The Genom of Homo sapiens.pdf

396 GOLDSTEIN, CAVALLERI, AND AHMADIFigure 1. Pooled odds ratios for 25 polymorphisms reported asassociated with common diseases for variants that are (darkblue) and are not (red) replicated by meta-analysis (data fromLohmueller et al. 2003). Note that the values indicated are forthe associations studied in Lohmueller et al. (2003), and notnecessarily for the condition with which the given polymorphismis most strongly associated. For example, the odds ratioestimate for APOE relates to schizophrenia, not Alzheimer’sdisease.this proportion might be. It thus seems premature to castthe search as one for a small number of susceptibilitygenes for each condition.We therefore think that study of the genetics of commondiseases is better cast not as a search for “disease”genes, but rather as an assessment of how polymorphismsin the human genome influence disease risk and drug responsesin specific environments and genetic backgrounds.This may sound like a subtle or pedantic distinction,but there are implications both in the wayresearch is motivated and interpreted, and in how it is perceivedby the public. For example, current interpretationsof association studies tend to be typological in assessingwhether a variant really is or is not a risk factor. We believethat this is too driven by the Mendelian experience,and that the real point is to understand the effects of thepolymorphism, as opposed to reach conclusive agreementthat a polymorphism has some effect. Understandingwhether and how polymorphisms have their effects ingiven environments and genetic backgrounds is a longtermprocess that does not have any clear stopping point.Similarly, the Mendelian experience leads the public tothe view that there are good and bad variants, whereaswhat little we know about complex disease causation suggeststhat few variants are uniformly bad or uniformlygood.Looked at this way, the research program ahead of usis open-ended, and we should be clear about how hard itwill be. There are more than 10 million polymorphic sitesout of the approximately 3 billion bases in our genome intypical human populations (where polymorphic is definedas having a minor allele frequency greater than 1%;cf. Kruglyak and Nickerson 2001). The job is not to findthe few disease polymorphisms out of these 10 million forany given condition. Instead, the challenge is to understandhow these 10 million polymorphisms collectivelymoderate disease risk, and how they influence responseto treatment, in their appropriate genetic and environmentalbackgrounds.A focus on even this large set of sites assumes that themost important genetic contributions are due to sites thatcarry classic polymorphisms (i.e., minor allele frequencygreater than 1%). Clearly, these polymorphic sites contributeprimarily to genetic differences among people onaverage, but there is considerable debate about their importancerelative to less common variation in the geneticsof common disease predisposition (Wright and Hastie2001). When the problem is cast this way, it is apparentthat very large scale and high-quality genetic associationstudies, with all their problems (Zondervan and Cardon2004), are indispensable in the effort to understand thegenetics of common diseases and variable drug reactions.AND HOW TO FIND ITMost genes causing Mendelian diseases have beenidentified using linkage analyses that assess the co-inheritanceof disease and genetic markers in pedigrees. Linkageanalyses, however, are known to have limited powerto identify variants that have only modest effects on diseaserisks. Risch and Merikangas (1996) showed that forvariants of only modest effects, association studies,which assess correlations between diseases and genevariations in random population samples, have considerablygreater power than linkage-based methods (Rischand Merikangas 1996). Two basic strategies have beenadvocated. Botstein and Risch have recently argued for aSequence-based or direct approach, which focuses on theidentification of all variants in and near exons, and in corepromoter regions of genes (Botstein and Risch 2003).They argue that the experience of Mendelian disease suggeststhat this is where most of the important variants willreside. All identified variants in these putatively functionalgenomic regions would then be genotyped in individualsof known phenotype, and correlations assessed ina case-control or related design.The alternative, Map-based approach, refrains frommaking any assumptions about the precise genomic locationof causal variants. Instead, the aim is to identify amap of markers that is designed to be sufficient to be associatedwith all other variants in a gene or gene region ofinterest. Both approaches have strengths and weaknesses.The principal advantages of the map-based approach areeconomy and the fact that it does not require strong assumptionsabout where the important variants are. Theprincipal weakness is that indirect approaches will havedifficulty in accurately representing variants with low minorallele frequency, as discussed below. For these reasons,we see the two approaches as more complementarythan competing. Here, we describe the development ofmap-based approaches.LINKAGE DISEQUILIBRIUM GENE MAPPINGLinkage disequilibrium (LD) gene mapping was firstused successfully to fine-localize Mendelian mutations