang, L., Hawkins, R. D., and Kandel, E. R. (1996). Control of memory formation through regulated expression of a CaMKII transgene. Science 274, 1678–1683. Medina, M., Marinescu, R. C., Overhauser, J., and Kosik, K. S. (2000). Hemizygosity of deltacatenin (CTNND2) is associated with severe mental retardation in cri-du-chat syndrome. Genomics 63, 157–164. Meneses, A. (1999). 5-HT system and cognition. Neurosci. Biobehav. Rev. 23, 1111–1125.


131

Ming, G., Song, H., Berninger, B., Inagaki, N., Tessier-Lavigne, M., and Poo, M. (1999). Phospholipase C-gamma and phosphoinositide 3-kinase mediate cytoplasmic signaling in nerve growth cone guidance. Neuron 23, 139–148. Molchan, S. E., Sunderland, T., McIntosh, A. R., Herscovitch, P., and Schreurs, B. G. (1994). A functional anatomical study of associative learning in humans. Proc. Natl. Acad. Sci. USA 91, 8122–8126. Morgan, D., Holcomb, L., Saad, I., Gordon, M., and Mahin, M. (1998). Impaired spatial navigation learning in transgenic mice over-expressing heme oxygenase-1. Brain Res. 808, 110–112. Mott, D. D., and Lewis, D. V. (1994). The pharmacology and function of central GABAB receptors. Int. Rev. Neurobiol. 36, 97–223. Nakazato, F., Tada, T., Sekiguchi, Y., Murakami, K., Yanagisawa, S., Tanaka, Y., and Hongo, K. (2002). Disturbed spatial learning of rats after intraventricular administration of transforming growth factor-beta 1. Neurol. Med. Chir (Tokyo) 42, 151–156. Nieto-Bona, M. P., Garcia-Segura, L. M., and Torres-Aleman, I. (1997). Transynaptic modulation by insulin-like growth factor I of dendritic spines in Purkinje cells. Int. J. Dev. Neurosci. 15, 749–754. Noordewier, M. O., and Warren, P. V. (2001). Gene expression microarrays and the integration of biological knowledge. Trends Biotechnol. 19, 412–415. Ohbayashi, N., Hoshikawa, M., Kimura, S., Yamasaki, M., Fukui, S., and Itoh, N. (1998). Structure and expression of the mRNA encoding a novel fibroblast growth factor, FGF-18. J. Biol. Chem. 273, 18161–18164. Osborne, L. R., Campbell, T., Daradich, A., Scherer, S. W., and Tsui, L. C. (1999). Identification of a putative transcription factor gene (WBSCR11) that is commonly deleted in Williams-Beuren syndrome. Genomics 57, 279–284. Oyake, T., Itoh, K., Motohashi, H., Hayashi, N., Hoshino, H., Nishizawa, M., Yamamoto, M., and Igarashi, K. (1996). Bach proteins belong to a novel family of BTB-basic leucine zipper transcription factors that interact with MafK and regulate transcription through the NF-E2 site. Mol. Cell. Biol. 16, 6083–6095. Park, C. R. (2001). Cognitive eVects of insulin in the central nervous system. Neurosci. Biobehav. Rev. 25, 311–323. Pasinetti, G. M. (2001). Use of cDNA microarray in the search for molecular markers involved in the onset of Alzheimer’s disease dementia. J. Neurosci. Res. 65, 471–476. Pasinetti, G. M., and Ho, L. (2001). From cDNA microarrays to high-throughput proteomics. Implications in the search for preventive initiatives to slow the clinical progression of Alzheimer’s disease dementia. Restor. Neurol. Neurosci. 18, 137–142. Rachal, P. C., Fleshner, M., Watkins, L. R., Maier, S. F., and Rudy, J. W. (2001). The immune system and memory consolidation: A role for the cytokine IL-1beta. Neurosci. Biobehav. Rev. 25, 29–41. Rampon, C., Jiang, C. H., Dong, H., Tang, Y. P., Lockhart, D. J., Schultz, P. G., Tsien, J. Z., and Hu, Y. (2000). EVects of environmental enrichment on gene expression in the brain. Proc. Natl. Acad. Sci. USA 97, 12880–12884. Rissman, E. F., Heck, A. L., Leonard, J. E., Shupnik, M. A., and Gustafsson, J. A. (2002). Disruption of estrogen receptor beta gene impairs spatial learning in female mice. Proc. Natl. Acad. Sci. USA 99, 3996–4001. Saito, T., Greenwood, A., Sun, Q., and Anderson, D. J. (1995). Identification by diVerential RT-PCR of a novel paired homeodomain protein specifically expressed in sensory neurons and a subset of their CNS targets. Mol. Cell. Neurosci. 6, 280–292. Sanchez-Andres, J. V., and Alkon, D. L. (1991). Voltage-clamp analysis of the eVects of classical conditioning on the hippocampus. J. Neurophysiol. 65, 796–807.

132

CAVALLARO

Schreurs, B. G., McIntosh, A. R., Bahro, M., Herscovitch, P., Sunderland, T., and Molchan, S. E. (1997). Lateralization and behavioral correlation of changes in regional cerebral blood flow with classical conditioning of the human eyeblink response. J. Neurophysiol. 77, 2153–2163. Schreurs, B. G., Sanchez-Andres, J. V., and Alkon, D. L. (1991). Learning-specific diVerences in Purkinje-cell dendrites of lobule HVI (Lobulus simplex): Itracellular recording in a rabbit cerebellar slice. Brain Res. 548, 18–22. Seabra, M. C., Mules, E. H., and Hume, A. N. (2002). Rab GTPases, intracellular traYc and disease. Trends Mol. Med. 8, 23–30. Serra-Pages, C., Medley, Q. G., Tang, M., Hart, A., and Streuli, M. (1998). Liprins, a family of LAR transmembrane protein-tyrosine phosphatase-interacting proteins. J. Biol. Chem. 273, 15611–15620. Siomi, H., Siomi, M. C., Nussbaum, R. L., and Dreyfuss, G. (1993). The protein product of the fragile X gene, FMR1, has characteristics of an RNA-binding protein. Cell 74, 291–298. Smith, T. A. (2001). Type A gamma-aminobutyric acid (GABAA) receptor subunits and benzodiazepine binding: Significance to clinical syndromes and their treatment. Br. J. Biomed. Sci. 58, 111–121. Sterneck, E., Paylor, R., Jackson-Lewis, V., Libbey, M., Przedborski, S., Tessarollo, L., Crawley, J. N., and Johnson, P. F. (1998). Selectively enhanced contextual fear conditioning in mice lacking the transcriptional regulator CCAAT/enhancer binding protein delta. Proc. Natl. Acad. Sci. USA 95, 10908–10913. Stork, O., and Welzl, H. (1999). Memory formation and the regulation of gene expression. Cell Mol. Life Sci. 55, 575–592. Strandmann, E. P., Senkel, S., and RyVel, G. U. (1998). The bifunctional protein DCoH/PCD, a transcription factor with a cytoplasmic enzymatic activity, is a maternal factor in the rat egg and expressed tissue specifically during embryogenesis. Int. J. Dev. Biol. 42, 53–59. Stubley-Weatherly, L., Harding, J. W., and Wright, J. W. (1996). EVects of discrete kainic acidinduced hippocampal lesions on spatial and contextual learning and memory in rats. Brain Res. 716, 29–38. Sweatt, J. D. (2001). The neuronal MAP kinase cascade: A biochemical signal integration system subserving synaptic plasticity and memory. J. Neurochem. 76, 1–10. Taubenfeld, S. M., Wiig, K. A., Monti, B., Dolan, B., Pollonini, G., and Alberini, C. M. (2001). Fornix-dependent induction of hippocampal CCAAT enhancer-binding protein [beta] and [delta] Co-localizes with phosphorylated cAMP response element-binding protein and accompanies long-term memory consolidation. J. Neurosci. 21, 84–91. The Dutch-Belgian Fragile X Consortium (1994). Fmr1 knockout mice: A model to study fragile X mental retardation. Cell 78, 23–33. Tischmeyer, W., and Grimm, R. (1999). Activation of immediate early genes and memory formation. Cell Mol. Life Sci. 55, 564–574. Tudor, M., Akbarian, S., Chen, R. Z., and Jaenisch, R. (2002). Transcriptional profiling of a mouse model for Rett syndrome reveals subtle transcriptional changes in the brain. Proc. Natl. Acad. Sci. USA 99, 15536–15541. Verkerk, A. J., Pieretti, M., SutcliVe, J. S., Fu, Y. H., Kuhl, D. P., Pizzuti, A., Reiner, O., Richards, S., Victoria, M. F., and Zhang, F. P. (1991). Identification of a gene (FMR-1) containing a CGG repeat coincident with a breakpoint cluster region exhibiting length variation in fragile X syndrome. Cell 65, 905–914. Wendler, W. M., Kremmer, E., Forster, R., and Winnacker, E. L. (1997). Identification of pirin, a novel highly conserved nuclear protein. J. Biol. Chem. 272, 8482–8489. Winston, J. T., Koepp, D. M., Zhu, C., Elledge, S. J., and Harper, J. W. (1999). A family of mammalian F-box proteins. Curr. Biol. 9, 1180–1182.


133

Woolf, N. J., Zinnerman, M. D., and Johnson, G. V. (1999). Hippocampal microtubule-associated protein-2 alterations with contextual memory. Brain Res. 821, 241–249. Wright, J. W., Kramar, E. A., Meighan, S. E., and Harding, J. W. (2002). Extracellular matrix molecules, long-term potentiation, memory consolidation and the brain angiotensin system. Peptides 23, 221–246. Xiao, B., Tu, J. C., and Worley, P. F. (2000). Homer: A link between neural activity and glutamate receptor function. Curr. Opin. Neurobiol. 10, 370–374. Yamada, K., Noda, Y., Nakayama, S., Komori, Y., Sugihara, H., Hasegawa, T., and Nabeshima, T. (1995). Role of nitric oxide in learning and memory and in monoamine metabolism in the rat brain. Br. J. Pharmacol. 115, 852–858. Yamanouchi, H., Jay, V., Otsubo, H., Kaga, M., Becker, L. E., and Takashima, S. (1998). Early forms of microtubule-associated protein are strongly expressed in cortical dysplasia. Acta Neuropathol. (Berl) 95, 466–470. Yao, P. J., Zhu, M., Pyun, E. I., Brooks, A. I., Therianos, S., Meyers, V. E., and Coleman, P. D. (2003). Defects in expression of genes related to synaptic vesicle traYcking in frontal cortex of Alzheimer’s disease. Neurobiol. Dis. 12, 97–109. Yeo, C. H., Hardiman, M. J., and Glickstein, M. (1985). Classical conditioning of the nictitating membrane response of the rabbit. II. Lesions of the cerebellar cortex. Exp. Brain Res. 60, 99–113. Young, W. J., Lee, Y. F., Smith, S. M., and Chang, C. (1998). A bidirectional regulation between the TR2/TR4 orphan receptors (TR2/TR4) and the ciliary neurotrophic factor (CNTF) signaling pathway. J. Biol. Chem. 273, 20877–20885. Zhang, W., Vazquez, L., Apperson, M., and Kennedy, M. B. (1999). Citron binds to PSD-95 at glutamatergic synapses on inhibitory neurons in the hippocampus. J. Neurosci. 19, 96–108. Zhang, W. R., Hashimoto, N., Ahmad, F., Ding, W., and Goldstein, B. J. (1994). Molecular cloning and expression of a unique receptor-like protein-tyrosine-phosphatase in the leucocyte-commonantigen–related phosphate family. Biochem. J. 302(pt. 1), 39–47. Zhao, W., Meiri, N., Xu, H., Cavallaro, S., Quattrone, A., Zhang, L., and Alkon, D. L. (2000). Spatial learning induced changes in expression of the ryanodine type II receptor in the rat hippocampus. FASEB J. 14, 290–300. Zhen, X., Du, W., Romano, A. G., Friedman, E., and Harvey, J. A. (2001). The p38 mitogenactivated protein kinase is involved in associative learning in rabbits. J. Neurosci. 21, 5513–5519. Zhou, J., Liyanage, U., Medina, M., Ho, C., Simmons, A. D., Lovett, M., and Kosik, K. S. (1997). Presenilin 1 interaction in the brain with a novel member of the Armadillo family. Neuroreport 8, 2085–2090.


MICROARRAY ANALYSIS OF HUMAN NERVOUS SYSTEM GENE EXPRESSION IN NEUROLOGICAL DISEASE

Steven A. Greenberg Department of Neurology, Division of Neuromuscular Disease Brigham and Women’s Hospital Children’s Hospital Informatics Program Harvard Medical School Boston, Massachusetts 02115

I. II. III. IV.

Introduction Disease Pathophysiology Microarray-Based Disease Classification Human Neurological Diseases Studied by Microarray Technology A. Multiple Sclerosis B. Brain Tumors C. Neuromuscular Disorders D. Alzheimer’s Disease V. Conclusion References

I. Introduction

In this chapter, we focus on the use of microarrays applied to human tissues for the study of neurological disease. Several issues related to tissue processing should first be noted. Using human tissues for microarray studies results in significant limitations not present with the use of animal tissue. Most human brain tissue available for research comes from autopsy specimens; such tissue is usually obtained after significant premortem agonal and postmortem periods that likely result in significant changes to the transcriptome. The eVects of such events in human brain tissue on the transcriptome have not been systematically studied as it applies to large-scale analyses. Such studies have been performed in human bowel tissue and suggest that significant changes result from ischemia (Huang et al., 2001). Furthermore, brain tissue is typically fixed in formalin, which preserves histochemical properties but likely alters nucleic acids. Fixatives that preserve RNA are available but are not routinely used. Studies comparing complementary DNA (cDNA) experimental reproducibility suggest that flashfrozen tissue oVers high reproducibility, with formalin-fixed tissue considerably less reproducible (Specht et al., 2000). Methods for the extraction and analysis of INTERNATIONAL REVIEW OF NEUROBIOLOGY, VOL. 60

135


136

GREENBERG

RNA from fixed tissues have been discussed by Van Deerlin et al. (2002). In this chapter, we also discuss the unique challenges relating to isolation of RNA from individual cells by laser capture microdissection. The accessibility of human tissue is a major determinant of the applicability of microarray techniques in the study of human neurological disease. Diseases of the peripheral nervous system aVecting muscle and peripheral nerve appear more easily studied than diseases aVecting brain and spinal cord. Studies of peripheral blood mononuclear cells (PBMCs) in systemic non-blood–related diseases suggest that there is much to be learned from these cells in multiple sclerosis (MS) (Ramanathan et al., 2001; Sturzebecher et al., 2003), rheumatoid arthritis (RA) (Bennett et al., 2003), and systemic lupus erythematosus (SLE) (Baechler et al., 2003; Bennett et al., 2003; Han et al., 2003; Ye et al., 2003). We next focus on issues relating to experimental design and data analysis. Microarray gene expression technology has two broadly defined but distinct applications: the study of disease pathophysiology and the study of disease classification and prediction. The methods applied to the analysis of microarray data for each of these applications diVer. Microarray-based disease pathophysiology studies use fold ratios and statistical significance testing to determine the diVerential gene expression among tissue samples and use this information to understand the disease process. Microarray-based disease classification studies use class methods applied to expression profiles to compare diagnostic categories, discover disease subtypes, correlate phenotypical variables with gene expression patterns, and build models that predict clinically relevant variables, which might include diagnosis or prognosis. We briefly discuss these distinctions and then review the published literature of applications specifically to human nervous system expression.

II. Disease Pathophysiology

Scientists use most microarray studies to understand the mechanisms of disease. To understand disease pathophysiology, we must accurately identify genes that are diVerentially expressed across diVerent conditions. By studying the role of these genes, we can better understand the molecular mechanisms behind the disease. The most frequently used measure of diVerential gene expression is the fold ratio. Given the high level of noise in microarray data, fold ratios with noise model–based error bars may be more appropriate, though cumbersome. Calculation of fold ratios alone, however, does not provide a statistical measure of significance as familiar to medical researchers as a p value. A number of statistical approaches have been used to compute p values for microarray data; the most commonly used method is the two-sample t test, and

MICROARRAY ANALYSIS OF GENE EXPRESSION IN NEUROLOGICAL DISEASE

137

its many variations. The t statistic measures the likelihood that the mean values of two distributions are distinct. It is computed by scaling the diVerence between the means by the standard deviation, usually with the assumption that the variances are diVerent in the two groups. However, when the sample size is not large enough, we prefer to use a modified version of the t test in which a small factor is added to the computed standard deviation to guard against having a small standard deviation simply by chance. Such cases can occur when the estimated variance is much smaller than the true variance. A nonparametric approach, such as the Wilcoxon test, or a Bayesian approach, such as a Bayesian version of a regularized t test (Baldi and Long, 2001) or an Empirical Bayes (Lonnstedt and Speed, 2002) method can be used also. These methods can partially compensate for small sample sizes. The principal problem with the use of t tests and other statistical tests providing p values for microarray data pertains to multiple hypothesis testing. Microarray experiments test multiple—typically thousands—of independent hypotheses simultaneously. The established level of significance in medical research testing of a single hypothesis is p < .05. In a microarray experiment with 10,000 gene expression values measured, one expects by chance alone 10,000 0.05 ¼ 500 genes to have p values of less than 0.05. This cutoV of statistical significance is clearly inadequate and it is truly remarkable how many authors and reviewers of published microarray-based papers have neglected this basic statistical problem. Improved approaches to this problem have been studied. The Bonferroni method divides the nominal cutoV by the number of hypothesis tests; however, this method is much too conservative and does not appear to be the best way to view the underlying problem. Methods to deal with this include the Westfall–Young approach (Dudoit et al., 2002), a permutation-based approach (Tusher et al., 2001), and use of the false discovery rate (Reiner et al., 2003; Storey and Tibshirani, 2003). In any event, microarray-based studies of diVerential gene expression must always address this issue using at least some appropriate method.

III. Microarray-Based Disease Classification

The approach to the use of microarray data for diagnostic and prognostic classification can be understood through the concepts of the ‘‘expression profile’’ and class comparison, class prediction, and class discovery (Simon et al., 2003) (Table I). A tissue’s ‘‘expression profile’’ is an ordered list of numbers, otherwise known as a one-dimensional vector. Each number represents the level of expression of a given gene in the tissue. By mathematically manipulating and organizing

138

GREENBERG

TABLE I Classification Methods in the Application of Microarrays to Disease Diagnosis and Prognosis Class method

Goal

Computational tools

Comparison

Descriptive compare/ contrast

Prediction

Predict class for new sample

Cluster analysis, supervised learning, fold analysis Supervised learning

Discovery

Discover new disease subtypes

Cluster analysis

Example Distinct profiles of myopathy subtypes Predict diagnosis, prognosis in breast cancer patients Discovery of unrecognized lymphoma subtypes

expression profiles of multiple tissue samples, one can form ‘‘classes.’’ Three distinct class methods are used in microarray studies, discussed in the following paragraph. Under certain circumstances, the gene expression profile of a tissue can be referred to as a gene expression signature. Furthermore, like an ordinary handwritten signature, this gene expression signature must be reproducible and unique to a tissue under defined circumstances. For its practical use, some method of reliably ‘‘reading’’ this expression signature is also required. A study reporting ‘‘gene expression signatures’’ of a disease must establish all three features of reproducibility of the expression profiles, uniqueness of the profile to the disease, and a reliable method to ‘‘read’’ this profile to determine whether it is indeed diVerent than other expression profiles. By performing multiple microarray experiments on tissue samples from patients with diVerent diseases, one can combine data into data sets with each row representing a gene and each column a microarray experiment, with the number values for row i and column j being the expression level measured for gene i in tissue j. Microarray-based disease classification is essentially the study of such data sets and methods to organize the columns (individual tissue microarray experiments) into classes with meaningful clinical distinctions (Fig. 1). The three distinct classification methods—comparison, prediction, and discovery—diVer principally in their goals and are discussed next in turn. In class comparison analysis, the data set consists of ‘‘labeled’’ specimens, each of which has a predefined class assignment, and the goal is to understand whether diVerent classes have diVerent expression profiles and to compare and contrast the gene expression diVerences among the classes. An example is analysis of a gene expression data set consisting of individual gene expression profiles from each of a number of individuals with distinct myopathies (Greenberg et al., 2002). In class prediction analysis, classes are predefined, but the goal is to use the data to build a model to predict the correct class assignment of a new sample (Fig. 2). Such models


139

Fig. 1. Schematic approach to clustering of tissue samples based on similarity of gene expression patterns. Tissue specimens with similar expression patterns are clustered on the right. (See Color Insert.)

typically take the form of a multivariate function of any number of gene expression measurements but can be as simple as a single threshold gene expression value. In class discovery, the goal is to examine the gene expression correlates of heterogeneity within a single class (Fig. 3). This is fundamentally diVerent than class comparison and prediction approaches. Instead, subgroups of patterns of gene expression are sought, so samples within a subgroup are suYciently similar to each other and suYciently distinct from the rest of the group. Class discovery has as its goal the establishment of previously unrecognized subgroups of gene expression profiles that also have important and clinically relevant phenotypical accompaniments. Because disease heterogeneity is nearly universal and poses significant diYculties in the management of individual patients, class discovery has great potential for immediate contributions to the daily practice of medicine. Computational approaches to disease classification include cluster analysis and supervised learning techniques (Kohane et al., 2002; Quackenbush, 2001). Cluster analysis partitions expression data into groups using a measure of similarity and an organization structure that represents this similarity. Cluster analysis methods diVer in the similarity measure chosen (e.g., Euclidean distance, Pearson

140

GREENBERG

Fig. 2. Class prediction: building a predictor. (1) From data, choose a ‘‘gene set’’ that will discriminate among classes. (2) Choose a prediction function that when applied to a new expression profile produces a real number. (3) Choose a prediction rule that classifies a sample based on the output of the prediction function after application to it. (4) Validate the model.

Fig. 3. Class discovery.

correlation coeYcient, and mutual information) (Greenberg, 2001b), the organizational structure used to represent the partition [e.g., hierarchical clustering produces a dendrogram or treelike structure (Eisen et al., 1998), k-means produce groups in a multidimensional surface, relevance networks (Butte et al., 2000) produce graphs with connected subgraphs), and the algorithm used to achieve this partition (e.g., single linkage, average linkage]. Hierarchical cluster analysis is widely used in published microarray studies for class comparison and discovery (Alizadeh et al., 2000; Bittner et al., 2000; Greenberg et al., 2002; Hedenfalk et al., 2001; Nielsen et al., 2002; Perou et al., 2000; Sorlie et al., 2001; van’t Veer et al., 2002). The approach with class discovery is to perform hierarchical classification


141

on tissues from a single class and to examine the resulting tree structure for natural divisions that might reflect subclasses. One then needs to find a meaningful phenotypical diVerence that is statistically diVerent between the subclasses, such as survival times. Supervised learning techniques are used for class prediction (Golub et al., 1999; Radmacher et al., 2002; Simon et al., 2003). Data are used to construct a model (‘‘predictor’’) that predicts the proper class when a new sample is presented to it. Construction of such a predictor requires the choice of a subset of ‘‘informative’’ genes whose variability among classes is essential, a multivariate prediction function combining these measurements of informative genes, and a prediction rule stating which values of the prediction function define a sample into a particular class. Methods of choosing informative genes (e.g., classification trees, correlation coeYcients), creating predictor functions (e.g., linear weighting functions) (Golub et al., 1999), support vector machines (Brown et al., 2000; Ramaswamy et al., 2001), neural networks (Khan et al., 2001), and choices of prediction rules (e.g., threshold values for optimal sensitivity and specificity) account for significant variability with supervised methods. The principal challenge with supervised methods for class prediction is avoiding overfitting of the data, which typically results in good performance on the data set from which it was constructed but poor performance as a predictor for new data. Although class prediction through supervised learning has an enormous potential impact on clinical medicine, it is important to realize that important papers in this area often contain seriously flawed analyses that in our opinion are not yet appropriate for application to clinical medicine. The lack of technical expertise among investigators and journal reviewers remains a serious problem in this field (Simon et al., 2003). For example, the Netherlands Cancer Institute in Amsterdam is reportedly using expression levels of 70 genes from microarray-generated tumor profiles of patients with breast cancer together with a class prediction model (van de Vijver et al., 2002) to determine which women will receive adjuvant treatment after surgery (Schubert, 2003). Although these investigators’ model had a reported accuracy of 73% for prediction of breast cancer outcome based on gene expression profiles, the methodology was biased by incomplete performance of cross-validation techniques; the unbiased estimate of accuracy was in fact 59%, marginally better than a coin flip alone (Simon et al., 2003).

IV. Human Neurological Diseases Studied by Microarray Technology

Published microarray studies of human tissue and neurological disease exist in MS (Baranzini and Hauser, 2002; Dyment and Ebers, 2002; Lock et al., 2002; Mycko et al., 2003; Ramanathan et al., 2001; Steinman, 2001; Steinman and

142

GREENBERG

Zamvil, 2003; Sturzebecher et al., 2003; Tompkins and Miller, 2002; Whitney et al., 1999, 2001), central nervous system tumors (Chopra et al., 2003; Hernan et al., 2003; MacDonald et al., 2001; Mayanil et al., 2001; Nutt et al., 2003; Pomeroy et al., 2002; Yoon et al., 2002), AD (Colangelo et al., 2002; Ginsberg et al., 2000; Hata et al., 2001; Ho et al., 2001; Loring et al., 2001; Pasinetti, 2001; Walker et al., 2001), neuronal ceroid lipofuscinosis (Cooper, 2003), amyotrophic lateral sclerosis (Ishigaki et al., 2002; Malaspina et al., 2001), inflammatory myopathies (Greenberg, 2001a; Tezak et al., 2002), congenital myopathies (Sanoudou et al., 2003), and muscular dystrophies (Campanaro et al., 2002; Haslett et al., 2002; Noguchi et al., 2003; Tsukahara and Arahata, 2003). Several reviews pertinent to human neurological disease have been written as well (Greenberg, 2001a; Luo and Geschwind, 2001; Mirnics, 2001; Shoemaker and Linsley, 2002; Sturla et al., 2003). We review several of these areas next.

A. Multiple Sclerosis Gene expression changes in brain lesions from patients with MS have been studied in small numbers of patients. A study by Lock et al. (2002) identified relevant genes in postmortem tissue from four patients with MS compared to two without. Although acute and ‘‘silent’’ lesions were compared and said to show upregulation of diVerent genes, the number of specimens is so small that whether these diVerences are anything other than experimental noise is unclear. Of particular note, though, these investigators found upregulation of granulocyte colony-stimulating factor (G-CSF) in acute lesions and demonstrated that treatment of mice with G-CSF before onset of experimental autoimmune encephalitis (EAE) decreased severity of the disease. Other even smaller studies have also looked at gene expression in MS brain tissue (Mycko et al., 2003; Whitney et al., 1999). In addition, several investigators have turned to studying PBMCs in patients with MS, which are far more easily accessible than brain tissue. Expression studies of PBMCs have been revealing in other autoimmune disorders including lupus erythematosus and RA, as noted earlier. Sturzebecher et al. (2003) studied expression profiles of PBMCs in patients treated with interferon- (IFN-). This study included a design in which such PBMCs were exposed to IFN- in vitro (called ex vivo experiments by the authors). Ten female patients in three classes were studied: six responders to IFN-, two initial responders who developed neutralizing antibodies (Nab) and lost their responsiveness, and two who initially failed to respond (INR). The definition of significant diVerential expression was nonstandard: either twofold in at least three of the six responders and not twofold decreased in four nonresponders (in so-called ex vivo experiments), or twofold in at


143

least three of the eight patients in the combined responders þ Nab group and not twofold decreased in two patients who INR (in so-called in vitro experiments). Using these criteria, the authors found 112 genes significant: 25 ex vivo and 107 in vitro (20 in common). The complexity and ad hoc methods of the design and analysis in this study results in uncertainties about interpreting the conclusions. Rather than use an average-fold ratio across a class, the authors use individual experiment’s fold ratios subject to additional constraints. There is no statement about reproducibility studies performed by the authors and the very small numbers and particular definition of significance make it diYcult to accept the results confidently. In addition, the title of this report is highly inappropriate (‘‘Expression Profiling Identifies Responder and Non-responder Phenotypes to Interferon- in Multiple Sclerosis’’). The claim made is that of class prediction—that measuring an expression profile allows one to identify whether a patient is a responder or not. Such a claim is diYcult to understand given that no class prediction methods (Table I) were used in this study. It is easy to make thousands of measurements and find one or a few that distinguish two classes by chance; it is quite another matter to demonstrate that one of these measurements allows correct class prediction (identification) of new specimens. As an analogy, what if one looked at the serum calcium of 10 patients with MS, 6 responders and 4 nonresponders, and found a cutoV, say, of 10.2, so three responders all had levels more than 10.2 and the nonresponders less than 10.2. Would it be acceptable to then claim that a serum calcium level of more than or less than 10.2 identifies responders from nonresponders in a study with only 10 patients? Now consider this problem even further confounded by an initial search of more than at least 6000 such clinical measurements to find one such variable that distinguishes six responders from four nonresponders and the claim that such a measure identifies the correct phenotype is not believable. A similar study of in vitro IFN exposure of PBMCs in one patient with MS and two normal controls found approximately 500 genes of 6432 diVerentially regulated (Wandinger et al., 2001). This study is so small and used a fold ratio of 2.0 as the determination of significance that it is diYcult to be confident of these results. In another study of PBMCs in MS (Ramanathan et al., 2001), misreported as a study of brain tissue in a subsequent review (Baranzini and Hauser, 2002), samples from 15 patients with MS and 15 normal controls were hybridized to cDNA arrays with 5184 probes representing 4000 genes. This study was well designed with two simple classes and a t test used to establish significance. No microarray reproducibility data were provided. However, the level of significance was set at p < .05. This ignores the multiple-hypothesis problem—the testing of at least 4000 hypotheses simultaneously. By definition, the number of genes expected with a p value less than .05 on a t test applied to 4000 genes is

144

GREENBERG

4000 0.05 ¼ 200. This study found only 34 genes to be significant, with the lowest p value of .01. This number is so much less than what would result by chance alone that it seems likely that random and systematic noise is highly represented in this data set. As noted previously, a p value cutoV of .05 is highly inappropriate. Methods to compute appropriate p values for microarray data do exist (Brody et al., 2002; Dudoit et al., 2002; Reiner et al., 2003; Storey and Tibshirani, 2003; Tusher et al., 2001). Another study of PBMCs in patients with MS looked at the gene expression changes over time in 13 patients treated with IFN-1b. This impressive study appears to have generated high-quality data, as 14 of the 21 genes identified as significantly (by a Bayesian modified t test) diVerentially expressed were previously known to be IFN regulated. In the area of disease classification and prediction, this author is unaware of any studies in MS, the study by Sturzebecher et al. (2003) not withstanding, as discussed earlier.

B. Brain Tumors The use of microarray data for cancer classification, diagnosis, and prediction is a rapidly growing area that has been referred to as predictive molecular pathology (Sauter and Simon, 2002). This approach has been particularly remarkable in its contribution to breast cancer (van de Vijver et al., 2002; van’t Veer et al., 2002) and lymphoma (Alizadeh et al., 2000; Copur et al., 2002; Magrath, 2002; Rosenwald et al., 2002). Glioma (Caskey et al., 2000; Kim et al., 2002; Mukasa et al., 2002; Nutt et al., 2003; Rickman et al., 2001; Watson et al., 2001), medulloblastoma (MacDonald et al., 2001), and other embryonal tumors of the central nervous system (Pomeroy et al., 2002) have been studied with this approach. Using class prediction methods, investigators have been able to construct highly accurate predictive models on the small data sets on which they have been tested. Microarray studies of glioma appear to oVer significant predictive power beyond current clinical variables (Nutt et al., 2003). Methodological problems remain an important limitation in some of these studies, however. For example, MacDonald et al. (2001) claimed a 72% ‘‘accuracy’’ of their predictor of metastatic versus nonmetastatic medulloblastoma; however, this predictor was an ambiguous one and did not make a definite choice for each specimen, instead leaving several unclassified (the model ‘‘chose’’ not to make a prediction). The data actually demonstrate that their model was correct for only 57% of predictions and incorrect for 22% (the claim of 72% ignored the samples not classified). Furthermore, the validation set consisted of only nonmetastatic samples; the performance of the predictor for metastatic samples was not tested and could be very diVerent.


145

C. Neuromuscular Disorders The relative ease of obtaining fresh frozen muscle tissue and to a lesser extent peripheral nerve tissue from patients in comparison to brain tissue, provides significant power to the value of microarray studies in human neuromuscular disorders. In one of the very few microarray-based disease classification studies of noncancerous diseases, Greenberg et al. (2002) have shown the ability of expression profiles to distinguish among the various subtypes of myopathy and in particular subtypes of inflammatory myopathy. A previously unrecognized role for type I IFN-stimulated genes has also been identified in the pathogenesis of dermatomyositis (Greenberg et al., 2002; Tezak et al., 2002). Models of pathogenesis for Duchenne’s muscular dystrophy (Haslett et al., 2002; Noguchi et al., 2003), dysferlinopathies (Campanaro et al., 2002), X-linked Emery–Dreifuss muscular dystrophy (Tsukahara and Arahata, 2003), and nemaline myopathies (Sanoudou et al., 2003) have also been constructed principally from microarray data. In contrast to muscle disease, no substantial work on large-scale gene expression in patients with human peripheral nerve disease has been performed. The limited accessibility of peripheral nerve cell bodies containing RNA may be a factor in this regard, as clinical performed nerve biopsies sample neuronal axons, though the transcriptome of Schwann’s cells is reflected in such biopsies. Lastly, two studies have addressed diVerential gene expression changes in the spinal cord of patients with amyotrophic lateral sclerosis (Ishigaki et al., 2002; Malaspina et al., 2001); it is highly likely that ischemia was a significant factor aVecting transcriptional changes in these studies.

D. Alzheimer’s Disease Gene expression studies in AD have focused on discrete brain regions, particularly the hippocampus (Colangelo et al., 2002; Ginsberg et al., 2000) and cholinergic nuclei (Mufson et al., 2002), but also more generally in the brain (Hata et al., 2001; Ho et al., 2001; Loring et al., 2001; Pasinetti, 2001; Walker et al., 2001). These studies have focused on disease pathophysiology and have yielded many results that will require further biological validation.

V. Conclusion

Microarrays can be used to study disease pathophysiology and disease classification. These are distinct applications and one should be mindful of which category a given application is focused on. Disease pathophysiology uses fold

146

GREENBERG

ratios and statistical significance testing to determine the diVerential gene expression among tissue samples and use this information to understand the disease process. Disease classification uses class methods applied to expression profiles to compare diagnostic categories, discover disease subtypes, correlate phenotypical variables with gene expression patterns, and build models that predict clinical relevant variables, which might include diagnosis or prognosis.

References

Alizadeh, A. A., Eisen, M. B., Davis, R. E., Ma, C., Lossos, I. S., Rosenwald, A., Boldrick, J. C., Sabet, H., Tran, T., Yu, X., Powell, J. I., Yang, L., Marti, G. E., Moore, T., Hudson, J., Jr., Lu, L., Lewis, D. B., Tibshirani, R., Sherlock, G., Chan, W. C., Greiner, T. C., Weisenburger, D. D., Armitage, J. O., Warnke, R., Staudt, L. M., et al. (2000). Distinct types of diVuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503–511. Baechler, E. C., Batliwalla, F. M., Karypis, G., GaVney, P. M., Ortmann, W. A., Espe, K. J., Shark, K. B., Grande, W. J., Hughes, K. M., Kapur, V., Gregersen, P. K., and Behrens, T. W. (2003). Interferon-inducible gene expression signature in peripheral blood cells of patients with severe lupus. Proc. Natl. Acad. Sci. USA 100, 2610–2615. Baldi, P., and Long, A. D. (2001). A Bayesian framework for the analysis of microarray expression data: Regularized t-test and statistical inferences of gene changes. Bioinformatics 17, 509–519. Baranzini, S. E., and Hauser, S. L. (2002). Large-scale gene-expression studies and the challenge of multiple sclerosis. Genome. Biol. 3, 1027. Bennett, L., Palucka, A. K., Arce, E., Cantrell, V., Borvak, J., Banchereau, J., and Pascual, V. (2003). Interferon and granulopoiesis signatures in systemic lupus erythematosus blood. J. Exp. Med. 197, 711–723. Bittner, M., Meltzer, P., Chen, Y., Jiang, Y., Seftor, E., Hendrix, M., Radmacher, M., Simon, R., Yakhini, Z., Ben-Dor, A., Sampas, N., Dougherty, E., Wang, E., Marincola, F., Gooden, C., Lueders, J., Glatfelter, A., Pollock, P., Carpten, J., Gillanders, E., Leja, D., Dietrich, K., Beaudry, C., Berens, M., Alberts, D., and Sondak, V. (2000). Molecular classification of cutaneous malignant melanoma by gene expression profiling. Nature 406, 536–540. Brody, J. P., Williams, B. A., Wold, B. J., and Quake, S. R. (2002). Significance and statistical errors in the analysis of DNA microarray data. Proc. Natl. Acad. Sci. USA 99, 12975–12978. Brown, M. P., Grundy, W. N., Lin, D., Cristianini, N., Sugnet, C. W., Furey, T. S., Ares, M., Jr., and Haussler, D. (2000). Knowledge-based analysis of microarray gene expression data by using support vector machines. Proc. Natl. Acad. Sci. USA 97, 262–267. Butte, A. J., Tamayo, P., Slonim, D., Golub, T. R., and Kohane, I. S. (2000). Discovering functional relationships between RNA expression and chemotherapeutic susceptibility using relevance networks. Proc. Natl. Acad. Sci. USA 97, 12182–12186. Campanaro, S., Romualdi, C., Fanin, M., Celegato, B., Pacchioni, B., Trevisan, S., Laveder, P., De Pitta, C., Pegoraro, E., Hayashi, Y. K., Valle, G., Angelini, C., and Lanfranchi, G. (2002). Gene expression profiling in dysferlinopathies using a dedicated muscle microarray. Hum. Mol. Genet. 11, 3283–3298. Caskey, L. S., Fuller, G. N., Bruner, J. M., Yung, W. K., Sawaya, R. E., Holland, E. C., and Zhang, W. (2000). Toward a molecular classification of the gliomas: Histopathology, molecular genetics, and gene expression profiling. Histol. Histopathol. 15, 971–981.


147

Chopra, A., Brown, K. M., Rood, B. R., Packer, R. J., and MacDonald, T. J. (2003). The use of gene expression analysis to gain insights into signaling mechanisms of metastatic medulloblastoma. Pediatr. Neurosurg. 39, 68–74. Colangelo, V., Schurr, J., Ball, M. J., Pelaez, R. P., Bazan, N. G., and Lukiw, W. J. (2002). Gene expression profiling of 12633 genes in Alzheimer hippocampal CA1: Transcription and neurotrophic factor down-regulation and up-regulation of apoptotic and pro-inflammatory signaling. J. Neurosci. Res. 70, 462–473. Cooper, J. D. (2003). Progress towards understanding the neurobiology of Batten disease or neuronal ceroid lipofuscinosis. Curr. Opin. Neurol. 16, 121–128. Copur, M. S., Ledakis, P., and Bolton, M. (2002). Molecular profiling of lymphoma. N. Engl. J. Med. 347, 1376–1377. Dudoit, S., Yang, Y. H., Speed, T. P., and Callow, M. J. (2002). Statistical methods for identifying diVerentially expressed genes in replicated cDNA microarray experiments. Statistica Sinica 12, 111–139. Dyment, D. A., and Ebers, G. C. (2002). An array of sunshine in multiple sclerosis. N. Engl. J. Med. 347, 1445–1447. Eisen, M. B., Spellman, P. T., Brown, P. O., and Botstein, D. (1998). Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. USA 95, 14863–14868. Ginsberg, S. D., Hemby, S. E., Lee, V. M., Eberwine, J. H., and Trojanowski, J. Q. (2000). Expression profile of transcripts in Alzheimer’s disease tangle-bearing CA1 neurons. Ann. Neurol. 48, 77–87. Golub, T. R., Slonim, D. K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J. P., Coller, H., Loh, M. L., Downing, J. R., Caligiuri, M. A., Bloomfield, C. D., and Lander, E. S. (1999). Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring. Science 286, 531–537. Greenberg, S. A. (2001a). DNA microarray gene expression analysis technology and its application to neurological disorders. Neurology 57, 755–761. Greenberg, S. A. (2001b). Theory and application of metric choices for DNA microarray expression analysis technologies. Masters Thesis, Massachusetts Institute of Technology. Greenberg, S. A., Sanoudou, D., Haslett, J. N., Kohane, I. S., Kunkel, L. M., Beggs, A. H., and Amato, A. A. (2002). Molecular profiles of inflammatory myopathies. Neurology 59, 1170–1182. Han, G. M., Chen, S. L., Shen, N., Ye, S., Bao, C. D., and Gu, Y. Y. (2003). Analysis of gene expression profiles in human systemic lupus erythematosus using oligonucleotide microarray. Genes. Immun. 4, 177–186. Haslett, J. N., Sanoudou, D., Kho, A. T., Bennett, R. R., Greenberg, S. A., Kohane, I. S., Beggs, A. H., and Kunkel, L. M. (2002). Gene expression comparison of biopsies from Duchenne muscular dystrophy (DMD) and normal skeletal muscle. Proc. Natl. Acad. Sci. USA 99, 15000–15005. Hata, R., Masumura, M., Akatsu, H., Li, F., Fujita, H., Nagai, Y., Yamamoto, T., Okada, H., Kosaka, K., Sakanaka, M., and Sawada, T. (2001). Up-regulation of calcineurin Abeta mRNA in the Alzheimer’s disease brain: Assessment by cDNA microarray. Biochem. Biophys. Res. Commun. 284, 310–316. Hedenfalk, I., Duggan, D., Chen, Y., Radmacher, M., Bittner, M., Simon, R., Meltzer, P., Gusterson, B., Esteller, M., Kallioniemi, O. P., Wilfond, B., Borg, A., and Trent, J. (2001). Gene-expression profiles in hereditary breast cancer. N. Engl. J. Med. 344, 539–548. Hernan, R., Fasheh, R., Calabrese, C., Frank, A. J., Maclean, K. H., Allard, D., Barraclough, R., and Gilbertson, R. J. (2003). ERBB2 up-regulates S100A4 and several other prometastatic genes in medulloblastoma. Cancer. Res. 63, 140–148. Ho, L., Guo, Y., Spielman, L., Petrescu, O., Haroutunian, V., Purohit, D., Czernik, A., Yemul, S., Aisen, P. S., Mohs, R., and Pasinetti, G. M. (2001). Altered expression of a-type but not b-type

148

GREENBERG

synapsin isoform in the brain of patients at high risk for Alzheimer’s disease assessed by DNA microarray technique. Neurosci. Lett. 298, 191–194. Huang, J., Qi, R., Quackenbush, J., Dauway, E., Lazaridis, E., and Yeatman, T. (2001). EVects of ischemia on gene expression. J. Surg. Res. 99, 222–227. Ishigaki, S., Niwa, J., Ando, Y., Yoshihara, T., Sawada, K., Doyu, M., Yamamoto, M., Kato, K., Yotsumoto, Y., and Sobue, G. (2002). DiVerentially expressed genes in sporadic amyotrophic lateral sclerosis spinal cords–screening by molecular indexing and subsequent cDNA microarray analysis. FEBS Lett. 531, 354–358. Khan, J., Wei, J. S., Ringner, M., Saal, L. H., Ladanyi, M., Westermann, F., Berthold, F., Schwab, M., Antonescu, C. R., Peterson, C., and Meltzer, P. S. (2001). Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks. Nat. Med. 7, 673–679. Kim, S., Dougherty, E. R., Shmulevich, L., Hess, K. R., Hamilton, S. R., Trent, J. M., Fuller, G. N., and Zhang, W. (2002). Identification of combination gene sets for glioma classification. Mol. Cancer. Ther. 1, 1229–1236. Kohane, I. S., Kho, A., and Butte, A. J. (2002). Microarrays for an Integrative Genomics. MIT Press, Cambridge. Lock, C., Hermans, G., Pedotti, R., Brendolan, A., Schadt, E., Garren, H., Langer-Gould, A., Strober, S., Cannella, B., Allard, J., Klonowski, P., Austin, A., Lad, N., Kaminski, N., Galli, S. J., Oksenberg, J. R., Raine, C. S., Heller, R., and Steinman, L. (2002). Gene-microarray analysis of multiple sclerosis lesions yields new targets validated in autoimmune encephalomyelitis. Nat. Med. 8, 500–508. Lonnstedt, I., and Speed, T. (2002). Replicated microarray data. Statistica Sinica 12, 31–46. Loring, J. F., Wen, X., Lee, J. M., Seilhamer, J., and Somogyi, R. (2001). A gene expression profile of Alzheimer’s disease. DNA Cell Biol. 20, 683–695. Luo, Z., and Geschwind, D. H. (2001). Microarray applications in neuroscience. Neurobiol. Dis. 8, 183–193. MacDonald, T. J., Brown, K. M., LaFleur, B., Peterson, K., Lawlor, C., Chen, Y., Packer, R. J., Cogen, P., and Stephan, D. A. (2001). Expression profiling of medulloblastoma: PDGFRA and the RAS/MAPK pathway as therapeutic targets for metastatic disease. Nat. Genet. 29, 143–152. Magrath, I. (2002). Molecular characteristics of diVuse large-B-cell lymphoma. N. Engl. J. Med. 346, 1998–1999. Malaspina, A., Kaushik, N., and de Belleroche, J. (2001). DiVerential expression of 14 genes in amyotrophic lateral sclerosis spinal cord detected using gridded cDNA arrays. J. Neurochem. 77, 132–145. Mayanil, C. S., George, D., Freilich, L., Miljan, E. J., Mania-Farnell, B., McLone, D. G., and Bremer, E. G. (2001). Microarray analysis detects novel Pax3 downstream target genes. J. Biol. Chem. 276, 49299–49309. Mirnics, K. (2001). Microarrays in brain research: The good, the bad and the ugly. Nat. Rev. Neurosci. 2, 444–447. Mufson, E. J., Counts, S. E., and Ginsberg, S. D. (2002). Gene expression profiles of cholinergic nucleus basalis neurons in Alzheimer’s disease. Neurochem. Res. 27, 1035–1048. Mukasa, A., Ueki, K., Matsumoto, S., Tsutsumi, S., Nishikawa, R., Fujimaki, T., Asai, A., Kirino, T., and Aburatani, H. (2002). Distinction in gene expression profiles of oligodendrogliomas with and without allelic loss of 1p. Oncogene 21, 3961–3968. Mycko, M. P., Papoian, R., Boschert, U., Raine, C. S., and Selmaj, K. W. (2003). cDNA microarray analysis in multiple sclerosis lesions: Detection of genes associated with disease activity. Brain 126, 1048–1057. Nielsen, T. O., West, R. B., Linn, S. C., Alter, O., Knowling, M. A., O’Connell, J. X., Zhu, S., Fero, M., Sherlock, G., Pollack, J. R., Brown, P. O., Botstein, D., and van de Rijn, M. (2002). Molecular characterisation of soft tissue tumours: A gene expression study. Lancet 359, 1301–1307.


149

Noguchi, S., Tsukahara, T., Fujita, M., Kurokawa, R., Tachikawa, M., Toda, T., Tsujimoto, A., Arahata, K., and Nishino, I. (2003). cDNA microarray analysis of individual Duchenne muscular dystrophy patients. Hum. Mol. Genet. 12, 595–600. Nutt, C. L., Mani, D. R., Betensky, R. A., Tamayo, P., Cairncross, J. G., Ladd, C., Pohl, U., Hartmann, C., McLaughlin, M. E., Batchelor, T. T., Black, P. M., von Deimling, A., Pomeroy, S. L., Golub, T. R., and Louis, D. N. (2003). Gene expression-based classification of malignant gliomas correlates better with survival than histological classification. Cancer Res. 63, 1602–1607. Pasinetti, G. M. (2001). Use of cDNA microarray in the search for molecular markers involved in the onset of Alzheimer’s disease dementia. J. Neurosci. Res. 65, 471–476. Perou, C. M., Sorlie, T., Eisen, M. B., van de Rijn, M., JeVrey, S. S., Rees, C. A., Pollack, J. R., Ross, D. T., Johnsen, H., Akslen, L. A., Fluge, O., Pergamenschikov, A., Williams, C., Zhu, S. X., Lonning, P. E., Borresen-Dale, A. L., Brown, P. O., and Botstein, D. (2000). Molecular portraits of human breast tumours. Nature 406, 747–752. Pomeroy, S. L., Tamayo, P., Gaasenbeek, M., Sturla, L. M., Angelo, M., McLaughlin, M. E., Kim, J. Y., Goumnerova, L. C., Black, P. M., Lau, C., Allen, J. C., Zagzag, D., Olson, J. M., Curran, T., Wetmore, C., Biegel, J. A., Poggio, T., Mukherjee, S., Rifkin, R., Califano, A., Stolovitzky, G., Louis, D. N., Mesirov, J. P., Lander, E. S., and Golub, T. R. (2002). Prediction of central nervous system embryonal tumour outcome based on gene expression. Nature 415, 436–442. Quackenbush, J. (2001). Computational analysis of microarray data. Nat. Rev. Genet. 2, 418–427. Radmacher, M. D., McShane, L. M., and Simon, R. (2002). A paradigm for class prediction using gene expression profiles. J. Comput. Biol. 9, 505–511. Ramanathan, M., Weinstock-Guttman, B., Nguyen, L. T., Badgett, D., Miller, C., Patrick, K., Brownscheidle, C., and Jacobs, L. (2002). In vivo gene expression revealed by cDNA arrays: The pattern in relapsing-remitting multiple sclerosis patients compared with normal subjects. J. Neuroimmunol. 116, 213–219. Ramaswamy, S., Tamayo, P., Rifkin, R., Mukherjee, S., Yeang, C. H., Angelo, M., Ladd, C., Reich, M., Latulippe, E., Mesirov, J. P., Poggio, T., Gerald, W., Loda, M., Lander, E. S., and Golub, T. R. (2001). Multiclass cancer diagnosis using tumor gene expression signatures. Proc. Natl. Acad. Sci. USA 98, 15149–15154. Reiner, A., Yekutieli, D., and Benjamini, Y. (2003). Identifying diVerentially expressed genes using false discovery rate controlling procedures. Bioinformatics 19, 368–375. Rickman, D. S., Bobek, M. P., Misek, D. E., Kuick, R., Blaivas, M., Kurnit, D. M., Taylor, J., and Hanash, S. M. (2001). Distinctive molecular profiles of high-grade and low-grade gliomas based on oligonucleotide microarray analysis. Cancer Res. 61, 6885–6891. Rosenwald, A., Wright, G., Chan, W. C., Connors, J. M., Campo, E., Fisher, R. I., Gascoyne, R. D., Muller-Hermelink, H. K., Smeland, E. B., Giltnane, J. M., Hurt, E. M., Zhao, H., Averett, L., Yang, L., Wilson, W. H., JaVe, E. S., Simon, R., Klausner, R. D., Powell, J., DuVey, P. L., Longo, D. L., Greiner, T. C., Weisenburger, D. D., Sanger, W. G., Dave, B. J., Lynch, J. C., Vose, J., Armitage, J. O., Montserrat, E., Lopez-Guillermo, A., Grogan, T. M., Miller, T. P., LeBlanc, M., Ott, G., Kvaloy, S., Delabie, J., Holte, H., Krajci, P., Stokke, T., and Staudt, L. M. (2002). The use of molecular profiling to predict survival after chemotherapy for diVuse large-B-cell lymphoma. N. Engl. J. Med. 346, 1937–1947. Sanoudou, D., Haslett, J. N., Kho, A. T., Guo, S., Gazda, H. T., Greenberg, S. A., Lidov, H. G., Kohane, I. S., Kunkel, L. M., and Beggs, A. H. (2003). Expression profiling reveals altered satellite cell numbers and glycolytic enzyme transcription in nemaline myopathy muscle. Proc. Natl. Acad. Sci. USA 100, 4666–4671. Sauter, G., and Simon, R. (2002). Predictive molecular pathology. N. Engl. J. Med. 347, 1995–1996. Schubert, C. M. (2003). Microarray to be used as routine clinical screen. Nat. Med. 9, 9. Shoemaker, D. D., and Linsley, P. S. (2002). Recent developments in DNA microarrays. Curr. Opin. Microbiol. 5, 334–337.

150

GREENBERG

Simon, R., Radmacher, M. D., Dobbin, K., and McShane, L. M. (2003). Pitfalls in the use of DNA microarray data for diagnostic and prognostic classification. J. Natl. Cancer. Inst. 95, 14–18. Sorlie, T., Perou, C. M., Tibshirani, R., Aas, T., Geisler, S., Johnsen, H., Hastie, T., Eisen, M. B., van de Rijn, M., JeVrey, S. S., Thorsen, T., Quist, H., Matese, J. C., Brown, P. O., Botstein, D., Eystein Lonning, P., and Borresen-Dale, A. L. (2001). Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc. Natl. Acad. Sci. USA 98, 10869–10874. Specht, K., Richter, T., Muller, U., and Walch, A. (2000). Quantitative gene expression analysis in microdissected archival tissue by real-time RT-PCR. J. Mol. Med. 78, B27. Steinman, L. (2001). Multiple sclerosis and gene expression profiling. Adv. Exp. Med. Biol. 490, 109–112. Steinman, L., and Zamvil, S. (2003). Transcriptional analysis of targets in multiple sclerosis. Nat. Rev. Immunol. 3, 483–492. Storey, J. D., and Tibshirani, R. (2003). Statistical significance for genomewide studies. Proc. Natl. Acad. Sci. USA 100, 9440–9445. Sturla, L. M., Fernandez-Teijeiro, A., and Pomeroy, S. L. (2003). Application of microarrays to neurological disease. Arch. Neurol. 60, 676–682. Sturzebecher, S., Wandinger, K. P., Rosenwald, A., Sathyamoorthy, M., Tzou, A., Mattar, P., Frank, J. A., Staudt, L., Martin, R., and McFarland, H. F. (2003). Expression profiling identifies responder and non-responder phenotypes to interferon-beta in multiple sclerosis. Brain 126, 1419–1429. Tezak, Z., HoVman, E. P., Lutz, J. L., Fedczyna, T. O., Stephan, D., Bremer, E. G., Krasnoselska-Riz, I., Kumar, A., and Pachman, L. M. (2002). Gene expression profiling in DQA1*0501þ children with untreated dermatomyositis: A novel model of pathogenesis. J. Immunol. 168, 4154–4163. Tompkins, S. M., and Miller, S. D. (2002). An array of possibilities for multiple sclerosis. Nat. Med. 8, 451–453. Tsukahara, T., and Arahata, K. (2003). A comparative gene expression analysis of Emery-Dreifuss muscular dystrophy using a cDNA microarray. Methods Mol. Biol. 217, 253–262. Tusher, V. G., Tibshirani, R., and Chu, G. (2001). Significance analysis of microarrays applied to the ionizing radiation response. Proc. Natl. Acad. Sci. USA 98, 5116–5121. van de Vijver, M. J., He, Y. D., van’t Veer, L. J., Dai, H., Hart, A. A., Voskuil, D. W., Schreiber, G. J., Peterse, J. L., Roberts, C., Marton, M. J., Parrish, M., Atsma, D., Witteveen, A., Glas, A., Delahaye, L., van der Velde, T., Bartelink, H., Rodenhuis, S., Rutgers, E. T., Friend, S. H., and Bernards, R. (2002). A gene-expression signature as a predictor of survival in breast cancer. N. Engl. J. Med. 347, 1999–2009. Van Deerlin, V., Gill, L., and Nelson, P. (2002). Optimizing Gene Expression Analysis in Archival Brain Tissue. Neurochemical Research 27, 993–1003. van ’t Veer, L. J., Dai, H., van de Vijver, M. J., He, Y. D., Hart, A. A., Mao, M., Peterse, H. L., van der Kooy, K., Marton, M. J., Witteveen, A. T., Schreiber, G. J., Kerkhoven, R. M., Roberts, C., Linsley, P. S., Bernards, R., and Friend, S. H. (2002). Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530–536. Walker, D. G., Lue, L. F., and Beach, T. G. (2001). Gene expression profiling of amyloid beta peptide-stimulated human post-mortem brain microglia. Neurobiol. Aging 22, 957–966. Wandinger, K. P., Sturzebecher, C. S., Bielekova, B., Detore, G., Rosenwald, A., Staudt, L. M., McFarland, H. F., and Martin, R. (2001). Complex immunomodulatory eVects of interferon-beta in multiple sclerosis include the upregulation of T helper 1-associated marker genes. Ann. Neurol. 50, 349–357. Watson, M. A., Perry, A., Budhjara, V., Hicks, C., Shannon, W. D., and Rich, K. M. (2001). Gene expression profiling with oligonucleotide microarrays distinguishes World Health Organization grade of oligodendrogliomas. Cancer Res. 61, 1825–1829.


151

Whitney, L. W., Becker, K. G., Tresser, N. J., Caballero-Ramos, C. I., Munson, P. J., Prabhu, V. V., Trent, J. M., McFarland, H. F., and Biddison, W. E. (1999). Analysis of gene expression in mutiple sclerosis lesions using cDNA microarrays. Ann. Neurol. 46, 425–428. Whitney, L. W., Ludwin, S. K., McFarland, H. F., and Biddison, W. E. (2001). Microarray analysis of gene expression in multiple sclerosis and EAE identifies 5-lipoxygenase as a component of inflammatory lesions. J. Neuroimmunol. 121, 40–48. Ye, S., Pang, H., Gu, Y. Y., Hua, J., Chen, X. G., Bao, C. D., Wang, Y., Zhang, W., Qian, J., Tsao, B. P., Hahn, B. H., Chen, S. L., Rao, Z. H., and Shen, N. (2003 Oct). Protein interaction for an interferon-inducible systemic lupus associated gene, IFIT1. Rheumatology (Oxford) 42(10), 1155–1163. Yoon, J. W., Kita, Y., Frank, D. J., Majewski, R. R., Konicek, B. A., Nobrega, M. A., Jacob, H., Walterhouse, D., and Iannaccone, P. (2002). Gene expression profiling leads to identification of GLI1-binding elements in target genes and a role for multiple downstream pathways in GLI1induced cell transformation. J. Biol. Chem. 277, 5548–5555.


DNA MICROARRAY ANALYSIS OF POSTMORTEM BRAIN TISSUE

Ka´roly Mirnics,*,{ Pat Levitt,*,{ and David A. Lewis*,x *Department of Psychiatry Department of Neurobiology and x Department of Neuroscience, School of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania 15261 { John F. Kennedy Center for Human Development and Department of Pharmacology Vanderbilt University, Nashville, Tennessee 37203 {

I. Introduction II. Challenges in Studies of the Postmortem Human Brain A. The Starting Material is Postmortem Tissue B. Samples Originate from a Diverse Genetic Background C. A Lifetime of Events Shaped the Transcriptome D. Treatment of Disease Influences Gene Expression E. Molecular Complexity of Human Brain Tissue F. Limited Sample Size and Sample Diversity G. Other Confounds: Technical and Biological III. Microarray Analysis of Human Brain Disorders A. Neurological Disorders B. Substance Abuse and Addiction Research C. Psychiatric Disorders D. Molecular Similarities between Brain Disorders IV. Where Do We Go from Here? References

I. Introduction

DNA microarray analyses of the transcriptome have played a pivotal role for analyzing the molecular biology of cancer and have become a very powerful exploratory tool in the study of central nervous system (CNS) (Luo and Geschwind, 2001; Marcotte et al., 2001; Mirnics et al., 2001b; Pollock, 2002). Systematic analysis of tens of thousands of transcripts simultaneously in a microarray experiment represents a true data-driven approach; that is, it allows but does not require a specific hypothesis. As such, the obtained data reach beyond the tested hypotheses, allowing the discovery of novel unanticipated relationships that are rarely revealed in hypothesis-driven research. This approach is especially important in psychiatric disorders, where our ability to form reasonable hypotheses about a disease process is often limited by the substantial diversity of cellular INTERNATIONAL REVIEW OF NEUROBIOLOGY, VOL. 60

153


154

MIRNICS et al.

phenotypes in the brain, as well as their heterogeneous molecular content and connectivity. We have witnessed the evolution of several sequence-based transcript analysis methods, including total gene expression analysis (TOGA) (SutcliVe et al., 2000), serial analysis of gene expression (SAGE) (Velculescu et al., 1995), and multiple parallel sequencing (MPSS) (Brenner et al., 2000). Although these and other sequence-based technologies oVer a very accurate analysis of the whole transcriptome and give rise to readily cross-comparable data, they are labor intensive and relatively low throughput. In addition, the costs associated with these experiments are prohibitive for most individual laboratories. As a result, these experiments are most often performed by analyzing pooled samples. This experimental design does not account for the individuality of each sample and thus allows for extreme outliers to significantly bias the resulting data. As a result of these limitations, the use of sequence-based genomic approaches in neuroscience remains limited. In contrast, complementary hybridization-based methods (primarily represented by DNA microarrays) (DeRisi et al., 1996; Lockhart et al., 1996; Schena et al., 1995) continue to be at the forefront of the transcriptome analysis of the brain tissue; at moderate cost, they can be relatively easily performed in an academic laboratory setting. Furthermore, they can rapidly assess the gene expression of most of the genome, creating a unique transcript profile for each sample. In human brain disorders, this information is essential, and using multiple molecular markers that give rise to complex gene expression patterns, it can uncover substratification within the experimental samples, which cannot be achieved in a pooled experimental design.

II. Challenges in Studies of the Postmortem Human Brain

Analyzing the human brain transcriptome and interpreting the outcome of these experiments is a challenging task. Although many of the following considerations are not unique for genomic experiments and apply in general to postmortem brain research, they have a major impact on planning and interpreting the outcome of microarray experiments.

A. The Starting Material is Postmortem Tissue In the absence of viable animal models for most of the psychiatric disorders, analysis of human postmortem tissue remains imperative. Although these samples are characterized by relatively long postmortem intervals (PMIs) that

MICROARRAYS IN PSYCHIATRY

155

typically range from 4 to 30 hours, in most cases they contain high-integrity messenger RNA (mRNA) that is suitable for microarray analysis. It is widely accepted that recombinant RNA (rRNA) 28s:18s ratio is a good predictor of mRNA integrity, and that intensity ratio less than 1 usually indicate the presence of degraded mRNA that is not suitable for microarray analysis. However, the most precise assessment of mRNA integrity is routinely performed using the AVymetrix GeneChip technology (Lipshutz et al., 1999; Lockhart and Barlow, 2001a), which using a multitude of short oligonucleotide probes assesses the 30 :50 integrity ratio for multiple housekeeping genes (e.g., actin and GAPDH). We believe that AVymetrix GeneChip 30 :50 ratios of more than 3 for actin and more than 2 for GAPDH have a significant impact on software decisions regarding diVerential gene expression, resulting in artificial overexpression reports of the more intact sample. Hence, samples not meeting these criteria should be excluded from the experimental series. Indeed, sample integrity remains one of the most critical aspects in postmortem brain research. In concordance with previous literature reports (Harrison et al., 1995), we found little evidence that samples with shorter PMIs would be more suitable for microarray analysis than those with longer PMIs. Rather, RNA integrity appears to be more dependent on the circumstances of death. Currently, the best surrogate measurement of agonal state and RNA integrity appears to be brain pH values. In general, samples with pH values less than 6.25 have a high probability of containing significant amounts of degraded RNA, thereby preventing a high-quality microarray analysis of these samples.

B. Samples Originate from a Diverse Genetic Background The genetic background of each brain (except monozygotic twins) is unique, and this has a major impact on the interpretation of microarray findings (Mirnics and Lewis, 2001). Gene single nucleotide polymorphisms (SNPs) exceed the number of genes by several orders of magnitudes, and for many genes, they represent one of the key determinants of expression levels. Although most studies have focused on the eVects of SNPs in the coding region of the mRNA, ample evidence exists that polymorphisms in noncoding regions can have significant eVects on transcriptional responses and mRNA levels; regulatory elements for most genes are in multiple locations, including several kilobases upstream of the gene in the 50 upstream region and in intronic sequences (D’Sa et al., 2002). Unfortunately, we do not have good strategies to distinguish SNP-driven expression changes from those that are downstream results of SNP-triggered changes or those that arise from epigenetic influences (see later discussion). In addition, no systematic studies of the human brain have addressed the relative frequency of expression changes resulting from SNPs versus epigenetic

156

MIRNICS et al.

transcriptome alterations; however, studies of transgenic animals clearly suggest that epigenetic and adaptational changes greatly outnumber SNP-driven or gene deletion–induced expression changes (Bunney et al., 2003; D’Agata et al., 2002). In the human brain, we can speculate that SNP-driven expression changes would be less common and more robust, present in all cell types expressing the gene, and present only in few subjects manifesting the disease. In contrast, epigenetic and adaptational gene expression responses are likely to be characteristic of most subjects with a disorder and they may diVerentially aVect diVerent brain structures and/or cell types. However, also plausible may be to expect that some expression changes have a dual origin; in some subjects they are determined by genetic influences, whereas in other individuals they are epigenetic. This combination of genetic and epigenetic-adaptational changes may sort subjects with a given clinical psychiatric syndrome into a spectrum of molecular phenotypes, which will allow a more precise molecular subclassification of psychiatric disorders. In contrast, common gene expression changes characteristic of all subjects within a given disorder may represent convergence points in cellular–molecular pathways that are putatively related to the common clinical features of the syndrome. As a result, these shared expression changes could represent more universal drug targets in the future than the more subject-specific SNP-driven transcriptome changes that are potentially involved in the primary pathogenesis of the disease. As we begin to better understand the molecular transcript networks, genetic and genomic research of psychiatric disorders will become more interactive, and only together will they be able to understand the precise molecular events associated with psychiatric disorders.

C. A Lifetime of Events Shaped the Transcriptome Microarray studies have unequivocally proven that in rodents environmental enrichment significantly shapes the brain transcriptome (Rampon et al., 2000), and this is undoubtedly even more pronounced in the human brain (Francis et al., 2002). The lifetime experiences from our surroundings constantly shape our transcriptome profile. This process starts from the intrauterine formation of the CNS and continues until death. Some of these influences will leave behind long-lasting structural changes that are obvious many years after the removal of the primary insults (Stanwood et al., 2001), whereas others are rapidly reversible and related to environmental influences relatively close to the time of death. Smoking, drug abuse, physical activity, and many other lifestyle events are all potent modulators of the CNS transcriptome, and these influences interact with gene expression diversity that occurs as a result of diVerences in age, sex, and race.


157

Unfortunately, the transcriptome profile obtained in microarray analyses of human postmortem brain tissue does not speak to these dynamics of the molecular events associated with the disorder; it provides only a transcript snapshot. Precise reconstruction of disease-associated dynamic events, even using these molecular snapshots from many subjects, is extremely diYcult. However, the gathered information allows the generation of informed hypotheses about the pathophysiological processes associated with the disease, and this allows testing these specific hypotheses in animal models. Although there are no perfect animal models of psychiatric disorders, tissue culture and animal experiments allow us to model aspects of the critical physiological processes underlying psychiatric disorders (Dirks et al., 2003; Halim et al., 2003; Lipska et al., 2003; Van den Buuse et al., 2003). For example, transgenic animal studies greatly complement postmortem studies, because they (1) allow testing of causality by targeted manipulation of the genome, (2) minimize sample individuality and provide a comparison to a true control population that has similar genetic background and was exposed to similar epigenetic influences, and (3) provide information about the dynamics of the process studied through the analysis of diVerent time points. The interplay of postmortem and animal studies is a two-way exchange of critical information. The outcome of animal studies can be verified in postmortem brain tissue studies, and based on the obtained anatomical-regional findings, this information will generate a new hypothesis, which will in turn start a new round of the discovery process using animal models.

D. Treatment of Disease Influences Gene Expression Most subjects with major psychiatric disorders were receiving psychotropic medications at the time of death. This potentially represents a major confounding factor in postmortem brain research. Drug treatment is a potent modifier of CNS gene expression (Gunther et al., 2003; Kontkanen et al., 2002; Yamada et al., 2002); however, molecular changes induced by treatment are not well understood. Furthermore, transcriptome changes induced by drugs may diVerentially aVect various brain regions. In addition, the eYcacy of drug treatment can strongly depend on the genetic background of the treated individual (Geschwind, 2003; Shilling and Kelsoe, 2002). Our understanding of the drug-induced transcriptome changes is mostly derived from tissue culture and rodent experiments, yet these models may not model some of the critical aspects of the drug actions; the most developed brain regions (e.g., prefrontal cortex) show significant diVerences between the rodent and primate brain regions. These diVerences include (but are not limited to) projection neuron/interneuron ratio, relative thickness of laminae, local connectivity, and dopaminergic innervation. Thus,

158

MIRNICS et al.

targeted evaluation of medication eVects on nonhuman primates is a well justified and integral part of the psychiatric transcriptome research. Gene expression profiles obtained in nonhuman primate experiments (combined with information obtained on humans that were oV medication at the time of death) will greatly help us diVerentiate medication-induced eVects from disease eVects (Middleton et al., 2002; Mirnics et al., 2000, 2001c; Pierri et al., 1999). However, we must not forget that not all medication eVects are mediated through the transcription machinery; we can also expect significant medication eVects at the level of protein–protein interactions.

E. Molecular Complexity of Human Brain Tissue Most of the microarray experiments conducted were performed on bulk tissue, so they have limited anatomical resolution. However, the remarkable cellular and molecular diversity of the CNS tissue creates an additional challenge in analyzing brain tissue (Mirnics et al., 2001a,b). In the context of this diversity, in psychiatric disorders (1) not all cell types expressing a certain gene are aVected with a disease process and (2) the magnitude of transcript changes in the aVected cell population is usually moderate. As a result, biologically important transcript changes that occur in a subpopulation of cells are often masked by the unaVected sources of the RNA pool. This clearly limits the power of bulk analysis methods (including microarrays and real-time reverse transcriptase polymerase chain reaction [RT-PCR]) to detect these ‘‘diluted’’ expression changes. Furthermore, even for gene expression changes uncovered by microarrays, we do not know where the gene expression changes occur, and this information is critical for the interpretation of the obtained data (Pongrac et al., 2002). For example, a change in three diVerent transcripts (e.g., a receptor, a G-protein subunit, and a cellular eVector) will have a very diVerent biological interpretation if they occur in the same cell or if they are distributed across non-overlapping cell populations. Reference information about gene expression profiles of subpopulations of cells are virtually nonexistent, greatly limiting our capability to interpret microarray data in the context of cellular phenotypes. We do not even know the gene expression profiles of diVerent brain regions (but see eVorts by Huminiecki et al., 2003) or individual cortical laminae. However, the invention of laser capturedissection microscopy (LCM) (Bonner et al., 1997) and the development of singlecell transcriptome profiling (Ginsberg and Che, 2002; Ginsberg et al., 2000; Hemby et al., 2002; Kamme and Erlander, 2003; Luo et al., 1999; WittliV and Erlander, 2002) are altering the design of microarray experiments because they permit the analysis of cellular subpopulations and specific laminae or subnuclei (Bonaventure et al., 2002). This is especially important for the microarray analysis


159

of psychiatric disorders, where our current biological knowledge suggests that certain cell types and laminae are preferentially aVected by the disease process (Pierri et al., 2001; Rajkowska et al., 1998, 2001, 2002). However, in situ hybridization (ISH) provides the highest anatomical resolution of all transcript analysis methods and it can pinpoint changes that occur in neuronal or glial subpopulations or those specific to laminae or subnuclei. Despite being a labor-intensive and relatively low-throughput method, this precision makes ISH a preferred choice in following up bulk transcriptome data: It not only verifies the microarray-generated data but also puts the findings into a precise anatomical context, which can provide further insight into the functional meaning of the observed changes.

F. Limited Sample Size and Sample Diversity Comprehensive studies of postmortem tissue obtained from subjects with psychiatric disorders are limited by challenges involved in obtaining brain tissue donations from well-characterized subjects (Lewis, 2002). In addition, these precious gifts are diverse with respect to a range of factors, including age, race, PMI, medication history, and lifestyle of the deceased. Furthermore, psychiatric disorders are heterogeneous, and the clinical diversity within a disorder (e.g., paranoid schizophrenia vs schizoaVective disorder) may correspond to molecular subclasses that can introduce further variance into the studied cohorts (Mirnics and Lewis, 2001). In addition, co-morbidity with other disorders represents a significant challenge in interpretation of the obtained data. This co-morbidity includes both diseases that are known to aVect gene expression in the nervous system (e.g., substance abuse) and systemic diseases (e.g., diabetes or systemic lupus) (Green et al., 2003) for which the influences on the brain transcriptome have not been assessed. It is also important to realize that certain co-morbidities (e.g., nicotine abuse) occur at a much higher rate in subjects with psychiatric disorders (Batel, 2000; Chambers et al., 2001), which may lead to cohort biases. As a result, for now we will have to continue to perform experimental series on limited sample sizes with a considerable intradiagnostic variability and comorbidity. We hope that the future standardization of microarrays experimental protocols (e.g., sample harvest, storage, RNA isolation, reverse transcription and amplification procedures, labeling, microarray platforms, and analysis strategies) will enable direct data set comparisons originating from diVerent groups of investigators (Mirnics, 2001a, 2002). These data sets, if they become truly transparent and can be analyzed as a single experiment, will have the power to define molecular subclasses (or the transcript continuum) even for the most complex psychiatric disorders.

160

MIRNICS et al.

G. Other Confounds: Technical and Biological Confounds related to microarray technology can be avoided by careful experimental planning (Mirnics, 2001b). These include making sure that the samples are parallel processed, using the same harvesting techniques and storage procedures, isolating RNA and performing the reverse transcription and labeling reactions according to strict protocols. Choosing microarrays from the same lot in a single study also appears to further reduce experimental variability. In the data analysis, segmentation and microarray image analysis should be performed blinded to the sample class. Finally, replication of the microarray study on a second set of arrays is strongly recommended, because it provides feedback about the assay noise and eliminates most of the technical confounds and type I errors. Unfortunately, it is impossible to control for all the disease-unrelated variables that may shape the brain transcriptome and can potentially confound the results. In postmortem studies, investigators try to match PMI, age, race, gender, and other factors between the experimental and control cohorts. Using a pairwise experimental design (in which each experimental sample is matched as closely as possible to a control sample) eliminates many of the gene expression diVerences that are related to the factors used for matching. However, many putative variables are unknown and they may be unevenly distributed across our control and experimental groups. These factors, depending on the strength of their transcriptome-shaping influences, may result in partially or fully confounded data sets, underscoring the importance of replication studies in psychiatric postmortem brain research. These validation studies are ideally performed on a diVerent subject cohorts group and by a diVerent group of investigators.

III. Microarray Analysis of Human Brain Disorders

To date, most of the brain microarray experiments have focused on animal models. There is a substantial body of literature on gene expression changes observed in in vivo and in vitro disease models (Grunblatt et al., 2001; Toyooka et al., 2002; Zhou et al., 2003), eVects of drug treatment (Kontkanen et al., 2002; Yamada et al., 2001, 2002), addiction research (Ammon et al., 2003; Thibault et al., 2000, 2001; Yuferov et al., 2003), transcriptome changes in the developing brain (Karsten et al., 2003; Kornblum and Geschwind, 2001; Lockhart and Barlow, 2001b; Sandberg et al., 2000), analysis of transgenic animals (D’Agata et al., 2002; Dirks et al., 2003; Tudor et al., 2002), phenotypical/genotypical influences on brain (TabakoV et al., 2003; Wang et al., 2003), and aging (Blalock et al., 2003). These experiments greatly complement the rapidly growing


161

transcriptome profiling data obtained from postmortem tissue but cannot replace performing experiments on the diseased human brain; the human brain transcriptome is diVerent from the expression pattern seen in the animal brains, including the transcript profile of the nonhuman primates (Caceres et al., 2003).

A. Neurological Disorders Postmortem microarray research of neurological disorders has been very productive over the last several years. In particular, transcriptome profiling of Rett’s syndrome, Alzheimer’s disease (AD), and multiple sclerosis (MS) has been at the forefront of these analyses, and the results are providing a fundamentally new view of these disorders. Analysis of the frontal cortex in subjects with Rett’s syndrome revealed that mutation of transcriptional repressor methylCpG–binding protein-2 (MECP-2) leads to alterations in the mRNA levels NMDA-NR1, MAP-2, and synaptic vesicle proteins ( Johnston et al., 2001), as well as increased expression of glial markers (Colantuoni et al., 2001). Studies of AD also revealed a complex set of multiregional expression changes. Studies in the amygdala and cingulate cortex revealed upregulated transcripts related to chronic inflammation, cell adhesion, cell proliferation, and protein synthesis, whereas signal transduction, energy metabolism, stress response, synaptic vesicle synthesis and function, calcium binding, and cytoskeleton-related transcripts were downregulated (Loring et al., 2001). Comparing the neurofibrillary tangle–containing hippocampus of AD subjects to the non–tanglebearing parietal cortex within the same brains revealed a robust increase in calcineurin A mRNA in the pyramidal cells of most diseased subjects relative to control brains (Hata et al., 2001). In the analysis of AD brains Pasinetti (2001) found mRNA expression changes suggesting that protein and amino acid metabolism, cytoskeleton integrity, and fatty acid metabolism are involved in early phases of AD dementia. Most notably, this study also suggested that neurotransmitter-released transcripts, including synapsin, may be diVerentially regulated in the brains of cases at high risk for dementia (Ho et al., 2001). Finally, in a GeneChip study of CA1 region Colangelo et al. (2002) found widespread transcriptional alterations, misregulation of RNAs involved in metal ion homeostasis, trophic factor signaling deficits, decreases in neurotrophic support, and activated apoptotic and neuroinflammatory signaling in moderately aVected AD hippocampal CA1. Although most of the AD microarray studies focused on bulk tissue, elegant studies of Ginsberg et al. (1999) successfully identified the expression profiles of neurofibrillary tangles and CA1 tangle–bearing projection neurons (Ginsberg et al., 2000). Furthermore, they reported that in AD subjects, anterior nucleus basalis neurons undergo selective alterations in gene expression,

162

MIRNICS et al.

including upregulation of cathepsin D and downregulation of synaptophysin, synaptotagmin, and protein phosphatases (PP1) transcripts (Mufson et al., 2002b). In addition, they were able to correlate postmortem single-cell expression profiles in nucleus basalis (Mufson et al., 2002a,b) with the degree of cognitive impairment seen in the premortem subjects with AD: The initial findings suggest that alterations in neurofilament and tau gene expression occur in NB neurons at early stages of cognitive decline. In one of the first cDNA array studies performed on the human postmortem brain tissue, Whitney et al. (1999) analyzed the expression pattern of more than 5000 genes and compared the gene expression profile of normal white matter to that found in acute lesions from the brain of a single patient with MS. Using a radioactive sample labeling technology, this study identified 62 diVerentially expressed genes, including the DuVy chemokine receptor, interferon regulatory factor-2, and tumor necrosis factor- receptor-2 among others. In another microarray analysis of MS lesions Lock et al. (2002) reported increased transcripts of genes encoding interleukin-6 (IL-6) and IL-17, interferon-, and associated downstream cascades. This study also observed significant expression diVerences between acute lesions with inflammation versus ‘‘silent’’ lesions without inflammation. For example, granulocyte colony-stimulating factor is upregulated in acute but not in chronic MS lesions. Based on the human expression findings, this study also evaluated the amelioration of experimental autoimmune encephalomyelitis in mice, finding that knocking out the immunoglobulin Fc primarily ameliorated changes associated with the chronic form of MS. Most recently, Mycko et al. (2003) found major gene expression diVerences in a microarray analysis of the regions of pathologically proven diVerent activity of MS lesions. Namely, the lesion margin and lesion center in active lesions reported 57 and 69 diVerentially expressed genes, whereas the margins and centers of silent lesions showed only 11 and 2 diVerentially expressed genes. To compare diVerences between chronic active and silent lesions, the investigators also performed a comparison of the pooled data from both types of lesions. Perhaps not unexpectedly, many of the genes with changed expression encoded proteins that are involved in inflammation/immune response. This microarray analysis has also identified a novel set of genes associated with lesion activity in MS, many of them not previously linked to the disease. Finally, McDonough et al. (2003) have profiled MS cortex rather than white matter and have analyzed non–lesioned cortex and lesioned areas in area 4 of motor cortex. Preliminary results suggest that there is a robust decrease in GABAergic neurotransmission including GAD67 (a synthetic enzyme for the inhibitory neurotransmitter GABA), and GABA receptor 1 and 3 subunits. Furthermore, initial data suggest that transcript decreases in nuclear encoded mitochondrial genes of the respiratory chain are also strongly associated with MS in the motor cortex.


163

B. Substance Abuse and Addiction Research Thanks to the mostly conserved responses to various chemical substances between the human and animal brain tissue, animal model studies of substance abuse and addiction are very informative and greatly outnumber the experiments performed on human brain tissue. Transcriptomic studies in this field are spearheaded by ethanol and cocaine abuse analysis (Ang et al., 2001; Freeman et al., 2001, 2002a,b,c; Thibault et al., 2000; Yuferov et al., 2003), both in disease models and on postmortem tissue (Albertson et al., 2003; Lewohl et al., 2000b, 2001). However, the importance of analysis of human postmortem tissue cannot be overestimated; the human brain is more complex than the rodent brain and diVers from it in connectivity, variability of genetic background, interneuron/ projection neuron ratio, and numerous other aspects. This is underlined by an elegant genetic association study on microarrays (Uhl et al., 2001) that marked the ADH, BDNF, and seven other loci linked to vulnerability to nicotine or alcohol abuse. One of the first postmortem microarray expression studies also implicated many novel genes that may be associated with alcohol abuse. Lewohl et al. (2000a) used both cDNA and oligonucleotide platforms to assess the eVect of chronic ethanol exposure on the transcriptome of the postmortem human prefrontal cortex and found a prominent downregulation of myelin-related genes in the experimental samples. Furthermore, in this study, cell cycle genes and several neuronal genes also reported reproducible transcript level changes in the cortices of alcoholics. Using a more complex array platform (Mayfield et al., 2002), in addition to transcript changes in genes encoding myelination proteins, these investigators also found mRNA changes in genes involved in calcium, cyclic adenosine monophosphate (cAMP), and thyroid-signaling pathways. Interestingly, preliminary studies by Albertson et al. (2003) also suggest a profound myelin dysregulation in human cocaine abusers, suggesting a possible common mechanism between some eVects of cocaine and alcohol abuse. Furthermore, a cDNA array study by Tang et al. (2003) compared gene and protein expression patterns between cocaine overdose victims and age-matched controls in the ventral tegmental area (VTA) and lateral substantia nigra (lSN). Interestingly, whereas the lSN showed no significant changes in gene expression between the overdose victims and matched controls, VTA analysis revealed significant upregulation of NMDAR1, GluR2, GluR5, and KA2 receptor mRNAs both at the transcript and at the protein level. C. Psychiatric Disorders Postmortem brain microarray studies of psychiatric disorders have primarily focused on schizophrenia, although there is increased interest in microarray analysis of postmortem tissue from subjects with major depression, bipolar disorder and autism.

164

MIRNICS et al.

In a study of autism, Purcell et al. (2001) analyzed cerebellar samples using two microarray platforms from 10 individuals with autism and 23 matched controls. The mRNA levels of several genes were significantly increased in autism, including excitatory amino acid transporter-1 and glutamate receptor AMPA-1, which was also verified at a protein level. Based on this study, the authors concluded that subjects with autism might have specific abnormalities in the AMPA-type glutamate receptors and glutamate transporters in the cerebellum. Although several comprehensive and very promising studies are in progress, perhaps because of the phenotypical and molecular complexity of major depression, microarray analysis of this disorder is still at a preliminary stage. In one such initial microarray study on medication-free depressed patients who died by suicide, Sibille et al. (2002) found that expression of 100–150 genes appears to segregate patients into two distinct molecular subtypes. The observed diVerences in gene expression were consistent across Brodmann areas 9 and 47 and included gene transcripts encoding genes involved in monoaminergic/glutamatergic synaptic neurotransmission and neurotrophin-dependent tyrosine phosphorylation. In a separate preliminary study, Evans et al. (2002) used DNA microarrays to study expression profiles of human postmortem brains from patients diagnosed with major depressive or bipolar disorder in the anterior cingulate cortex, dorsolateral prefrontal cortex, and the cerebellum. The findings suggest that the dorsal lateral prefrontal cortex from patients with a major depressive disorder is characterized by coordinated alterations of gene expression in growth factor pathways. 1. Transcriptome Changes in Schizophrenia Of all psychiatric disorders, transcriptome changes in schizophrenia are the best characterized. The lessons learned in this transcriptome assessment process are directly applicable to microarray analysis of other psychiatric disorders, allowing us to more carefully design and carry out transcriptome experiments. Schizophrenia is a complex multigenic brain disorder characterized by a constellation of psychotic, negative, and cognitive features (Carpenter and Buchanan, 1994; Lewis and Lieberman, 2000). In addition to genetic factors, environmental factors influencing brain development greatly contribute to risk for the disease, which has a typical clinical onset around puberty/young adulthood. The prefrontal cortex of subjects with schizophrenia shows both anatomical and physiological changes, and prefrontal dysfunction is thought to underlie at least some of the cognitive deficits in schizophrenia. Not surprisingly, microarrays studies have focused mostly on complex expression changes that occur in area 9 of the dorsolateral prefrontal cortex (DLPFC). a. Existing Microarray Studies: Results. The existing microarray studies of schizophrenia provide a solid foundation and can be further explored in specific


165

hypothesis-driven research. These early studies diVered considerably in experimental design, including diVerent approaches vis-a`-vis sample harvest, material pooling, RNA amplification and labeling, DNA array platforms, and data analysis approaches. Furthermore, the studied cohorts ranged from elderly chronically hospitalized with end-stage schizophrenia to outpatient cohorts that also included individuals with schizoaVective disorder. Not surprisingly, the molecular profiles obtained in these experiments are somewhat diVerent, though not mutually exclusive. In our initial study, we preformed a cDNA microarray expression profiling of prefrontal cortex (area 9) from matched pairs of schizophrenic and control subjects (Mirnics et al., 2000). A biological pathway-related analysis revealed that genes encoding proteins involved in presynaptic secretory release (PSYN) were decreased in all subjects with schizophrenia, albeit with specific pattern of altered transcripts diVering across subjects. A similar, robust decrease pattern was also observed for transcripts encoding genes involved in GABAergic and glutaminergic transmission. Selected microarray observations were verified by ISH. At the ‘‘most changed’’ analysis level, N-ethylmaleimide–sensitive factor (NSF) and synapsin II (SYN2) were robustly decreased in all subjects with schizophrenia. Furthermore, we observed expression reduction in GAD67 and AMPA-2 receptors, replicating previously reported literature findings (Eastwood et al., 1995; Volk et al., 2000). None of these changes were observed in monkeys chronically treated with haloperidol, suggesting that the obtained expression changes were part of the disease process, rather than a consequence of treatment with antipsychotic medication. Furthermore, an expanded study that included the original data set, revealed statistically significant expression alterations in 5 of 71 assessed metabolic pathways (Middleton et al., 2002). Reductions in expression were observed for gene transcripts that are part of the ornithine and polyamine metabolism, mitochondrial malate shuttle system, transcarboxylic acid cycle, aspartate and alanine metabolism, and ubiquitination pathway. Metabolic genes showing the most decreased expression included cytosolic malate dehydrogenase (MAD), mitochondrial glutamate-oxaloacetate transaminase type 2, ornithine decarboxylase antizyme inhibitor, and ornithine aminotransferase. Most of these genes that were consistently decreased across subjects with schizophrenia were not similarly decreased in haloperidol-treated monkeys. In contrast, the transcript encoding the cytosolic form of MAD displayed increases in expression in chronic haloperidol-treated monkeys. This increase was most prominent in the deep cortical layers, the cortical regions that have the highest concentration of the D2 receptors in the DLPFC. These molecular analyses implicate a highly specific pattern of metabolic alterations in the DLPFC of subjects with schizophrenia and raise the possibility that antipsychotic medications may exert a therapeutic eVect, at least in part, by normalizing some of these expression changes.

166

MIRNICS et al.

Fig. 1. Substratification and putative co-regulatory patterns in the prefrontal cortex of subjects with schizophrenia. (A) Two-way hierarchical clustering of complementary DNA (cDNA) microarray data obtained by comparing 10 pairs of subjects with schizophrenic and matched controls. The 4096 expressed genes are clustered in rows; matched subject pairs are clustered in columns. For each gene, z scores are continuously color coded from red (decreased expression) to green (increased expression) in schizophrenic subjects. Two-way clustering was performed by average linkage based on euclidean distance using cluster. Note that there is a significant substratification of the data set: In addition to


167

In a further analysis of the expanded microarray data set, hierarchical clustering revealed an unexpected result: We observed a molecular substratification of the studied subjects with schizophrenia. In Fig. 1A, of the more than 4000 expressed genes (rows) in 10 array comparisons (columns), in addition to about 300 consistently underexpressed (Fig. 1B) and about 70 overexpressed genes, there are two extensive mirror-image data clusters (Fig. 1C), arguing for a presence of two major subclasses of schizophrenic subjects within the data set. The physiological meaning of these major clusters is not easily interpretable, but it suggests that follow-up analysis should also assess the relationship to this substratification. For example, if this gene expression substratification is a result of specific genetic vulnerability, our power to identify the susceptibility genes will be greatly enhanced if we subdivide and analyze the studied cohort according to a molecular substratification. Finally, we decided to test a hypothesis that at least some of the gene transcripts consistently decreased across all subjects with schizophrenia (cluster in Fig. 1B) may share a common underlying molecular drive. To test this, we performed a batch retrieval of 2kB/gene from the EZRetrieve database for the 50 most changed genes in our study. This region presumably contains both the promoter site and multiple gene transcription regulatory elements. After obtaining these sequences, they were reformatted with a custom made perl script and imported into a SRMS database (Silico Informatics Systems, Santa Clara, California). This was followed by a motif analysis on the MEME server (Bailey and Elkan, 1995), and the obtained results were once again inserted into the SRMS database. Graphical visualization of the motifs, patterns, and sequences was performed in SRMS. Interestingly, the genes with consistently decreased expression shared multiple DNA motifs (Fig. 1D) that may represent common regulatory sequences. Even when corrected for multiple comparisons, the occurrence of multiple motifs was highly significant ( p < .0001).

about 200 genes showing a consistent downregulation across all comparisons (denoted by red vertical bar) and about 50 genes reporting a consistent upregulation (vertical green bar), we observed two major clusters of genes that show an inverted expression pattern (vertical blue bars). (B) To visualize the individual genes and their expression ratio z scores, part of the cluster denoting consistently decreased genes is enlarged. (C) Pairwise expression z scores for the two clusters showing inverted gene expression patterns are plotted on the y-axis; x-axis denotes matched schizophrenia–control subject pairs. Note that for this subset of genes the expression patterns are inverted. (D) Lines represent 50 DNA region of several genes selected from the cluster of Fig. 1B. Colored boxes denote common motifs in the promoter region. Note that the same conserved motifs occur in multiple genes, suggesting a putative common co-regulation of these transcripts. (E) Targeted assessment of coregulation of the functionally related N-ethylmaleimide sensitive factor (NSF)-SNAP and NSF are highly co-regulated patterns across the 10 comparisons, whereas -SNAP shows a less correlated downregulation pattern. (See Color Insert.)

168

MIRNICS et al.

To assess gene expression patterns in the cerebellum, prefrontal cortex, and middle temporal gyrus of subjects with schizophrenia, Vawter et al. (2001) used a pooled experimental design in conjunction with membrane-based cDNA microarrays containing about 1100 brain-biased probes. In the cerebellum and PFC of drug-treated subjects, 21 genes showed diVerential expression, compared to only 5 genes for drug-naive patients. DiVerentially expressed gene products were related to synaptic signaling and proteolytic functions, some of which also showed diVerential expression in the middle temporal gyrus (tyrosine-3-monooxygenase/ tryptophan 5-monooxygenase activation protein, eta polypeptide; sialyl transferase; proteasome subunit, alpha type 1; ubiquitin carboxyl-terminal esterase L1; and solute carrier family 10, member 1). In a separate study, using the same DNA array platform and multiple pools of RNA from subjects with schizophrenia and matched controls (BA9 PFC), Vawter et al. (2002) found three genes that showed consistently decreased expression in schizophrenia by both z-ratio diVerences and decreased normalized numerical ratios. These were histidine triad nucleotidebinding protein (HINT), ubiquitin-conjugating enzyme E2N (UBE2N) and glutamate receptor, ionotropic, AMPA-2 (GRIA2). Interestingly, the results confirmed many of the gene expression decreases we reported in our initial study. These genes included multiple gene products belonging to the presynaptic secretory, glutamatergic, and GABAergic pathways. Using a groupwise experimental design and AVymetrix GeneChip oligonucleotide arrays with about 6000 probe sets, Hakak et al. (2001) performed a transcriptome analysis of the PFC of subjects with elderly, hospitalized subjects with schizophrenia, and matched controls. Most notably the results identified a set of oligodendrocyte- and myelination-related genes that were underexpressed in the diseased subjects. In a further linear discriminant analysis (Schadt et al., 2001), the 35 myelination-related genes (including the most decreased MAL, MAG, transferrin, gelsolin, and Her-3 transcripts) perfectly separated out the schizophrenic subjects from the matched controls. Mimmack et al. (2002), using diVerent cohorts, also performed a DNA array analysis of the PFC of subjects with schizophrenia. In this study the investigators employed a custom-made cDNA array platform with 300 gene probes, which were selected based on their likeliness to be involved in the pathophysiology of schizophrenia. This study, which was cross-validated across three independent cohorts of subjects, found that several members of the apolipoprotein L (apoL) family showed increased mRNA levels in subjects with schizophrenia. Importantly, the apoL proteins belong to the group of high-density lipoproteins, with all six apoL genes located in proximity to each other on chromosome 22q12, a confirmed susceptibility locus for schizophrenia. The same laboratory performed an indexing-based diVerential display PCR study and GeneChip analysis of BA9 PFC of subjects with schizophrenia, bipolar disorder, and matched controls (Tkachev et al., 2003). These oligonucleotide arrays contained more than


169

20,000 unique gene probe sets, providing an in-depth view into the transcriptome changes associated with these two diseases. Results of diVerential display and quantitative PCR analysis, as well as the microarray study, showed a reduction of key oligodendrocyte-related and myelin-related transcripts in subjects with schizophrenia and bipolar disorder. Importantly, the expression changes for both disorders showed a high degree of overlap. In an elegant analysis of entorhinal cortex layer II stellate neurons from postmortem samples of schizophrenic and age-matched control brains, Hemby et al. (2002) found marked diVerences in expression of various G-protein–coupled receptor signaling transcripts, glutamate receptor subunits, synaptic proteins, and other transcripts. In a secondary screen of these entorhinal cortex layer II stellate neurons, schizophrenia-associated decreases were observed in levels of G-protein subunit i(alpha)1, glutamate receptor 3, NMDA receptor 1, synaptophysin, SNAP23, and SNAP25. b. Existing Microarray Studies: Common Findings. At the simplest level of data analysis, these studies identified the ‘‘genes with most changed expression.’’ Although the findings are somewhat diverse (mostly as a result of specificities in experimental design), it is also important to point out that there are consistent findings between these expression studies. Genes that reported expression changes in the PFC of subjects with schizophrenia across multiple studies (Mirnics et al., 2000, 2001c; Vawter et al., 2001, 2002; Petryshen et al., 2003; Pongrac et al., 2002) (performed on multiple cohorts in diVerent laboratories) include several neural genes such as regulator of G-protein signaling 4 (RGS4), neuroserpine, AMPA-2 receptor, GAD67, AF1q, NSF, 14-3-3 isoforms, MAD-1, as well as multiple oligodendrocyte-related genes (proteolipid protein 1, ErbB3, transferrin, myelin-associated glycoprotein, and gelsolin (Hakak et al., 2001; Hof et al., 2002; Mimmack et al., 2002; Pongrac et al., 2002; Tkachev et al., 2003). These molecules paint a complex molecular picture of schizophrenia, one that involves molecular disturbances in various cell types. When all the results from the schizophrenia transcriptome studies are combined, it appears that the aVected transcripts in schizophrenia are associated with the processes of synaptic release, cell signaling, second messenger systems, energy metabolism, protein turnover, and myelination. These changes are distributed in a complex pattern across diVerent cell populations, including projection neurons, interneurons, and oligodendrocytes. How are these changes orchestrated? Although we can generate informed and testable hypotheses about co-regulations of some transcripts (e.g., BDNF-TRKB-GAD67-parvalbumin (Hashimoto et al., 2003) or OLIG1-SOX10-PLP1-Her-3-MBP-MOG-MAG (Tkachev et al., 2003), the connection between other systems is not obvious, and in its complexity, it will exceed the power of transcriptome analysis methods, thereby emphasizing the need for follow-up experiments in various biological models.

170

MIRNICS et al.

c. Existing Microarrays Studies: Differences. Although a significant portion of the variable findings can be explained by the diVerent methodological approaches, the diversity of the molecular phenotypes subsumed under schizophrenia may hold a key to others. In this context, all studied cohorts may be biased to preferentially include diVerent molecular subphenotypes of subjects with schizophrenia. For example, chronically hospitalized subjects who responded poorly to antipsychotic medications may have a more severe (or even diVerent) molecular phenotype than subjects with schizophrenia who were living in the community and responded well to treatment. This explanation is consistent with the current view of the genetics of schizophrenia, which suggests that there are a large number of susceptibility genes, each of which has a relatively small eVect (Pulver, 2000; Tsuang, 2000). As a result, the potential combination of the genetic susceptibility factors is huge, and this can undoubtedly result in a broad spectrum of transcriptome phenotypes of schizophrenia. d. Expression Changes vs Susceptibility Genes in Schizophrenia. In our initial microarray study, we found regulator of G-protein signaling 4 (RGS4) as the gene with a most prominent expression reduction across all subjects with schizophrenia (Mirnics et al., 2001c). This finding was of a particular interest to us, because the RGS4 protein limits the duration of signaling from multiple G-protein receptors, including the ones that are the targets of atypical antipsychotic agents. Verification by ISH revealed that this decrease in the RGS4 transcript was present across multiple cortical regions, including the prefrontal, primary motor, and visual cortices. Furthermore, this change was not observed in monkeys treated with chronic antipsychotic medication or in subjects with major depression. These results raised the possibility that expression changes in RGS4 might reflect a primary genetic abnormality and that variants in RGS4 might confer increased risk for schizophrenia. To test this idea, we conducted genetic association and linkage studies (Chowdari et al., 2002) using samples ascertained independently in Pittsburgh and New Delhi and by the NIMH Collaborative Genetics Initiative. Using the transmission disequilibrium test, we observed significant transmission distortion in the Pittsburgh and NIMH samples. Among SNPs spanning approximately 300 kb, significant associations involved four SNPs localized to a 10-kb region at RGS4, although the associated haplotypes diVered. Two other recent research groups have obtained results confirming an association between RGS4 SNPs and schizophrenia (Morris et al., 2003; Williams et al., 2003), suggesting that RGS4 represents a novel schizophrenia susceptibility gene. Conforming RGS4 as a susceptibility gene has important implications by providing proof of principle that microarray-discovered transcriptome changes may identify underlying susceptibility genes. Indeed, further evidence is now emerging that expression studies are providing valuable leads for genetic associations: GAD-67 (one of the genes with the most consistently observed altered expression in schizophrenia) (Addington et al., 2003; Straub et al., 2003) and


171

GABA-A receptor 3 (Lo et al., 2003) and 2 (Turunen et al., 2003) subunits (both showing expression decreases in our data set) have been implicated in preliminary genetic studies as putative schizophrenia susceptibility genes.

D. Molecular Similarities between Brain Disorders Similarities in transcriptome changes across multiple psychiatric disorders are not entirely unexpected in the context of the partially overlapping susceptibility loci. Indeed, the emerging microarray data suggest that certain molecular events may be characteristic for more than one brain disease. This knowledge will be essential to understanding the disease-associated molecular pathophysiology; however, these relationships are not well explored. Although they are clinically very diVerent diseases, schizophrenia and MS appear to share common deficits in expression of myelination genes, critical GABA system transcripts, mitochondrial genes, and RGS4 expression (Hakak et al., 2001; Lock et al., 2002; McDonough et al., 2003; Middleton et al., 2002; Mirnics et al., 2000; Mycko et al., 2003; Pongrac et al., 2002; Steinman and Zamvil 2003; Tkachev et al., 2003; Whitney et al., 1999). However, these molecular changes occur in diVerent brain regions, thus potentially defining diVerent phenotypical manifestations of the disease. Bipolar disorder and schizophrenia also share a common deficit in the myelination-related genes (Tkachev et al., 2003). Interestingly, similar to bipolar disorder and schizophrenia, cocaine abuse and ethanol abuse may also be associated with deficits in gene transcripts responsible for myelination (Albertson et al., 2003; Lewohl et al., 2000b; Mayfield et al., 2002). Autism, major depression, and schizophrenia share a common mechanism of altered glutaminergic gene expression (Mirnics et al., 2000; Purcell et al., 2001; Sibille et al., 2002). AD and schizophrenia are both characterized by reduction in synaptic markers (Ho et al., 2001; Mirnics et al., 2000; Pasinetti 2001; Vawter et al., 2002). In contrast, the absence of similar expression changes may also be somewhat informative vis-a`-vis the disease process. Perhaps surprisingly, at the level of expression changes, major depression and bipolar disorder appear to have less in common than expected (Evans et al., 2003; Petryshen et al., 2003; Tomita et al., 2003).

IV. Where Do We Go from Here?

Although much has been done already, microarray studies of human brain disorders are in still in the data collection phase; we are identifying genes that are consistently changed across subjects and trying to provide a mechanistic

172

MIRNICS et al.

explanation for these expression changes. Sometime soon, we will see further evolution of postmortem transcriptome studies: Microarray data collection and reporting will be standardized: Although microarray experiments have already uncovered important processes associated with psychiatric diseases, our data are not readily cross-comparable. The Pritzker Foundation funded joint expression profiling eVorts between University of California at Irvine, the University of Michigan, and the University of California at Davis represent a prototype as to how these experiments should be conducted in the future (Fell, 2001). Sample preparation, experimental designs, microarrays, equipment, and data analysis methods are all standardized, giving rise to a truly transparent data set generated across three academic centers. On the technical end, recently defined standards for reporting microarray data (MGED/MIAME) (Ball et al., 2002; Brazma et al., 2001; Pollock 2002; Spellman et al., 2002) and the newly developed microarray data repositories (e.g., Gene Expression Omnibus— GEO [Edgar et al., 2002]) will provide a good framework for the disclosure and sharing of the obtained data. Microarray expression data across diVerent regions will be correlated with available premortem clinical information: Studies thus far have focused on a single brain region, but soon multiple brain regions within subjects will be simultaneously analyzed and compared across diseased and control subjects. This will enable us to define molecular changes that are characteristic for individual brain regions. Based on these data, we will generate falsifiable hypotheses about the molecular underpinnings of the brain diseases and their region-related symptomatology (e.g., cognitive symptoms relative to PFC expression changes or auditory hallucinations relative to temporal cortex transcriptome alterations in schizophrenia). The eVects of psychotropic medications on the transcriptome will be systematically tested: Antipsychotics, antidepressants, anticonvulsants, lithium, and other commonly used medications will be tested for the ability to modify the transcriptome. This will be performed in various acute and chronic models that range from cell culture systems to nonhuman primates. This information will be invaluable to separate the disease eVects from those that may be a result of the medication treatment. Based on the transcriptome data, novel in vivo and in vitro models of the psychiatric disorders will be generated: Over the last 3 years, we and others have identified multiple schizophrenia susceptibility genes. Unfortunately, studies relating the expression findings to the disease mechanism are sparse and rather speculative. For most of the genes, critical basic science information is not available regarding coexpression patterns of gene family members and binding partners, cellular compartmentalization, or modulation potential of physiological responses. Production and systematic analysis of genetically engineered mice, especially


173

conditional knockout systems and hypomorphs, will provide us with novel tools that will help us understand the molecular function of susceptibility genes vis-a`-vis psychiatric disorders. Based on expression data, we will continue to search for and identify susceptibility genes: Others and we have identified more than 50 genes with changed expression that may be associated with schizophrenia, and it is likely that a similar consensus information will be soon available for other brain disorders. Based on the data obtained, we expect that many of the genes showing expression alterations will be located in disease-specific cytogenetic regions. Although some of them will likely represent novel susceptibility genes, most of these genes will have to be evaluated in genetic studies. We will uncover transcript co-regulatory networks: Most expression changes, regardless of their origin (genetic determinants, adaptational responses, or epigenetic influences), are interdependent. With the advancement of analytical tools and the generation of large data sets, we will uncover the transcripts that may represent co-regulated components of a well-defined molecular response. Once we identify these putative interactions, they will be validated in more simplified and better controlled systems, providing us with a view of dynamic changes occurring over time.

Acknowledgments

We would like to thank Dr. Zeljka Korade Mirnics for her thoughtful comments on the manuscript. We are thankful for all members of our laboratories for their involvement and dedication to this project. Supported by R01 MH067234 (KM) and NIMH Conte CNMD (P50 MH45156) Projects 1(DAL), 2 (KM) and 4(PL).

References

Addington, A., Gornick, M., Sporn, A., Gogtay, N., Greenstein, D., Lenane, M., Gochman, P., Weinberger, D., Rapoport, J., and Straub, R. (2003). Polymorphisms in the 2q31.1 gene GAD1, which encodes glutamate decarboxylase 1 (GAD 67), are associated with childhood onset schizophrenia. XI WCPG Proceedings—American Journal of Medical Genetics 122, Abstract P279. Albertson, D., Pruetz, B., Schmidt, C., Kuhn, D., Kapatos, G., and Bannon, M. (2003). Decreased expression of myelin-related genes in human cocaine abusers. In Society for Neuroscience Annual Meeting (http://sfn.scholarone.com/itin.2003/index.html), New Orleans. Ammon, S., Mayer, P., Riechert, U., Tischmeyer, H., and Hollt, V. (2003). Microarray analysis of genes expressed in the frontal cortex of rats chronically treated with morphine and after naloxone precipitated withdrawal. Brain Res. Mol. Brain Res. 112, 113–125.

174

MIRNICS et al.

Ang, E., Chen, J., Zagouras, P., Magna, H., Holland, J., SchaeVer, E., and Nestler, E. J. (2001). Induction of nuclear factor-kappaB in nucleus accumbens by chronic cocaine administration. J. Neurochem. 79, 221–224. Bailey, T. L., and Elkan, C. (1995). The value of prior knowledge in discovering motifs with MEME. Proc. Int. Conf. Intell. Syst. Mol. Biol. 3, 21–29. Ball, C. A., Sherlock, G., Parkinson, H., Rocca-Sera, P., Brooksbank, C., Causton, H. C., Cavalieri, D., Gaasterland, T., Hingamp, P., Holstege, F., Ringwald, M., Spellman, P., Stoeckert, C. J., Jr., Stewart, J. E., Taylor, R., Brazma, A., and Quackenbush, J. (2002). Standards for microarray data. Science 298, 539. Batel, P. (2000). Addiction and schizophrenia. Eur. Psychiatry 15, 115–122. Blalock, E. M., Chen, K. C., Sharrow, K., Herman, J. P., Porter, N. M., Foster, T. C., and Landfield, P. W. (2003). Gene microarrays in hippocampal aging: Statistical profiling identifies novel processes correlated with cognitive impairment. J. Neurosci. 23, 3807–3819. Bonaventure, P., Guo, H., Tian, B., Liu, X., Bittner, A., Roland, B., Salunga, R., Ma, X. J., Kamme, F., Meurers, B., Bakker, M., Jurzak, M., Leysen, J. E., and Erlander, M. G. (2002). Nuclei and subnuclei gene expression profiling in mammalian brain. Brain Res. 943, 38–47. Bonner, R. F., Emmert-Buck, M., Cole, K., Pohida, T., Chuaqui, R., Goldstein, S., and Liotta, L. A. (1997). Laser capture microdissection: Molecular analysis of tissue. Science 278, 1481–1483. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeckert, C., Aach, J., Ansorge, W., Ball, C. A., Causton, H. C., Gaasterland, T., Glenisson, P., Holstege, F. C., Kim, I. F., Markowitz, V., Matese, J. C., Parkinson, H., Robinson, A., Sarkans, U., Schulze-Kremer, S., Stewart, J., Taylor, R., Vilo, J., and Vingron, M. (2001). Minimum information about a microarray experiment (MIAME)—toward standards for microarray data. Nat. Genet. 29, 365–371. Brenner, S., Johnson, M., Bridgham, J., Golda, G., Lloyd, D. H., Johnson, D., Luo, S., McCurdy, S., Foy, M., Ewan, M., Roth, R., George, D., Eletr, S., Albrecht, G., Vermaas, E., Williams, S. R., Moon, K., Burcham, T., Pallas, M., DuBridge, R. B., Kirchner, J., Fearon, K., Mao, J., and Corcoran, K. (2000). Gene expression analysis by massively parallel signature sequencing (MPSS) on microbead arrays. Nat. Biotechnol. 18, 630–634. Bunney, W. E., Bunney, B. G., Vawter, M. P., Tomita, H., Li, J., Evans, S. J., Choudary, P. V., Myers, R. M., Jones, E. G., Watson, S. J., and Akil, H. (2003). Microarray technology: A review of new strategies to discover candidate vulnerability genes in psychiatric disorders. Am. J. Psychiatry 160, 657–666. Caceres, M., Lachuer, J., Zapala, M. A., Redmond, J. C., Kudo, L., Geschwind, D. H., Lockhart, D. J., Preuss, T. M., and Barlow, C. (2003). Elevated gene expression levels distinguish human from non–human primate brains. Proc. Natl. Acad. Sci. USA 100(22), 13030–13035. Carpenter, W. T., and Buchanan, R. W. (1994). Schizophrenia. N. Engl. J. Med. 330, 681–690. Chambers, R. A., Krystal, J. H., and Self, D. W. (2001). A neurobiological basis for substance abuse comorbidity in schizophrenia. Biol. Psychiatry 50, 71–83. Chowdari, K. V., Mirnics, K., Semwal, P., Wood, J., Lawrence, E., Bhatia, T., Deshpande, S. N., Thelma, B. K., Ferrell, R. E., Middleton, F. A., Devlin, B., Levitt, P., Lewis, D. A., and Nimgaonkar, V. L. (2002). Association and linkage analyses of RGS4 polymorphisms in schizophrenia. Hum. Mol. Genet. 11, 1373–1380. Colangelo, V., Schurr, J., Ball, M. J., Pelaez, R. P., Bazan, N. G., and Lukiw, W. J. (2002). Gene expression profiling of 12633 genes in Alzheimer hippocampal CA1: Transcription and neurotrophic factor down-regulation and up-regulation of apoptotic and pro-inflammatory signaling. J. Neurosci. Res. 70, 462–473. Colantuoni, C., Jeon, O. H., Hyder, K., Chenchik, A., Khimani, A. H., Narayanan, V., HoVman, E. P., Kaufmann, W. E., Naidu, S., and Pevsner, J. (2001). Gene expression profiling in postmortem


175

Rett syndrome brain: DiVerential gene expression and patient classification. Neurobiol. Dis. 8, 847–865. D’Agata, V., Warren, S. T., Zhao, W., Torre, E. R., Alkon, D. L., and Cavallaro, S. (2002). Gene expression profiles in a transgenic animal model of fragile X syndrome. Neurobiol. Dis. 10, 211–218. DeRisi, J., Penland, L., Brown, P. O., Bittner, M. L., Meltzer, P. S., Ray, M., Chen, Y., Su, Y. A., and Trent, J. M. (1996). Use of a cDNA microarray to analyze gene expression patterns in human cancer. Nat. Genet. 14, 457–460. Dirks, A., Groenink, L., Westphal, K. G., Olivier, J. D., Verdouw, P. M., van der Gugten, J., Geyer, M. A., and Olivier, B. (2003). Reversal of startle gating deficits in transgenic mice overexpressing corticotropin-releasing factor by antipsychotic drugs. Neuropsychopharmacology 28, 1790–1798. D’Sa, C., Tolbert, L. M., Conti, M., and Duman, R. S. (2002). Regulation of cAMP-specific phosphodiesterases type 4B and 4D (PDE4) splice variants by cAMP signaling in primary cortical neurons. J. Neurochem. 81, 745–757. Eastwood, S. L., McDonald, B., Burnet, P. W., Beckwith, J. P., Kerwin, R. W., and Harrison, P. J. (1995). Decreased expression of mRNAs encoding non-NMDA glutamate receptors GluR1 and GluR2 in medial temporal lobe neurons in schizophrenia. Brain Res. Mol. Brain Res. 29, 211–223. Edgar, R., Domrachev, M., and Lash, A. E. (2002). Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucl. Acids Res. 30, 207–210. Evans, S., Choudary, P., Vawter, M., Tomita, H., Li, J., Meador-WoodruV, J., Lopez, J., Myers, R., Jones, E., Bunney, W., Watson, S., Akil, H. (2002). DNA microarray analysis of dorsal lateral prefrontal cortex from patients with major depressive disorder reveals coordinated alteration of gene expression in a growth factor pathway. In Society for Neuroscience Annual Meeting, Orlando, Fla. (http://sfn.scholarone.com/itin.2003/index.html). Evans, S., Vawter, M., Choudary, P., Li, J., Tomita, H., Atz, M., Lopez, J., Thompson, R., Bolstad, B., Speed, T., Myers, R., Bunney, W., Jones, E., Watson, S., Akil, H. (2003). Comparison of gene expression profiles between bipolar disorder, major depression and schizophrenia in cortical brain regions. In Society for Neuroscience Annual Meeting, New Orleans. (http://sfn.scholarone.com/ itin.2003/index.html). Fell, A. (2001). The disordered mind. UC Davis Magazine 19, online. Francis, D. D., Diorio, J., Plotsky, P. M., and Meaney, M. J. (2002). Environmental enrichment reverses the eVects of maternal separation on stress reactivity. J. Neurosci. 22, 7840–7843. Freeman, W. M., Dougherty, K. E., Vacca, S. E., and Vrana, K. E. (2002a). An interactive database of cocaine-responsive gene expression. Scientific World Journal 2, 701–706. Freeman, W. M., Brebner, K., Lynch, W. J., Robertson, D. J., Roberts, D. C., and Vrana, K. E. (2001). Cocaine-responsive gene expression changes in rat hippocampus. Neuroscience 108, 371–380. Freeman, W. M., Brebner, K., Patel, K. M., Lynch, W. J., Roberts, D. C., and Vrana, K. E. (2002b). Repeated cocaine self-administration causes multiple changes in rat frontal cortex gene expression. Neurochem. Res. 27, 1181–1192. Freeman, W. M., Brebner, K., Lynch, W. J., Patel, K. M., Robertson, D. J., Roberts, D. C., and Vrana, K. E. (2002c). Changes in rat frontal cortex gene expression following chronic cocaine. Brain Res. Mol. Brain Res. 104, 11–20. Geschwind, D. H. (2003). DNA microarrays: Translation of the genome from laboratory to clinic. Lancet Neurol. 2, 275–282. Ginsberg, S. D., and Che, S. (2002). RNA amplification in brain tissues. Neurochem. Res. 27, 981–992. Ginsberg, S. D., Hemby, S. E., Lee, V. M., Eberwine, J. H., and Trojanowski, J. Q. (2000). Expression profile of transcripts in Alzheimer’s disease tangle-bearing CA1 neurons. Ann. Neurol. 48, 77–87.

176

MIRNICS et al.

Ginsberg, S. D., Crino, P. B., Hemby, S. E., Weingarten, J. A., Lee, V. M., Eberwine, J. H., and Trojanowski, J. Q. (1999). Predominance of neuronal mRNAs in individual Alzheimer’s disease senile plaques. Ann. Neurol. 45, 174–181. Green, A. I., Canuso, C. M., Brenner, M. J., and Wojcik, J. D. (2003). Detection and management of comorbidity in patients with schizophrenia. Psychiatr. Clin. North Am. 26, 115–139. Grunblatt, E., Mandel, S., Maor, G., and Youdim, M. B. (2001). Gene expression analysis in N-methyl-4-phenyl-1,2,3,6-tetrahydropyridine mice model of Parkinson’s disease using cDNA microarray: EVect of R-apomorphine. J. Neurochem. 78, 1–12. Gunther, E. C., Stone, D. J., Gerwien, R. W., Bento, P., and Heyes, M. P. (2003). Prediction of clinical drug eYcacy by classification of drug-induced genomic expression profiles in vitro. Proc. Natl. Acad. Sci. USA 100, 9608–9613. Hakak, Y., Walker, J. R., Li, C., Wong, W. H., Davis, K. L., Buxbaum, J. D., Haroutunian, V., and Fienberg, A. A. (2001). Genome-wide expression analysis reveals dysregulation of myelinationrelated genes in chronic schizophrenia. Proc. Natl. Acad. Sci. USA 98, 4746–4751. Halim, N. D., Weickert, C. S., McClintock, B. W., Hyde, T. M., Weinberger, D. R., Kleinman, J. E., and Lipska, B. K. (2003). Presynaptic proteins in the prefrontal cortex of patients with schizophrenia and rats with abnormal prefrontal development. Mol. Psychiatry 8, 797–810. Harrison, P. J., Heath, P. R., Eastwood, S. L., Burnet, P. W., McDonald, B., and Pearson, R. C. (1995). The relative importance of premortem acidosis and postmortem interval for human brain gene expression studies: Selective mRNA vulnerability and comparison with their encoded proteins. Neurosci. Lett. 200, 151–154. Hashimoto, T., Volk, D. W., Eggan, S. M., Mirnics, K., Pierri, J. N., Sun, Z., Sampson, A. R., and Lewis, D. A. (2003). Gene expression deficits in a subclass of GABA neurons in the prefrontal cortex of subjects with schizophrenia. J. Neurosci. 23, 6315–6326. Hata, R., Masumura, M., Akatsu, H., Li, F., Fujita, H., Nagai, Y., Yamamoto, T., Okada, H., Kosaka, K., Sakanaka, M., and Sawada, T. (2001). Up-regulation of calcineurin Abeta mRNA in the Alzheimer’s disease brain: Assessment by cDNA microarray. Biochem. Biophys. Res. Commun. 284, 310–316. Hemby, S. E., Ginsberg, S. D., Brunk, B., Arnold, S. E., Trojanowski, J. Q., and Eberwine, J. H. (2002). Gene expression profile for schizophrenia: Discrete neuron transcription patterns in the entorhinal cortex. Arch. Gen. Psychiatry 59, 631–640. Ho, L., Guo, Y., Spielman, L., Petrescu, O., Haroutunian, V., Purohit, D., Czernik, A., Yemul, S., Aisen, P. S., Mohs, R., and Pasinetti, G. M. (2001). Altered expression of a-type but not b-type synapsin isoform in the brain of patients at high risk for Alzheimer’s disease assessed by DNA microarray technique. Neurosci. Lett. 298, 191–194. Hof, P. R., Haroutunian, V., Copland, C., Davis, K. L., and Buxbaum, J. D. (2002). Molecular and cellular evidence for an oligodendrocyte abnormality in schizophrenia. Neurochem. Res. 27, 1193–1200. Huminiecki, L. B., Lloyd, A. T., and Wolfe, K. (2003). Congruence of tissue expression profiles from Gene Expression Atlas, SAGEmap and TissueInfo databases. BMC Genomics 4, 31. Johnston, M. V., Jeon, O. H., Pevsner, J., Blue, M. E., and Naidu, S. (2001). Neurobiology of Rett syndrome: A genetic disorder of synapse development. Brain Dev. 23(Suppl. 1), S206–S213. Kamme, F., and Erlander, M. G. (2003). Global gene expression analysis of single cells. Curr. Opin. Drug Discov. Dev. 6, 231–236. Karsten, S. L., Kudo, L. C., Jackson, R., Sabatti, C., Kornblum, H. I., and Geschwind, D. H. (2003). Global analysis of gene expression in neural progenitors reveals specific cell-cycle, signaling, and metabolic networks. Dev. Biol. 261, 165–182. Kontkanen, O., Toronen, P., Lakso, M., Wong, G., and Castren, E. (2002). Antipsychotic drug treatment induces diVerential gene expression in the rat cortex. J. Neurochem. 83, 1043–1053.


177

Kornblum, H., and Geschwind, D. (2001). The use of representational diVerence analysis and cDNA microarrays in neural repair research. Restor. Neurol. Neurosci. 18, 89–94. Lewis, D. A. (2002). The human brain revisited: Opportunities and challenges in postmortem studies of psychiatric disorders. Neuropsychopharmacology 26, 143–154. Lewis, D. A., and Lieberman, J. A. (2000). Catching up on schizophrenia: Natural history and neurobiology. Neuron 28, 325–334. Lewohl, J. M., Dodd, P. R., Mayfield, R. D., and Harris, R. A. (2001). Application of DNA microarrays to study human alcoholism. J. Biomed. Sci. 8, 28–36. Lewohl, J. M., Wang, L., Miles, M. F., Zhang, L., Dodd, P. R., and Harris, R. A. (2000a). Gene expression in human alcoholism: Microarray analysis of frontal cortex. Alcohol Clin. Exp. Res. 24, 1873–1882. Lewohl, J. M., Wang, L., Miles, M. F., Zhang, L., Dodd, P. R., and Harris, R. A. (2000b). Gene expression in human alcoholism: Microarray analysis of frontal cortex. Alcohol Clin. Exp. Res. 24, 1873–1882. Lipshutz, R. J., Fodor, S. P., Gingeras, T. R., and Lockhart, D. J. (1999). High density synthetic oligonucleotide arrays. Nat. Genet. 21, 20–24. Lipska, B. K., Lerman, D. N., Khaing, Z. Z., Weickert, C. S., and Weinberger, D. R. (2003). Gene expression in dopamine and GABA systems in an animal model of schizophrenia: EVects of antipsychotic drugs. Eur. J. Neurosci. 18, 391–402. Lo, W., Lau, C., Xuan, Z., Chan, C., Feng, G., He, L., Cao, Z., Liu, H., Luan, Q., and Xue, H. (2003). Association of SNPs and haplotypes in GABAA receptor beta-2 gene with schizophrenia. XI WCPG Proc. Am. J. Med. Genet. 122, abstract P112. Lock, C., Hermans, G., Pedotti, R., Brendolan, A., Schadt, E., Garren, H., Langer-Gould, A., Strober, S., Cannella, B., Allard, J., Klonowski, P., Austin, A., Lad, N., Kaminski, N., Galli, S. J., Oksenberg, J. R., Raine, C. S., Heller, R., and Steinman, L. (2002). Genemicroarray analysis of multiple sclerosis lesions yields new targets validated in autoimmune encephalomyelitis. Nat. Med. 8, 500–508. Lockhart, D. J., and Barlow, C. (2001a). Neural gene expression analysis using DNA arrays. In ‘‘Methods in genomic neuroscience’’ (H. Chin and S. Moldin, Eds.). CRC Press LLC, Boca Raton, Fla. Lockhart, D. J., and Barlow, C. (2001b). Expressing what’s on your mind: DNA arrays and the brain. Nat. Rev. Neurosci. 2, 63–68. Lockhart, D. J., Dong, H., Byrne, M. C., Follettie, M. T., Gallo, M. V., Chee, M. S., Mittmann, M., Wang, C., Kobayashi, M., Horton, H., and Brown, E. L. (1996). Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat. Biotechnol. 14, 1675–1680. Loring, J. F., Wen, X., Lee, J. M., Seilhamer, J., and Somogyi, R. (2001). A gene expression profile of Alzheimer’s disease. DNA Cell Biol. 20, 683–695. Luo, L., Salunga, R. C., Guo, H., Bittner, A., Joy, K. C., Galindo, J. E., Xiao, H., Rogers, K. E., Wan, J. S., Jackson, M. R., and Erlander, M. G. (1999). Gene expression profiles of lasercaptured adjacent neuronal subtypes. Nat. Med. 5, 117–122. Luo, Z., and Geschwind, D. H. (2001). Microarray applications in neuroscience. Neurobiol. Dis. 8, 183–193. Marcotte, E. R., Srivastava, L. K., and Quirion, R. (2001). DNA microarrays in neuropsychopharmacology. Trends Pharmacol. Sci. 22, 426–436. Mayfield, R. D., Lewohl, J. M., Dodd, P. R., Herlihy, A., Liu, J., and Harris, R. A. (2002). Patterns of gene expression are altered in the frontal and motor cortices of human alcoholics. J. Neurochem. 81, 802–813. McDonough, J., Dutta, R., Gudz, T., Perin, M., Macklin, W., Mirnics, K., and Trapp, B. (2003). Decreases in GABA and mitochondrial genes are implicated in MS cortical pathology through

178

MIRNICS et al.

microarray analysis of postmortem MS cortex. In Society for Neuroscience Annual Meeting, New Orleans. (http://sfn.scholarone.com/itin.2003/index.html). Middleton, F. A., Mirnics, K., Pierri, J. N., Lewis, D. A., and Levitt, P. (2002). Gene expression profiling reveals alterations of specific metabolic pathways in schizophrenia. J. Neurosci. 22, 2718–2729. Mimmack, M. L., Ryan, M., Baba, H., Navarro-Ruiz, J., Iritani, S., Faull, R. L., McKenna, P. J., Jones, P. B., Arai, H., Starkey, M., Emson, P. C., and Bahn, S. (2002). Gene expression analysis in schizophrenia: Reproducible up-regulation of several members of the apolipoprotein L family located in a high-susceptibility locus for schizophrenia on chromosome 22. Proc. Natl. Acad. Sci. USA 99, 4680–4685. Mirnics, K. (2001a). Microarrays in brain research: The good, the bad and the ugly. Nat. Rev. Neurosci. 2, 444–447. Mirnics, K., (2001b). Gene expression analysis of the brain: It is all about design! In ‘‘DNA microarray syllabus—SFN 2001 Short Course’’ (D. H. Geschwind, Ed.). Society For Neuroscience Annual Meeting, San Diego. Mirnics, K. (2002). Microarrays in brain research: Data quality and limitations. Curr. Genomics 3, 122–136. Mirnics, K., and Lewis, D. A. (2001). Genes and subtypes of schizophrenia. Trends Mol. Med. 7, 281–283. Mirnics, K., Lewis, D. A., and Levitt, P. (2001a). DNA microarrays and human brain disorders. In ‘‘Methods in genomic neuroscience’’ (H. Chin and S. O. Moldin, Eds.). CRC Press, Boca Raton, Fla. Mirnics, K., Middleton, F. A., Lewis, D. A., and Levitt, P. (2001b). Analysis of complex brain disorders with gene expression microarrays: Schizophrenia as a disease of the synapse. Trends Neurosci. 24, 479–486. Mirnics, K., Middleton, F. A., Marquez, A., Lewis, D. A., and Levitt, P. (2000). Molecular characterization of schizophrenia viewed by microarray analysis of gene expression in prefrontal cortex. Neuron 28, 53–67. Mirnics, K., Middleton, F. A., Stanwood, G. D., Lewis, D. A., and Levitt, P. (2001c). Disease-specific changes in regulator of G-protein signaling 4 (RGS4) expression in schizophrenia. Mol. Psychiatry 6, 293–301. Morris, D. W., Rodgers, A., McGhee, K. A., Schwaiger, S., Scully, P., Quinn, J., Meagher, D., Waddington, J. L., Gill, M., and Corvin, A. P. (2004). Confirming RGS4 as a susceptibility gene for schizophrenia. Am. J. Med. Genet. 125B(1), 50–53. Mufson, E., Counts, S., Che, S., and Ginsberg, S. (2002a). Expression profiles of cytoskeletal mRNAs in cholinergic nucleus basalis neurons in people with mild cognitive impairment and Alzheimer’s disease. In Society for Neuroscience Annual Meeting Orlando, Fla. Available online. Mufson, E. J., Counts, S. E., and Ginsberg, S. D. (2002b). Gene expression profiles of cholinergic nucleus basalis neurons in Alzheimer’s disease. Neurochem. Res. 27, 1035–1048. Mycko, M. P., Papoian, R., Boschert, U., Raine, C. S., and Selmaj, K. W. (2003). cDNA microarray analysis in multiple sclerosis lesions: Detection of genes associated with disease activity. Brain 126, 1048–1057. Pasinetti, G. M. (2001). Use of cDNA microarray in the search for molecular markers involved in the onset of Alzheimer’s disease dementia. J. Neurosci. Res. 65, 471–476. Petryshen, T., O’Leary, S., Lehar, J., Mootha, V., Raad, R., Subramanian, A., Tsan, G., Lander, E., and Sklar, P. (2003). Identification of altered gene pathways in prefrontal cortex and cerebellum of schizophrenia, bipolar disorder, and depression patients. In Society for Neuroscience Annual Meeting, New Orleans. (http://sfn.scholarone.com/itin.2003/index.html). Pierri, J. N., Chaudry, A. S., Woo, T. U., and Lewis, D. A. (1999). Alterations in chandelier neuron axon terminals in the prefrontal cortex of schizophrenic subjects. Am. J. Psychiatry 156, 1709–1719.


179

Pierri, J. N., Volk, C. L., Auh, S., Sampson, A., and Lewis, D. A. (2001). Decreased somal size of deep layer 3 pyramidal neurons in the prefrontal cortex of subjects with schizophrenia. Arch. Gen. Psychiatry 58, 466–473. Pollock, J. D. (2002). Gene expression profiling: Methodological challenges, results, and prospects for addiction research. Chem. Phys. Lipids 121, 241–256. Pongrac, J., Middleton, F. A., Lewis, D. A., Levitt, P., and Mirnics, K. (2002). Gene expression profiling with DNA microarrays: Advancing our understanding of psychiatric disorders. Neurochem. Res. 27, 1049–1063. Pulver, A. E. (2000). Search for schizophrenia susceptibility genes. Biol. Psychiatry 47, 221–230. Purcell, A. E., Jeon, O. H., Zimmerman, A. W., Blue, M. E., and Pevsner, J. (2001). Postmortem brain abnormalities of the glutamate neurotransmitter system in autism. Neurology 57, 1618–1628. Rajkowska, G., Selemon, L. D., and Goldman-Rakic, P. S. (1998). Neuronal and glial somal size in the prefrontal cortex: A postmortem morphometric study of schizophrenia and Huntington disease. Arch. Gen. Psychiatry 55, 215–224. Rajkowska, G., Halaris, A., and Selemon, L. D. (2001). Reductions in neuronal and glial density characterize the dorsolateral prefrontal cortex in bipolar disorder. Biol. Psychiatry 49, 741–752. Rajkowska, G., Miguel-Hidalgo, J. J., Makkos, Z., Meltzer, H., Overholser, J., and Stockmeier, C. (2002). Layer-specific reductions in GFAP-reactive astroglia in the dorsolateral prefrontal cortex in schizophrenia. Schizophr. Res. 57, 127–138. Rampon, C., Jiang, C. H., Dong, H., Tang, Y. P., Lockhart, D. J., Schultz, P. G., Tsien, J. Z., and Hu, Y. (2000). EVects of environmental enrichment on gene expression in the brain. Proc. Natl. Acad. Sci. USA 97, 12880–12884. Sandberg, R., Yasuda, R., Pankratz, D. G., Carter, T. A., Del Rio, J. A., Wodicka, L., Mayford, M., Lockhart, D. J., and Barlow, C. (2000). Regional and strain-specific gene expression mapping in the adult mouse brain. Proc. Natl. Acad. Sci. USA 97, 11038–11043. Schadt, E. E., Li, C., Ellis, B., and Wong, W. H. (2001). Feature extraction and normalization algorithms for high-density oligonucleotide gene expression array data. J. Cell Biochem. Suppl. 37, 120–125. Schena, M., Shalon, D., Davis, R. W., and Brown, P. O. (1995). Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270, 467–470. Shilling, P. D., and Kelsoe, J. R. (2002). Functional genomics approaches to understanding brain disorders. Pharmacogenomics 3, 31–45. Sibille, E., Arango, V., Galfalvy, H., Pavlidis, P., and Mann, J. (2002). Molecular subtypes in depression and suicide. In Society for Neuroscience Annual Meeting, Orlando, Fla. Available online. Spellman, P. T., Miller, M., Stewart, J., Troup, C., Sarkans, U., Chervitz, S., Bernhart, D., Sherlock, G., Ball, C., Lepage, M., Swiatek, M., Marks, W. L., Goncalves, J., Markel, S., Iordan, D., Shojatalab, M., Pizarro, A., White, J., Hubley, R., Deutsch, E., Senger, M., Aronow, B. J., Robinson, A., Bassett, D., Stoeckert, C. J., Jr., and Brazma, A. (2002). Design and implementation of microarray gene expression markup language (MAGE-ML). Genome Biol. 3, Research 0046. Stanwood, G. D., Washington, R. A., and Levitt, P. (2001). Identification of a sensitive period of prenatal cocaine exposure that alters the development of the anterior cingulate cortex. Cereb. Cortex 11, 430–440. Steinman, L., and Zamvil, S. (2003). Transcriptional analysis of targets in multiple sclerosis. Nat. Rev. Immunol. 3, 483–492. Straub, R., Egan, M., Goldberg, T., Callicott, J., Hariri, A., Vakkalanka, R., Balkissoon, R., and Weinberger, D. (2003). GAD1, which encodes glutamate decarboxylase 1 (GAD 67), is associated with adult onset schizophrenia in two independent samples. XI WCPG Proc. Am. J. Med. Genet. 122, Abstract P361.

180

MIRNICS et al.

SutcliVe, J. G., Foye, P. E., Erlander, M. G., Hilbush, B. S., Bodzin, L. J., Durham, J. T., and Hasel, K. W. (2000). TOGA: An automated parsing technology for analyzing expression of nearly all genes. Proc. Natl. Acad. Sci. USA 97, 1976–1981. TabakoV, B., Bhave, S. V., and HoVman, P. L. (2003). Selective breeding, quantitative trait locus analysis, and gene arrays identify candidate genes for complex drug-related behaviors. J. Neurosci. 23, 4491–4498. Tang, W. X., Fasulo, W. H., Mash, D. C., and Hemby, S. E. (2003). Molecular profiling of midbrain dopamine regions in cocaine overdose victims. J. Neurochem. 85, 911–924. Thibault, C., Wang, L., Zhang, L., and Miles, M. F. (2001). DNA arrays and functional genomics in neurobiology. Int. Rev. Neurobiol. 48, 219–253. Thibault, C., Lai, C., Wilke, N., Duong, B., Olive, M. F., Rahman, S., Dong, H., Hodge, C. W., Lockhart, D. J., and Miles, M. F. (2000). Expression profiling of neural cells reveals specific patterns of ethanol-responsive gene expression. Mol. Pharmacol. 58, 1593–1600. Tkachev, D., Mimmack, M. L., Ryan, M. M., Wayland, M., Freeman, T., Jones, P. B., Starkey, M., Webster, M. J., Yolken, R. H., and Bahn, S. (2003). Oligodendrocyte dysfunction in schizophrenia and bipolar disorder. Lancet 362, 798–805. Tomita, H., Vawter, M., Evans, S., Choudary, P., Li, J., Bolstad, B., Speed, T., Myers, R., Jones, E., Watson, S., Akil, H., and Bunney, W. (2003). Gene expression profiles in postmortem brains of mood disorder patients. In Society for Neuroscience Annual Meeting, New Orleans. (http:// sfn.scholarone.com/itin.2003/index.html). Toyooka, K., Usui, M., Washiyama, K., Kumanishi, T., and Takahashi, Y. (2002). Gene expression profiles in the brain from phencyclidine-treated mouse by using DNA microarray. Ann. NY Acad. Sci. 965, 10–20. Tsuang, M. (2000). Schizophrenia: Genes and environment. Biol. Psychiatry 47, 210–220. Tudor, M., Akbarian, S., Chen, R. Z., and Jaenisch, R. (2002). Transcriptional profiling of a mouse model for Rett syndrome reveals subtle transcriptional changes in the brain. Proc. Natl. Acad. Sci. USA 99, 15536–15541. Turunen, J., Paunio, T., Ekelund, J., Suhonen, J., Varilo, T., Partonen, T., Jokiaho, A., Hennah, W., Parker, A., Meyer, J., Lo¨nnqvist, J., and Peltonen, L. (2003). Association of GABRG2 gene variants with susceptibility to schizophrenia. XI WCPG Proc. Am. J. Med. Genet. 122, Abstract O4. Uhl, G. R., Liu, Q. R., Walther, D., Hess, J., and Naiman, D. (2001). Polysubstance abusevulnerability genes: Genome scans for association, using 1,004 subjects and 1,494 singlenucleotide polymorphisms. Am. J. Hum. Genet. 69, 1290–1300. Van den Buuse, M., Garner, B., and Koch, M. (2003). Neurodevelopmental animal models of schizophrenia: EVects on prepulse inhibition. Curr. Mol. Med. 3, 459–471. Vawter, M. P., Crook, J. M., Hyde, T. M., Kleinman, J. E., Weinberger, D. R., Becker, K. G., and Freed, W. J. (2002). Microarray analysis of gene expression in the prefrontal cortex in schizophrenia: A preliminary study. Schizophr. Res. 58, 11–20. Vawter, M. P., Barrett, T., Cheadle, C., Sokolov, B. P., Wood, W. H., 3rd, Donovan, D. M., Webster, M., Freed, W. J., and Becker, K. G. (2001). Application of cDNA microarrays to examine gene expression diVerences in schizophrenia. Brain Res. Bull. 55, 641–650. Velculescu, V. E., Zhang, L., Vogelstein, B., and Kinzler, K. W. (1995). Serial analysis of gene expression. Science 270, 484–487. Volk, D. W., Austin, M. C., Pierri, J. N., Sampson, A. R., and Lewis, D. A. (2000). Decreased glutamic acid decarboxylase67 messenger RNA expression in a subset of prefrontal cortical gamma-aminobutyric acid neurons in subjects with schizophrenia. Arch. Gen. Psychiatry 57, 237–245. Wang, H., Zhu, Y. Z., Wong, P. T., Farook, J. M., Teo, A. L., Lee, L. K., and Moochhala, S. (2003). cDNA microarray analysis of gene expression in anxious PVG and SD rats after cat-freezing test. Exp. Brain Res. 149, 413–421.


181

Whitney, L. W., Becker, K. G., Tresser, N. J., Caballero-Ramos, C. I., Munson, P. J., Prabhu, V. V., Trent, J. M., McFarland, H. F., and Biddison, W. E. (1999). Analysis of gene expression in mutiple sclerosis lesions using cDNA microarrays. Ann. Neurol. 46, 425–428. Williams, N. M., Preece, A., Spurlock, G., Norton, N., Williams, W. J., McCreadie, R. G., Buckland, P., Sharkey, V., Chowdari, K. V., Zammit, S., Nimgaonkar, V. L., Kirov, G., Owen, M. J., and O’Donovan, M. C. (2004). Support for RGS4 as a susceptibility gene for schizophrenia. Biol. Psychiatry 55(2), 192–195. WittliV, J. L., and Erlander, M. G. (2002). Laser capture microdissection and its applications in genomics and proteomics. Methods Enzymol. 356, 12–25. Yamada, M., Takahashi, K., Tsunoda, M., Nishioka, G., Kudo, K., Ohata, H., Kamijima, K., Higuchi, T., and Momose, K. (2002). DiVerential expression of VAMP2/synaptobrevin-2 after antidepressant and electroconvulsive treatment in rat frontal cortex. Pharmacogenomics J. 2, 377–382. Yamada, M., Yamazaki, S., Takahashi, K., Nara, K., Ozawa, H., Yamada, S., Kiuchi, Y., Oguchi, K., Kamijima, K., Higuchi, T., and Momose, K. (2001). Induction of cysteine string protein after chronic antidepressant treatment in rat frontal cortex. Neurosci. Lett. 301, 183–186. Yuferov, V., Kroslak, T., Laforge, K. S., Zhou, Y., Ho, A., and Kreek, M. J. (2003). DiVerential gene expression in the rat caudate putamen after ‘‘binge’’ cocaine administration: Advantage of triplicate microarray analysis Synapse 48, 157–169. Zhou, R., Damschroder-Williams, P., Yuan, P., Chen, G., Du, J., and Manji, H. (2003). Microarray studies reveal a novel target for the long-term treatment of bipolar disorder: The anti-apoptotic, GR chaperone protein, Bag-1. In Society for Neuroscience Annual Meeting, New Orleans. (http:// sfn.scholarone.com/itin.2003/index.html).


INDEX

A Adenosine receptors, in learning/memory, 115 Adrenergic receptors in cell signaling, 122 Affymetrix, 79, 81, 155 algorithms from, 55, 82 arrays from, 26–29, 42–45, 62, 113, 168 Aging, 117 Alzheimer’s disease, 2, 109–110, 117, 124, 147, 161–162 Amino-allyl labeling, 11, 12 Angelman syndrome, 124 Animal models conditioned animals for, 119–123 conditioned stimulus to animals for, 119–122 FMR1 knockout mice, in learning/memory with, 123–125 transgenics in, 157 unconditioned stimulus to animals for, 119–122 Apolipoprotein L (apoL), in schizophrenia, 168 Apoptosis, 110, 118, 123 App gene, in gene expression, 86 Arraying robot microarray platforms and, 8 Association networks, in gene expression, 87 Associative memory storage ryanodine receptor type 2, role in, 98, 109, 116 Autism, 163–164, 171 B Bach 2 gene, 105–108 Background correction microarray data analysis and, 45–47 RMA and, 46–47 Bayesian network analysis, for gene expression, 85, 139 modified t test, 146

Bonferroni analysis, in gene expression, genetics, 139 Box plot, 39, 41–42 BXD strain set, 85–86, 88 C CA. See Conditioned animals Calmodulin-dependent protein kinases, in learning/memory, 116 Cancer, 2 brain tumors and, 144–146 delta-Catenin in synaptogenesis, 117 CD156, transmembrane glycoprotein, 108 cDNA. See Complementary DNA cDNA arrays, 1–2, 9, 145 cohybridization of samples for, 29–30 constraints of, 29–30, 98 Celera Genomics Discovery System, 81 Cell signaling adrenergic receptors in, 122 dopamine receptors in, 122 G-protein signaling 4 receptor in, 169–171 Homer proteins in, 116–117 learning related genes and, 110 in learning/memory, 110, 114–117, 122–123 ligand-gated ion channels in, 115, 122 microarrays, animal models of learning/ memory and, 110, 114–117, 122–123 serotonin receptors in, 122 Shank proteins in, 117 transforming growth factor in, 122 Cerebellum, 105, 168 lobule HVI, 104, 109 Citron, rho target molecule in synaptogenesis, 117 Cliques, in gene expression, 87 Clusters, in gene expression, 87, 89, 141–143 agglomerative, 102 analysis of, 119–121, 125

183

184

INDEX

Clusters, in gene expression (cont.) clustering algorithms of, 101 divisive, 102 hierarchical, 100, 102, 141–143, 166–167 k-Means, 100, 102–103 numerical, 103 semantic, 100, 102–103 similarity coefficient of, 101 visual representation, 100, 102 Complementary DNA (cDNA), 1, 6–10 Complex trait analysis, in gene expression, 62–63, 66 Conditioned animals (CA), 119–123 Conditioned stimulus to animals (CSTA), 119–122 CSTA. See Conditioned stimulus to animals Cyanine dye labeling of RNA, 12 Cyclic AMP, 163 Cytokines, in brain interleukins, 114, 162 D Data analysis Bayesian network analysis, for gene expression and, 85, 139 cluster analysis and, 100–104, 119–121, 125, 141–143 gene expression vectors in, 101 for microarray platforms, 12–13, 32–45, 99–101 multiple power analysis in, 36–45 normalization of microarray data and, 38–45, 82, 99–101, 138 power calculations with, 34–36 scatter plot for, 107, 125 t test, in microarrays, 32, 35, 37, 138–139 time-series analysis, for gene expression, 85 Databases Celera Genomics Discovery System and, 81 EZRetrieve, 167 for nucleotide probes/microarray platforms, 10 Silico Informatics Systems, 167 Digoxygenin (DIG) for probe labeling, 15 DLPFC. See Dorso-lateral prefrontal cortex Dopamine receptors in cell signaling, 122 in learning/memory, 114

Dorso-lateral prefrontal cortex (DLPFC), in brain, 164, 165 Down syndrome, 124 E Epilepsy, 2, 117, 124 Expression Analysis Systematic Explorer (EASE), 14 F F-box protein, 107 FGF-18. See Fibroblast growth factor-18 Fibroblast growth factor-18 (FGF-18), 110, 113–114, 118–119 Fluorophores, in labeling, 3–5, 98 Cy3, 30, 100 Cy3-dCTP, 3–4 Cy5, 30, 100 Cy5-dCTP, 3–4 FMR1 knockout mice, in learning/memory, 123–125 Fragile X syndrome, 117, 123–124 Free radical metabolism, 118 Frequenin, in learning/memory, 116 G GABA receptors in learning/memory, 114–116, 122, 170–171 in MS, 162 in schizophrenia, 165 GCSF. See Granulocyte colony-stimulating factor Genbank, 111–112 Gene expression, 119 App gene in, 87 association networks in, 87 cliques in, 87 clusters in, 87, 89 drugs, effect on, 157–158 exogenous factors in, 61 genetic variation and, 61 Hoxd8 gene and, 87, 90 memory suppressor genes in, 105 microarray data summaries of, 45–48 microarray platforms in, 1–3 regulation factors in, 105

INDEX

strain differences in, 65 Viaat gene in, 86–88 Gene expression, genetics, 59–60 aging differences in, 65 Bayesian network analysis and, 85, 139 Bonferroni analysis and, 139 cliques, clusters, association networks in, 87, 89 complex traits in, 62, 63, 66 functional correlates in, 85–90 gene-environment interactions in, 75–76 gene-gene interactions in, 74–75 high-throughput mRNA assays for, 60 mapping studies in, 65, 70–71, 76 models for, 90–91 modification of, 61 normalization of data in, 82 polymorphism detection in, 62, 76–77 probe level effects/variation in, 79–81, 83 QTL in, 66–79, 81, 84–85, 87 recombinant inbred lines in, 71 redundancy/overlap of results in, 81, 83 regulatory networks in, 60, 83, 85 sample size limitations in, 63–64 sex differences in, 65 signal transduction cascade in, 60 steady-state abundance in, 62 strain differences in, 63, 65 time-series analysis in, 85 trait correlates and, 87–88, 90 transcript abundance and, 66–69, 74 transcription factors in, 60, 108 transcription regulation in, 60–62, 72–74, 105–106, 108, 110, 123 variation in, 61–62, 66–67, 69, 76 Viaat gene in, 86–88 Westfall-Young analysis and, 139 Gene Ontology Consortium gene classification from, 14 Gene upregulation/downregulation models of learning/memory and, 99, 104–109, 117, 161 Genomic DNA, 2 Glutamate ionotropic receptors, in learning/memory, 115 G-protein signaling 4 receptor (RGS4) in schizophrenia, 169–171 Granulocyte colony-stimulating factor (GCSF), 144

185

H Hippocampus, 97, 104–105, 107, 121, 147 Hippostasin, 107 Homer proteins in cell signaling, 116–117 Hoxd8 gene, in gene expression, 87, 90 Hybridization, 2 comparative, 3 in situ, 14–15, 105, 159, 165, 170 subtractive, 6 I IGF-I. See Insulin-like growth factor-I IHS. See Hybridization, in situ Inositol triphosphate receptors, in learning/memory, 116 Insulin receptors, in learning/memory, 115 Insulin-like growth factor-I (IGF-I), 105–107 Interferon, 138, 147, 162 therapy with, 144–146 Interleukins, 114, 162 Iteratively re-weighted least squares (IRLS), 48, 52 K Kainate ionotropic receptors, in learning/memory, 115 Kolmogorov-Smirnov test, 56 L Laser capture microdissection (LCM) in microarray analysis, 138, 158 Lateral substantia nigra (ISN), in brain, 163–164 Learning related genes (LRG) in apoptosis, 110, 118 in cell signaling, 110 in cell-cell interaction/cytoskeletal protein regulation, 110 in enzyme regulation, 110 in synapse protein regulation, 110 in translation/transcription regulation, 110, 123 Leukocyte common antigen-related (LAR), 107 Ligand-gated ion channels, in learning/memory, 115, 122 Liprin-beta, 2, 107–108

186

INDEX

Lobule HVI, of cerebellum, 104, 107 Long term memory (LTM) animal models of, 104 dependence on protein synthesis in, 97 inhibition by protein inhibitors and, 105 transcription factors in, 105, 108 LRG. See Learning related genes LTM. See Long term memory Lupus erythematosis, 138, 144 M MA plot, 38–40, 43–44 MAP-2. See Microtubule-associated protein-2 Memory related genes (MRG), 109–110, 113, 119, 121 Microarray analysis, in neurological disease, 137–138 brain tumors and, 144–146 disease classification/prediction with, 138–143 experimental design/data analysis and, 138 interferon activity and, 138, 145–147, 162 laser capture microdissection and, 138, 158 methodology limitations, 146 neuromuscular disorders and, 147 peripheral nerve diseases and, 138 predictor/model approach, 143 RNA isolation for, 138 Microarray analysis, in psychiatry, 153–154 animal models v. human tissue in, 160–161 common findings/deviations in, 169–170 complementary hybridization testing in, 154 data driven approach, importance in, 153–154 disorder models and, 172–173 drugs, effect on gene expression in, 157–158 expanded studies in, 172 molecular complexity of brain tissue and, 158–159 neurological disorders and, 161–162 postmortem tissue, importance in, 154–155 psychiatric disorders and, 163–171 sample size/diversity in, 159 schizophrenia in, 164–171 SNP-driven expression in, 155–156 standardized data collection/reporting in, 172 substance abuse/addiction and, 163, 171 technical considerations of, 160 therapeutic testing and, 172

transcript analysis testing in, 154 transcript regulatory networks and, 173 transcriptome modulation and, 156–157 upregulation/downregulation of genes and, 165–166 Microarray experimental design/data analysis, 25–26 Affymetrix algorithms in, 55, 82 Affymetrix arrays in, 26–29, 42–45, 62, 113, 168 background correction in, 45–47 Box plot, 39, 41–42 cDNA arrays in, 29–32, 38–42, 54, 145 cohybridization of samples for, 29–30, 100 direct v. indirect comparison in, 30–31 local control in, 28 MA plot, 38–40, 43–44 normalization of data in, 38–45 NUSE in, 50–51 power calculations in, 32–35 quality assessment of data in, 47–54 randomization, replication in, 26–28, 31–32 relative expression in, 52–54 residual variance in, 48–52 RMA and, 46–47, 57 sample pooling, biological averaging in, 29, 40–42 sample size in, 32–35 statistical limitations in, 34–35 statistical solutions for, 35–45 variability, summarization, data quality in, 26–28, 31–32, 37, 38–45, 47–54 Microarray platforms, 1–3 acceptance of, 1–2 arraying robot with, 8 cDNA arrays and, 1–2, 9 commercial sources of, 15 commercial v. in-house slides for, 7–10 data analysis for, 12–14 data verification for, 14–15 databases, for nucleotide probes and, 10 design considerations of, 5–6, 11 detection limits of, 2, 8, 10 disease detection with, 1–2, 11 experimental flow with, 3–6 experimental time frame of, 5 gene expression/high-throughput monitoring and, 1, 3 labeling techniques with, 11–12

INDEX

methylation-specific oligonucleotide arrays for, 3 multiple splice variants with, 10 oligonucleotide arrays and, 1–2, 8 online tools for, 14, 16–17 polyadenylation variants with, 10 quantitative assays with, 3 sample preparation for, 11 sample source considerations with, 11 signal v. sample amplification with, 11–12 software for, 13 variation sources from, 12–13 Microarray sample pool (MSP), 40–41 Microarrays, animal models of learning/ memory, 97–98 apoptosis and, 118, 123 array intensity v. expression in, 99 associative memory storage and, 98, 109 cell signaling in, 110, 114–117, 122–123 cell-cell interactions/cytoskeletal proteins, 110, 117–118 cerebellum/lobule HVI and, 104, 107 data analysis of, 99–101, 106, 110–113, 120–121, 125 Dopamine receptors and, 114 enzymatic regulation in, 108, 118 fluorescent-tagged cDNA and, 98 FMR1 knockout mice, as models for, 123–125 GABA receptors and, 114–116, 122, 170–171 gene upregulation/downregulation and, 99, 104–109, 117 interleukens and, 114 long term memory and, 97, 104–105 memory-related genes, 109 normalization of data in, 99 pathology and, 123–126 physical activity-related genes, 109 physiology of learning/memory and, 104–123 protein modification in, 105–106, 108–109 rabbit nictitating response, as model for, 104–109 rat passive avoidance, as model for, 119–123 rat water maze, as model for, 109–119 RNA fingerprinting and, 97 signal transduction in, 105, 108, 116 synaptic protein regulation in, 110, 118–119, 123 transcription regulation, 60, 105–106, 108, 110, 123 translation regulation in, 118

187

Microtubule-associated protein-2 (MAP-2) in cell-cell interactions, 117 Mismatch (MM), 82 Mitogen-activated protein kinases, in learning/ memory, 116 MPSS. See Multiple parallel sequencing MRG. See Memory related genes MSP. See Microarray sample pool Multiple parallel sequencing (MPSS), 154 Multiple sclerosis (MS), 138, 144–146, 161, 162 Muscular dystrophy, 147 N Narg1 gene, in gene expression, 86 Natural cell adhesion molecule (NCAM), 107 NCAM. See Natural cell adhesion molecule Neuraxin, in cell-cell interactions, 117 Neurotransmitter transport, in learning/ memory, 116 Nictitating membrane response (NMR), in rabbit, 104–109 Nitric oxide synthase (iNOS), in learning/ memory, 116 NMR. See Nictitating membrane response Normalization of data in microarray data analysis, 38–45, 100–101 array intensity levels in, 99 mismatch in, 82 normalization constants with, 99 perfect match in, 82 positional-dependent nearest neighbor in, 82 quality control in, 99 quartile, 44–45 robust multichip average in, 82 sample-sample fold comparison in, 99 statistical group comparisons and, 99 Northern blot in microarray data confirmation, 4, 6, 14 NUSE. See Unscaled standard errors NUSE, in microarray data analysis, 50–51 O Oncology, 2 P p value, in microarray data analysis, 99, 138–139, 146 PARG. See Physical activity-related genes

188

INDEX

Pathology, of learning/memory, 123–126 PBMC. See Peripheral blood mononuclear cells PCR. See Polymerase chain reaction PDNN. See Positional-dependent nearest neighbor Perfect match (PM), 42–43, 45, 55–56, 82 Peripheral blood mononuclear cells (PBMC), 138, 144–146 Permutation testing analysis, 78–79, 139 Phoecin, 108 Physical activity-related genes (PARG), 109, 113 Pirin, 107 Pleiotropy, 86 Polymerase chain reaction (PCR), 2, 11, 168. See also RT-PCR Polymorphisms in gene expression, genetics, in brain, 62 Positional-dependent nearest neighbor (PDNN), in normalization of data, 82 Power calculations/analyses in microarray experimental design/data analysis, 32–35 Prefrontal cortex, in brain, 164–165, 168 Q Quantitative trait locus (QTL), 66–79, 81, 84–85, 87 analysis, 77–79 R Randomization, replication biological, 31–32 in microarray experimental design, 26–28, 31–32 technical, 31, 38 Receiver operator characteristic (ROC), 56–57 Redundancy/overlap of results, 81, 83 Resonance light scattering (RLS), 12 Rett syndrome, 117, 124, 161 Reverse transcriptase PCR. See RT-PCR RGS4. See G-protein signaling 4 receptor Rheumatoid arthritis (RA), 138, 144 RMA. See Robust multi-array analysis RNA isolation/purification, 3, 5, 11 direct labeling in, 11–12 RNA polymerase, 83 Robust multi-array analysis (RMA), 46–47, 57, 82

ROC. See Receiver operator characteristic RT-PCR, 6 quantitative, 14–15, 76, 158, 169 verification of array results with, 14–15 Ryanodine receptor associative memory storage, role of, 98, 109, 116 in learning/memory, 116 S SAGE. See Serial analysis of gene expression Sample pooling, biological averaging in microarray experimental design, 29 Sample-sample fold comparison in normalization of data, 99, 138 Scatter plot, for data analysis, 107, 125 Schizophrenia, 159, 163 Apolipoprotein L and, 168 microarray analysis, in psychiatry and, 164–171 RGS4 in, 169–171 SD. See Standard deviation Serial analysis of gene expression (SAGE), 7, 154 Serotonin receptors in cell signaling, 122 Shank proteins in cell signaling, 117 Signal transduction cascade, in brain, 60 Single nucleotide polymorphism (SNP), 2, 81, 155–156, 170 SNP. See Single nucleotide polymorphism Standard deviation (SD), 32–35 Statistical limitations false discovery rates with, 64–65 false positives in microarray data analysis and, 36, 99 in microarray data analysis, 34–35, 63–64 normalization of, 38–45 p value, in microarray data analysis and, 99 redundancy/overlap of results and, 81, 83 t test, in microarray data analysis and, 32, 35, 37, 138–139 Statistical solutions, in microarray experimental design, 35–45 Striatin, 108 Substance abuse/addiction microarray analysis, in psychiatry and, 163, 171

INDEX

Susceptibility genes, 170, 173 Synapsin II (SYN2), 165 Synaptic protein regulation, in learning/ memory, 110, 118–119, 123 Synaptogenesis delta-Catenin and, 117 Citron, Rho target molecule in, 117 T t test, in microarray data analysis, 32, 35, 37, 138–139 T7-directed amplification, 11–12 TGF. See Transforming growth factor Time-series analysis, for gene expression, 85 Tiramide signal amplification, 11 Tln gene, in gene expression, 86 TOGA. See Total gene expression analysis Total gene expression analysis (TOGA) in microarray assays, 76, 154 in transcript abundance assays, 76, 154 Transcript abundance, in gene expression, genetics, 66–69, 74 Transcription factors, 60, 108 Transcription regulation in gene expression, genetics, 60–62, 72–74, 105–106, 108, 110, 123 in gene expression, in brain, 60, 105–106, 108, 110, 118, 123 regulatory networks in, 83, 85, 173

189

Transcriptome, 7, 60–61, 69, 77, 137, 153, 155–157 Transcriptome-QTL analysis empirical significance tests and, 78 multiple testing analysis and, 78 permutation testing analysis and, 78–79 Transforming growth factor (TGF) in cell signaling, 122 Transgenic animal models, 157 beta-Tubulin, in cell-cell interactions, 117 U Unconditioned stimulus to animals (USTA), 119–122 Unscaled standard errors (NUSE), 50, 51 USTA. See Unconditioned stimulus to animals V Variability, summarization, data quality experimental design and, 26–28, 31–32, 37–45 Ventral tegmental area (VTA), in brain, 163 Viaat gene, in gene expression, 86–88 W Westfall-Young analysis and in gene expression, genetics, 139 Williams-Beuren syndrome, 108


CONTENTS OF RECENT VOLUMES

Volume 37

Memory and Forgetting: Long-Term and Gradual Changes in Memory Storage Larry R. Squire

Section I: Selectionist Ideas and Neurobiology in

Implicit Knowledge: New Perspectives on Unconscious Processes Daniel L. Schacter

Population Thinking and Neuronal Selection: Metaphors or Concepts? Ernst Mayr

Section V: Psychophysics, Psychoanalysis, and Neuropsychology

Selectionist and Neuroscience Olaf Sporns

Instructionist

Ideas

Selection and the Origin of Information Manfred Eigen

Phantom Limbs, Neglect Syndromes, Repressed Memories, and Freudian Psychology V. S. Ramachandran

Section II: Populations

Neural Darwinism and a Conceptual Crisis in Psychoanalysis Arnold H. Modell

Development

and

Neuronal

Morphoregulatory Molecules and Selectional Dynamics during Development Kathryn L. Crossin

A New Vision of the Mind Oliver Sacks

Exploration and Selection in the Early Acquisition of Skill Esther Thelen and Daniela Corbetta

index

Population Activity in the Control of Movement Apostolos P. Georgopoulos

Volume 38

Section III: Functional Integration in the Brain

Segregation

and

Reentry and the Problem of Cortical Integration Giulio Tononi Coherence as an Organizing Principle of Cortical Functions Wolf Singerl Temporal Mechanisms in Perception Ernst Po¨ppel

Regulation of GABAA Receptor Function and Gene Expression in the Central Nervous System A. Leslie Morrow Genetics and the Organization of the Basal Ganglia Robert Hitzemann, Yeang Olan, Stephen Kanes, Katherine Dains, and Barbara Hitzemann

Section IV: Memory and Models

Structure and Pharmacology of Vertebrate GABAA Receptor Subtypes Paul J. Whiting, Ruth M. McKernan, and Keith A. Wafford

Selection versus Instruction: Use of Computer Models to Compare Brain Theories George N. Reeke, Jr.

Neurotransmitter Transporters: Biology, Function, and Regulation Beth Borowsky and Beth J. Hoffman 191

Molecular

192


Presynaptic Excitability Meyer B. Jackson

Volume 40

Monoamine Neurotransmitters in Invertebrates and Vertebrates: An Examination of the Diverse Enzymatic Pathways Utilized to Synthesize and Inactivate Biogenic Amines B. D. Sloley and A. V. Juorio

Mechanisms of Nerve Cell Death: Apoptosis or Necrosis after Cerebral Ischemia R. M. E. Chalmers-Redman, A. D. Fraser, W. Y. H. Ju, J. Wadia, N. A. Tatton, and W. G. Tatton

Neurotransmitter Systems in Schizophrenia Gavin P. Reynolds

Changes in Ionic Fluxes during Cerebral Ischemia Tibor Kristian and Bo K. Siesjo

Physiology of Bergmann Glial Cells Thomas Mu¨ller and Helmut Kettenmann

Techniques for Examining Neuroprotective Drugs in Vitro A. Richard Green and Alan J. Cross

index Volume 39

Techniques for Examining Neuroprotective Drugs in Vivo Mark P. Goldberg, Uta Strasser, and Laura L. Dugan

Modulation of Amino Acid-Gated Ion Channels by Protein Phosphorylation Stephen J. Moss and Trevor G. Smart

Calcium Antagonists: Their Role in Neuroprotection A. Jacqueline Hunter

Use-Dependent Regulation Receptors Eugene M. Barnes, Jr.

GABAA

Sodium and Potassium Channel Modulators: Their Role in Neuroprotection Tihomir P. Obrenovich

Synaptic Transmission and Modulation in the Neostriatum David M. Lovinger and Elizabeth Tyler

NMDA Antagonists: Their Role in Neuroprotection Danial L. Small

of

The Cytoskeleton and Neurotransmitter Receptors Valerie J. Whatley and R. Adron Harris

Development of the NMDA Ion-Channel Blocker, Aptiganel Hydrochloride, as a Neuroprotective Agent for Acute CNS Injury Robert N. McBurney

Endogenous Opioid Regulation of Hippocampal Function Michele L. Simmons and Charles Chavkin

The Pharmacology of AMPA Antagonists and Their Role in Neuroprotection Rammy Gill and David Lodge

Molecular Neurobiology of the Cannabinoid Receptor Mary E. Abood and Billy R. Martin

GABA and Neuroprotection Patrick D. Lyden

Genetic Models in the Study of Anesthetic Drug Action Victoria J. Simpson and Thomas E. Johnson Neurochemical Bases of Locomotion and Ethanol Stimulant Effects Tamara J. Phillips and Elaine H. Shen Effects of Ethanol on Ion Channels Fulton T. Crews, A. Leslie Morrow, Hugh Criswell, and George Breese index

Adenosine and Neuroprotection Bertil B. Fredholm Interleukins and Cerebral Ischemia Nancy J. Rothwell, Sarah A. Loddick, and Paul Stroemer Nitrone-Based Free Radical Traps as Neuroprotective Agents in Cerebral Ischemia and Other Pathologies Kenneth Hensley, John M. Carney, Charles A. Stewart, Tahera Tabatabaie, Quentin Pye, and Robert A. Floyd


Neurotoxic and Neuroprotective Roles of Nitric Oxide in Cerebral Ischemia Turgay Dalkara and Michael A. Moskowitz

Sensory and Cognitive Functions Lawrence M. Parsons and Peter T. Fox

A Review of Earlier Clinical Studies on Neuroprotective Agents and Current Approaches Nils-Gunnar Wahlgren

Skill Learning Julien Doyon

index

193

Section V: Clinical and Neuropsychological Observations Executive Function and Motor Skill Learning Mark Hallett and Jordon Grafman

Volume 41 Section I: Historical Overview

Verbal Fluency and Agrammatism Marco Molinari, Maria G. Leggio, and Maria C. Silveri

Rediscovery of an Early Concept Jeremy D. Schmahmann

Classical Conditioning Diana S. Woodruff-Pak

Section II: Anatomic Substrates

Early Infantile Autism Margaret L. Bauman, Pauline A. Filipek, and Thomas L. Kemper

The Cerebrocerebellar System Jeremy D. Schmahmann and Deepak N. Pandya Cerebellar Output Channels Frank A. Middleton and Peter L. Strick Cerebellar-Hypothalamic Axis: Basic Circuits and Clinical Observations Duane E. Haines, Espen Dietrichs, Gregory A. Mihailoff, and E. Frank McDonald Section III. Physiological Observations Amelioration of Aggression: Response to Selective Cerebellar Lesions in the Rhesus Monkey Aaron J. Berman Autonomic and Vasomotor Regulation Donald J. Reis and Eugene V. Golanov Associative Learning Richard F. Thompson, Shaowen Bao, Lu Chen, Benjamin D. Cipriano, Jeffrey S. Grethe, Jeansok J. Kim, Judith K. Thompson, Jo Anne Tracy, Martha S. Weninger, and David J. Krupa

Olivopontocerebellar Atrophy and Friedreich’s Ataxia: Neuropsychological Consequences of Bilateral versus Unilateral Cerebellar Lesions The´re`se Botez-Marquard and Mihai I. Botez Posterior Fossa Syndrome Ian F. Pollack Cerebellar Cognitive Affective Syndrome Jeremy D. Schmahmann and Janet C. Sherman Inherited Cerebellar Diseases Claus W. Wallesch and Claudius Bartels Neuropsychological Abnormalities in Cerebellar Syndromes—Fact or Fiction? Irene Daum and Hermann Ackermann Section VI: Theoretical Considerations Cerebellar Microcomplexes Masao Ito

Visuospatial Abilities Robert Lalonde

Control of Sensory Data Acquisition James M. Bower

Spatial Event Processing Marco Molinari, Laura Petrosini, and Liliana G. Grammaldo

Neural Representations of Moving Systems Michael Paulin

Section IV: Functional Neuroimaging Studies Linguistic Processing Julie A. Fiez and Marcus E. Raichle

How Fibers Subserve Computing Capabilities: Similarities between Brains and Machines Henrietta C. Leiner and Alan L. Leiner

194


Cerebellar Timing Systems Richard Ivry

Volume 43

Attention Coordination and Anticipatory Control Natacha A. Akshoomoff, Eric Courchesne, and Jeanne Townsend

Early Development of the Drosophila Neuromuscular Junction: A Model for Studying Neuronal Networks in Development Akira Chiba

Context-Response Linkage W. Thomas Thach

Development of Larval Body Wall Muscles Michael Bate, Matthias Landgraf, and Mar Ruiz Gmez Bate

Duality of Cerebellar Motor and Cognitive Functions James R. Bloedel and Vlastislav Bracha Section VII: Future Directions Therapeutic and Research Implications Jeremy D. Schmahmann

Volume 42 Alzheimer Disease Mark A. Smith Neurobiology of Stroke W. Dalton Dietrich Free Radicals, Calcium, and the Synaptic Plasticity-Cell Death Continuum: Emerging Roles of the Trascription Factor NFB Mark P. Mattson AP-I Transcription Factors: Short- and LongTerm Modulators of Gene Expression in the Brain Keith Pennypacker

Development of Electrical Properties and Synaptic Transmission at the Embryonic Neuromuscular Junction Kendal S. Broadie Ultrastructural Correlates of Neuromuscular Junction Development Mary B. Rheuben, Motojiro Yoshihara, and Yoshiaki Kidokoro Assembly and Maturation of the Drosophila Larval Neuromuscular Junction L. Sian Gramates and Vivian Budnik Second Messenger Systems Underlying Plasticity at the Neuromuscular Junction Frances Hannan and Yi Zhong Mechanisms of Neurotransmitter Release J. Troy Littleton, Leo Pallanck, and Barry Ganetzky Vesicle Recycling at the Drosophila Neuromuscular Junction Daniel T. Stimson and Mani Ramaswami Ionic Currents in Larval Muscles of Drosophila Satpal Singh and Chun-Fang Wu

Ion Channels in Epilepsy Istvan Mody

Development of the Adult Neuromuscular System Joyce J. Fernandes and Haig Keshishian

Posttranslational Regulation of Ionotropic Glutamate Receptors and Synaptic Plasticity Xiaoning Bi, Steve Standley, and Michel Baudry

Controlling the Motor Neuron James R. Trimarchi, Ping Jin, and Rodney K. Murphey

Heritable Mutations in the Glycine, GABAA, and Nicotinic Acetylcholine Receptors Provide New Insights into the Ligand-Gated Ion Channel Receptor Superfamily Behnaz Vafa and Peter R. Schofield

Volume 44

index

Human Ego-Motion Perception A. V. van den Berg Optic Flow and Eye Movements M. Lappe and K.-P. Hoffman


The Role of MST Neurons during Ocular Tracking in 3D Space K. Kawano, U. Inoue, A. Takemura, Y. Kodaka, and F. A. Miles Visual Navigation in Flying Insects M. V. Srinivasan and S.-W. Zhang Neuronal Matched Filters for Optic Flow Processing in Flying Insects H. G. Krapp A Common Frame of Reference for the Analysis of Optic Flow and Vestibular Information B. J. Frost and D. R. W. Wylie Optic Flow and the Visual Guidance of Locomotion in the Cat H. Sherk and G. A. Fowler Stages of Self-Motion Processing in Primate Posterior Parietal Cortex F. Bremmer, J.-R. Duhamel, S. B. Hamed, and W. Graf Optic Flow Perception C. J. Duffy

Analysis

for

Self-Movement

Neural Mechanisms for Self-Motion Perception in Area MST R. A. Andersen, K. V. Shenoy, J. A. Crowell, and D. C. Bradley Computational Mechanisms for Optic Flow Analysis in Primate Cortex M. Lappe Human Cortical Areas Underlying the Perception of Optic Flow: Brain Imaging Studies M. W. Greenlee

195

Brain Development and Generation of Brain Pathologies Gregory L. Holmes and Bridget McCabe Maturation of Channels and Receptors: Consequences for Excitability David F. Owens and Arnold R. Kriegstein Neuronal Activity and the Establishment of Normal and Epileptic Circuits during Brain Development John W. Swann, Karen L. Smith, and Chong L. Lee The Effects of Seizures of the Hippocampus of the Immature Brain Ellen F. Sperber and Solomon L. Moshe Abnormal Development and Catastrophic Epilepsies: The Clinical Picture and Relation to Neuroimaging Harry T. Chugani and Diane C. Chugani Cortical Reorganization and Seizure Generation in Dysplastic Cortex G. Avanzini, R. Preafico, S. Franceschetti, G. Sancini, G. Battaglia, and V. Scaioli Rasmussen’s Syndrome with Particular Reference to Cerebral Plasticity: A Tribute to Frank Morrell Fredrick Andermann and Yuonne Hart Structural Reorganization of Hippocampal Networks Caused by Seizure Activity Daniel H. Lowenstein Epilepsy-Associated Plasticity in gammaAmniobutyric Acid Receptor Expression, Function and Inhibitory Synaptic Properties Douglas A. Coulter

What Neurological Patients Tell Us about the Use of Optic Flow L. M. Vaina and S. K. Rushton

Synaptic Plasticity and Secondary Epileptogenesis Timothy J. Teyler, Steven L. Morgan, Rebecca N. Russell, and Brian L. Woodside

index

Synaptic Plasticity in Epileptogenesis: Cellular Mechanisms Underlying Long-Lasting Synaptic Modifications that Require New Gene Expression Oswald Steward, Christopher S. Wallace, and Paul F. Worley

Volume 45 Mechanisms of Brain Plasticity: From Normal Brain Function to Pathology Philip. A. Schwartzkroin

Cellular Correlates of Behavior Emma R. Wood, Paul A. Dudchenko, and Howard Eichenbaum

196


Mechanisms of Neuronal Conditioning David A. T. King, David J. Krupa, Michael R. Foy, and Richard F. Thompson

Biosynthesis of Neurosteroids and Regulation of Their Synthesis Synthia H. Mellon and Hubert Vaudry

Plasticity in the Aging Central Nervous System C. A. Barnes

Neurosteroid 7-Hydroxylation Products in the Brain Robert Morfin and Luboslav Sta´rka

Secondary Epileptogenesis, Kindling, and Intractable Epilepsy: A Reappraisal from the Perspective of Neuronal Plasticity Thomas P. Sutula Kindling and the Mirror Focus Dan C. McIntyre and Michael O. Poulter Partial Kindling and Behavioral Pathologies Robert E. Adamec The Mirror Epileptogenesis B. J. Wilder

Focus

and

Secondary

Hippocampal Lesions in Epilepsy: A Historical Review Robert Naquet Clinical Evidence for Secondary Epileptogensis Hans O. Luders Epilepsy as a Progressive (or Nonprogressive ‘‘Benign’’) Disorder John A. Wada Pathophysiological Aspects of Landau-Kleffner Syndrome: From the Active Epileptic Phase to Recovery Marie-Noelle Metz-Lutz, Pierre Maquet, Annd De Saint Martin, Gabrielle Rudolf, Norma Wioland, Edouard Hirsch, and Chriatian Marescaux Local Pathways of Seizure Propagation in Neocortex Barry W. Connors, David J. Pinto, and Albert E. Telefeian Multiple Subpial Assessment C. E. Polkey

Transection:

A

Clinical

The Legacy of Frank Morrell Jerome Engel, Jr. Volume 46 Neurosteroids: Beginning of the Story Etienne E. Baulieu, P. Robel, and M. Schumacher

Neurosteroid Analysis Ahmed A. Alomary, Robert L. Fitzgerald, and Robert H. Purdy Role of the Peripheral-Type Benzodiazepine Receptor in Adrenal and Brain Steroidogenesis Rachel C. Brown and Vassilios Papadopoulos Formation and Effects of Neuroactive Steroids in the Central and Peripheral Nervous System Roberto Cosimo Melcangi, Valerio Magnaghi, Mariarita Galbiati, and Luciano Martini Neurosteroid Modulation of Recombinant and Synaptic GABAA Receptors Jeremy J. Lambert, Sarah C. Harney, Delia Belelli, and John A. Peters GABAA -Receptor Plasticity during LongTerm Exposure to and Withdrawal from Progesterone Giovanni Biggio, Paolo Follesa, Enrico Sanna, Robert H. Purdy, and Alessandra Concas Stress and Neuroactive Steroids Maria Luisa Barbaccia, Mariangela Serra, Robert H. Purdy, and Giovanni Biggio Neurosteroids in Learning and Processes Monique Valleé, Willy Mayo, George F. Koob, and Michel Le Moal

Memory

Neurosteroids and Behavior Sharon R. Engel and Kathleen A. Grant Ethanol and Neurosteroid Interactions in the Brain A. Leslie Morrow, Margaret J. VanDoren, Rebekah Fleming, and Shannon Penland Preclinical Development of Neurosteroids as Neuroprotective Agents for the Treatment of Neurodegenerative Diseases Paul A. Lapchak and Dalia M. Araujo


Clinical Implications of Circulating Neurosteroids Andrea R. Genazzani, Patrizia Monteleone, Massimo Stomati, Francesca Bernardi, Luigi Cobellis, Elena Casarosa, Michele Luisi, Stefano Luisi, and Felice Petraglia Neuroactive Steroids and Central Nervous System Disorders Mingde Wang, Torbjo¨rn Ba¨ckstro¨m, Inger Sundstro¨m, Go¨ran Wahlstro¨m, Tommy Olsson, Di Zhu, Inga-Maj Johansson, Inger Bjo¨rn, and Marie Bixo Neuroactive Steroids in Neuropsychopharmacology Rainer Rupprecht and Florian Holsboer Current Perspectives on the Role of Neurosteroids in PMS and Depression Lisa D. Griffin, Susan C. Conrad, and Synthia H. Mellon

197

Processing Human Brain Tissue for in Situ Hybridization with Radiolabelled Oligonucleotides Louise F. B. Nicholson In Situ Hybridization of Astrocytes and Neurons Cultured in Vitro L. A. Arizza-McNaughton, C. De Felipe, and S. P. Hunt In Situ Hybridization on Organotypic Slice Cultures A. Gerfin-Moser and H. Monyer Quantitative Analysis of in Situ Hybridization Histochemistry Andrew L. Gundlach and Ross D. O’Shea Part II: Nonradioactive in Situ hybridization

index

Nonradioactive in Situ Hybridization Using Alkaline Phosphatase-Labelled Oligonucleotides S. J. Augood, E. M. McGowan, B. R. Finsen, B. Heppelmann, and P. C. Emson

Volume 47

Combining Nonradioactive in Situ Hybridization with Immunohistological and Anatomical Techniques Petra Wahle

Introduction: Studying Gene Expression in Neural Tissues by in Situ Hybridization W. Wisden and B. J. Morris

Nonradioactive in Situ Hybridization: Simplified Procedures for Use in Whole Mounts of Mouse and Chick Embryos Linda Ariza-McNaughton and Robb Krumlauf

Part I: In Situ Hybridization with Radiolabelled Oligonucleotides In Situ Hybridization with Oligonucleotide Probes Wl. Wisden and B. J. Morris

index

Cryostat Sectioning of Brains Victoria Revilla and Alison Jones

Volume 48

Processing Rodent Embryonic and Early Postnatal Tissue for in Situ Hybridization with Radiolabelled Oligonucleotides David J. Laurie, Petra C. U. Schrotz, Hannah Monyer, and Ulla Amtmann

Assembly and Intracellular GABAA Receptors Eugene Barnes

Trafficking

of

Processing of Retinal Tissue for in Situ Hybridization Frank Mu¨ller

Subcellular Localization and Regulation of GABAA Receptors and Associated Proteins Bernhard Lu¨scher and Jean-Marc Fritschy D1 Dopamine Receptors Richard Mailman

Processing the Spinal Cord for in Situ Hybridization with Radiolabelled Oligonucleotides A. Berthele and T. R. To¨lle

Molecular Modeling of Ligand-Gated Ion Channels: Progress and Challenges Ed Bertaccini and James R. Trudel

198


Alzheimer’s Disease: Its Diagnosis and Pathogenesis Jillian J. Kril and Glenda M. Halliday DNA Arrays and Functional Genomics in Neurobiology Christelle Thibault, Long Wang, Li Zhang, and Michael F. Miles index

The Treatment of Infantile Spasms: An Evidence-Based Approach Mark Mackay, Shelly Weiss, and O. Carter Snead III ACTH Treatment of Infantile Spasms: Mechanisms of Its Effects in Modulation of Neuronal Excitability K. L. Brunson, S. Avishai-Eliner, and T. Z. Baram

Volume 49

Neurosteroids and Infantile Spasms: The Deoxycorticosterone Hypothesis Michael A. Rogawski and Doodipala S. Reddy

What Is West Syndrome? Olivier Dulac, Christine Soufflet, Catherine Chiron, and Anna Kaminski

Are there Specific Anatomical and/or Transmitter Systems (Cortical or Subcortical) That Should Be Targeted? Phillip C. Jobe

The Relationship between encephalopathy and Abnormal Neuronal Activity in the Developing Brain Frances E. Jensen

Medical versus Surgical Treatment: Which Treatment When W. Donald Shields

Hypotheses from Functional Neuroimaging Studies Csaba Juha´sz, Harry T. Chugani, Ouo Muzik, and Diane C. Chugani

Developmental Outcome with and without Successful Intervention Rochelle Caplan, Prabha Siddarth, Gary Mathern, Harry Vinters, Susan Curtiss, Jennifer Levitt, Robert Asarnow, and W. Donald Shields

Infantile Spasms: Unique Sydrome or General Age-Dependent Manifestation of a Diffuse Encephalopathy? M. A. Koehn and M. Duchowny

Infantile Spasms versus Myoclonus: Is There a Connection? Michael R. Pranzatelli

Histopathology of Brain Tissue from Patients with Infantile Spasms Harry V. Vinters

Tuberous Sclerosis as an Underlying Basis for Infantile Spasm Raymond S. Yeung

Generators of Ictal and Interictal Electroencephalograms Associated with Infantile Spasms: Intracellular Studies of Cortical and Thalamic Neurons M. Steriade and I. Timofeev

Brain Malformation, Epilepsy, and Infantile Spasms M. Elizabeth Ross

Cortical and Subcortical Generators of Normal and Abnormal Rhythmicity David A. McCormick Role of Subcortical Structures in the Pathogenesis of Infantile Spasms: What Are Possible Subcortical Mediators? F. A. Lado and S. L. Moshe´ What Must We Know to Develop Better Therapies? Jean Aicardi

Brain Maturational Aspects Relevant to Pathophysiology of Infantile Spasms G. Auanzini, F. Panzica, and S. Franceschetti Gene Expression Analysis as a Strategy to Understand the Molecular Pathogenesis of Infantile Spasms Peter B. Crino Infantile Spasms: Criteria for an Animal Model Carl E. Stafstrom and Gregory L. Holmes index


Volume 50 Part I: Primary Mechanisms How Does Glucose Generate Oxidative Stress In Peripheral Nerve? Irina G. Obrosova Glycation in Diabetic Neuropathy: Characteristics, Consequences, Causes, and Therapeutic Options Paul J. Thornalley Part II: Secondary Changes Protein Kinase C Changes in Diabetes: Is the Concept Relevant to Neuropathy? Joseph Eichberg Are Mitogen-Activated Protein Kinases Glucose Transducers for Diabetic Neuropathies? Tertia D. Purves and David R. Tomlinson Neurofilaments in Diabetic Neuropathy Paul Fernyhough and Robert E. Schmidt Apoptosis in Diabetic Neuropathy Aviva Tolkovsky Nerve and Ganglion Blood Flow in Diabetes: An Appraisal Douglas W. Zochodne Part III: Manifestations Potential Mechanisms of Neuropathic Pain in Diabetes Nigel A. Calcutt Electrophysiologic Measures of Diabetic Neuropathy: Mechanism and Meaning Joseph C. Arezzo and Elena Zotova Neuropathology and Pathogenesis of Diabetic Autonomic Neuropathy Robert E. Schmidt Role of the Schwann Cell in Diabetic Neuropathy Luke Eckersley

199

Nerve Growth Factor for the Treatment of Diabetic Neuropathy: What Went Wrong, What Went Right, and What Does the Future Hold? Stuart C. Apfel Angiotensin-Converting Enzyme Inhibitors: Are there Credible Mechanisms for Beneficial Effects in Diabetic Neuropathy? Rayaz A. Malik and David R. Tomlinson Clinical Trials for Drugs Against Diabetic Neuropathy: Can We Combine Scientific Needs With Clinical Practicalities? Dan Ziegler and Dieter Luft index

Volume 51 Energy Metabolism in the Brain Leif Hertz and Gerald A. Dienel The Cerebral Glucose-Fatty Acid Cycle: Evolutionary Roots, Regulation, and (Patho) physiological Importance Kurt Heininger Expression, Regulation, and Functional Role of Glucose Transporters (GLUTs) in Brain Donard S. Dwyer, Susan J. Vannucci, and Ian A. Simpson Insulin-Like Growth Factor-1 Promotes Neuronal Glucose Utilization During Brain Development and Repair Processes Carolyn A. Bondy and Clara M. Cheng CNS Sensing and Regulation of Peripheral Glucose Levels Barry E. Levin, Ambrose A. Dunn-Meynell, and Vanessa H. Routh

Part IV: Potential Treatment

Glucose Transporter Protein Syndromes Darryl C. De Vivo, Dong Wang, Juan M. Pascual, and Yuan Yuan Ho

Polyol Pathway and Diabetic Peripheral Neuropathy Peter J. Oates

Glucose, Stress, and Hippocampal Neuronal Vulnerability Lawrence P. Reagan

200


Glucose/Mitochondria in Neurological Conditions John P. Blass Energy Utilization in the Ischemic/Reperfused Brain John W. Phillis and Michael H. O’Regan Diabetes Mellitus and the Central Nervous System Anthony L. McCall Diabetes, the Brain, and Behavior: Is There a Biological Mechanism Underlying the Association between Diabetes and Depression? A. M. Jacobson, J. A. Samson, K. Weinger, and C. M. Ryan Schizophrenia and Diabetes David C. Henderson and Elissa R. Ettinger Psychoactive Drugs Affect Glucose Transport and the Regulation of Glucose Metabolism Donard S. Dwyer, Timothy D. Ardizzone, and Ronald J. Bradley index

Neural Control of Salivary S-IgA Secretion Gordon B. Proctor and Guy H. Carpenter Stress and Secretory Immunity Jos A. Bosch, Christopher Ring, Eco J. C. de Geus, Enno C. I. Veerman, and Arie V. Nieuw Amerongen Cytokines and Depression Angela Clow Immunity and Schizophrenia: Autoimmunity, Cytokines, and Immune Responses Fiona Gaughran Cerebral Lateralization and the Immune System Pierre J. Neveu Behavioral Conditioning of the Immune System Frank Hucklebridge Psychological and Neuroendocrine Correlates of Disease Progression Julie M. Turner-Cobb The Role of Psychological Intervention in Modulating Aspects of Immune Function in Relation to Health and Well-Being J. H. Gruzelier index

Volume 52

Volume 53

Neuroimmune Relationships in Perspective Frank Hucklebridge and Angela Clow Sympathetic Nervous System Interaction with the Immune System Virginia M. Sanders and Adam P. Kohm Mechanisms by Which Cytokines Signal the Brain Adrian J. Dunn Neuropeptides: Modulators of Responses in Health and Disease David S. Jessop

Immune

Brain–Immune Interactions in Sleep Lisa Marshall and Jan Born Neuroendocrinology of Autoimmunity Michael Harbuz Systemic Stress-Induced Th2 Shift and Its Clinical Implications Ibia J. Elenkov

Section I: Mitochondrial Structure and Function Mitochondrial DNA Structure and Function Carlos T. Moraes, Sarika Srivastava, Ilias Kirkinezos, Jose Oca-Cossio, Corina van Waveren, Markus Woischnick, and Francisca Diaz Oxidative Phosphorylation: Structure, Function, and Intermediary Metabolism Simon J. R. Heales, Matthew E. Gegg, and John B. Clark Import of Mitochondrial Proteins Matthias F. Bauer, Sabine Hofmann, and Walter Neupert Section II: Primary Respiratory Chain Disorders Mitochondrial Disorders of the Nervous System: Clinical, Biochemical, and Molecular Genetic Features Dominic Thyagarajan and Edward Byrne


Section III: Secondary Respiratory Chain Disorders Friedreich’s Ataxia J. M. Cooper and J. L. Bradley Wilson Disease C. A. Davie and A. H. V. Schapira

201

The Mitochondrial Theory of Aging: Involvement of Mitochondrial DNA Damage and Repair Nadja C. de Souza-Pinto and Vilhelm A. Bohr index

Hereditary Spastic Paraplegia Christopher J. McDermott and Pamela J. Shaw Cytochrome c Oxidase Deficiency Giacomo P. Comi, Sandra Strazzer, Sara Galbiati, and Nereo Bresolin Section IV: Toxin Induced Mitochondrial Dysfunction Toxin-Induced Mitochondrial Dysfunction Susan E. Browne and M. Flint Beal Section V: Neurodegenerative Disorders Parkinson’s Disease L. V. P. Korlipara and A. H. V. Schapira Huntington’s Disease: The Mystery Unfolds? A˚sa Peterseń and Patrik Brundin Mitochondria in Alzheimer’s Disease Russell H. Swerdlow and Stephen J. Kish Contributions of Mitochondrial Alterations, Resulting from Bad Genes and a Hostile Environment, to the Pathogenesis of Alzheimer’s Disease Mark P. Mattson Mitochondria and Amyotrophic Lateral Sclerosis Richard W. Orrell and Anthony H. V. Schapira

Volume 54

Unique General Anesthetic Binding Sites Within Distinct Conformational States of the Nicotinic Acetylcholine Receptor Hugo R. Ariaas, William, R. Kem, James R. Truddell, and Michael P. Blanton Signaling Molecules and Receptor Transduction Cascades That Regulate NMDA ReceptorMediated Synaptic Transmission Suhas. A. Kotecha and John F. MacDonald Behavioral Measures of Alcohol Self-Administration and Intake Control: Rodent Models Herman H. Samson and Cristine L. Czachowski Dopaminergic Mouse Mutants: Investigating the Roles of the Different Dopamine Receptor Subtypes and the Dopamine Transporter Shirlee Tan, Bettina Hermann, and Emiliana Borrelli Drosophila melanogaster, A Genetic Model System for Alcohol Research Douglas J. Guarnieri and Ulrike Heberlein index

Section VI: Models of Mitochondrial Disease Models of Mitochondrial Disease Danae Liolitsa and Michael G. Hanna

Volume 55

Section VII: Defects of Oxidation Including Carnitine Deficiency

Section I: Virsu Vectors For Use in the Nervous System

Defects of Oxidation Including Carnitine Deficiency K. Bartlett and M. Pourfarzam

Non-Neurotropic Adenovirus: a Vector for Gene Transfer to the Brain and Gene Therapy of Neurological Disorders P. R. Lowenstein, D. Suwelack, J. Hu, X. Yuan, M. Jimenez-Dalmaroni, S. Goverdhama, and M.G. Castro

Section VIII: Mitochondrial Involvement in Aging

202


Adeno-Associated Virus Vectors E. Lehtonen and L. Tenenbaum Problems in the Use of Herpes Simplex Virus as a Vector L. T. Feldman Lentiviral Vectors J. Jakobsson, C. Ericson, N. Rosenquist, and C. Lundberg Retroviral Vectors for Gene Delivery to Neural Precursor Cells K. Kageyama, H. Hirata, and J. Hatakeyama Section II: Gene Therapy with Virus Vectors for Specific Disease of the Nervous System The Principles of Molecular Therapies for Glioblastoma G. Karpati and J. Nalbatonglu

Processing and Representation of SpeciesSpecific Communication Calls in the Auditory System of Bats George D. Pollak, Achim Klug, and Eric E. Bauer Central Nervous System Control of Micturition Gert Holstege and Leonora J. Mouton The Structure and Physiology of the Rat Auditory System: An Overview Manuel Malmierca Neurobiology of Cat and Human Sexual Behavior Gert Holstege and J. R. Georgiadis index

Volume 57

Oncolytic Herpes Simplex Virus J. C. C. Hu and R. S. Coffin

Cumulative Subject Index of Volumes 1-25

Recombinant Retrovirus Vectors for Treatment of Brain Tumors N. G. Rainov and C. M. Kramm

Volume 58

Adeno-Associated Viral Vectors for Parkinson’s Disease I. Muramatsu, L. Wang, K. Ikeguchi, K-i Fujimoto, T. Okada, H. Mizukami, Y. Hanazono, A. Kume, I. Nakano, and K. Ozawa HSV Vectors for Parkinson’s Disease D. S. Latchman Gene Therapy for Stroke K. Abe and W. R. Zhang Gene Therapy for Mucopolysaccharidosis A. Bosch and J. M. Heard index

Volume 56 Behavioral Mechanisms and the Neurobiology of Conditioned Sexual Responding Mark Krause NMDA Receptors in Alcoholism Paula L. Hoffman,

Cumulative Subject Index of Volumes 26–50

Volume 59 Loss of Spines and Neuropil Liesl B. Jones Schizophrenia as a Disorder of Neuroplasticity Robert E. McCullumsmith, Sarah M. Clinton, and James H. Meador-Woodruff The Synaptic Pathology of Schizophrenia: Is Aberrant Neurodevelopment and Plasticity to Blame? Sharon L. Eastwood Neurochemical Basis for an Epigenetic Vision of Synaptic Organization E. Costa, D. R. Grayson, M. Veldic, and A. Guidotti Muscarinic Receptors in Schizophrenia: Is There a Role for Synaptic Plasticity? Thomas J. Raedler


Serotonin and Brain Development Monsheel S. K. Sodhi and Elaine Sanders-Bush Presynaptic Proteins and Schizophrenia William G. Honer and Clint E. Young Mitogen-Activated Protein Kinase Signaling Svetlana V. Kyosseva Postsynaptic Density Scaffolding Proteins at Excitatory Synapse and Disorders of Synaptic Plasticity: Implications for Human Behavior Pathologies Andrea de Bartolomeis and Germano Fiore Prostaglandin-Mediated Signaling in Schizophrenia S. Smesny Mitochondria, Synaptic Plasticity, and Schizophrenia Dorit Ben-Shachar and Daphna Laifenfeld Membrane Phospholipids and Cytokine Interaction in Schizophrenia Jeffrey K. Yao and Daniel P. van Kammen Neurotensin, Schizophrenia, and Antipsychotic Drug Action Becky Kinkead and Charles B. Nemeroff

203

Schizophrenia, Vitamin D, and Brain Development Alan Mackay-Sim, François Fe´ron, Darryl Eyles, Thomas Burne, and John McGrath Possible Contributions of Myelin and Oligodendrocyte Dysfunction to Schizophrenia Daniel G. Stewart and Kenneth L. Davis Brain-Derived Neurotrophic Factor and the Plasticity of the Mesolimbic Dopamine Pathway Oliver Guillin, Nathalie Griffon, Jorge Diaz, Bernard Le Foll, Erwan Bezard, Christian Gross, Chris Lammers, Holger Stark, Patrick Carroll, Jean-Charles Schwartz, and Pierre Sokoloff S100B in Schizophrenic Psychosis Matthias Rothermundt, Gerald Ponath, and Volker Arolt Oct-6 Transcription Factor Maria Ilia NMDA Receptor Function, Neuroplasticity, and the Pathophysiology of Schizophrenia Joseph T. Coyle and Guochuan Tsai index

DNA Arrays in Neurobiology (International Review of Neurobiology, Vol 60)

The Neurobiology of Painting - International Review of Neurobiology Vol 74

INTERNATIONAL REVIEW NEUROBIOLOGY

The Neurobiology of Painting, Volume 74: International Review of Neurobiology

International Review of Neurobiology, Volume 6

International Review of Neurobiology Volume 25

International Review of Neurobiology Volume 38

International Review of Neurobiology Volume 18

International Review of Neurobiology Volume 31

International Review of Neurobiology Volume 14

International Review of Neurobiology, Volume 23

International Review of Neurobiology Volume 1

International Review of Neurobiology Volume 13

International Review of Neurobiology Volume 5

International Review of Neurobiology, Volume 21

International Review of Neurobiology Volume 36

International Review of Neurobiology Volume 16

International Review of Neurobiology Volume 32

International Review of Neurobiology Volume 12

International Review of Neurobiology Volume 3

International Review of Neurobiology, Volume 48

International Review of Neurobiology, Volume 20

International Review of Neurobiology, Volume 73

International Review of Neurobiology Volume 29

International Review of Neurobiology, Volume 8

International Review of Neurobiology Volume 30

International Review of Neurobiology Volume 17

International Review of Neurobiology Volume 4

International Review of Neurobiology, Volume 63

International Review of Neurobiology Volume 2

International Review of Neurobiology Volume 24

DNA Arrays in Neurobiology (International Review of Neurobiology, Vol 60)

The Neurobiology of Painting - International Review of Neurobiology Vol 74

INTERNATIONAL REVIEW NEUROBIOLOGY

The Neurobiology of Painting, Volume 74: International Review of Neurobiology

International Review of Neurobiology, Volume 6

International Review of Neurobiology Volume 25

International Review of Neurobiology Volume 38

International Review of Neurobiology Volume 18

International Review of Neurobiology Volume 31

International Review of Neurobiology Volume 14

International Review of Neurobiology, Volume 23

International Review of Neurobiology Volume 1

International Review of Neurobiology Volume 13

International Review of Neurobiology Volume 5

International Review of Neurobiology, Volume 21

International Review of Neurobiology Volume 36

International Review of Neurobiology Volume 16

International Review of Neurobiology Volume 32

International Review of Neurobiology Volume 12

International Review of Neurobiology Volume 3

International Review of Neurobiology, Volume 48

International Review of Neurobiology, Volume 20

International Review of Neurobiology, Volume 73

International Review of Neurobiology Volume 29

International Review of Neurobiology, Volume 8

International Review of Neurobiology Volume 30

International Review of Neurobiology Volume 17

International Review of Neurobiology Volume 4

International Review of Neurobiology, Volume 63

International Review of Neurobiology Volume 2

International Review of Neurobiology Volume 24

Recommend Documents