Chapter 64: Pharmacogenomics

The previous chapter discussed mechanisms underlying variability in drug action, highlighting pharmacokinetic and pharmacodynamic pathways to beneficial and adverse drug events. Work in the past several decades has defined how genetic variation can play a prominent role in modulating these pathways. Initial studies described unusual drug responses due to single genetic variants in individual subjects, defining the field of pharmacogenetics. A more recent view extends this idea to multiple genetic variants across populations, and the term “pharmacogenomics” is often used. Understanding the role of genetic variation in drug response could improve the use of current drugs, avoid drug use in those at increased risk for adverse drug reactions (ADRs), guide development of new drugs, and even be used as a lens through which to understand mechanisms of diseases themselves. This chapter will outline the principles of pharmacogenomics, the evidence as currently available that genetic factors play a role in variable drug actions, and outline areas of controversy and future work.

A goal of traditional Mendelian genetics is to identify DNA variants associated with a distinct phenotype in multiple related family members (Chap. 457) (See Also Chaps. 456 and 457). However, it is unusual for a drug response phenotype to be accurately measured in more than one family member, let alone across a kindred. Some clinical studies have examined drug disposition traits (such as urinary drug excretion after a fixed test dose) in twins, and have in some instances shown greater concordance in monozygotic compared to dizygotic pairs, supporting a genetic contribution to the trait under study. However, in general, non-family-based approaches are generally used to identify and validate DNA variants contributing to variable drug actions.

Types of Genetic Variants Influencing Drug Response

The commonest type of genetic variant is a single nucleotide polymorphism (SNP), and nonsynonymous SNPs (i.e., those that alter primary amino acid sequence encoded by a gene) are a common cause of variant function in genes regulating drug responses, often termed pharmacogenes (Table 64-1). Small insertions and deletions can similarly alter protein function, or lead to functionally important splice variation. Examples of synonymous coding region variants altering pharmacogene function have also been described; the postulated mechanism is an alteration in the rate of RNA translation, and hence in folding of the nascent protein. Variation in pharmacogene promoters has been described, and copy number variation (gene deletion or multiple copies of the same gene) is also well described.

Table 64-1 lists examples of individual types of genomic variation and the impact they can have on function of pharmacogenes. Multiple genotyping approaches may be needed to detect important variants; for example, SNP assays may fail to detect large gene duplications, and highly polymorphic regions (such as human leukocyte antigens, HLA-B) are currently best evaluated by sequencing.

Table 64-1 highlights the fact that the frequency of important variation across pharmacogenes can vary strikingly by ancestry, with the result that certain ethnic groups may be at unusually high risk of displaying variant response to specific drugs.

Candidate Gene Approaches

Most studies to date have used an understanding of the molecular mechanisms modulating drug action to identify candidate genes in which variants could explain variable drug responses. One very common scenario is that variable drug actions can be attributed to variability in plasma drug concentrations. When plasma drug concentrations vary widely (e.g., more than an order of magnitude), especially if their distribution is non-unimodal as in Fig. 64-1, variants in single genes controlling drug concentrations often contribute. In this case, the most obvious candidate genes are those responsible for drug metabolism and elimination. Other candidate genes are those encoding the target molecules with which drugs interact to produce their effects or molecules modulating that response, including those involved in disease pathogenesis.

Genome-Wide Association Studies

The field has also had some success with “unbiased” approaches such as genome-wide association (GWA) (Chap. 456), particularly in identifying single variants associated with high risk for certain forms of drug toxicity. GWA studies have identified variants in the HLA-B locus that are associated with high risk for severe skin rashes during treatment with the anticonvulsant carbamazepine and hepatotoxicity with flucloxacillin, an antibiotic never marketed in the United States. A GWA study of simvastatin-associated myopathy identified a single noncoding SNP in SLCO1B1, encoding OATP1B1, a drug transporter known to modulate simvastatin uptake into the liver, which accounts for 60% of myopathy risk. GWA approaches have also implicated interferon variants in antileukemic responses and in response to therapy in hepatitis C. African-American subjects are known to have higher dose requirements to achieve stable anticoagulation with warfarin, due in part to variation in CYP2C9 and VKORC1, discussed below. In addition, a GWA study identified novel SNPs near CYP2C9 that contribute to this effect in African Americans.

Clinically important genetic variants have been described in multiple molecular pathways of drug disposition (Table 64-2). A distinct multimodal distribution of drug disposition (as shown in Fig. 64-1) argues for a predominant effect of variants in a single gene in the metabolism of that substrate. Individuals with two alleles (variants) encoding for nonfunctional protein make up one group, often termed poor metabolizers (PM phenotype). For most genes, many variants can produce such a loss of function, and assessing whether they are on the same or different alleles (i.e., the diplotype) can complicate the use of genotyping in clinical practice. Furthermore, some variants produce only partial loss of function, and the presence of more than one variant may be required to define a specific allele. Individuals with one functional allele, or multiple reduction of function alleles, make up a second (intermediate metabolizers) and may or may not be distinguishable from those with two functional alleles (normal metabolizers, often termed extensive metabolizers, EMs). Ultra-rapid metabolizers (UMs) with especially high enzymatic activity (occasionally due to gene duplication; Table 64-1 and Fig. 64-1) have also been described for some traits. Many drugs in widespread use can inhibit specific drug disposition pathways (see Chap. 63, Table 63-1), and so EM individuals receiving such inhibitors can respond like PM patients (phenocopying). Polymorphisms in genes encoding drug uptake or drug efflux transporters may be other contributors to variability in drug delivery to target sites and, hence, in drug effects.

CYP3A

Members of the CYP3A family (CYP3A4, CYP3A5) metabolize the greatest number of drugs in therapeutic use. CYP3A4 activity is highly variable (up to an order of magnitude) among individuals, but non-synonymous coding region polymorphisms (those that change the encoded amino acid) are rare. Thus, the underlying mechanism likely reflects genetic variation in regulatory regions.

Most subjects of European or Asian origin carry a polymorphism that disrupts splicing in the closely related CYP3A5 gene. As a result, these individuals display reduced CYP3A5 activity whereas CYP3A5 activity tends to be greater in subjects of African origin. Decreased efficacy of the antirejection agent tacrolimus in subjects of African origin has been attributed to more rapid CYP3A5-mediated elimination and a lower risk of vincristine-associated neuropathy has been reported in CYP3A5 “expressers.”

CYP2D6

CYP2D6 is second to CYP3A4 in the number of commonly used drugs that it metabolizes. CYP2D6 activity is polymorphically distributed, with 5–10% of European- and African-derived populations (but very few Asians) displaying the PM phenotype (Fig. 64-1). Dozens of loss-of-function variants in CYP2D6 have been described; the PM phenotype arises in individuals with two such alleles. In addition, ultra-rapid metabolizers with multiple functional copies of CYP2D6 have been identified especially in East Africa, the Middle East, and Oceania. PMs have slower elimination rates and lower clearance of substrate drugs; as a consequence (Fig. 64-1B), steady state concentrations are higher and the time taken to achieve steady state is longer than in EMs (see Chap. 63). Conversely, UMs display very low steady state parent drug concentrations and an abbreviated time to steady state.

Codeine is biotransformed by CYP2D6 to the potent active metabolite morphine, so its effects are blunted in PMs and exaggerated in UMs. Deaths due to respiratory depression in children given codeine after tonsillectomy have been attributed to the UM trait, and the U.S. Food and Drug Administration (FDA) has revised the package insert to include a prominent “black box” warning against its use in this setting. In the case of drugs with beta-blocking properties metabolized by CYP2D6, greater signs of beta blockade (e.g., bronchospasm, bradycardia) are seen in PM subjects than in EMs. This can be seen not only with orally administered beta blockers such as metoprolol and carvedilol, but also with ophthalmic timolol and with the sodium channel–blocking antiarrhythmic propafenone, a CYP2D6 substrate with beta-blocking properties. Ultra-rapid metabolizers may require very high dosages of nortriptyline and other tricyclic antidepressants to achieve a therapeutic effect. Tamoxifen undergoes CYP2D6-mediated biotransformation to an active metabolite, so its efficacy may be in part related to this polymorphism. In addition, the widespread use of selective serotonin reuptake inhibitors (SSRIs) to treat tamoxifen-related hot flashes may also alter the drug’s effects because many SSRIs, notably fluoxetine and paroxetine, are also CYP2D6 inhibitors.

CYP2C19

The PM phenotype for CYP2C19 is common (20%) among Asians and rarer (2–3%) in other populations. The impact of polymorphic CYP2C19-mediated metabolism has been demonstrated with the proton pump inhibitor omeprazole, where ulcer cure rates with “standard” dosages were much lower in EM patients (29%) than in PMs (100%). Thus, understanding the importance of this polymorphism would have been important in developing the drug, and knowing a patient’s CYP2C19 genotype should improve therapy. CYP2C19 is responsible for bioactivation of the antiplatelet drug clopidogrel, and several large retrospective studies have documented decreased efficacy (e.g., increased myocardial infarction after placement of coronary stents or increased stroke or transient ischemic attacks) among subjects with one or two reduction of function alleles. In addition, some studies suggest that omeprazole and possibly other proton pump inhibitors phenocopy this effect by inhibiting CYP2C19.

CYP2C9

There are common variants in CYP2C9 that encode proteins with reduction or loss of catalytic function. These variant alleles are associated with increased rates of neurologic complications with phenytoin, hypoglycemia with glipizide, and reduced warfarin dose required to maintain stable anticoagulation. Rare patients homozygous for loss of function alleles may require very low warfarin dosages. Up to 50% of the variability in steady-state warfarin dose requirement is attributable to polymorphisms in CYP2C9 and in the promoter of VKORC1, which encodes the warfarin target with lesser contributions by genes controlling vitamin K metabolism such as CYP4F2. The angiotensin-receptor blocker losartan is a prodrug that is bioactivated by CYP2C9; as a result, PMs and those receiving inhibitor drugs may display little response to therapy.

DPYD

Individuals homozygous for loss of function alleles in dihydropyrimidine dehydrogenase, encoded by DPYD, are at high risk for severe toxicity when exposed to the substrate anticancer drug 5-Fluorouracil (5-FU), as well as to capecitabine and tegafur, which are metabolized to 5-FU. Dose reductions have been recommended in intermediate metabolizers.

Transferase Variants

Thiopurine S-methyltransferase (TPMT) bioinactivates the antileukemic drug 6-mercaptopurine (6-MP) and 6-MP is itself an active metabolite of the immunosuppressive azathioprine. Homozygotes for alleles encoding inactive TPMT (1/300 individuals) predictably exhibit severe and potentially fatal pancytopenia on standard doses of azathioprine or 6-MP. On the other hand, homozygotes for fully functional alleles may display less anti-inflammatory or antileukemic effect with standard doses of the drugs.

N-acetylation is catalyzed by hepatic N-acetyl transferase (NAT), which represents the activity of two genes, NAT1 and NAT2. Both enzymes transfer an acetyl group from acetyl coenzyme A to the drug; polymorphisms in NAT2 are thought to underlie individual differences in the rate at which drugs are acetylated and thus define “rapid acetylators” and “slow acetylators.” Slow acetylators make up ∼50% of European and African populations but are less common among East Asians. Slow acetylators have an increased incidence of the drug-induced lupus syndrome during procainamide and hydralazine therapy and of hepatitis with isoniazid.

Individuals homozygous for a common promoter polymorphism that reduces transcription of uridine diphosphate glucuronosyltransferase (UGT1A1) have benign hyperbilirubinemia (Gilbert’s syndrome; Chap. 330). This variant has also been associated with diarrhea and increased bone marrow depression with the antineoplastic prodrug irinotecan, whose active metabolite is normally detoxified by UGT1A1-mediated glucuronidation. The antiretroviral atazanavir is a UGT1A1 inhibitor, and individuals with the Gilbert’s variant develop higher bilirubin levels during treatment. While this is benign, the hyperbilirubinemia can complicate clinical care because it may raise the question of whether coexistent hepatic injury is present.

Transporter Variants

The risk for myotoxicity with simvastatin and possibly other statins appears increased with variants in SLCO1B1. Variants in ABCB1, encoding the drug efflux transporter P-glycoprotein, may increase digoxin toxicity. Variants in the uptake transporters MATE1 and MATE2 have been reported to modulate metformin’s glucose-lowering activity.

A variant in the VKORC1 promoter, especially common in Asian subjects (Table 64-1), reduces transcriptional activity and warfarin dose requirement. Multiple polymorphisms identified in the β₂-adrenergic receptor appear to be linked to specific phenotypes in asthma and congestive heart failure, diseases in which β₂-receptor function might be expected to determine prognosis. Polymorphisms in the β₂-receptor gene have also been associated with response to inhaled β₂-receptor agonists, while those in the β₁-adrenergic receptor gene have been associated with variability in heart rate slowing and blood pressure lowering. In addition, in heart failure, the arginine allele of the common β₁-adrenergic receptor gene polymorphism R389G has been associated with decreased mortality and decreased incidence of atrial fibrillation during treatment with the investigational beta blocker bucindolol.

Drugs may also interact with genetic pathways of disease to elicit or exacerbate symptoms of the underlying conditions. In the porphyrias, CYP inducers are thought to increase the activity of enzymes proximal to the deficient enzyme, exacerbating or triggering attacks (Chap. 409). Deficiency of glucose-6-phosphate dehydrogenase (G6PD), most often in individuals of African, Mediterranean, or South Asian descent, increases the risk of hemolytic anemia in response to the antimalarial primaquine (Chap. 96) and the uric acid-lowering agent rasburicase, which does not cause hemolysis in patients with normal amounts of the enzyme. Patients with mutations in RYR1 encoding the skeletal muscle intracellular release calcium (also termed type 1 ryanodine receptor) are asymptomatic until exposed to certain general anesthetics, which can trigger the rare syndrome of malignant hyperthermia. Certain antiarrhythmics and other drugs can produce marked QT prolongation and torsades de pointes (Chap. 241), and in some patients, this adverse effect represents unmasking of previously subclinical congenital long QT syndrome.

Immunologically Mediated Drug Reactions

The Stevens-Johnson syndrome (SJS) and toxic epidermal necrolysis (TEN) are potentially fatal skin reactions now increasingly recognized to be linked to specific HLA alleles (see Table 64-2). Some cases of hepatotoxicity have also been linked to variants in this region. The frequency of risk alleles often varies by ancestry (Table 64-1). The HLA risk alleles appear to be necessary but not sufficient to elicit these reactions. For example, HLA-B*57:01 is a risk allele for abacavir-related SJS/TEN and flucloxacillin-related hepatotoxicity. However, while 55% of abacavir-exposed subjects will develop reaction, only 1/10,000 subjects exposed to flucloxacillin develop hepatotoxicity. Thus, a third factor, the nature of which has not yet been established, seems necessary.

Tumor and Infectious Agent Genomes

The actions of drugs used to treat infectious or neoplastic disease may be modulated by variants in these nonhuman germline genomes. Genotyping tumors is a rapidly evolving approach to target therapies to underlying mechanisms and to avoid potentially toxic therapy in patients who would derive no benefit (Chap. 67). Trastuzumab, which potentiates anthracycline-related cardiotoxicity, is ineffective in breast cancers that do not express the herceptin receptor. Imatinib targets a specific tyrosine kinase, BCR-Abl1, that is generated by the translocation that creates the Philadelphia chromosome typical of chronic myelogenous leukemia (CML). BCR-Abl1 is not only active but may be central to the pathogenesis of CML; use of imatinib and other BCR-Abl1 inhibitors has resulted in remarkable efficacy not only in CML but also in other BCR-Abl1-positive tumors such as gastrointestinal stromal tumors (see Chap. 67). Similarly, the anti-epidermal growth factor receptor (EGFR) antibodies cetuximab and panitumumab appear especially effective in colon cancers in which K-ras, a G protein in the EGFR pathway, is not mutated. Vemurafenib does not inhibit wild-type BRAF but is active against the V600E mutant form of the kinase. Crizotinib is highly effective in non-small cell lung cancers harboring anaplastic lymphoma kinase (ALK) mutations.

The discovery of common variant alleles with relatively large effects on drug response raises the prospect that these variants could be used to guide therapy. Desired outcomes could be better ways of choosing likely effective drugs and dosages, or avoiding drugs that are likely to produce severe adverse drug events or be ineffective in individual subjects. Indeed, the FDA now incorporates pharmacogenetic data into package inserts meant to guide prescribing. A decision to adopt pharmacogenetically guided dosing for a given drug depends on multiple factors. The most important are the magnitude and clinical importance of the genetic effect and the strength of evidence linking genetic variation to variable drug effects (e.g., anecdote versus post-hoc analysis of clinical trial data versus randomized clinical trial, RCT). The evidence can be strengthened if statistical arguments from clinical trial data are complemented by an understanding of underlying physiologic mechanisms. Cost versus expected benefit may also be a factor.

Reactive versus Preemptive Approaches

Two approaches to pharmacogenetic implementation have been put in place at both “early adopter” institutions and are currently being evaluated. In the first, variant-specific assays are ordered at the time of drug prescription and delivered rapidly (often within an hour or two) and the results then used to guide therapy with that specific drug. The alternative to this “reactive” approach is a “preemptive” approach in which pharmacogenetic testing for large numbers of potential variants across many drugs is undertaken prior to prescription of any such drug. The data are then available in electronic health record (EHR) systems and coupled to real time clinical decision support (CDS). When a drug whose effects are known to be influenced by pharmacogenetic variants is prescribed, the EHR system looks up whether variants likely to affect response are present; if so, CDS will alert healthcare providers that an alternate drug or a different dose may be required.

Challenges

There are multiple challenges in putting in place either system. Assay validity and reproducibility have been issues in the past, but are less likely now. National consortia are now being put in place to develop standards for pharmacogenetic CDS. While common variants in genes such as those listed in Table 64-1 have been clearly associated with variable drug responses, the effect of rare variants, now readily discoverable by large scale sequencing, is unknown. The extent to which a dose adjustment might be recommended may vary depending on whether zero, one, or two variant alleles are present, and whether such variants are reduction of function, loss-of-function, or gain of function. The Clinical Pharmacogenetics Implementation Consortium (CPIC) has developed and published guidelines for multiple drug-gene pairs focusing on the question of what might be an appropriate drug dose adjustment given the availability of genetic data. CPIC does not, however, address the question of when or how such genetic testing should be undertaken.

Developing Evidence that Pharmacogenetic Testing Alters Drug Outcomes

A major issue is whether pharmacogenetic testing affects important drug response outcomes. When the evidence is compelling, alternate therapies are not available, and there are clear recommendations for dosage adjustment in subjects with variants, there is a strong argument for deploying genetic testing as a guide to prescribing; HLA-B*57:01 testing for abacavir is an example described below. In other situations, the arguments are less compelling: the magnitude of the genetic effect may be smaller, the consequences may be less serious, alternate therapies may be available, or the drug effect may be amenable to monitoring by other approaches.

One school argues that the physiology and pharmacology are known, and that RCTs are, therefore, unnecessary (and conceivably unethical). The analogy is sometimes drawn to well-recognized dose adjustment of renally excreted drugs in the presence of renal dysfunction. RCTs have not been conducted and the idea of such dose adjustment is well accepted in the medical community and recommended in FDA-approved drug labels. Others have argued that the effect of genetic variants is generally modest and variability in drug actions has many non-genetic sources, so genetic testing might provide marginal benefit at best.

Efforts to demonstrate the value of pharmacogenetic testing have met with mixed results. An RCT clearly showed that HLA-B*57:01 testing eliminates SJS/TEN due to abacavir. Similarly, regulatory authorities in some countries in Southeast Asia mandated HLA-B*15:02 testing prior to initiation of carbamazepine; however, in this case, an unfortunate outcome was that while the use of carbamazepine dropped, it was often substituted by phenytoin (another drug associated with SJS/TEN), so the incidence of the severe ADR was unchanged.

RCTs evaluating the effect of using pharmacogenetically guided therapy to optimize warfarin treatment have shown either no effect or a modest benefit of incorporating genetic information into prescribing the drug. These RCTs focused on time in therapeutic range in the first 4–12 weeks of treatment, and were not powered to examine outcomes such as recurrent thrombosis or bleeding. Retrospective analyses of bleeding cases vs non-bleeding controls in EHRs and administrative databases have suggested a role for CYP2C9*3 or the variants in V433M variant in CYP4F2 in mediating this risk.

While large retrospective analyses indicate that CYP2C19 loss of function variants decrease clopidogrel efficacy, RCTs are difficult to design: many argue that it is unethical to randomize individuals known to be homozygous for loss of function alleles, since administering clopidogrel is then tantamount to administering placebo. However, trials examining outcomes in only heterozygotes might require very large numbers of subjects.

New effective alternate therapies to warfarin and clopidogrel that appear to lack important pharmacogenetic variants have emerged. One approach to therapy, therefore, is to use pharmacogenetic testing to identify subjects in whom variants are absent and therefore standard doses of the conventional inexpensive drugs are likely to be effective and reserve alternate more expensive therapies for subjects likely to have variant responses to warfarin or clopidogrel. As price drops and as experience grows with newer agents, it is likely that clopidogrel and warfarin will be largely supplanted.

Genetic tools are now being increasingly used to identify or validate new drug targets. Initial studies this field suggest that a new drug development program is more likely to succeed if evidence from human genetics supports the role of a possible drug target in disease pathogenesis and suggests that the risk of toxicity due to high risk pharmacokinetics or other mechanisms is small.

Finding Protective Alleles Can Identify Drug Targets

One example of using genetics to identify a new drug target started with the discovery that very rare gain of function variants in PCSK9 are a rare cause of familial hypercholesterolemia. Subsequently, population studies showed that carriers of loss of function SNPs (2.5% of African Americans) had decreased low-density lipoprotein, decreased incidence of coronary artery disease, and no deleterious consequences in other organ systems. These data triggered the development of PCSK9 antagonists which were marketed less than 10 years after the initial population studies. Other targets implicated by similar population genetic studies include SLC30A8 for the prevention of type 2 diabetes and APOC3 for hypertriglyceridemia. In the latter examples, the identification of an apparently protective effect of rare loss-of-function alleles required very large datasets (>100,000) coupling DNA to longitudinal clinical information; long-term epidemiologic studies like the Framingham Heart Study or EHR systems are now being harnessed to address this opportunity.

Cancer

In cancer, tumor sequencing has identified new targets for drug development, often constitutively active kinases. A problem in this area has been the rapid emergence of drug resistance, often after extraordinary initial responses. For example, 40% of melanomas appear to be driven by the V600E mutant form of BRAF, and the specific inhibitor vemurafenib can produce clinically spectacular remission. However, durable responses are rare, and it is now apparent that combination therapy, often with inhibitors of the MEK pathway, can provide improved therapy. Another approach that is rapidly gaining wide use in cancer are drugs that reverse immune system inhibition (Chap. 69). In some patients the release of this “break” can provide durable remissions, whereas in others, severe adverse events, including colitis, pneumonitis, and myocarditis, have been reported. Understanding the mechanisms underlying variability to these therapies is a major emerging challenge in the field.

Using Multiple Data Types

The development of methods to understand associations across multiple large datasets is another approach that is being explored in drug development. For example, a GWA of risk of rheumatoid arthritis identified multiple risk loci and many encode proteins that are known targets for intervention in the disease. Interestingly, others encode proteins that are targets for drugs used in other conditions, such as certain cancers, raising the question of whether such drugs could be “repurposed” for rheumatoid arthritis. An extension of this approach is the broader issue of systems pharmacology, in which multiple sources of data are used to identify potential molecules or pathways that would be amenable to treatment, by new drugs or by existing agents, using analysis of genomic, transcriptomic, proteomic, and other large datasets. Similar approaches are being developed to predict toxicity expected from targeting specific genes or disease pathways.

The science of pharmacogenomics has evolved from isolated examples of rare adverse drug actions to a more comprehensive view of the role of genetic variation in mediating the effects of most drugs. Current principles include:

• Genetic variants with an important effect on drug actions can be common and their frequencies often vary by ancestry.

• One common mechanism is modulation of drug concentrations.

• No practitioner can be expected to remember all variants important for all drugs. Electronic data systems can now be accessed to describe this information. Ultimately, this information will be used by linking individual pharmacogenetic data to smart electronic health record systems.

• Incorporating genetic approaches into drug development projects hold the promise of more rapid development of targeted, safe, and effective therapies.