ARTICLESignificant variation in haplotype block structure butconservation in tagSNP patterns among globalpopulationsSheng Gu1, Andrew J Pakstis1, Hui Li1, William C Speed1, Judith R Kidd1andKenneth K Kidd*,11Department of Genetics, Yale University School of Medicine, New Haven, CT, USAThe initial belief that haplotype block boundaries and haplotypes were largely shared across populationswas a foundation for constructing a haplotype map of the human genome using common SNP markers.The HapMap data document the generality of a block-like pattern of linkage disequilibrium (LD) withregions of low and high haplotype diversity but differences among the populations. Studies of manyadditional populations demonstrate that LD patterns can be highly variable among populations bothacross and within geographic regions. Because of this variation, emphasis has shifted to thegeneralizability of tagSNPs, those SNPs that capture the bulk of variation in a region. We have examinedthe LD and tagSNP patterns based upon over 2000 individual samples in 38 populations and 134 SNPs in 10genetically independent loci for a total of 517 kb with an average density of 1 SNP/5 kb. Four different‘block’ definitions and the pairwise LD tagSNP selection algorithm have been applied. Our results not onlyconfirm large variation in block partition among populations from different regions (agreeing withprevious studies including the HapMap) but also show that significant variation can occur amongpopulations within geographic regions. None of the block-defining algorithms produces a consistentpattern within or across all geographic groups. In contrast, tagSNP transferability is much greater than thesimilarity of LD patterns and, although not perfect, some generalizations of transferability are possible.The analyses show an asymmetric pattern of tagSNP transferability coinciding with the subsetting ofvariation attributed to the spread of modern humans around the world.European Journal of Human Genetics (2007) 15, 302–312. doi:10.1038/sj.ejhg.5201751; published online 3 January 2007Keywords: haplotype; haplotype block; linkage disequilibrium; tagSNPsIntroductionHaplotype and linkage disequilibrium (LD) analyses arewidely used for mapping disease genes and understandinghuman population history.1–4Some haplotype studieshave suggested that the human genome can be parsedobjectively into haplotype blocks in which there is littleevidence for historical recombination.5–9In recognition ofa few common haplotypes capturing most of the geneticvariation across regions encompassing multiple SNPs,Gabriel et al7proposed testing only the minimum numberof SNPs sufficient to define the common haplotypes,haplotype tag SNPs (htSNPs or tagSNPs). Defining thegenomic regions of high LD, that is, haplotype ‘blocks’, wasan initial focus and continued to be an interest of theInternational HapMap project.8,9The effort to construct a haplotype map or maps of thehuman genome based on limited population samples hasReceived 21 July 2006; revised 26 September 2006; accepted 25 October2006; published online 3 January 2007*Correspondence: Dr KK Kidd, Department of Genetics, Yale UniversitySchool of Medicine, 333 Cedar Street, PO Box 208005, New Haven, CT06520-8005, USA.Tel: þ 1 203 785 2654; Fax: þ 1 203 785 6568;E-mail: [email protected] Journal of Human Genetics (2007) 15, 302–312&2007 Nature Publishing Group All rights reserved 1018-4813/07$30.00www.nature.com/ejhgan underlying assumption of extensive sharing of patternsof LD such that tagSNPs will show good ‘transferability’(how well tags selected in one or more analysis panels willtransfer to disease studies performed in these or otherpopulations9) among populations, especially populationswithin the same geographic regions. Despite ongoingcontroversy over some issues,10 – 12the original belief thathaplotype blocks represent a fundamental aspect of thehuman genome appears to be an oversimplified view ofgenome organization.Few studies have examined LD in a global set ofpopulations. One study used the Human Genome DiversityProject-CEPH panel to examine variation in LD at oneregion on chromosome 22. It concluded that a widecommunality in LD patterns exists in human populationsfrom different continental regions despite differences intheir demographic histories.13However, this conclusion wasbased on only 12 SNPs across a 1.78 Mb region. In manypopulations, there was little LD, which resulted in similartagSNP sets. Another much more extensive study of theportability of tagSNPs across global populations at highermarker density focused just on tagSNP transferability.14Itconcluded that tagSNPs from one population are highlyinformative in other populations within each continentalgroup. However, other studies have reported significantvariation among populations in block structure andtagSNPs. Sawyer et al15studied three loci in 16 diversepopulations with an emphasis on African and Europeanpopulations. They found significant quantitative and qua-litative variation in LD among populations both across andwithin geographic groups, and no group showed consis-tency in patterns of LD for all three loci under study. Liuet al16reached a similar conclusion with respect to tagSNPs.The HapMap data do not address directly the question ofhow tags selected in one or more of the populationsstudied will transfer to disease studies performed in theseor other populations, and thus the general applicability ofthe HapMap data needs to be confirmed in samples fromseveral local populations.8,9We are examining a set ofpopulations comparable to the HGDP-CEPH panel used byGonza´lez-Neira et al;13,14indeed, many of the populationsamples in that panel originated in our lab. However, oursample sizes are, on average, twice as large as in the HGDP-CEPH panel. Using data that represent an even morediverse sampling of LD patterns among populations thaninitially planned for extensions of the HapMap effort, weconsider consistency of blocks and tagSNPs among popula-tions and the similarity of different methods in definingblocks. Our data on 38 populations widely distributedaround the world (Figure 1) illustrate considerable diversityin the patterns of LD suggesting that HapMap data do notgeneralize in this respect. However, the transferability oftagSNPs is, in general, quite high, in agreement withGonza´lez-Neira et al.14Figure 1 The graphical distribution of the 38 population samples used in this study.