The Genom of Homo sapiens.pdf

Structure of Linkage Disequilibrium in Humans:Genome Factors and Population StratificationJ. BERTRANPETIT, F. CALAFELL, D. COMAS, A. GONZÁLEZ-NEIRA, AND A. NAVARROUnitat de Biologia Evolutiva, Universitat Pompeu Fabra, 08003 Barcelona, Catalonia, SpainHuman genomes differ at, on average, one per thousandbases; this statement, which has been repeated adnauseam, takes on a whole new dimension when thephysical organization of the human genome is taken intoaccount. Thus, each allele at each polymorphic site doesnot exist independently, but is physically linked to the alleleoccurring at the next polymorphic site, and so on, allalong each chromosome. A haplotype is any combinationof allelic states at linked sites. Such physical links arequite strong and are only broken by recombination viacrossing-over or similar DNA exchange processes suchas gene conversion. Recombination, the shuffling ofchromosome segments of maternal and paternal originduring gamete formation, happens at such a broad scalethat relatively wide DNA portions tend to travel togetherfrom one generation to the next. In fact, the averagechance of two adjacent bases being separated by recombinationin one generation is in the order of one in a hundredmillion (10 –8 ), or, if considering polymorphic sitesat a density of one per kilobase, their probability of notstaying together in one generation would be roughly 10 –5 .Frequently, haplotype frequencies do not derive fromthe random assortment of alleles at each locus; preferentialassociations do exist. The departure of haplotype frequenciesfrom the expectation under random associationof their integrating alleles is called linkage disequilibrium(LD). LD arises by several mechanisms, including randomgenetic drift, mutation, migration, changes in populationsize, and selection due to genome and populationfactors (Bertranpetit and Calafell 2001). All this newlycreated LD will be eroded by recombination, which willbe more efficient in reshuffling the alleles at the most distantloci. Thus, LD is expected to decay with physical distance.In summary, sites that are physically close willtend to carry particular combinations of alleles.From a purely numerical point of view, LD can becharacterized in two ways: as a direct measure or as theresult of a formal hypothesis test. For the sake of simplicity,we will refer to LD between a pair of polymorphisms,although the LD concept can be extended, notwithout some difficulty, to the relationships among alarger number of markers. LD ranges between two possibleextremes: On one hand, knowledge of the allele in onelocus may not convey any information about the contentof the other locus; both are independent and LD is, in fact,nil. On the other hand, LD can be complete in the sensethat knowing which allele is at one locus determineswhich allele is at the other. Measures of LD (derived fromallele and haplotype frequencies) take extreme values(conveniently, 0 and 1) for these two extreme cases andthe appropriate intermediate value for intermediate situations.The two most popular measures, D´ and r 2 , arebased on the departure from the expected haplotype frequenciesunder linkage equilibrium and on the correlationcoefficient between haplotype frequencies, respectively.However, a given D´ or r 2 value does not reveal by itselfthe statistical significance of LD; for that, proper testingis needed, usually by means of a chi-square statistic or byFisher’s exact test.LD has a clear role to play in biomedical research: Thephysical position of a gene contributing to a phenotype(usually a disease) can be deduced from the polymorphicmarkers with which it is found at high LD. Most genes involvedin the causation of Mendelian disorders were notfound via LD, but with linkage mapping, in which the locationof a disease gene is investigated by observing, infamilies with affected and nonaffected individuals, thepatterns of joint inheritance of the phenotype and each ofthe polymorphic markers in a genetic map. Accordingly,the events that are useful to reject a genome segment ascontaining the disease locus are recombinations happeningwithin the families collected for the study. Alternatively,if LD is assessed in the whole population, thewhole history of recombination between the disease locusand the map markers can be used to locate the gene. Thisapproach was used first by Hastbacka et al. (1992) to refinethe position of the gene for diastrophic dysplasiawithin the interval provided by linkage analysis.LD is implicitly embedded in association mapping, oneof the most popular approaches in the search for the geneticcomponents of complex diseases. In associationstudies, the allelic frequencies in one or more polymorphismsare compared between affected individuals andhealthy controls. If frequencies are significantly different,there are two possible explanations: Either the polymorphismitself contributes to the phenotype (which can oftenbe ruled out given that most polymorphisms used aresynonymous or otherwise neutral variants), or the polymorphismis in linkage disequilibrium with genetic variantsactually involved in the disease phenotype.Association studies often involve polymorphisms in oraround candidate genes that, given their known biologicalCold Spring Harbor Symposia on Quantitative Biology, Volume LXVIII. © 2003 Cold Spring Harbor Laboratory Press 0-87969-709-1/04. 79