The Genom of Homo sapiens.pdf

GENETICS OF COMMON DISEASES 397within genomic regions implicated through linkage analyses,as for example, in the cloning of the gene responsiblefor cystic fibrosis (Riordan et al. 1989).Two factors greatly increased interest in the applicationof LD mapping to common diseases. One was the demonstrationof Risch and Merikangas (1996) of the greaterpower of association mapping over linkage studies. Theyshowed that for modest effect sizes, affected sib pair analyses,for example, would require unrealistically largesample sizes. The other was the rapidly expanding coverageof the human genome with polymorphic markers.The potential application of LD mapping for commondisease generated intense interest in patterns of linkagedisequilibrium in human populations. Until 2001, thiswork was predicated in many ways on expectations ofgradual, though noisy, decay of LD with physical distancein the genome. It had long been appreciated thatcertain genes, such as the β-globin gene cluster (Chakravatiet al. 1984; Jeffries et al. 2000), harbored hot spots ofrecombination—regions of intense homologous recombination—andalso that LD decay was often poorly correlatedwith distance, either because of uneven recombinationin some regions or because of the stochasticeffects.Nevertheless, many empirical studies looked at the averagedecay in LD with distance over different genomicregions, effectively ignoring the precise pattern of decaywithin a given region. In addition, models used to generateexpectation of the pattern of LD assumed uniform recombination.In his influential report, Leonid Kruglyakestablished a sort of null model for the expected extent ofLD by assuming an idealized human demographic historyand assuming homogeneous recombination rates. Underthese assumptions, Kruglyak (1999) showed that usablelevels of LD (at that time set at d 2 >0.1, which correspondsto a r 2 5%) across 500 kb of Chromosome 5q31, Daly etal. (2001) argued that the pattern was better viewed asdiscrete, with stretches of sequence showing little or noLD breakdown interspersed by regions of sharper LDbreakdown. The stretches of limited haplotype diversitywere called blocks, and within them up to 95% of the observedchromosomes were accounted for by 3 or 4 haplotypes(Daly et al. 2001). Similar patterns were reported byJohnson et al., who observed a similar block-like patternof LD among 122 SNPs across 135 kb of 9 genes in Europeanpopulations. Haplotype diversity within blockswas low, with a maximum of 6 common haplotypes (frequency>5%) observed within any one block (Johnson etal. 2001).These publications led directly to the idea that blocksof LD, within which haplotype diversity is limited, are aprevailing characteristic of the genome. This in turn led tothe concept that a set of SNPs could represent, or tag,each of the common haplotypes in a given region. TheseSNPs were first referred to as haplotype-tagging SNPs(htSNPs). In an accompanying supplement to the Johnsonet al. study, David Clayton introduced an approach for selectinghtSNPs that focused on the proportion of the haplotypediversity that could be explained by a set of tags(Johnson et al. 2001). This definition of htSNPs was notexplicitly tied to the idea of a block, but given its focus ontagging a small number of haplotypes, it has often beenperceived in the community as applying to the selectionof htSNPs within blocks (see below).In the same issue, Jeffries et al. published data relatedto a possible cause of uneven decay of LD. Through singlesperm typing in the class II major histocompatibilitycomplex (MHC) region, they showed recombinationevents to be clustered in narrow, discrete regions (Jeffreyset al. 2001).This shows that at least in one genomic region the locationof recombination hot spots corresponds with, anddrives discontinuities in, the pattern of LD. There is currentlya great deal of debate about the extent to which LDdiscontinuities coincide with recombination rate hotspots. However, this should not distract from the centralcontribution of the Johnson et al., Daly et al., and Jeffrieset al. papers. In breaking with the strong tradition ofviewing LD decay as relatively smooth and homogeneous,this work represented a striking and importantconceptual advance. After these papers, it was immediatelyunacceptable to focus on descriptions of LD that ignoredthe precise pattern of decay within a region, andsimilarly unacceptable to make predictions based onmodels that assumed uniform recombination rates.Whatever the causes of sharp LD discontinuities, theyhave important implications for the design and interpretationof association studies. The first large-scale studyexplicitly incorporating the idea of very uneven patternsof LD decay, or blocks of LD, was reported less than ayear later by Gabriel et al. (2002), who analyzed patternsof haplotype diversity from over 3700 SNPs in 51 regionsspanning 13 megabases of the genome in four populationgroups (European, African-American, Nigerian, and Chinese).The same apparent block-like pattern of LD withaccompanying lack of within-block haplotype diversityas reported by Daly et al. (2001) was observed across all