bringing gwas finding to clinical use

Genome-wide association studies (GWASs) have revo-lutionized the identification of genomic regions associ-ated with complex diseases. In the past 7years, more than 1,600 publications have identified ~2,000 robust associations with more than 300 complex diseases and traits. These numbers are orders of magnitude greater than those of replicable linkage and candidate gene association findings to date for complex diseases. Initial euphoria at this veritable hoard of reliable associa-tions, totalling nearly 100 or more for some traits1,2, has dimmed somewhat with the recognition that GWAS-defined loci, singly or in aggregate, typically explain only a small proportion of trait heritability3,4. This miss-ing heritability, which is also reflected in the generally small odds ratios and limited predictive value5,6 of these variants, has raised questions about the ultimate appli-cability of these findings to risk prediction in particular and to clinical medicine in general7,8.

Impatience with the seemingly slow pace of imple-menting genomic findings in clinicalcare9,10 may have been fuelled in part by the frequency and enthusiasm with which so many new genome-wide associations have been announced. Optimistic scenarios for the clinical use of genomic information have no doubt contributed to high expectations11,12. Most GWAS findings are less than 3years old, however, whereas clinical adoption of scientific discoveries in general has been estimated to take up to 17years13. In addition, these timeframes are derived from translational experience with proven pre-ventive or therapeutic strategies, however the nature of GWAS-defined associations may substantially prolong this lead time. This is because GWAS findings typically

suggest only that a disease-causing variant resides some-where in a specific genomic region that may include doz-ens of nearby genes and variants. And although some GWAS findings have identified previously unknown pathogenic pathways, such as complement-mediated inflammation in macular degeneration14 and autophagy in Crohns disease15,16, these have yet to find major clinical applications.

Despite the advent of newer technologies, particu-larly sequencing-based methods for identifying disease-associated variants, as discussed below, GWAS-based findings will continue, for sometime, to be essential for designing effective clinical applications. This is in part because of the mass of GWAS data that has already been accumulated that can continue to be mined for addi-tional trait associations and because of the unflagging pace with which new GWAS findings are still being published (FIG.1). More importantly, the continued value of GWAS comes from the efficiency of the method for interrogating low-frequency variants (those with minor allele frequencies (MAFs) of 0.55%) that are initially identified through sequencing, using dense genotyping platforms that capture a large proportion of genomic variation underlying complex traits17,18 (see also Illumina products for whole-genome genotyping and copy number variation analysis).

This Review is intended first to summarize briefly the criticisms and limitations of the GWAS method and how they affect the potential value of GWAS find-ings in clinical care. Although several approaches are available to address these limitations such as impu-tation, fine mapping and evaluation of genegene and

Division of Genomic Medicine, National Human Genome Research Institute, 5635 Fishers Lane, Room 4113, MSC 9305, Rockville, Maryland 20892-9305, USA.e-mail: [email protected]:10.1038/nrg3523Published online 9 July 2013

HeritabilityThe proportion of the total phenotypic variation in a trait that can be attributed to genetic effects.

Odds ratiosA measure of effect size. Defined as the ratio of the odds (that is, the probability of disease divided by 1 minus the probability) of a disease being observed in one group of genotypes and the odds of a disease being observed in another group.

Minor allele frequencies(MAFs). The frequency of the less common allele of a polymorphism. It has a value between 0 and 0.5 and can vary between populations.

Bringing genome-wide association findings into clinical useTeri A.Manolio

Abstract | Genome-wide association studies (GWASs) have been heralded as a major advance in biomedical discovery, having identified ~2,000 robust associations with complex diseases since 2005. Despite this success, they have met considerable scepticism regarding their clinical applicability; this scepticism arises from such aspects as the modest effect sizes of associated variants and their unclear functional consequences. There are, however, promising examples of GWAS findings that will or that may soon be translated into clinical care. These examples include variants identified through GWASs that provide strongly predictive or prognostic information or that have important pharmacological implications; these examples may illustrate promising approaches to wider clinical application.

T R A N S L AT I O N A L G E N E T I C S

Nature Reviews Genetics | AOP, published online 9 July 2013; doi:10.1038/nrg3523 R E V I E W S

NATURE REVIEWS | GENETICS ADVANCE ONLINE PUBLICATION | 1

2013 Macmillan Publishers Limited. All rights reserved

Negative selectionA form of natural selection that suppresses alternative genetic variants in favour of the ancestral type.

Enhancer elementsA regulatory DNA element that usually binds several transcription factors and can activate transcription from a promoter at great distance and in an orientation-independent manner.

geneenvironment interactions these are beyond the scope of this Review. It next explores several illustra-tive examples of how, despite these issues, GWAS find-ings have been or probably will be valuable in disease prediction, biomarker identification and disease subclas-sification, treatment selection and drug dosing. This Review is not intended as a comprehensive nor even as a representative presentation of all the scenarios in which GWAS findings may be useful in clinical care, but rather it is as an exploration of promising findings that can or should be developed further for clinical translation. Finally, it identifies several challenges in the clinical trans-lation of GWAS findings that must be resolved before the GWAS revolution can be integrated into clinicalcare.

Controversy over potential clinical valueCommon versus rare polymorphisms. Criticisms of the GWAS approach include its emphasis on common polymorphisms (that is, those with an MAF of 5% or more) and consequent incomplete tagging of rarer, potentially causal variants, particularly among people of non-European ancestry19,20. Both characteristics tend to obscure associations with variants that have a lower frequency but a potentially higher phenotypic impact, in which much of genomically mediated risk is believed to reside21,22. As genomic technologies improve, detec-tion of associations with variants of lower frequency is increasingly becoming possible, as discussedbelow.

Small effect sizes and missing heritability. A substan-tial surprise that was evident from some of the earli-est GWAS findings for complex diseases was that their associated odds ratios were small (often less than 1.3). In addition, even in combination, they explained very

little (often 10% or less) of the complex trait heritability that is estimated from familial clustering3,23. Small effect sizes present particular problems for clinical applica-tions that might involve disease prediction or risk clas-sification, as very large odds ratios are typically needed to improve predictive accuracy above that provided by clinical information5,7.

Some of this missing heritability has been attrib-uted to imperfect tagging (such that a weak GWAS signal might be indicating a variant that is much more strongly associated and that lies some genomic distance away21, as noted above) or to a lack of ascertainment of potentially important underlying structural variation24,25 (see below). A not uncommon view is that GWAS was a useful early technology that should largely be supplanted by sequencing approaches to detect rarer variants of large effect22. Deeper sequencing-based characteriza-tion of genomic variation17,18, fine mapping26,27, impu-tation28,29 and denser single-nucleotide polymorphism (SNP) arrays are extending the reach of GWAS to ever lower ranges of minor allele frequency. At the same time, the costs and interpretive challenges of whole-genome sequencing currently put it out of reach for all but the smallest association studies. In addition, sequencing is best suited to detecting rare variants underlying con-ditions that have been under strong negative selection, which acts to keep their frequencies low. By contrast, many complex diseases of clinical importance occur later in life when selection is less likely to have been operative. Increases in effect sizes, heritability explained and risk prediction accuracy based on GWAS findings may be hoped for with improved understanding of non-coding variation, in which the vast majority of GWAS associations are localized30.

Assessment of structural variation. Although GWAS arrays are extraordinarily efficient at genotyping SNPs, they are less effective at capturing structural variation, such as insertions, deletions, inversions and copy num-ber variants, which commonly occur in the human genome24. Such variants have been shown to have strong associations with several conditions, particu-larly neurodevelopmental diseases, such as autism and schizophrenia31, and they clearly have a role in inter-individual variations in drug metabolism32 but can be challenging to detect by GWAS methods25. Clinical applications of GWAS findings will thus be limited to the degrees that structural variants have a role in disease diagnosis, prevention and treatment and that they are under-represented in GWASarrays.

Signals in non-coding regions. The high proportion (80% or more) of GWAS-defined signals falling in non-coding regions of the genome30 has frustrated efforts to identify their potential causal roles and has slowed attempts to build on these findings for clinical use. Recent data dem-onstrating strong correlations of GWAS-defined signals with enhancer elements and other regulatory regions33,34 (FIG.2), however, have renewed enthusiasm for exploring functional implications of these signals that at present are but dimly understood.

Figure 1 | Pace of genome-wide association study publications since 2005. The pace of genome-wide association study (GWAS) publications has increased dramatically over 7years and shows no signs of slowing. The figure is based on data from the US National Human Genome Research Institute Catalog of Published Genome-Wide Association Studies.

Nature Reviews | Genetics

2005 2006 2007 2008 2009 2010 2011 2012

Calendar quarter

Tota

l num

ber o

f pub

licat

ions

0

200

400

600

800

1,000

1,200

1,400

1,600

R E V I E W S

2 | ADVANCE ONLINE PUBLICATION www.nature.com/reviews/genetics


Linkage disequilibrium(LD). The nonrandom association of alleles at two or more loci. The pattern of LD in a given genomic region reflects the history of natural selection, mutation, recombination, genetic drift and other demographic and evolutionary forces.

Expression quantitative trait locus(eQTL). A locus at which genetic allelic variation is associated with variation in gene expression levels.

SensitivityThe proportion of true positives that are accurately identified as such (for example, the percentage of cases that are diagnosed using a questionnaire). A sensitivity of 100% means that all cases are correctly identified.

SpecificityThe proportion of true negatives that are classified as negatives. For example, a diagnostic test with a specificity of 100% means that all healthy people have been identified as healthy.

Positive predictive value(PPV). The probability that an individual who tests positive truly has the condition (true positive). A measure of how well a screening or diagnostic test distinguishes true positives from false positives that do not have the disease.

Difficulty dissecting linkage disequilibrium. Linkage disequilibrium (LD) presents another formidable obstacle to the interpretation and application of GWAS findings. Although LD made the GWAS approach possible (by allowing a single tag SNP to serve as a proxy for much of the surrounding genetic variation with which it is inher-ited), parsing such signals to a single causative variant (if one exists) can require substantial additional research. An early GWAS for childhood-onset asthma, for example, identified associations with multiple markers in a 206 kb region on chromosome 17q21 that contains 19 genes35. Expression data implicated a gene of unknown function, ORM1-like 3 (ORMDL3), as potentially responsible for the asthma association signal because the SNPs that underlie the GWAS signal were most strongly associated with ORMDL3 expression in relevant immune tissues35. Subsequent GWASs, however, showed that the strong-est association signal in this region lies in the nearby gasdermin B (GSDMB) gene and also implicated by LD both ORMDL3 and the neighbouring GSDMA gene. Furthermore, the expression quantitative trait locus (eQTL) in the region that had been previously associated with ORMDL3 expression was found also to influence the expression of GSMDB36,37. This signal has yet to be iso-lated to one of these plausible and tightly linked genes in the 17q21 locus, delaying the understanding of the poten-tial role, if any, of these genes in asthma pathogenesis and the potential application of this finding to diagnostic or therapeutic decisionmaking.

LD in GWAS-defined associations can sometimes, however, be used to implicate or to exclude strong candidate genes. An example is the case of the strong GWAS associations with inflammatory bowel disease that were found to be clustered around interleukin 23 receptor (IL23R) and not in the neighbouring IL12RB2 locus16,38,39. Specific inhibitors of IL23 are currently under development and clinical investigation.

Assessment and prediction of risk. Despite the numerous terms used to describe and to quantify disease risk, few are interpreted in the same way by each patient or clini-cian, and none captures the concept entirely. Several of these, such as sensitivity, specificity and positive predictive value, have repeatedly been described (for example, see REF.40). The metric most often used for assessing the added value of a predictive factor is the increase in area under the receiver operator characteristic (ROC) curve41. The ROC curve plots the true-positive fraction (sensitiv-ity) of various cut points of a continuous measure, such as fasting glucose to detect a disease such as diabetes, against their false-positive fractions (1 minus the speci-ficity). The area under this curve (AUC) is frequently used to describe the ability of a measure to discriminate between those with and without disease. Ideal classifiers would have a high true-positive rate, even at low rates of false positives and would produce almost square curves, whereas poor classifiers would increase true positives at about the same rate as false positives, looking more like a diagonal curve. AUC thus ranges from 0.5 (providing no discrimination between those with and without the condition) to 1.0 (perfect discrimination).

For many common conditions, even multiple GWAS-defined variants considered together often fail to increase the AUC much over clinical risk factors42,43. This is because, as noted above, SNPs associated with common diseases typically have small odds ratios (

Major histocompatibility complex(MHC). A large complex of tightly linked genes on human chromosome 6, many of which are involved in the immune response. The human leukocyte antigen genes are located within the MHC.

odds ratios of 200 or more may be necessary for meaning-ful risk prediction in individuals44 (FIG.3a), although the predictive value of multiple genetic variants considered jointly may improve with identification of a larger number of SNPs that explain a greater proportion of variance44,45 (FIG.3b). Even for a condition such as age-related macular degeneration, however, in which the three major variants produce an AUC of 0.79, the correct detection of 80% of cases results in a false-positive rate of >40%5.

As noted above, currently known variants typically explain too little variability in disease occurrence to be of much predictive value, but it can be anticipated that as more risk variants are identified, the predictive value of cumulative genotype scores will increase5. Care must also be taken when evaluating the predictive value of genetic models as they often perform best in the population from which they were developed, and their performance can be affected by differences between populations in genotype frequencies, phenotypic effect sizes and disease incidence46.

Clinically relevant GWAS findingsAreas in which GWAS findings are leading to clinical applications include risk prediction, disease classification, drug development and drug toxicity. Illustrative examples first identified by GWAS or that capitalize on the strengths of this method are discussed in some detail below, and other examples on the horizon are briefly listed thereafter.

Risk prediction. Clinical scenarios in which GWAS findings might find ready application in risk prediction

would include those in which early (before disease onset) identification of high-risk individuals is impor-tant owing to the availability of a treatment that either prevents disease entirely or that is most effective in improving outcome when it is instituted before clinical abnormalities are detected or before a firm clinical diag-nosis can be made. Type1 diabetes mellitus is a condi-tion of typically childhood onset that may meet these criteria after promising immune-modulating therapies have been fully developed. It results from immune-mediated destruction of the insulin-producing -islet cells of the pancreas. It causes substantial morbidity and mortality, requires life-long insulin treatment and is highly heritable: it has an estimated sibling relative risk of 15 (REFS47,48). By the time that the disease is detected clinically, the -cells are almost completely destroyed, and no known treatment can restore them. Prevention by immune-modulating therapies is thus likely to be the best near-term approach to reducing the burden of this disease, and such treatments are under active development49.

Although nearly half of the sibling relative risk of type1 diabetes is explained by associations within the major histocompatibility complex (MHC) locus, which were identified well before the GWAS era, GWAS inves-tigations have increased the number of loci associated with this condition to over 50 (REFS48,50; see also the Type1 Diabetes Consortium website). Together with the MHC, these loci are estimated to explain two-thirds to three-quarters of familial risk and possibly more if famil-ial risk is over-estimated, as has been suggested48. This is

Figure 3 | Use of odds ratios in risk prediction. a | Area under receiver operator characteristic (ROC) curve for predicting a hypothetical condition with variants carrying modest (1.5), sizeable (10) and large (50) odds ratios, showing false-positive fractions at 80% sensitivity (dotted line; fractions are >75%, >25% and 40%. The figure is reproduced, with permission, from REF.5 (2009) Jakobsdottir et al.


0 0.2 0.4 0.6 0.8 1.00

0.2

0.4

0.6

0.8

1.0

False-positive frequency

True

-pos

itiv

e fr

eque

ncy

0 0.2 0.4 0.6 0.8 1.00

0.2

0.4

0.6

0.8

1.0a b

False-positive frequency

True

-pos

itiv

e fr

eque

ncy

1.5

10

50

Three-gene modelCFH + LOC modelCFH modelLOC modelC2 modelNull model

R E V I E W S



far greater than the proportion of heritability explained for any other complex disease, making type1 diabetes unique in this regard. Predictive models incorporating all of these loci have AUCs close to 0.9 (REFS51,52), the highest known for any condition.

Although highly predictive biomarkers are also available in the form of islet cell autoantibodies53, screening the population at large for such a rare condi-tion using genetic variation or biomarkers is not fea-sible, as false positives would vastly outnumber true positives. Targeting autoantibody screening, however, to those with a known genetic predisposition measured by a genetic risk score or positive family history, or both, narrows the screening population and substantially increases positive predictive value and yield48. When effective immune-modulating therapies become avail-able, type1 diabetes would meet the criteria that are noted above for effective genetic risk prediction: early identification is important owing to effective preven-tive or disease-altering therapies, heritability is high, identified genetic loci explain a large portion of the risk, and genotyping scores show high discriminative value53. The only weakness in type1 diabetes as a model for clinical application at present is in the availability of effective therapies; although immune modulation has shown promise in animal models and transient preser-vation of -cell function has been achieved in human studies54, definitive therapies have yet to be developed. When they become available, combining a genetic risk score with family history (which is often predictive above and beyond identified variants55) could be used in clinical care to identify a high-risk group of individuals in whom interventions could be considered. At present genetic findings are being used to select participants with high-risk type1 diabetes alleles for prevention and treatment studies and to exclude those with protective alleles, which is an important prelude to use of these variants in clinicalcare.

Another potential application of GWAS-defined risk variants is in reclassification of subjects who are initially assigned to a low-risk category on the basis of established risk factors (such as age, sex and hypertension for car-diovascular disease risk). Subjects can be reclassified to a higher or lower category after the inclusion of an addi-tional measure such as a genetic variant. Reclassification can be particularly important for clinical settings in which risk thresholds have been defined and guidelines for intervention have been developed, with the recog-nition that such thresholds, although useful clinically, are by nature arbitrary and debatable7. In a cohort that had been classified for risk of cardiovascular events, a combination of genetic variants was used to develop a genotype score for reclassification56. As a result, of the 26% of the study cohort that was initially estimated to be at intermediate risk, 35% (that is, 9% of the total cohort) were reclassified into low- or high-risk categories; this showed significant improvement in reclassification sta-tistics compared to established risk factors alone (FIG.4). Reclassification can be of particular value in clinical deci-sion making in people defined as intermediate risk by standard guidelines, although actual reclassification typi-cally involves only minor increments in risk score that shift borderline subjects between adjacent categories7.

Disease classification. Genetic variants can also help to identify useful biomarkers, which can themselves be used as risk or prognostic indicators in certain circum-stances. Genetic or biomarker risk information may be particularly valuable when its assessment is easier, cheaper, more available, more reliable and/or more sen-sitive than other clinical indicators of risk; these are con-ditions that to date have rarely been met, in part because of the small effect sizes and limited predictive value of GWAS findings discussed earlier.

Monogenic forms of diabetes often known collec-tively as maturity-onset diabetes of the young (MODY)

Figure 4 | Reclassification of cardiovascular risk based on genotype score. A cohort of 4,232 people was classified into low (1020%; red) 10-year risk of cardiovascular disease before and after applying genotype risk score. a | Before incorporation of the genotype score, standard risk factors define 3,739 people as low-risk, 370 as intermediate-risk and 123 as high-risk. b | Of the 3,739 people initially defined as low-risk, 68 are defined by genotyping score as intermediate risk; of the 370 people initially defined as intermediate-risk, 61 are defined as low-risk, and 34 are defined as high-risk; and of the 123 initially defined as high risk, genotyping defines 23 as intermediate-risk. c | After incorporation of the genotype score, reclassification statistics and outcome data show improvement in classification: 3,732 at low-risk, 366 at intermediate-risk and 134 at high-risk. The figure is based on data from REF.56.


After

Within

Before

c

b

a1233739 370

1003671 2756168

1343732 366

34

23

R E V I E W S



Missense variantA variant that results in the substitution of an amino acid in a protein.

Splice variantA variant, usually found at the intronexon boundary, that alters the splicing of an exon to its surrounding exons.

account for 12% of all diabetes cases, but they are com-monly misdiagnosed as type1 or type2 diabetes57. These typically non-insulin-dependent disorders of -cell function are dominantly inherited but are clinically heterogeneous in regard to age of onset, severity and causative genes and mutations. One of the most com-mon forms is due to mutations in HNF1 homeobox A (HNF1A)58,59. Identification of mutations in this gene has important implications for patients and their rela-tives, as many patients with HNF1A-MODY are better managed with low-dose sulphonylureas than with met-formin or insulin, and their first-degree relatives are at a 50% risk of inheriting the same mutation. Genetic testing remains difficult, however, because of the high cost and limited availability of tests, the low prevalence of the condition (and hence low predictive value), the lack of familiarity with available genetic tests and their interpretation among clinicians, and highly debated ethical concerns about the risks of genetic information. Patients are typically selected for molecular testing on the basis of atypical course or parental history, but such criteria tend to be poor discriminators, and many cases are missed57. As -cell dysfunction is present in all forms of diabetes, availability of a biomarker that is specific to non-pancreatic manifestations may be most useful in indicating the presence of HNF1A mutations59.

Although mutations in HNF1A that underlie MODY were first identified through linkage studies rather than through GWASs, the GWAS approach also demon-strated associations of common variants in HNF1A with levels of C-reactive protein (CRP)60, which is a potential biomarker of the condition. HNF1A encodes the transcription factor HNF1, binding sites for which are present in the CRP promoter. Variants resulting in loss of HNF1 binding lead to loss of CRP expression61. CRP levels are in fact markedly lower in patients with HNF1A-MODY than in other forms of diabetes and are indeed significantly lower than in non-diabetics59,62. A recent multi-centre meta-analysis demonstrated AUCs of 0.79 to 0.91 across centres for distinguishing HNF1A-MODY from young-adult-onset type2 diabe-tes with high sensitivity (78%) and specificity (80%)62. Combining low CRP and the clinical criterion of age at diagnosis

RhabdomyolysisThe rapid breakdown of skeletal muscle tissue due to injury, drugs, toxins or metabolic disease. This leads to electrolyte release and high concentrations of myoglobin in plasma and urine that are toxic to the kidneys and can cause renal failure and death.

MethotrexateA folic acid antagonist used as a chemotherapeutic and immunosuppressant drug.

the GWAS approach, in which a common SNP (MAF for rs6051702 = 19%) tags less common functional vari-ants. It also demonstrates the large effect sizes that may be detected for pharmacogenetic variants, as the combi-nation of these two functional alleles explained 1929% of the variability in haemoglobin decline across three major ethnic groups63. The potential clinical applica-tions of these recently identified associations and causal relationships in prognostication and drug development have yet to be explored, but they could be substantial. As the genetic basis of action and toxicity of other drugs becomes better understood through GWAS of treatment response and other techniques, it might be anticipated that similar loss-of-function variants could be attractive targets for pharmacological manipulation to avoid or to reduce toxicity.

Drug toxicity. Pharmacological inhibitors of 5-hydroxy-3-methylglutaryl co-enzyme A (HMG-CoA) reductase, collectively known as statins, are the most effective known agents for reducing low-density lipoprotein cho-lesterol and are widely and increasingly prescribed67. An important adverse effect of these drugs that can lead to non-adherence to therapy is myopathy, which is char-acterized by muscle pain that occurs in 15% of treated patients68,69, weakness and elevated muscle enzyme lev-els. Rarely, this myopathy can progress to rhabdomyolysis, renal failure and death. The mechanism is poorly under-stood and may involve induction of pro-apoptotic path-ways or immune-mediated necrotizing myopathy70. Risk is increased with advanced age, low body mass, higher doses, concomitant drug administration and conditions that interfere with hepatic metabolism. Risk of myopathy

is highest with simvastatin, which is the most commonly prescribed member of this drug class and the third most commonly prescribed drug in the United States in 2010 (REFS69,70).

In 2008, a GWAS was carried out for 85 patients with myopathy and 90 matched controls from the 12,000-patient-strong Study of the Effectiveness of Additional Reductions in Cholesterol and Homocysteine (SEARCH) trial. This study detected an association with a single non-coding SNP in solute carrier organic anion transporter family member 1B1 (SLCO1B1; which encodes an organic anion transporting polypeptide (OATP1B1) that has been shown to regulate the hepatic uptake of statins)71. This non-coding SNP is in nearly complete LD with a nonsynonymous SNP (r 2 = 0.97) that encodes a valine-to-alanine substitution and that confers a 4.5-fold increased risk of myopathy per copy of the minor C allele (FIG.5). This variant and others in this highly polymorphic gene have been demonstrated to reduce hepatic uptake and to increase serum levels of statins7174. Observed areas under the concentration time curve, which is a measure of invivo simvastatin exposure, are 2.2-fold greater in rs4149056 CC homozygotes com-pared with TT homozygotes69. Roughly 25% of the popu-lation carry the C allele, either alone or in combination with other SLCO1B1 variants, all of which appear to affect simvastatin levels similarly. The effects of these variants differ quantitatively across the seven clinically available statins and are greatest for simvastatin75. Although myo-pathy risk is strongly dose-dependent, the effect of dose also appears to be greatest for simvastatin.

Interestingly, GWASs have also shown SLCO1B1 variants to be associated with reduced clearance of methotrexate and increased gastrointestinal toxicity in children with acute lymphoblastic leukaemia76,77. Here again effect sizes are large: SLCO1B1 variants explain 10% of methotrexate clearance (only dosage regimen explained more) and carry a 15-fold increased risk of toxicity. GWASs have also shown SLCO1B1 variants to be associated with bilirubin levels78, presumably reflecting the role of this transporter in hepaticuptake.

Guidelines for managing simvastatin-induced myo-pathy risk in the context of SLCO1B1 genotyping have recently been published69. Although genotyping is not currently required in conjunction with treatment, it is commercially available as a single assay or in combi-nation with other pharmacokinetic variants (see The Pharmacogenomics Knowledgebase). Heterozygotes for the C allele are expected to have an intermediate activity phenotype and to be at intermediate risk of myopathy, whereas homozygotes have low activity and are at highest risk (although other variants, concomi-tant drugs and other factors may modify these expecta-tions). The Clinical Pharmacogenomics Implementation Consortium has recommended dose reductions or alter-native therapy in C allele carriers69. Serial monitoring of muscle enzymes in CC homozygotes may be beneficial even at lower doses, although this remains to be proved69.

With the availability of clinical guidelines, genotyping of SLCO1B1 before initiating simvastatin therapy will probably gain broader clinical usage. Pilot studies are

Figure 5 | Risk of myopathy in chronic simvastatin use. Estimated cumulative risk of myopathy associated with high-dose simvastatin by solute carrier organic anion transporter family member 1B1 (SLCO1B1) rs4149056 genotype. The figure is modified from REF.71 (2008) Massachusetts Medical Society.

Years since starting 80 mg of simvastatin

Cum

ulat

ive

perc

enta

ge o

f pat

ient

s w

ho h

ave

had

a m

yopa

thy

0

5

10

15

20

0 1 2 3 4 5 6

CC genotypeCT genotypeTT genotype


R E V I E W S



Decision support toolsSoftware tools providing intelligently filtered and appropriately timed medical information specific to a given patient to aid in clinical decision making at the point of care. Examples include computerized alerts of potential adverse effects of a proposed treatment or reminders of overdue screening tests.

underway to examine effective modes of implementa-tion, acceptance by clinicians and patients, and impact on adherence and lipid-lowering efficacy69. It is an instructive example because the genetically mediated risk is not readily identified by other means, the vari-ants are common and the effects are large, and treatment modifications in the form of increased monitoring or alternative therapy are available. What remains to be tackled before implementation can be effective is wider availability of clinically licensed testing, more compelling evidence of efficacy and smoother integration of phar-macogenetic information into clinical decision making. Smoother integration is probably best achieved through integration with electronic medical records and decision support tools that alert clinicians when they are about to prescribe simvastatin in a patient at genetically increased risk of myopathy69. Such tools are currently under devel-opment and testing and are likely to be adaptable to a host of pharmacogenetic indications.

This SLCO1B1 finding demonstrates that GWASs for treatment response are powerful approaches for identifying variants signalling increased risk of adverse effects and that these variants may become indications for closer monitoring or alternative therapies. Another interesting recent example is in erectile dysfunction fol-lowing radiation therapy, in which initial models sug-gest that GWAS-defined SNPs substantially increase the AUC for predicting this common side effect of treatment79. Similarly, a recent report of antipsychotic-drug-induced extreme weight gain associatedwith mel-anocortin 4 receptor (MC4R) variants may also point the way towards alternative treatments80.

Other findings on the horizon. This Review focuses on in-depth exploration of fairly mature and illustrative examples of GWAS findings that are ready for translation in four specific areas, but several others that are under development are likely to be applied soon. Findings on the clinical horizon that relate to risk prediction include recent reclassification studies in breast cancer81 and osteoporotic fractures82, demonstrating increased dis-crimination of clinical risk with the inclusion of genomic markers. These findings are similar to those discussed earlier for coronary disease, whereas predictive models incorporating GWAS findings are now being formally tested for their impact on coronary-disease-related outcomes in a randomized clinical trial83. Other cogent examples of GWAS findings in risk prediction include complement-related variants in predicting macular degeneration5 and APOL1 variants predicting hyperten-sive renal disease84,85. Given the limitations of genomic risk measures in the prediction of almost all diseases except type1 diabetes, in the short term, predictive vari-ants are likely to be used to identify people at particular genetic risk for targeting or intensifying interventions proven effective in diseasecases.

In the area of disease classification and subtyping, GWAS findings have recently been used to develop more refined molecular taxonomies for breast cancer86, lung cancer87 and inflammatory bowel disease88. More recent GWAS-augmented efforts that relate to drug develop-ment include IL23R inhibition, which has been identi-fied as a potential therapeutic strategy for inflammatory bowel disease through examination of LD patterns in some of the earliest GWAS38. Further, because IL23R has

Table 1 | Characteristics of GWAS findings that make them readily translatable to clinical care

Application Key characteristics Example

Risk prediction Heritability is high T1DM loci

Large proportion of heritability is explained

Genotyping can be targeted to high-risk group (such as that defined by positive family history)

Genotyping scores substantially increase predictive value

Early detection is important

Preventive strategies are available

Disease subtyping or classification

Clinical syndrome has multiple subtypes with varying prognosis or response to treatment aetiologies

CRP (or HNF1A typing) for MODY

Subtypes are not readily discernible by clinical examination

Compared to other indicators, assay of biomarker (genetic or non-genetic) is easier, cheaper, more available, more reliable and more sensitive

Subtyping affects treatment selection

Subtyping identifies increased risk in family members

Drug development Loss of function variants have a useful biological effect ITPA variants and ribavirin toxicity

Drug toxicity Variants are common SLCO1B1 variants and statins

Effect sizes are large

Alternative drugs or increased monitoring are available

CRP, C-reactive protein; GWAS, genome-wide association study; HNF1A, HNF1 homeobox A; ITPA, inosine triphosphatase; MODY, maturity-onset diabetes of the young; SLCO1B1, solute carrier organic anion transporter family member 1B1; T1DM, type 1 diabetes mellitus.

R E V I E W S



multiple variants, some associated with increased risk and some with decreased risk, identification of specific variants in a given patient may eventually be useful in determining the appropriateness of IL23specific therapy. Complement-related findings are also being therapeuti-cally explored for macular degeneration89. In the area of drug toxicity, the GWAS approach has recently been used to identify variants predisposing to paclitaxel-induced peripheral neuropathy90 and aspirin-exacerbated res-piratory disease91. Additional clinical applications are expected as the thousands of GWAS results from the past 7years are more deeply explored and as additional GWAS are conducted. As genomic approaches to clinical care become more commonplace and disseminate beyond major medical centres92,93, it may be expected that the pace of identifying clinically relevant GWAS findings and translating them to the clinic will accelerate.

Conclusions: pathways to translationThese four examples in risk prediction, disease clas-sification, drug development and drug selection dem-onstrate the potential for rapid application of selected GWAS findings in clinical care. Finding unifying char-acteristics among these four that promote successful translation is challenging (TABLE1). Indeed, translational potential may be driven as much by the clinical scenario (such as importance of early detection, availability of alternative treatments and accessibility of genotyping) as by the variants involved or the relevant genetic architec-ture. The examples cited here tend to share some char-acteristics, such as moderate allele frequencies and large effect sizes, but these may be more important in permit-ting initial identification of the association than for its translation. The HNF1A-MODY variants, for example, have low allele frequencies, whereas allele frequency

would largely be irrelevant were an effective treatment for ribavirin-induced anaemia to be found for carriers of the ancestral ITPAallele.

A key component in translating GWAS findings is linking initial genomic discoveries with clinicians who appreciate the clinical dilemmas that the findings could address, such as the importance of early prediction in type1 diabetes, molecular subtyping of type2 diabetes or seemingly unpredictable drug side effects. Although this is probably best done through the clinicians who define phenotypes and assemble samples for a given GWAS, their expertise may not extend to the mecha-nistic studies, therapeutic development, preventive approaches or evidence generation needed to produce a GWAS-initiated improvement in care. Having such improvements actually implemented clinically requires a panoply of additional capabilities, including: rapid, low-cost genotyping; point-of-care educational infor-mation and decision support tools; agreed-on evidence standards and practice guidelines; and institutional willingness to support the infrastructure needed for implementation92,93.

As with translation of many genomic discoveries, translating this work into direct health benefits will require interaction among a wide array of biomedical dis-ciplines, including genomics, molecular biology, clinical medicine, pharmacology, bioinformatics, dissemination and implementation research, and clinician education94. The first step, however, needs to be recognition that some GWAS findings go beyond simple genomic signposts and may yield themselves fairly directly to patient care appli-cations. The examples cited here not only demonstrate this but suggest characteristics that could be sought when mining the vast and growing array of GWAS discoveries for translatable results.

1. Lango Allen, H. etal. Hundreds of variants clustered in genomic loci and biological pathways affect human height. Nature 467, 832838 (2010).

2. Teslovich,T.M. etal. Biological, clinical and population relevance of 95 loci for blood lipids. Nature 466, 707713 (2010).

3. Manolio,T.A. etal. Finding the missing heritability of complex diseases. Nature 461, 747753 (2009).

4. Goldstein, D.B. Common genetic variation and human traits. N.Engl. J.Med. 360, 16961698 (2009).

5. Jakobsdottir,J., Gorin,M.B., Conley,Y.P., Ferrell,R.E. & Weeks,D.E. Interpretation of genetic association studies: markers with replicated highly significant odds ratios may be poor classifiers. PLoS Genet. 5, e1000337 (2009).This is a review of the predictive ability of strongly associated GWAS-defined SNPs in four diseases, demonstrating that high odds ratios (>50) are needed to improve prediction.

6. Aschard,H. etal. Inclusion of genegene and geneenvironment interactions unlikely to dramatically improve risk prediction for complex diseases. Am. J.Hum. Genet. 90, 962972 (2012).

7. Manolio, T.A. Genome-wide association studies and disease risk assessment. N.Engl. J.Med. 363, 166176 (2010).

8. Lopes,M.C., Zeggini,E. & Panoutsopoulou,K. Do genome-wide association scans have potential for translation? Clin. Chem. Lab. Med. 50, 255260 (2011).

9. Evans,J.P., Meslin,E.M., Marteau,T.M. & Caulfield,T. Deflating the genomic bubble. Science 331, 861862 (2011).

10. Varmus, H. Ten years on the human genome and medicine. N.Engl. J.Med. 362, 20282029 (2010).

11. Dulbecco, R. A turning point in cancer research: sequencing the human genome. Science 231, 10551056 (1986).

12. Collins, F. Shattuck lecture: medical and societal consequences of the human genome project. N.Engl. J.Med. 341, 2837 (1999).

13. Committee on Quality of Health Care in America, Institute of Medicine. Crossing the Quality Chasm: A New Health System for the 21st Century (National Academy Press, 2001).

14. Klein,R.J. etal. Complement factor H polymorphism in age-related macular degeneration. Science 308, 385389 (2005).

15. Rioux,J.D. etal. Genome-wide association study identifies new susceptibility loci for Crohn disease and implicates autophagy in disease pathogenesis. Nature Genet. 39, 596604 (2007).

16. Budarf,M.L., Labb,C., David,G. & Rioux,J.D. GWA studies: rewriting the story of IBD. Trends Genet. 25, 137146 (2009).

17. Genomes Project Consortium. A map of human genome variation from population-scale sequencing. Nature 467, 10611073 (2010).

18. Tennessen, J.A. et al. Evolution and functional impact of rare coding variation from deep sequencing of human exomes. Science 337, 6469 (2012).This is a summary report of rare variation identified in the US National Institutes of Health (NIH) Heart, Lung and Blood Institute Exome Sequencing Project for 15,585 human protein-coding genes in 2,440 individuals of European and African ancestry.

19. McClellan,J. & King,M.C. Genetic heterogeneity in human disease. Cell 141, 210217 (2010).

20. Bustamante,C.D., Burchard,E.G. & de la Vega,F.M. Genomics for the world. Nature 475, 163165 (2011).

21. Spencer,C., Hechter,E., Vukcevic,D. & Donnelly,P. Quantifying the underestimation of relative risks from genome-wide association studies. PLoS Genet. 7, e1001337 (2011).

22. Cirulli,E.T. & Goldstein,D.B. Uncovering the roles of rare variants in common disease through whole-genome sequencing. Nature Rev. Genet. 11, 415425 (2010).

23. Maher, B. Personal genomes: the case of the missing heritability. Nature 456, 1821 (2008).

24. Tuzun,E. etal. Fine-scale structural variation of the human genome. Nature Genet. 37, 727732 (2005).

25. McCarroll, S.A. Extending genome-wide association studies to copy-number variation. Hum. Mol. Genet. 17, R135R142 (2008).

26. Chung,C.C. & Chanock,S.J. Current status of genome-wide association studies in cancer. Hum. Genet. 130, 5978 (2011).

27. Travers,M.E. & McCarthy,M.I. Type 2 diabetes and obesity: genomics and the clinic. Hum. Genet. 130, 4158 (2011).

28. Mohlke,K.L. & Scott,L.J. What will diabetes genomes tell us? Curr. Diab. Rep. 12, 643650 (2012).

29. Servin,B. & Stephens,M. Imputation-based analysis of association studies: candidate regions and quantitative traits. PLoS Genet. 3, e114 (2007).

30. Hindorff, L. A. etal. Potential etiologic and functional implications of genome-wide association loci for human diseases and traits. Proc. Natl Acad. Sci. USA 106, 93629367 (2009).An overview of functional annotations for GWAS-defined SNPs in the first 3years of experience is presented here, and it demonstrates that a high proportion (>80%) of associations fall in non-coding regions.

R E V I E W S



31. Girirajan,S. etal. Phenotypic heterogeneity of genomic disorders and rare copy-number variants. N.Engl. J.Med. 367, 13211331 (2012).

32. He,Y., Hoskins,J.M. & McLeod,H.L. Copy number variants in pharmacogenetic genes. Trends Mol. Med. 17, 244251 (2011).

33. Ernst, J. etal. Mapping and analysis of chromatin state dynamics in nine human cell types. Nature 473, 4349 (2011).

34. The ENCODE Project Consortium. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 5774 (2012).This is the lead paper of 30 coordinated papers describing ENCODE findings of functional DNA sequences related to transcription, transcription factor association, chromatin structure and histone modification.

35. Moffatt,M.F. etal. Genetic variants regulating ORMDL3 expression contribute to the risk of childhood asthma. Nature 448, 470473 (2007).

36. Zhang,Y., Moffatt,M.F. & Cookson,W.O. Genetic and genomic approaches to asthma: new insights for the origins. Curr. Opin. Pulm. Med. 18, 613 (2012).

37. Ober,C. & Yao,T.C. The genetics of asthma and allergic disease: a 21st century perspective. Immunol. Rev. 242, 1030 (2011).

38. Duerr,R.H. etal. A genome-wide association study identifies IL23R as an inflammatory bowel disease gene. Science 314, 14611463 (2006).

39. Sarin,R., Wu,X. & Abraham,C. Inflammatory disease protective R381Q IL23 receptor polymorphism results in decreased primary CD4+ and CD8+ human T-cell functional responses. Proc. Natl Acad. Sci. USA 108, 95609565 (2011).

40. Craig,D.W. etal. Assessing and managing risk when sharing aggregate genetic variant data. Nature Rev. Genet. 12, 730736 (2011).

41. Kraft,P. etal. Beyond odds ratios communicating disease risk based on genetic profiles. Nature Rev. Genet. 10, 264269 (2009).

42. Cornelis,M.C. etal. Joint effects of common genetic variants on the risk for type 2 diabetes in U.S. men and women of European ancestry. Ann. Intern. Med. 150, 541550 (2009).

43. van der Net,J.B., Janssens,A.C., Sijbrands,E.J. & Steyerberg,E.W. Value of genetic profiling for the prediction of coronary heart disease. Am. Heart J. 158, 105110 (2009).

44. Ware J.H. The limitations of risk factors as prognostic tools. N.Engl. J.Med. 355, 26152617 (2006).

45. Kraft,P. & Hunter,D.J. Genetic risk predictionare we there yet? N.Engl. J.Med. 360, 17011703 (2009).

46. Wray,N.R., Goddard,M.E. & Visscher,P.M. Prediction of individual genetic risk of complex disease. Curr. Opin. Genet. Dev. 18, 257263 (2008).

47. Risch N. Assessing the role of HLA-linked and unlinked determinants of disease. Am. J.Hum. Genet. 40, 114 (1987).

48. Polychronakos,C. & Li,Q. Understanding type 1 diabetes through genetics: advances and prospects. Nature Rev. Genet. 12, 781792 (2011).This is a Review of allelic architecture of genetic susceptibility to type1 diabetes, based on GWASs, fine mapping and functional studies, and the potential for genetic prediction of T1D risk.

49. Chatenoud,L., Warncke,K. & Ziegler,A.G. Clinical immunologic interventions for the treatment of type1 diabetes. Cold Spring Harb. Perspect. Med. 2, a007716 (2012).

50. Bradfield,J.P. etal. A genome-wide meta-analysis of six type1 diabetes cohorts identifies multiple associated loci. PLoS Genet. 7, e1002293 (2011).

51. Jostins,L. & Barrett,J.C. Genetic risk prediction in complex disease. Hum. Mol. Genet. 20, R182R188 (2011).

52. Clayton, D.G. Prediction and interaction in complex disease genetics: experience in type1 diabetes. PLoS Genet. 5, e1000540 (2009).

53. Bingley, P.J. Clinical applications of diabetes antibody testing. J.Clin. Endocrinol. Metab. 95, 2533 (2010).

54. Gallagher,M.P., Goland,R.S. & Greenbaum,C.J. Making progress: preserving cells in type1 diabetes. Ann. NY Acad. Sci. 1234, 119134 (2011).

55. Dunlop,M.G. etal. Cumulative impact of common genetic variants and other risk factors on colorectal cancer risk in 42 103 individuals. Gut 62, 871881 (2013).

56. Kathiresan,S. etal. Polymorphisms associated with cholesterol and risk of cardiovascular events. N.Engl. J.Med. 358, 12401249 (2008).

57. Shields,B.M. etal. Maturity-onset diabetes of the young (MODY): how many cases are we missing? Diabetologia 53, 25042508 (2010).

58. Shepherd,M. etal. Predictive genetic testing in maturity-onset diabetes of the young (MODY). Diabet Med. 18, 417421 (2001).

59. Owen,K.R. etal. Assessment of high-sensitivity C-reactive protein levels as diagnostic discriminator of maturity-onset diabetes of the young due to HNF1A mutations. Diabetes Care 33, 19191924 (2010).

60. Reiner,A.P. etal. Polymorphisms of the HNF1A gene encoding hepatocyte nuclear factor-1 are associated with C-reactive protein. Am. J.Hum. Genet. 82, 11931201 (2008).One of two initial GWASs demonstrating association between HNF1A and C-reactive protein levels is presented here.

61. Toniatti,C., Demartis,A., Monaci,P., Nicosia,A. & Ciliberto,G. Synergistic trans-activation of the human C-reactive protein promoter by transcription factor HNF-1 binding at two distinct sites. EMBO J. 9, 44674475 (1990).

62. Thanabalasingham,G. etal. A large multi-centre European study validates high-sensitivity C-reactive protein (hsCRP) as a clinical biomarker for the diagnosis of diabetes subtypes. Diabetologia 54, 28012810 (2011).

63. Fellay,J. etal. ITPA gene variants protect against anaemia in patients treated for chronic hepatitis C. Nature 464, 405408 (2010).This is the first GWAS to demonstrate association between ITPA and ribavirin-induced anaemia.

64. Asselah,T., Pasmant,E. & Lyoumi,S. Unraveling the genetic predisposition of ribavirin-induced anaemia. J.Hepatol. 53, 971973 (2010).

65. Thompson,A.J. etal. Variants in the ITPA gene protect against ribavirin-induced hemolytic anemia and decrease the need for ribavirin dose reduction. Gastroenterology 139, 11811189 (2010).

66. Hitomi,Y. etal. Inosine triphosphate protects against ribavirin-induced adenosine triphosphate loss by adenylosuccinate synthase function. Gastroenterology. 140, 13141321 (2011).This functional study demonstrates that ITP substitutes for GTP for use by human adenylosuccinate synthase, thereby bypassing the ribavirin-induced depletion of GTP and subsequent haemolysis.

67. Carroll,M.D., Kit,B.K., Lacher,D.A., Shero,S.T. & Mussolino,M.E. Trends in lipids and lipoproteins in US adults, 19882010. JAMA 308, 15451554 (2012).

68. Thompson,P.D., Clarkson,P. & Karas,R.H. Statin-associated myopathy. JAMA 289, 16811690 (2003).

69. Wilke, R.A. etal. The clinical pharmacogenomics implementation consortium: CPIC guideline for SLCO1B1 and simvastatin-induced myopathy. Clin. Pharmacol. Ther. 92, 112117 (2010).This is a review of the impact of SLCO1B1 variants on patient response to statins and consensus guidelines for reducing the risk of simvastatin myopathy in variant carriers.

70. Mammen,A.L. & Amato,A.A. Statin myopathy: a review of recent progress. Curr. Opin. Rheumatol. 22, 644650 (2010).

71. SEARCH Collaborative Group. SLCO1B1 variants and statin-induced myopathya genomewide study. N.Engl. J.Med. 359, 789799 (2008).

72. Ghatak,A., Faheem,O. & Thompson,P.D. The genetics of statin-induced myopathy. Atherosclerosis 210, 337343 (2010).

73. Niemi,M., Pasanen,M.K. & Neuvonen,P.J. Organic anion transporting polypeptide 1B1: a genetically polymorphic transporter of major importance for hepatic drug uptake. Pharmacol. Rev. 63, 157181 (2011).

74. Voora,D. etal. The SLCO1B1*5 genetic variant is associated with statin-induced side effects. J.Am. Coll. Cardiol 54, 16091616 (2009).

75. Maggo,S.D., Kennedy,M.A. & Clark,D.W. Clinical implications of pharmacogenetic variation on the effects of statins. Drug Saf. 34, 119 (2011).

76. Trevio,L.R. etal. Germline genetic variation in an organic anion transporter polypeptide associated with methotrexate pharmacokinetics and clinical effects. J.Clin. Oncol. 27, 59725978 (2009).

77. Ramsey,L.B. etal. Rare versus common variants in pharmacogenetics: SLCO1B1 variation and methotrexate disposition. Genome Res. 22, 18 (2012).

78. Johnson,A.D. etal. Genome-wide association meta-analysis for total serum bilirubin levels. Hum. Mol. Genet. 18, 27002710 (2009).

79. Kerns,S.L. etal. A 2-stage genome-wide association study to identify single nucleotide polymorphisms associated with development of erectile dysfunction following radiation therapy for prostate cancer. Int. J.Radiat. Oncol. Biol. Phys. 85, e21e28 (2013).

80. Malhotra,A.K. etal. Association between common variants near the melanocortin 4 receptor gene and severe antipsychotic drug-induced weight gain. Arch. Gen. Psych. 69, 904912 (2012).

81. Comen,E. etal. Discriminatory accuracy and potential clinical utility of genomic profiling for breast cancer risk in BRCA-negative women. Breast Cancer Res. Treat. 127, 479487 (2011).

82. Nguyen,T.V. & Eisman,J.A. Genetics and the individualized prediction of fracture. Curr. Osteoporos Rep. 10, 236244 (2012).

83. Knowles,J.W. etal. Randomized trial of personal genomics for preventive cardiology: design and challenges. Circ. Cardiovasc. Genet. 5, 368376 (2012).

84. Kao, W.H. et al. Family investigation of nephropathy and diabetes research group. MYH9 is associated with nondiabetic end-stage renal disease in African Americans. Nature Genet. 40, 11851192 (2008).

85. Tzur,S. etal. Missense mutations in the APOL1 gene are highly associated with end stage kidney disease risk previously attributed to the MYH9 gene. Hum. Genet. 128, 345350 (2010).

86. Guedj,M. etal. A refined molecular taxonomy of breast cancer. Oncogene 31, 11961206 (2012).

87. Nevins, J.R. Pathway-based classification of lung cancer: a strategy to guide therapeutic selection. Proc. Am. Thorac Soc. 8, 180182 (2011).

88. Vermeire, S. Towards a novel molecular classification of IBD. Dig. Dis. 30, 425427 (2012).

89. Troutbeck,R., Al-Qureshi,S. & Guymer,R.H. Therapeutic targeting of the complement system in age-related macular degeneration: a review. Clin. Experiment Ophthalmol. 40, 1826 (2012).

90. Baldwin,R.M. etal. A genome-wide association study identifies novel loci for paclitaxel-induced sensory peripheral neuropathy in CALGB 40101. Clin. Cancer Res. 18, 50995109 (2012).

91. Park,B.L. etal. Genome-wide association study of aspirin-exacerbated respiratory disease in a Korean population. Hum. Genet. 132, 313321 (2013).

92. Manolio,T.A. etal. Implementing genomic medicine in the clinic: the future is here. Genet. Med. 15, 258267 (2013).This is a description of actively implemented genomic medicine programs at multiple US institutions, including common challenges, infrastructure and research needs. It outlines an implementation framework for investigating and introducing similar programmes elsewhere.

93. Crews,K.R., Hicks,J.K., Pui,C.H., Relling,M.V. & Evans,W.E. Pharmacogenomics and individualized medicine: translating science into practice. Clin. Pharmacol. Ther. 92, 467475 (2012).

94. Manolio,T.A. & Green,E.D. Genomics reaches the clinic: from basic discoveries to clinical impact. Cell 147, 1416 (2011).

Competing interests statementThe authors declare no competing financial interests.

FURTHER INFORMATION1000 Genomes Project: http://www.1000genomes.orgIllumina products for whole-genome genotyping and copy number variation analysis: http://www.illumina.com/applications/detail/snp_genotyping_and_cnv_analysis/whole_genome_genotyping_and_copy_number_variation_analysis.ilmnThe Pharmacogenomics Knowledgebase: http://www.PharmGKB.orgNational Human Genome Research Institute (NHGRI) Catalog of Published Genome-Wide Association Studies: http://www.genome.gov/gwastudiesNature Reviews Genetics Series on Genome-wide association studies: http://www.nature.com/nrg/series/gwas/index.htmlNature Reviews Genetics Series on Translational genetics: http://www.nature.com/nrg/series/translational/index.htmlType1 Diabetes Consortium: https://www.t1dgc.org/ home.cfm

ALL LINKS ARE ACTIVE IN THE ONLINE PDF

R E V I E W S



Abstract | Genome-wide association studies (GWASs) have been heralded as a major advance in biomedical discovery, having identified ~2,000 robust associations with complex diseases since 2005. Despite this success, they have met considerable scepticism reFigure 1 | Pace of genome-wide association study publications since 2005.The pace of genome-wide association study (GWAS) publications has increased dramatically over 7years and shows no signs of slowing. The figure is based on data from the US NationaControversy over potential clinical valueFigure 2 | Correlations of presumed regulatory regions with signals defined from genome-wide association studies.Findings from the ENCODE project indicate that single-nucleotide polymorphisms (SNPs) that are found to be associated with disease in genome-Figure 3 | Use of odds ratios in risk prediction.a | Area under receiver operator characteristic (ROC) curve for predicting a hypothetical condition with variants carrying modest (1.5), sizeable (10) and large (50) odds ratios, showing false-positive fraClinically relevant GWAS findingsFigure 4 | Reclassification of cardiovascular risk based on genotype score.A cohort of 4,232 people was classified into low (1020%; red) 10year risk of cardiovascular disease before and after applying genFigure 5 | Risk of myopathy in chronic simvastatin use.Estimated cumulative risk of myopathy associated with high-dose simvastatin by solute carrier organic anion transporter family member 1B1 (SLCO1B1) rs4149056 genotype. The figure is modified from RefTable 1 | Characteristics of GWAS findings that make them readily translatable to clinical careConclusions: pathways to translation

bringing gwas finding to clinical use

Documents