Cardiovascular pharmacogenomics: current status and future directions

Abstract

Drugs are widely used and highly effective in the treatment of heart disease. Nevertheless, in some instances, even drugs effective in a population display lack of efficacy or adverse drug reactions in individual patients, often in an apparently unpredictable fashion. This review summarizes the genomic factors now known to influence variability in responses to widely used cardiovascular drugs such as clopidogrel, warfarin, heparin and statins. Genomic approaches being used to discover new pathways in common cardiovascular diseases and thus potential new targets for drug development are described. Finally, the way in which this new information is likely to be used in an electronic medical record environment is discussed.

PHARMACOGENOMICS—BACKGROUND

It is axiomatic in clinical medicine that every drug treatment produces variable degrees of efficacy and a range of adverse events when a large population is exposed. Reduced efficacy or increased adverse events are a special problem in situations of decreased compliance, misdiagnosis or drug interactions that markedly elevate (or occasionally decrease) the concentration of the drug at a target molecular site of action. By linking variants in genes important for drug action to variable outcomes of drug therapy, pharmacogenomics attempts to improve therapy with available drugs and to point to avenues for new drug development. Initial discoveries in the field described adverse drug effects that clustered in specific families or in specific racial subgroups: hemolysis in African-American soldiers exposed to antimalarials in WWII (subsequently discovered to be related to G6PD deficiency); prolonged paralysis after succinylcholine due to pseudocholinesterase deficiency; or malignant hypothermia after exposure to general anesthesia now recognized to be due to mutations in the skeletal muscle ryanodine release channel encoded by RYR1. Thus, an initial focus in the field was on ‘outlier’ populations, and as genotyping and sequencing technologies matured it became clear that such outliers often reflect the presence of single variants with very large effect sizes. With the increasing sophistication of tools to study genomic diversity, the field has incorporated studies of multiple genetic variants, often with smaller individual effect sizes, as determinants of variability to drug response in an individual or across a population. Further, current analytic methods examining genomic variation often fail to fully explain variability in drug actions. Thus, tools such as transcriptional, proteomic and metabolomic profiling that may incorporate ‘downstream’ measures of the effects of multiple genes (often interacting in poorly understood ways) or of gene-environment interactions are now being incorporated into studies of variable drug actions.

Drugs used to treat cardiovascular conditions are among the most widely prescribed therapies worldwide. This review will build on recent summaries^1,2 to describe the current state of knowledge in cardiovascular pharmacogenomics and the directions in which future progress may be made in discovering the basis for variable drug responses and moving those advances to the bedside to improve clinical care.

CLASSES OF GENES MODULATING DRUG ACTION

Recognizing that drugs exert their effects by interacting with specific molecular targets, it follows that drug action will depend on drug concentration at the site of action, the interaction between drug and target, factors modifying the function of the target and the extent to which the target contributes to whole organ or whole-body pathophysiology. ‘Pharmacokinetics’ describes the time course of drug delivery to and removal from target and other (for example, plasma) sites, whereas ‘pharmacodynamics’ refers to the relationship between drug and its molecular, cellular, whole organ and whole-body effects independent of drug concentration. Variants in genes regulating these processes have been implicated as modulators of drug concentrations and drug effects.

Pharmacokinetics are typically analyzed in terms of variables affecting absorption, distribution, metabolism and elimination, each accomplished by specific gene products.³ The commonest route of metabolism is initial drug oxidation by specific member(s) of the cytochrome P450 (CYP) superfamily, and the most commonly implicated enzymes are CYP3A4, CYP2D6, CYP2C9 and CYP2C19. Drugs or metabolites may also be conjugated with methyl, sulfonyl, glucuronyl or acetyl groups by specific transferases, such as catechol-O-methylatransferase involved in metabolism of endogenous catecholamines or uridine glucuronyl transferase metabolizing bilirubin. Notably, drug metabolites may exert pharmacologic actions and these can be similar to, different from, or more potent than those of a parent drug. For some drugs, such as clopidogrel or codeine, metabolism is required to generate the active metabolites that actually produce pharmacologic effects.

The processes of absorption, distribution and elimination involve movement of drug (or metabolites) across cell membranes and can be passive or active. Just as specific CYPs are involved in metabolism of individual drugs, specific drug transport molecules accomplish drug uptake and/or efflux from specific cells. Examples of drug transport molecules modulating the pharmacokinetics of commonly used drugs include the multi-drug and toxin extrusion proteins 1 and 2 (MATE1 and MATE2), implicated in metformin concentrations and effects;⁴ SLCO1B1, encoding the hepatic uptake transporter OAT1B1, which mediates hepatocyte uptake of multiple statins;^5,6 and ABCB1, encoding P-glycoprotein, a widely expressed drug efflux pump important for bioavailability, central nervous system penetration and biliary elimination of many drugs, such as dabigatran and digoxin.^7,8

The best recognized examples of genetically determined variability in drug action have arisen from studies identifying one gene in which a loss-of-function variant can produce very large effects. One scenario under which this occurs is ‘high risk pharmacokinetics’, which describes a situation in which drug concentrations are dependent on the normal function of a single-drug metabolism pathway, either for bioactivation (for example, clopidogrel) or for elimination (for example, warfarin).⁹ In this situation, DNA variants (or drug interactions) that reduce function of the pathway can profoundly affect drug concentrations and, in turn, drug effects.

Large effect sizes may also be seen with variants that alter drug interactions with their target molecules. One example is non-synonymous variants in VKORC1, the target for warfarin action described further below. Another is common non-synonymous single-nucleotide polymorphisms (SNPs) in beta-1 and beta-2 adrenergic receptor genes, which modulate response to beta-blockers in cardiovascular therapy and to beta-agonists in asthma. For example, the arginine-variant at the R389G site in ADRB1 was reported to increase agonist-mediated increases in contractility compared with receptors with a glycine at position 389, and to increase survival among heart failure patients treated with bucindolol.¹⁰ A variation on this theme is now increasingly seen in cancer therapy, with drugs targeting specific pathways (often mutant kinases) expressed in the tumor; prescribing such medications in the absence of the target mutation is ineffective. Although examples of large effect sizes have defined the field to date, it is increasingly well-appreciated that variability in drug action may reflect more complex modulation of metabolism, drug-target interactions and overall target function. Analysis of the effects of multiple genetic variants may require very large data sets and may be facilitated by complementary ‘omics’ data including metabolomic, transcriptomic and proteomic profiling.¹¹

SPECIFIC DRUGS AND CLINICAL SETTINGS

A prominent role for genetic variation is now appreciated for a number of commonly used cardiovascular drugs, including clopidogrel, warfarin, simvastatin and other statins, and each serves as a model for the discovery and implementation aspects of pharmacogenetics. The current state of the art for these agents is briefly described here, followed by other areas in which pharmacogenetics may influence understanding variability in the action of cardiovascular drugs and in using this information clinically.

CLOPIDOGREL

Although clopidogrel was approved in the late 1990s as an antiplatelet drug, the fact that CYP2C19 function is critical for its bioactivation was first described in 2006.¹² The minor allele frequency for a common loss-of-function variant in CYP2C19, termed CYP2C19*2, is higher among Asians than among Caucasians, varying from ~ 15 to ~ 30%, and allele frequency for another loss-of-function variant, termed *3, varies from virtually absent in Africans to ~ 15% in East Asians.¹³ Resequencing the gene has identified dozens of rare, and heretofore uncharacterized, non-synonymous variants. The functional consequences of CYP2C19 variants on clopidogrel’s antiplatelet effects are greatest in individuals with loss-of-function variants on both CYP2C19 alleles (for example, homozygotes for CY2C19*2) and intermediate for individuals heterozygous for loss-of-function alleles. Variants that increase CYP2C19 metabolic activity (notably CYP2C19*17) have been associated with increased clopidogrel effect, and some evidence links them to increased bleeding risk.¹⁴ Analyses of outcomes by genotype in large placebo-controlled trials of clopidogrel action have supported the idea that homozygotes and to perhaps a lesser extent heterozygotes, demonstrate decreased clopidogrel efficacy, notably in acute coronary syndromes.¹⁵ Evidence for such an effect in other clinical settings, which clopidogrel’s antiplatelet effects may be less important is less compelling.^16,17 A genome-wide association study (GWAS) examining the extent to which clopidogrel-inhibited adenosine diphosphate-induced platelet aggregation in 429 Amish subjects identified a strong signal at the CYP2C19 locus, arising from the *2 allele.¹⁸ Interestingly, 73% of the variability in the trait was judged heritable, but only 12% could be attributed to CYP2C19*2. A further 10% was attributed to clinical variables, such as age and gender, and the remainder of the variability has not been explained. Thus, although the CYP2C19*2 allele exerts an unusually large genetic effect, its overall contribution to variability in clopidogrel action is unexpectedly small. No large randomized clinical trial has evaluated a standard approach to clopidogrel prescribing to a genotype-guided one. Some argue that the biologic data are sufficiently compelling that equipoise is lost and genotypes should be routinely incorporated into clopidogrel prescribing while others argue a randomized clinical trial is required. There are obviously major logistic difficulties in mounting a randomized clinical trial targeting 2–3% of the general population (that is, only those who are homozygotes for the *2 variant). Associations with variants in other genes, such as ABCB1¹⁹ or PON1 (encoding paraoxonase 1)²⁰ have been implicated as modulating clopidogrel activity. The PON1 finding was initially described in a very small association study and multiple larger studies have failed to replicate the result. This is probably an example of the dangers of false-positive associations in candidate gene studies when small numbers are evaluated.

WARFARIN

Warfarin is administered as a racemate, and the more active enantiomer, S-warfarin, is bioinactivated by CYP2C9.²¹ Initial studies identified two relatively common non-synonymous variants in the gene, one termed *2 reducing function and one termed *3 nearly completely eliminating function. In an initial study, individuals receiving steady-state dosages < 3 mg per day were more likely to carry *2 and *3 variants and were more likely to display bleeding complications during chronic therapy;²² the latter observation remains unexplained. In 2004, coding region variants in a vitamin K complex gene termed VKORC1 were identified as the cause of a rare familial syndrome of warfarin resistance, in which very large dosages of the drug produce no anticoagulant effect.²³ This finding identified the VKORC1 gene product as the molecular target with which warfarin interacts to exert its anticoagulant effect, and common variation in the VKORC1 promoter was subsequently associated with variability in hepatic VKORC1 transcripts and with steady-state warfarin dose requirement:²⁴ individuals expressing more VKORC1 required greater warfarin doses. These candidate gene findings were mirrored in GWAS, which identified variation at the VKORC1 and CYP2C9 loci as contributing up to 50% of the variability in steady-state warfarin dose requirement:^25,26 conditioning the GWAS analysis on these two loci identified variation in a third gene, CYP4F2, which had been implicated in vitamin K metabolism and warfarin response.²⁷

It is known that subjects of African ancestry require greater dosages of warfarin for anticoagulation at steady-state and this has been attributed to a greater frequency of high activity VKORC1 promoter variants.²⁸ CYP2C9*2 and *3 do occur in subjects of African origin, but at allele frequencies lower than those seen in Caucasians; other variants (*6, *8 and *11) have been found in African, but not Caucasian subjects.²⁹ As with CYP2C19, resequencing the gene in large number of subjects has identified dozens of new rare variants some of which are presumably loss-of-function. Similarly, rare VKORC1 non-synonymous variants have been described that markedly increase warfarin dose requirement; as with other rare variants, these may be enriched in specific ancestral populations.³⁰ In addition, a recent GWAS in African-American subjects identified variation near the CYP2C9 locus, but the association was unrelated to any known variant.³¹

Algorithms have been developed that combine clinical features, co-administered interacting drugs such as certain statins or amiodarone, and genetic variation to predict steady-state warfarin dose. In subjects whose steady-state dose is unusually high (> 7 mg per day) or unusually low (< 3 mg per day) algorithms that incorporate clinical factors plus genetic factors perform better than those incorporating only clinical factors.³² In late 2013, the results of three randomized clinical trials asking the question of whether routinely incorporating genetic variant information into initial warfarin dosing would improve time in the therapeutic range during the subsequent 30 or 90 days were published.^33–35 Two studies compared clinical algorithms to clinical plus genetic algorithms and showed no difference in time in therapeutic range, wherease a third compared a clinical plus genetic algorithm to an empiric (fixed dose) loading algorithm and demonstrated an improvement in time in therapeutic range. Interestingly, the genetic algorithm performed worse than the clinical only algorithm in the one study that included African-Americans and this observation remains unexplained. None of the trials were large enough or lasted long enough to examine clinical outcomes such as recurrent thrombosis or bleeding (the largest trial has 14 major bleeding events in 1015 patients). One more recent study has implicated CYP4F2 variants risk factor for bleeding during long-term (mean > 3 years in 265 cases versus 305 controls) warfarin therapy.³⁶ Another study examined candidate polymorphisms in 250 cases with major warfarin-related bleeding and 259 controls without bleeding in BioVU, the Vanderbilt DNA biobank. Most events occurred during long-term (> 30 days) therapy. The *3 variant in CYP2C9 was significantly more common in the group with bleeding (13.4%) than in controls (6.6%; P = 0.017); the adjusted odds ratio (OR) was 2.05 (95% confidence interval 1.04, 4.04).

New antiplatelet drugs such as prasugrel or ticagrelor and new anticoagulant drugs such as dabigatran, apixaban and rivaroxaban have become available as alternatives to clopidogrel and warfarin. The newer drugs do not appear to have this same pharmacogenomic variability as older ones, are currently more expensive, and some are not approved in certain subsets, such as the elderly. Thus, the extent to which and the rate at which clopidogrel and warfarin will be supplanted by newer drugs remains to be determined.

HEPARIN-INDUCED THROMBOCYTOPENIA

Mining BioVU identified 67 cases of this adverse drug reaction and 884 controls, and a GWAS identified a significant association (OR 18.52 (6.33–54.23), P = 3.18 × 10^{− 9}) with SNPs near the T-cell death-associated gene 8. These SNPs tag a missense variant in the gene that was associated with PF4/heparin antibody levels and positive PF4/heparin antibodies in non-heparin treated patients (OR 3.09 (1.14–8.13), P = 0.02). In a candidate gene study, SNPs in the HLA-DRA region were also associated with heparin-induced thrombocytopenia (OR 0.25 (0.15–0.44), P = 2.06 × 10^{− 6}).³⁷

STATINS

HMG-CoA reductase inhibitors (statins) are among the most widely prescribed cardiovascular drugs worldwide. Their undisputed efficacy at lowering low-density lipoprotein (LDL) and reducing cardiovascular events has led to their increasing use in cohorts at lower and lower cardiovascular risk. Pharmacogenetic studies of statins have focused on two broad areas: (1) efficacy judged by the extent to which LDL is lowered and, more recently, the extent to which myocardial infarction and other important vascular events are prevented and (2) myotoxicity. Examples are presented here.

Initial studies of efficacy focused on candidate genes, often those associated with baseline lipid abnormalities. An initial study examined 148 SNPs in 10 candidate genes in 1536 subjects receiving pravastatin, and identified two SNPs in the HMG CoA-reductase gene itself as associated with decreased drug efficacy (that is, smaller decreases in LDL).³⁸ Another study reported an association between a single relatively rare SNP in the HMG CoA-reductase gene and lipid lowering in 1601 diabetics.³⁹ Among the 3.3% of subjects who carried a single copy of the minor allele, 51% of patients failed to reach a cholesterol target, compared to 28% lacking this allele. In another candidate gene study, APOE genotype was strongly associated with the extent to which multiple statins lowered LDL in diabetics.⁴⁰ Another candidate gene study reported that the extent to which simvastatin lowered LDL was modulated by an intronic variant in CYP3A4 (CYP3A4*22) that reduces CYP3A4 function.⁴¹ In a group of Chinese subjects, a candidate gene study found associations between change in LDL with rosuvastatin and variants in the APOE locus, the transporter gene ABCG2, the flavin-containing monooxygenase 3 gene and the lipoprotein lipase gene.⁴² A combined analysis of three GWAS in 3932 subjects exposed to simvastatin, pravastatin or atorvastatin suggested that a variant within the calmin (CLMN) gene whose function remains uncertain and a variant in APOC1 near APOE modulated lipid lowering.⁴³

An initial study of events during statin therapy in 1214 cases of myocardial infarction or stroke and 2686 controls (all subjects received statins) examined the relationship between case–control status and polymorphisms in six candidate genes, chosen because they had been previously implicated as modulators of lipid levels or statin response. Although multiple associations were discovered, the number of observed associations was no greater than anticipated by chance.⁴⁴ Another study sought associations between SNPs in anti-inflammatory candidate genes in 668 cases of myocardial infarction and 1217 controls without myocardial infarction, all treated with statins. The strongest interaction was with a SNP in ADAMTS1; homozygotes had an OR of 0.10 (95% CI, 0.03–0.35).⁴⁵ A large study of subjects with diabetes ascertained from electronic medical records found a single SNP in LPA (G allele at rs10455872) to be a strong predictor of decreased LDL lowering and of cardiac events during treatment, with a P = 1.5 × 10^{− 29} in a meta-analysis involving 30 467 subjects. The authors concluded that the 15% of the population carrying the variant were at higher risk of coronary artery disease than non-carriers and that this effect was not altered by statin treatment.⁴⁰

A number of lines of evidence have implicated a common SNP in SLCO1B1 as a risk factor for myopathy during statin treatment, notably with high-dose simvastatin. Investigators in the SEARCH consortium performed a GWAS comparing 85 cases of incident myopathy during treatment with simvastatin at high doses (80 mg per day) to 90 controls, and identified a single SNP at genome-wide significance, in tight linkage disequilibrium with a known non-synonymous variant resulting in the V174A, and replicated the result in the Heart Protection Study among subjects receiving 40 mg per day simvastatin.⁴⁶ Homozygote carriers of the C-risk allele, 2.1% of the SEARCH population, were at ~ 20-fold increased risk for biochemically proven statin-related myopathy over 5 years treatment (OR 16.9), whereas homozygotes (24.7%) were at intermediate, approximately fourfold risk. Voora et al.⁴⁷ found that among users of simvastatin or atorvastatin, discontinuation rates were higher among variant carriers, whereas there was no effect among pravastatin users, who experienced little myopathy; the relationship of these findings to the development of myopathy remains uncertain. A study in Chinese subjects suggested that an intronic SNP in SLCO1B1 was also associated with LDL response among Chinese subjects receiving rosuvastatin or simvastatin.⁴⁸ After cerivastatin was withdrawn from the market because of an unusually high incidence of myotoxicity, a candidate gene study and GWAS examining 185 cases and 732 controls found a modest effect of the V174A SNP (OR 1.9) and implicated an intronic variant in the RYR2 gene, known to mediate calcium release in cardiac, but not skeletal muscle.⁴⁹

ARRHYTHMIA GENOMICS

A common framework for analyzing risk of drug-induced QT prolongation and risk for torsades de pointes is ‘reduced repolarization reserve'.⁵⁰ This idea suggests that repolarization is a complex, redundant and buffered process and that drug-induced long QT syndrome thus reflects multiple ‘hits’, commonly drug block of the potassium current I_Kr, encoded by KCNH2 (also known as HERG), combined with other factors such as hypokalemia, bradycardia and subclinical ion channel mutations. This concept has been supported by case reports and series of patients with clinically inapparent congenital long QT syndrome ‘exposed’ by drug block.^51–55 A number of studies have searched for common and rare variants increasing drug-induced long QT syndrome susceptibility. A large candidate gene survey in 176 cases and two groups of controls (n = 1044) found that rs1805128, a non-synonymous (D85N) variant in the potassium channel subunit KCNE1 was present in 8.6% of cases and 1.9–2.9% of controls (OR 9.0).⁵⁶ A GWAS of 216 cases and 771 controls (all Northern European) did not find any signals at genome-wide significance, arguing against a role for common variants in drug-induced long QT syndrome.⁵⁷ In an analysis of exomes from 65 cases and 148 controls, Weeke et al.⁵⁸ found a significant excess of rare variants in KCNE1 and ACN9 among cases; they also reported that 37% of cases harbored rare potassium channel variants compared with 21% of controls. Recent studies have suggested an alternate pathway to QT prolongation: decreased PI3-Kinase signaling resulting in altered ion channel expression or function (notably drug-induced increased ‘late’ sodium current);^59–62 The extent to which this pathway modulates arrhythmia susceptibility remains to be determined.

Atrial fibrillation (AF) remains the commonest arrhythmia-requiring therapy, and genome-wide approaches have identified common SNPs at 4q25 as risk factors;⁶³ the nearest gene, PITX2, encodes a transcription factor important for left-right cardiac differentiation and development of the ‘pulmonary’ myocardium,^64,65 an important site of AF initiation, and is thus a strong candidate. The SNP rs10033464 at the 4q25 locus was found to predict decreased response to antiarrhythmic drug therapy to maintain sinus rhythm in AF (OR 4.7, P = 0.0013).⁶⁶ In addition, carriers of the variant responded better to sodium channel blockers (class I antiarrhythmics), whereas variant carriers responded better to action potential prolonging (class III) drugs. SNPs at the 4q25 locus also predict recurrent AF after cardioversion⁶⁷ and AF ablation.⁶⁸ The common R389G variant in ADRB1, encoding the beta-1-adrenrgic receptor, has been associated with an increased likelihood that a rhythm control strategy would be effective in managing AF. If validated, these results could be the foundation for selecting therapeutic strategies and specific drugs in AF stratified by genotype.

GENOMIC APPROACHES TO IDENTIFYING NEW MECHANISMS FOR DRUG ACTION AND NEW TARGETS

Analyses of proteomic and transcriptional profiles in cell lines exposed to statins have been used to complement genomic approaches in statin myotoxicity. One study reported 247 differentially regulated transcripts in multiple biologic pathways including drug metabolism, fatty-acid metabolism, biometabolism and inflammation.⁶⁹ Another identified six expression quantitative trait loci interacting with simvastatin exposure and one, in GATM, encoding glycine amidinotransferase, was found to be associated with simvastatin myotoxicity in clinical trials.⁷⁰

Studies with PCSK9 that identified gain-of-function variants as a rare cause of familial hypercholesterolemia prompted Cohen et al.⁷¹ to examine the relationship between loss-of-function PCSK9 variants and heart disease. They identified two variants that encode stop codons at positions 142 and 679 in 85/3363 middle-aged African- Americans in the Atherosclerosis Risk in Communities study, and showed that variant carriers had 28% lower mean LDL values, and 88% less coronary artery disease risk over 15 years.⁷² These studies have been interpreted as identifying PCSK9 as a potential drug target in CAD, and two new agents are now being evaluated for marketing after phase 3 trials reported favorable outcomes not only on LDL values but also on coronary events.^73,74

The paradigm of using genetic approaches to link clear loss-of-function variants to desirable clinical outcomes (and thereby identify or validate drug targets) is now being applied more broadly in very large patient cohorts. For example, very rare loss-of-function variants in NPC1L1, encoding the ezetemibe target, were less common in subjects with coronary artery disease (11/29 954, 0.04%) compared with controls (71 of 83 140, 0.09%); these results provide evidence supporting the cardioprotective effect of ezetemibe.⁷⁵ Similarly, rare loss-of-function variants in APOC3 have been associated with lower triglyceride levels^76,77 and, in 110 971 subjects, a lower risk of CAD.⁷⁷ These results suggest that triglyceride lowering is a valid cardioprotective strategy.

Recent studies have also highlighted the relationship between metabolomic profiling and gut microbiota. Koeth and colleagues showed that plasma L-carnitine, a component of red meat, is metabolized by intestinal microbiota resulting in a proatherogenic molecule, trimethylamine-N-oxide (TAMO).⁷⁸ They reported that vegans produce less TAMO than did meat eaters, and that high plasma L-carnitine predicted risk for prevalent cardiovascular disease and cardiac events in a cohort of 2595 subjects, but the risk was confined to those who also had high TAMO levels. In mice, L-carnitine feeding increased TAMO concentration and this could be abrogated if intestinal microbiota were suppressed. In another study from the same group, TAMO levels were measured in human volunteers after a phosphatidylcholine challenge (hard boiled eggs plus labeled phosphatidylcholine) before and after administration of broad-spectrum antibiotics to suppress intestinal microbiota.⁷⁹ The increase in TAMO after phosphatidylcholine challenge was markedly suppressed after antibiotic treatment and in a population of 4007 subjects undergoing elective coronary angiography, TAMO levels predicted major adverse cardiovascular events over the subsequent 3 years. These data not only suggest novel biomarkers for CAD identification but also suggest the hypothesis that drugs inhibiting TAMO generation could be anti-atherogenic.

CLINICAL IMPLEMENTATION OF PHARMACOGENOMICS

The identification of genetic variants, especially those with large effect sizes, modulating drug effects alleles has raised the prospect of incorporating this information into routine clinical care to improve outcomes by ‘personalizing’ the choice of drug or drug dose to maximize efficacy and reduce the risk of serious toxicity. One approach is to genotype patients receiving target drugs at the point of care, whereas a second approach ‘preemptively’ imbeds multiple pharmacogenomic variants in the electronic medical record and provides advanced point of care decision support when a culprit drug is ordered in a patient with known genomic variants. There are considerable obstacles to implementing even well-documented genomic variants into a clinical care using either paradigm: incomplete evidence on appropriate therapeutic responses in patients with genomic variants; variable effect sizes of genetic variants; uncertainty over function of some variants, notably increasingly well-described rare variants in known target genes; logistic issues such as development of decision support or timely delivery of genetic variant information; and provider unfamiliarity with fundamental concepts in genomic medicine.⁸⁰

A number of centers are testing a ‘preemptive’ approach in which multiple genotypes relevant to the action of many drugs are embedded in electronic medical record systems long before the specific drugs such as clopidogrel, warfarin or tacrolimus are prescribed.^81–83 In addition, the National Human Genome Research Institute’s electronic Medical Records and Genomics network is using a targeted sequencing approach to explore the preemptive model in 9000 subjects.⁸⁴ An analysis of the first 10 000 subjects participating in Vanderbilt University’s PREDICT preemptive pharmacogenetic testing program⁸¹ highlighted the fact that the vast majority (91%) of subjects harbored a variant likely to affect the prescribing of one of five target drugs (warfarin, clopidogrel, simvastatin, tacrolimus or thiopurines), and that ~ 5% harbored high-risk genotypes such as CYP2C19*2/*2. Another potential application of genomic information is the development of multigenic (or multimarker) risk scores to reclassify individuals and thus better target treatments. For example, an individual with moderate coronary artery disease risk but a high genetic risk score might be a candidate for more aggressive prophylactic treatment; further informing the subject of high risk may modify behaviors.⁸⁵ These large experiments should provide further data on how best to implement genomic medicine in the electronic medical record context.