Will Investments in Biobanks, Prospective Cohorts, and Markers Of

Home , Rare disease

different single-nucleotide poly- Will investments in biobanks, morphisms for some bases within that segment). As such, nearby alleles in prospective cohorts, and markers the same haplotype are inherited to- gether more often than is expected by of common patterns of variation chance, a phenomenon that is called ‘linkage disequilibrium’. Owing to that benefit other populations for drug coinheritance, haplotypes offer a way of summarizing the genetic variation response and disease susceptibility found within them. Researchers need identify only a few SNPs (called ‘tag SNPs’) to identify the longer haplo- gene discovery? type, which greatly decreases genotyping costs for gene discovery.6 1 2 MW Foster and RR Sharp Those tag SNPs can be used to look for patterns or similarities in inter- 1Department of Anthropology, University of Oklahoma, Norman, OK, USA; individual genetic variation among 2Center for Medical Ethics and Health Policy, Baylor College of Medicine, those affected by a disease or drug Houston, TX, USA response when compared with un- affected controls (ie an association study). The rationale for this strategy is that those similarities (most of The Pharmacogenomics Journal (2005) 5, BACKGROUND which will not contribute to the drug 75–80. doi:10.1038/sj.tpj.6500295 The International HapMap Project, at response or disease in question) will Published online 25 January 2005 US $120 million perhaps the most allow researchers to narrow the possi- high-profile example of infrastructure ble chromosomal regions in which a investment in the next generation of contributing gene may be located.7,8 genetic research, is intended to pro- High-throughput technologies, de- Tag SNPs can be tested for their vide a set of markers to facilitate cost- creasing costs for genotyping, association with a drug response or effective association studies. To that and genomic resources that promise an increased susceptibility to a disease, end, the HapMap is using samples to reduce the number of genotypes with the idea that the actual contri- from four populations (Yoruba in required to scan the genome (such as buting gene(s) will be located nearby Ibadan, Nigeria; Japanese in Tokyo, the International HapMap Project) the associated marker(s). Other tech- Japan; Han Chinese in Beijing, China; soon will make association studies a niques, such as positional cloning, more successful strategy for identify- and Utah residents with ancestry from then can be used to identify the gene ing genetic contributors to disease Northern and Western Europe known or genes in question. susceptibility and drug response.1,2 as the CEPH samples) to construct a To have sufficient power to detect Accordingly, both nonprofit funders genomic resource that will identify the weak-to-intermediate genetic effects, and biotechnology companies cur- common (45%) haplotypes that are association studies will require very rently are investing in infrastructure estimated as structuring 80–85% of all large numbers of DNA samples linked 3,4 for association studies. Elements of genetic variation among humans. with phenotypic information. In addi- that emerging infrastructure such as Preliminary research suggests that the tion to traditional case–control or prospective cohorts and biobanks are same three to five common haplotypes cross-sectional samples, those data being developed with participants will be found at any given chromo- increasingly will come either from from a limited number of populations somal region in most if not all human prospective cohorts or from biobanks with the goal of applying their populations, although in different (which often include both retrospec- findings to many other populations. frequencies.5 tive and prospective data), the estab- However, not all phenotypes and A haplotype is a linear sequence of lishment and maintenance of which populations may be as suitable for bases on the same chromosome (usual- also will require significant funding this approach as others. Which ly varying in length from 100 to and effort. The advantage of biobanks populations and phenotypes will 100 000 bases) that is identifiable as and prospective cohorts over case– benefit and which will not in the one of a number of alternative struc- control designs is that the former coming era of large-scale association tures for a given segment of that provide longitudinal data for multiple studies? chromosome (by virtue of there being phenotypes. Investments in large-scale genomics project MW Foster and RR Sharp 76

The questions about the overall Project diverged from one another), and recombination events have added association study strategy, though, requiring that different tag SNPs be to those original variants, although are, first, the extent to which ‘com- developed for each population or the greater allelic diversity continues mon’ markers or patterns of variation sample studied.16,17 to be found in ancestral African popu- discovered in a few (albeit carefully These critiques will be neither gen- lations. Founder effects among those chosen) populations will be useful for erally proven nor disproved. Rather, who migrated out of Africa decreased gene discovery in other populations they will apply to some populations, the allelic diversity in non-African and, second, the applicability of gene some phenotypes, and some genomic populations, while subsequent popu- discoveries in a given population to regions but not to all. Such factors as lation bottlenecks also tended to re- other populations. Positive answers to the effect size of a disease susceptibi- duce diversity. Nevertheless, genetic both these questions often have been lity or drug response locus, the fre- drift can increase the frequency of linked because it makes no sense to quency of the disease or drug response otherwise rare genetic patterns and construct a HapMap without large allele(s), the frequency of the marker variants, particularly when those al- biobanks and prospective cohorts in allele(s), and the extent of linkage ready exist in the population. which to use tag SNPs. At the same disequilibrium between the marker In contrast, locally varying selection time, large biobanks and cohorts are and the disease susceptibility or drug pressures have been more responsible best utilized for association studies, response locus will figure into a for- for differences in frequencies of allelic the cost of which can become prohi- mula that can indicate the likelihood variants, with selection against a trait bitive without markers like tag SNPs of success for an association study for a tending to increase rare alleles and that can reduce the number of bases specific phenotype in a particular positive selection increasing the fre- that must be genotyped. population with a given study de- quency of adaptive alleles, while simi- The debate that has developed sign.15 We can, though, consider po- lar environmental conditions across around these two questions has fo- tential scenarios in which tag SNPs populations have tended to promote cused on the applicability of what has and disease susceptibility and drug– selection of alleles in similar frequen- been called the Common Disease/ response variants discovered using cies. Selection sometimes also can Common Variant (or CD/CV) hypo- prospective cohorts and biobanks affect the frequency of patterns of thesis,9 which supports the view that in some populations may be useful variation, such as when an ancestral one or a few genetic contributors in other populations. These scena- haplotype is nearby an allele on which account for significant numbers of rios more fully describe the range of there is a positive or negative selective cases of many common, complex dis- logical possibilities of which the CD/ pressure. eases in most or all populations10,11 CV hypothesis is but one. It is useful Indeed, the continued sorting and The CD/CV hypothesis, though, is not to consider that fuller range to think resorting of individuals over time into unique to disease expression and can more systematically about the ways in loosely bounded, transitory groupings be extended to other phenotypes such which investments in the infrastruc- (ie populations) each of which has as drug response. ture to support association studies experienced somewhat unique, dyna- Several critiques, however, have (and investigators’ and funders’ future mically varying selection and demo- been made of the CD/CV strategy choices increasingly to rely on those graphic events has resulted in the underlying the infrastructure being studies) may result in disparities in the different scenarios for contemporary developed around association studies. benefits of genetic research. populations described below. First, some have argued that common variants may be the exception rather Common Phenotype, Common than the rule, and that most common SCENARIOS pattern, Common Variant (CCC) diseases and drug responses may have Scenarios in which genetic discoveries This is what usually is meant by the multiple, rare (o1% frequency) genet- in one population may benefit other CD/CV hypothesis. SNPs, haplotypes, ic contributors that likely vary con- populations are best understood and variant alleles that are common siderably by population.12–14 Second, against the framework of human po- across contemporary populations most the effects of some common variants pulation history. The human popula- likely emerged from the original range may be relatively weak such that even tion as a whole is relatively young, of human variation prior to the con- association studies with very large with migrations out of Africa to other tinental diaspora. In addition, it may numbers of participants will not have continents beginning approximately be the case that some recombination sufficient power to detect them.15 100 000 years ago. Consequently, com- subsequent to the continental dia- Third, the tag SNPs identified by the mon patterns of variation and com- spora has tended to occur at the same HapMap Project may not be applicable mon variants that arose in Africa prior chromosomal ‘hotspots’ such that to other populations (due in part to to the continental diaspora still con- many independent changes in haplo- the effects of local selection pressures stitute a shared catalogue or map that type structure are nonetheless congru- and genetic drift that have affected is the ancestral basis for contemporary ent.18 Those common patterns haplotype frequencies since the po- population-specific patterns of sequ- (whether the result of ancestral haplo- pulations sampled for the HapMap ence variation. Subsequent mutation types or hotspot recombination) can

The Pharmacogenomics Journal Investments in large-scale genomics project MW Foster and RR Sharp 77

be identified in one population or set populations due to a variety of selec- founding or bottleneck event.19 Con- of populations and used in gene dis- tion pressures and demographic cir- sequently, populations that are most covery in other populations. More- cumstances, often in combination distantly related to the groups from over, common disease susceptibility with one another.19 Nonetheless, pat- which gene discovery resources such or drug response variants discovered terns of genetic variation that are Hapmap were constructed and that in one population will benefit all common across populations can be have experienced local selection pres- populations (to the extent that they used to discover otherwise rare var- sures or genetic drift (or both) may not are present in varying frequencies in iants that are specific to or more necessarily benefit fully from markers other groups). Mutations in APOE that frequent in particular populations such as tag SNPs derived from patterns increase risks for Alzheimer disease where the common pattern is in of genomic variation common to and heart disease often are cited as an association with the rare variant in other populations with different his- example of common variants present those populations at a similar fre- tories. However, due both to the in populations throughout the world.9 quency.15 Patterns of genetic variation shared African origin of all popula- In fact, although relatively few var- such as haplotypes, SNPs, or regions tions and to current members who iants that contribute to common dis- of linkage disequilibrium often are have some ancestors from other popu- eases have been confirmed to date, located on chromosomes nearby con- lations, at least some of the patterns most of the variants that are known tributing variants but do not neces- common to other populations still are common across populations (see sarily contain the variants such that will be evident in some frequency Table 1). Other common variants with it is possible for a population to have in populations that have higher fre- weak effects will require very large both common patterns of variation quencies of otherwise rare patterns. numbers of association study partici- (the result largely of demographic Thus, populations with some unique pants to be detected. In some cases, events that are neutral with respect patterns of variation still may benefit that larger number may become pro- to selection) and higher frequencies of from the use of common patterns hibitive such that a family linkage some otherwise rare variants (which for gene discovery (although these study or candidate gene approach is a may result either from positive selec- may be less efficient in part because more practical approach. tion or demographic circumstances or they are likely to occur at different both). Where a variant is rare in all frequencies than the disease and populations, of course, an association drug response alleles in question). Common Phenotype, Common study is unlikely to discover it and While these populations will be less Pattern, Rare Variant (CCR) family linkage studies (which require than ideal candidates for association In cases in which more than one gene fewer participants) are more appropri- studies, their members nonetheless or allele may contribute to suscept- ate. will benefit from such studies that ibility for the same disease or to discover common variants in other response to the same drug and in Common Phenotype, Rare Pattern, populations. which those alleles occur in different Common Variant (CRC) frequencies across populations, contri- Some populations (particularly rela- buting genes and alleles found in one tively small groups in which genetic Common Phenotype, Rare Pattern, population may not necessarily be the drift has affected frequencies) may Rare Variant (CRR) more common contributors for those have haplotypes, SNPs, or linkage Of course, some populations with diseases or drug responses in other disequilibrium regions that are more higher frequencies of otherwise rare populations. Otherwise rare variants frequent among their members but patterns of genetic variation also will can become more frequent in some rare in other populations. It has been have higher frequencies of otherwise suggested, for instance, that the com- rare variants that contribute to some mon haplotypes identified by the common diseases and drug responses. Table 1 Examples of common variants HapMap Project may not apply to Such populations are likely to be that contribute to complex diseases some chromosomal regions in Native strongly affected by relatively recent

27 American populations because those founder effects or bottlenecks (hence ApoE4 Alzheimers indigenous groups historically had the otherwise rare patterns of varia- Factor V Leiden Deep venous relatively small effective populations tion), while the frequency of otherwise thrombosis28 CCR5 HIV susceptibility29 sizes and were reproductively isolated rare variants may be amplified by Insulin VNTR Type 1 diabetes30 for at least 10 000 years from popula- locally varying selection pressures in 20 CTLA4 Type 1 diabetes, Graves tions sampled for the HapMap. In combination with the effects of demo- disease and autoimmune those circumstances, otherwise rare graphic events. These populations are thyroid disease30 patterns of variation could increase in least likely to benefit from the discov- PPARG P12A Type 2 diabetes31 frequency if they were nearby or ery of genetic patterns and variants 31 Kir6.2 E23K Type 2 diabetes included alleles under positive selec- common in other populations, at least 32,33 CETP Lipid levels tion or were among the less diverse in the cases of those rare variants, and 34 LPL Lipid levels range of variation inherited after a thus will benefit more from genetic

www.nature.com/tpj Investments in large-scale genomics project MW Foster and RR Sharp 78

studies of variation and disease or drug Rare Phenotype, Rare Pattern, large cohorts and biobanks. While response among their members. Rare Variant (RRR) the latter are not necessarily premised Populations that continue to be on the CD/CV hypothesis, justifica- strongly affected by localized selective tions for projects such as the UK Rare Phenotype, Common Pattern, pressures and/or by recent population Biobank and NIH-sponsored prospec- Common Variant (RCC) bottlenecks and founder effects may tive cohorts and business plans for Some diseases and drug responses are have higher frequencies of otherwise projects such as deCODE (a for-profit rare in most populations but occur at rare phenotypes, associated patterns of venture making use of DNA samples higher frequencies in a few popula- genetic variation, and genetic variants and medical records from large num- tions. Where those otherwise rare such that patterns of variation and bers of Icelandic citizens) have empha- phenotypes are more likely to occur, genetic variants common to other sized the numbers of people affected but are associated with patterns of populations are of relatively little by complex diseases such as cancer genetic variation and candidate var- benefit for disease susceptibility and and the benefits that gene discoveries iants that are also common in other drug response gene discovery. These in one population will have for other populations that have lower rates of constitute the ideal model of the populations.21,22 phenotypic expression, it is likely that ‘isolated’ population. Whether such The second is that additional invest- some environmental difference be- populations exist, though, in the pre- ments should be targeted for those tween populations is the more prob- sent context of multigenerational glo- populations that may not benefit from able cause for the higher frequencies of balization is uncertain. research in other populations because the rare phenotypes. In those cases, a of higher frequencies of otherwise rare common pattern of variation can be variants or rare patterns of variation used to indicate the need for more DISTRIBUTION OF BENEFITS (or both) among their members (ie extensive environmental investigation Of the seven scenarios outlined above, CRR and RRR). It is unknown how in populations with higher rates of three (CCC, CCR, and RCR) describe many populations and diseases fit this phenotypic expression. situations in which populations will profile, but it is probably a relatively benefit directly from gene discovery small number. Moreover, because Rare Phenotype, Common Pattern, resources such as the HapMap devel- small effective population size, geo- Rare Variant (RCR) oped in other populations, two (CCC graphic isolation, founder effects, and In other cases, a common pattern of and CRC) describe situations in which bottlenecks are likely to have contrib- genetic variation may be used to populations will benefit directly from uted to higher frequencies of other- identify an association with an other- gene discoveries in other populations, wise rare patterns and variants among wise rare variant that occurs at higher two (CRR and RRR) describe situations populations in this category, their frequencies in the populations with in which there are likely to be few if current numbers may be too small to higher rates of phenotypic expression. any direct benefits either from gene support association studies that often In other words, these are complex discovery resources or from gene dis- require significant numbers of partici- diseases or drug responses that have coveries in other populations, and one pants with a particular phenotype, not yet become common in most (RCC) describes situations in which especially to detect weaker genetic populations, suggesting that a higher environmental contributors are more effects. Thus, biobanks or cohorts localized prevalence may be due to a likely responsible than genetic contri- formed in those populations should relatively recent demographic event or butors. emphasize recruiting participants who to positive selective pressure for a The scenarios suggest two, perhaps, are genealogically related to one an- reproductive enhancement also con- competing principles of justice (at other to take advantage of opportu- ferred by the variant, as in the case of least insofar as funding resources are nities for family linkage studies. heterozygosity for the sickle cell muta- limited): Perhaps the greater disparity will be tion. Ashkenazi, Hutterite, and Old The first is that investments in in the cases of conditions that affect Order Amish populations are examples association study infrastructure should large numbers of people worldwide, in which a number of otherwise rare maximize returns for the benefit of the but for which there may be multiple, diseases occur more often due to high- most people. That principle would rare variants rather than one or a few er frequencies of otherwise rare var- favor investments that take advantage common variants. Even in the pre- iants.19 Still, while the variant may be of the CD/CV hypothesis (and its sence of common patterns of varia- rare in the populations used to devel- various corollaries detailed in scenar- tion, a number of population-specific op the common patterns, this does not ios CCC, CCR, RCR, and CRC) in studies will be required to identify diminish the fact that genetic varia- studies of diseases and drug responses. these variants and their frequencies. tion research in those other popula- Current trends in research infrastruc- However, improvements in the costs tions can be of benefit in discovering ture investment suggest that this prin- and efficiencies of gene resequencing the rare variant in populations in ciple is being applied through funding and other emerging technologies may which it and the phenotype to which for resources such as HapMap and the facilitate the discovery of multiple, it contributes are more frequent. formation of a growing number of rare variants where a candidate gene

The Pharmacogenomics Journal Investments in large-scale genomics project MW Foster and RR Sharp 79

or biological mechanism already has CONCLUSIONS Despite the potential for widespread been identified in another population. We are at the beginning of a shift in benefits, however, the emerging That possibility suggests that the genetic research from an historical emphasis on genetic variation infra- benefits of population-specific re- focus on studying rare conditions that structure and association studies search into contributors to drug affect relatively few people (ie Mende- could create some ongoing disparities response and disease susceptibility are lian traits) to giving more attention in the benefits from genetic research. not limited only to the discovery of and resources to conditions that affect Of course, there are not sufficient particular genetic variants. Some dis- the many (ie common, complex traits). resources to fund separate gene coveries can lead to development of a Previously, there would have been little discovery resources such as HapMap drug or a biological insight that is of point in evaluating funding for genetic or cohorts and biobanks in every broad relevance, even if the genetic studies based on numbers of people population or for all possible diseases variants are more restricted in distri- who are affected by a particular condi- and drug responses. As resources bution. For example, familial hyper- tion, by the size of the population continue to flow to the infrastructure cholesterolemia is a rare disease, but being studied, or by the extent to needed to support association studies, the discovery of the genetic variants which other populations are benefited; provisions for those populations underlying that condition led to our indeed, such criteria would have been that are least likely to benefit from understanding of the function of the detrimental to the development of studies in other populations (ie LDL receptor and cholesterol metabo- knowledge and technology that was CCR, CRR and RRR) should be con- lism,23 which in turn has led to the best nurtured in the context of rare, sidered. development of statins. Statins are monogenic conditions. Now, though, We suggest the following priorities: broadly useful for a number of more we are at the point at which those For populations in which otherwise common diseases,24 even if the genetic criteria are becoming more relevant. rare variants are more frequent, it will variation that began the process of Viewed in that historical light, the shift be important to make funding avail- discovery is rare. toward infrastructure investment in able to establish biobanks or cohorts Other indirect consequences of ge- association studies that emphasize re- so that association studies specific to netic investigation, though, may not sources and discoveries common to those populations will be possible. For be as beneficial. Larger biobanks or multiple populations is a logical con- populations with otherwise rare pat- cohorts will help identify or confirm sequence of improvements both in our terns of variation, funding for more environmental contributors that are scientific knowledge of the genome population-specific gene discovery more common among their members, and in technology. resources (ie population-specific but will be less likely to help identify Obviously, populations from which HapMaps) will be important, espe- those less common contributors that the association study infrastructure of cially for use in studies of phenotypes are rare among most study partici- biobanks, cohorts, and tag SNPs is that do not have common genetic pants. Indeed, the additional power being constructed—as well as others contributors. For both relatively that a larger biobank or cohort pro- with close historical relationships to small populations (many of which vides to detect common effects simul- them—will experience the clearest, will be economically disadvantaged) taneously can mask those contributors most immediate benefits from it. and for phenotypes and variants that are primarily localized within Nonetheless, based on these scenarios, that are truly rare in all populations, subsets of the larger sample, depend- the common patterns of variation continued funding for biobanks or ing on how participant information is identified by the HapMap and associa- cohorts enriched to support family collected and analyzed. In most cases, tion studies that discover common linkage studies and other smaller- because it is expensive to investigate patterns and variants in selected po- scale kinds of approaches should be environmental exposure histories in pulations have considerable potential protected from competing directly great detail for large numbers of for benefiting other populations, in- with large-scale, more expensive (and participants,25 large biobanks or co- cluding certain cases of rare diseases more trendy) projects. For both com- horts tend to collect participant in- and drug responses in which common mon and rare variants with weak formation through closed-end patterns of variation nonetheless may effects, developing smaller-scale alter- questions—that is, by giving partici- be useful in gene discovery (ie CCR natives to very large association stu- pants a range of predetermined an- and RCR).20 Even populations with dies may be a more cost-effective swers and forcing them to choose rare patterns of variation but with approach. Finally, we should be careful among them—and by taking fewer, common phenotypes and common not to extrapolate the broad applic- often less precise measurements of variants (ie CRC) will benefit from ability of common patterns of genetic continuous traits.26 The difficulty, research conducted in other popula- variation and common variants to though, is that such broad, decontex- tions (because of the common var- research on environmental contribu- tualized responses and streamlined iants), although they will not be ideal tors to those same phenotypes, since data collection make it less likely that candidates for association studies the latter will tend to be much more weak or localized effects will be de- using tag SNPs developed in other variable both within and between tected. populations. populations.

www.nature.com/tpj Investments in large-scale genomics project MW Foster and RR Sharp 80

ACKNOWLEDGEMENTS 7 Deloukas P, Bentley D. The HapMap Project 23 Brown MS, Goldstein JL. A receptor- This publication was made possible by and its applications to genetic studies of mediated pathway for cholesterol home- grant numbers ES11174 from the National drug response. Pharmacogenom J 2004; 4: ostasis. Science 1986; 232: 34–47. Institute on Environmental Health Sciences 88–90. 24 Maron DJ, Fazio S, Linton MF. Current and HG02691 the National Human Gen- 8 Zhao HY, Pfeiffer R, Gail MH. Haplotype perspectives on statins. Circulation 2000; analysis in population genetics and associa- 101: 207–213. ome Research Institute. Its contents are tion studies. Pharmacogenomics 2003; 4: 25 Barbour V. UK Biobank: a project in search solely the responsibility of the authors and 171–178. of a protocol? Lancet 2003; 361: 1734– do not necessarily represent the official 9 Pritchard JK, Cox NJ. The allelic architecture 1738. views of the NIEHS, NHGRI, or the National of human disease genes: common disease— 26 Wong MY, Day NE, Luan JA, Chan KP, Institutes of Health. We also are grateful for common variant y or not? Hum Mol Genet Wareham NJ. The detection of gene–envir- comments made by LD Brooks and an 2002; 11: 2417–2423. onment interaction for continuous traits: anonymous reviewer on earlier versions. 10 Chakravarti A. Population genetics—making should we deal with measurement error by sense out of sequence. Nat Genet 1999; 21: bigger studies or better measurement? Int J DUALITY OF INTEREST 56–60. Epidemiol 2003; 32: 51–57. 11 Reich DE, Lander ES. On the allelic spectrum 27 Weisgraber KH, Mahley RW. Human apoli- None declared. of human disease. Trends Genet 2001; 17: poprotein E: the Alzheimer’s disease con- 502–510. nection. FASEB J 1996; 10: 1485–1494. Correspondence should be sent to: 12 Weiss KM, Terwilliger JD. How many dis- 28 Rosendaal FR, Koster T, Vandenbroucke JP, Dr MW Foster, Department of eases does it take to map a gene with SNPs? Reitsma PH. High-risk of thrombosis in Nat Genet 2000; 26: 151–157. patients homozygous for factor-V Leiden Anthropology, University of Oklahoma, 13 Clark AG. Finding genes underlying risk of (activated protein-c resistance). Blood 1995; 455 W. Lindsey, Rm. 505C, Norman, complex disease by linkage disequilibrium 85: 1504–1508. Oklahoma 73019, USA. mapping. Curr Opin Genet Dev 2003; 13: 29 Martinson JJ, Chapman NH, Rees DC, Liu YT, Tel: þ 405 325 2491 296–302. Clegg JB. Global distribution of the CCR5 14 Smith DJ, Lusis AJ. The allelic structure of gene 32-basepair deletion. Nat Genet 1997; Fax: þ 405 325 7386 common disease. Hum Mol Genet 2002; 11: 16: 100–103. E-mail: [email protected] 2455–2461. 30 Hirschhorn JN, Lohmueller K, Byrne E, 15 Zondervan KT, Cardon LR. The complex Hirschhorn K. A comprehensive review of REFERENCES interplay among factors that influence genetic association studies. Genet Med 1 Carlson CS, Eberle MA, Kruglyak L, Nick- allelic association. Nat Rev Genet 2004; 5: 2002; 4: 45–61. erson DA. Mapping complex disease loci in 89–100. 31 McCarthy MI. Progress in defining the whole-genome association studies. Nature 16 Weiss KM, Clark AG. Linkage disequilibrium molecular basis of type 2 diabetes mellitus 2004; 429: 446–452. and the mapping of complex human traits. through susceptibility-gene identification. 2 Goldstein DB, Ahmadi KR, Weale ME, Wood Trends Genet 2002; 18: 19–24. Hum Mol Genet 2004; 13: R33–R41, (Sp. NW. Genome scans and candidate gene 17 Van Den Oord EJCG, Neale BM. Will Iss. 1). approaches in the study of common diseases haplotype maps be useful for finding genes? 32 Thompson JF, Lloyd DB, Lira ME, Milos PM. and variable drug responses. Trends Genet Mol Psychiatry 2004; 9: 227–236. Cholesteryl ester transfer protein pro- 2003; 19: 615–622. 18 Goldstein DB. Islands of linkage disequili- moter single-nucleotide polymorphisms 3 Wilson JF, Weale ME, Smith AC, Gratrix F, brium. Nat Genet 2001; 29: 109–111. in Sp1-binding sites affect transcription Fletcher B, Thomas MG et al. Population 19 Tishkoff SA, Verrelli BC. Patterns of human and are associated with high-density lipo- genetic structure of variable drug response. genetic diversity: implications for human protein cholesterol. Clin Genet 2004; 66: Nat Genet 2001; 29: 265–269. evolutionary history and disease. Ann Rev 223–228. 4 International HapMap Consortium. The In- Genom Hum Genet 2003; 4: 293–340. 33 Newman DL, Hoffjan S, Bourgain C, Abney ternational HapMap Project. Nature 2003; 20 Pe’er I, Beckmann JS. On the applicability of M, Nicolae RI, Profits ET et al. Are common 426: 789–796. a haplotype map to un-assayed populations. disease susceptibility alleles the same in 5 Gabriel SB, Schaffner SF, Nguyen H, Moore Hum Genet 2004; 114: 214–217. outbred and founder populations? Eur J JM, Roy J, Blumenstiel B et al. The structure 21 Collins FS. The case for a US prospective Hum Genet 2004; 12: 584–590. of haplotype blocks in the human genome. cohort study of genes and environment. 34 Morabia A, Cayanis E, Costanza MC, Ross Science 2002; 296: 2225–2229. Nature 2004; 429: 475–477. BM, Bernstein MS, Flaherty MS et al. 6 Johnson GCL, Esposito L, Barratt BJ, Smith 22 Gulcher J, Stefansson K. Population geno- Association between lipoprotein lipase AN, Heward J, Di Genova G et al. Haplotype mics: laying the groundwork for genetic (LPL) gene and blood lipids: a common tagging for the identification of common disease modeling and targeting. Clin Chem variant for a common trait? Genet Epidemiol disease genes. Nat Genet 2001; 29: 233–237. Lab Med 1998; 36: 523–527. 2003; 24: 309–321.

The Pharmacogenomics Journal