Pharmacogenetics in psychiatry: translating research into clinical practice

Abstract

Pharmacogenetic/pharmacogenomic (PGx) approaches to psychopharmacology aim to identify clinically meaningful predictors of drug efficacy and/or side-effect burden. To date, however, PGx studies in psychiatry have not yielded compelling results, and clinical utilization of PGx testing in psychiatry is extremely limited. In this review, the authors provide a brief overview on the status of PGx studies in psychiatry, review the commercialization process for PGx tests and then discuss methodological considerations that may enhance the potential for clinically applicable PGx tests in psychiatry. The authors focus on design considerations that include increased ascertainment of subjects in the earliest phases of illness, discuss the advantages of drug-induced adverse events as phenotypes for examination and emphasize the importance of maximizing adherence to treatment in pharmacogenetic studies. Finally, the authors discuss unique aspects of pharmacogenetic studies that may distinguish them from studies of other complex traits. Taken together, these data provide insights into the design and methodological considerations that may enhance the potential for clinical utility of PGx studies.

Pharmacogenetic/pharmacogenomic (PGx) approaches to psychopharmacology aim to identify clinically meaningful predictors of drug efficacy and/or side-effect burden. Given the clinical heterogeneity and prognostic uncertainty associated with most psychiatric disorders, the prospect of pharmacogenetically informed individualized treatment holds considerable appeal. Equally appealing are the pragmatic advantages of PGx approaches, including the ease of obtaining appropriate blood or saliva samples for DNA extraction, the immutability of genotype information and the increasingly low cost of genotyping assays. More broadly, PGx data may help illuminate the still-obscure molecular substrates of psychotropic drug action.

To date, however, PGx studies in psychiatry have not yielded compelling results, and clinical utilization of PGx testing in psychiatry is extremely limited. In this review, we will provide a brief overview on the status of PGx studies in psychiatry, review the commercialization process for PGx tests and then discuss methodological considerations that may enhance the potential for clinically applicable PGx tests in psychiatry.

Pharmacogenetic studies of psychotropic drug response

PGx studies have been carried out with multiple classes of psychotropic drugs, with a primary focus on the interindividual variation in drug efficacy. The vast majority of studies, which have been reviewed extensively elsewhere,^1–7 have utilized a candidate gene approach, usually based upon the receptor pharmacology of the psychotropic agent. The proof of principle for this approach is provided by PGx studies of the DRD2 gene, which codes for the dopamine D2 receptor, the common site of action of all commercially available drugs with antipsychotic potency.⁸ A recent meta-analysis indicates that a functional polymorphism in the DRD2 promoter region, which modulates levels of gene expression, significantly influences antipsychotic drug efficacy.⁹ Similarly, a promoter region polymorphism in the gene coding for the serotonin transporter, the common site of action for the selective serotonin reuptake inhibitor class of antidepressants, has been implicated by meta-analyses of selective serotonin reuptake inhibitor efficacy.¹⁰ In both instances, carriers of a common yet deficient genetic variant demonstrate between half and two-thirds of the response rate of noncarriers. Although these effect sizes are statistically robust, they do not yet yield adequate sensitivity and specificity to reliably guide clinical practice. Moreover, alternative treatment strategies for poor-prognosis genetic carriers have not been empirically tested, a task made more difficult by the fact that most widely used psychotropic medications in each major class have similar primary targets.

Pharmacokinetic genetic studies of psychotropic drug response are supported by the identification of multiple functional variants with well-defined effects on drug metabolism. For example, the gene for the cytochrome P450 2D6 enzyme, which is responsible for the metabolism of several psychotropic agents, contains > 100 genetic variants (as cataloged by the website: http://www.cypalleles.ki.se), many of which yield nonfunctional or reduced-function enzymes. There are four phenotypes of CYP2D6 produced by combinations of various alleles with different degrees of enzymatic activities: poor metabolizer, intermediate metabolizer, extensive metabolizer and ultrarapid metabolizer. Compared with extensive metabolizers with normal CYP2D6 enzyme activity, poor metabolizers and intermediate metabolizers have minimal or reduced activity, respectively. Ultrarapid metabolizers have duplicate or multiple copies of the gene that result in increased enzyme activity. Approximately 7–10% of Caucasians and 1–2% of Asians are poor metabolizers,¹¹ who tend to accumulate higher drug levels in blood and, theoretically, require lower doses to achieve therapeutic effects. Ultrarapid metabolizers, in contrast, consist of only 1% of the population and may require higher doses because of faster elimination of the drug.

Despite the functional significance of these variants, relatively fewer studies have focused on the CYP450 system in psychiatric PGx,⁵ as compared with pharmacodynamic studies. In part, this has been because of the lack of compelling empirical support for a relationship between plasma drug levels and psychotropic drug efficacy. Multiple early reports on the tricyclic antidepressant nortryptiline suggested a ‘therapeutic window’ for efficacy with plasma drug levels between 50 and 150 ng ml^–1;¹² for the prototypic atypical antipsychotic clozapine, plasma levels of at least 200 ng ml^–1 were considered a potential threshold level required for clinical response.¹³ However, consistent data in this regard and for other psycho-tropic drugs are limited. The lack of a clear relationship between drug levels and efficacy suggests that pharmacokinetic genetic variation may not significantly predict differential drug efficacy, although a potential role in side-effect prediction (discussed in greater detail below) has led some authors to suggest clinical guidelines for dosing based upon CYP450 genotype.¹⁴

A final approach that has been as yet uncommon in psychiatric PGx research is the utilization of genome-wide association studies (GWASs). A major challenge to GWASs in pharmacogenetics are the large sample sizes presumed to be necessary to overcome the statistical penalty incurred by the genotyping of hundreds of thousands of single-nucleotide polymorphisms.¹⁵ Although difficult to achieve for disease susceptibility studies, sample sizes in the thousands (or even tens of thousands) are essentially out of reach for prospective PGx trials. PGx data sets have traditionally been ‘piggybacked’ off of ongoing clinical trials, which are often not optimally designed from a pharmacogenetic perspective because of the inclusion of multiple treatment arms, ethnically and clinically heterogeneous samples and high discontinuation rates. An exception is the GENDEP (Genome-Based Therapeutic Drugs for Depression) project—which was specifically designed to achieve pharmacogenetic aims.¹⁶ All subjects were drawn from a single ethnic group, there were only two treatment arms (escitalopram and nortryptyline), drug selection was based upon distinct biological hypotheses on the mechanism of drug action (serotonergic versus noradrenergic) and the sample size (n = 811) was much larger than most academic clinical trials. Unfortunately, GENDEP, as with other GWAS data on antidepressant response from the STAR*D (Sequenced Treatment Alternatives to Relieve Depression) study¹⁷ (n = 1491) and the Munich Antidepressant Response Study¹⁸ (n = 339), did not detect any genome-wide significant or clinically useful predictors of antidepressant response. Similarly, the GWAS of lithium treatment response in the STEP-BD (Systematic Treatment Enhancement Program for Bipolar Disorder)¹⁹ (n = 1177) and several GWASs of antipsychotic drug responses from the CATIE (Clinical Antipsychotic Trials of Intervention Effectiveness)^20–23 (n = 738) did not find any genetic markers that can be readily used in clinical settings. The limitations of these studies include use of chronic patients, lack of medication adherence monitoring and ambiguity of phenotype definition.

Clinical pharmacogenetic testing in psychiatry

The lack of compelling data from pharmacogenetic studies has hampered the development of clinical PGx tests. To be approved by the US Food and Drug Administration (FDA), clinical tests must achieve higher levels of sensitivity and specificity than observed with PGx results to date. For this reason, most commercially available PGx tests in psychiatry have not undergone FDA approval, but have sought alternative paths into the commercial marketplace.

Currently, clinical genetic testing services in the United States are regulated by the Clinical Laboratory Improvements Amendments of 1988 (CLIA) (Code of Federal Regulations, Title 42, Part 493, 1995). CLIA certification is primarily focused on quality control and laboratory procedures, and does not provide databased review of the clinical utility of a particular test. Most PGx tests offered by various companies and laboratories achieved CLIA certification, but did not undergo FDA approval. Therefore, laboratories may develop tests (lab-developed tests) that are CLIA certified as biologically accurate assays, but may have relatively limited predictive power for any clinical phenotype. These tests may be marketed to physicians or, in some cases, directly to potential consumers, but cannot be sold for use by other laboratories.²⁴ The lab-developed tests are not routinely subject to post-marketing surveillance by federal authorities, and therefore limited data are available on the long-term impact of the introduction of these tests into clinical practice.

Alternatively, testing companies can pursue FDA approval for commercial testing products. Moderate-risk tests undergo 510(k) clearance, which just requires review of the data that support the clinical claim about the test. Higher-risk tests, in which a potentially serious intervention could be considered based upon test results, may be required to provide clinical trial evidence in support of the utility of the test. It should be noted, however, that the regulatory environment for genetic testing is undergoing scrutiny and new legislation may be proffered that significantly alters the current regulatory framework for PGx tests.²⁴

To date, there is only one FDA-approved pharmacogenetic test for use in psychiatry. Roche Diagnostics received FDA 510(k) clearance in 2004 for a micro-array-based product, the AmpliChip R CYP450 Test, which assesses 27 alleles in CYP2D6 and 3 alleles in CYP2C19. This product received approval based upon data indicating that CYP450 genotype can influence drug efficacy and safety via effects on drug metabolism.²⁵ The limited marketing upon the release of the test emphasized the relationship between these genotypes and clinical response to psychotropic drugs (such as the antipsychotic drug risperidone) that are primarily metabolized by CYP2D6 or CYP2C19, but the test was not associated with any specific drug or class of drugs and no specific clinical claims were granted. The clinical uptake of this product has been modest, perhaps because of clinician concerns about the interpretation of test results, the paucity of prospective data suggesting that test utilization influences clinical outcome and the lack of reimbursement for an expensive test. Moreover, with rapid progress in genotyping technology, several companies have produced genotyping platforms specifically designed for PGx, focusing on genetic markers associated with drug absorption, distribution, metabolism and enzymes (ADME). Illumina’s VeraCode ADME core panel assays markers in 34 genes, one-third of which code for cytochrome P450 enzymes (www.illumina.com/products/veracode_adme_core_panel.ilmn). The DMET Plus Premier Pack from Affymetrix provides 1936 drug metabolism markers in ~230 genes (www.affymetrix.com). Other companies have also developed lab-developed tests that assess CYP450 genotypes, and these may compete for market utilization. However, adoption of these platforms for clinical practice has been sparse. High cost and long turnaround time (at least 5–7 days) hinder their practical application. Several other non-FDA approved PGx tests that were specifically designed for psychotropic drug responses have been made commercially available, including tests for prediction of response to anti-depressant and antipsychotic drugs; however, these tests have also had limited utility and clinical applications.²⁶

Design considerations to enhance pharmacogenetic studies

Clearly, stronger data will be needed to achieve regulatory approval of clinically useful PGx testing in psychiatry. To accomplish this goal, it may be worthwhile to consider specific study design features that may enhance the power and precision of pharmacogenetic studies.

Focus on early phases of illness

The majority of PGx data sets in psychiatry are derived from ongoing clinical trials. Unfortunately, the patients who enter into these trials are often chronically ill with lengthy prior treatment histories; highly responsive patients may not present for these studies, as they do not seek changes to treatment if adequate response has already been attained. Study samples drawn from trials in chronic subjects may therefore be systematically biased toward inclusion of patients who are not fully responsive to treatment or who are nonadherent with treatment (or both), and not represent the full spectrum of treatment outcomes.²⁷ Consistent with this, overall response rates in most clinical trials of antipsychotic drugs are routinely lower than in trials that focus on patients in the earliest phases of illness, in which remission rates have been reported as high as 80% of enrolled subjects.^28,29 Chronically ill cohorts are also marked by increased duration of psychotic symptoms, substance abuse and functional/social disabilities; each factor may influence drug response rates and introduce increased variance into data analyses. For these reasons, studies that concentrate on early-phase or first-episode patients may provide enhanced power for pharmacogenetic studies of drug efficacy; for example, our recent antipsychotic PGx meta-analysis demonstrated a 50% greater effect size for the DRD2 promoter polymorphism in studies containing first-episode patients compared with studies of chronic patients.⁹

Utilization of additional phenotypes; adverse events

A drawback of pharmacogenetic studies that focus on clinical efficacy as the primary phenotype is the inherent method variance introduced by the use of clinical symptom ratings. Standard ratings scales are dependent upon subjective patient report and rater sensitivity, and tend to be decreasingly reliable as sample sizes and study site number increase. The effects of diminished rating reliability on study power were quantified in a report by Perkins et al.,³⁰ with dramatic results. As shown in Figure 1 (data interpolated from Perkins et al.³⁰), study power is reduced, and sample size requirements concomitantly increased, as the reliability of the dependent measure (phenotype) falls. These effects are equally valid for PGx studies that compare two groups defined by genotype as for clinical trials that compare two treatment conditions. Thus, contrary to current trends in genetics studies of other complex traits, it is possible that large PGx studies can be deceptively underpowered, whereas smaller trials with rigorously collected phenotypic information may be more robust.

Figure 1.

Effect of phenotypic rating reliability on statistical power given a fixed sample size (n = 100; red line, left axis) and required sample size to attain 80% power (blue line, right axis), given a fixed ‘true’ effect size (d = 0.4; adapted from Perkins et al.³⁰). ICC, intraclass correlation coefficient.

Another approach to enhancing signal to noise in PGx studies is to focus on treatment-related adverse effects as the phenotype of interest. In other branches of medicine, there have been notable successes in using PGx to identify powerful predictors of drug-induced adverse events. A specific human leukocyte antigen (HLA) allele markedly increases the risk for liver injury as a result of fluvoxacillin treatment (odds ratio (OR) = 80.6).³¹ Furthermore, the same HLA allele (HLA-B*5701) provides 100% specificity for development of an immunologically confirmed hypersensitivity reaction to abacavir, a widely used treatment for AIDS;³² the hypersensitivity reaction developed in every one of 23 carriers of the *5701 allele. In contrast, the overall occurrence rate is < 3% in noncarriers.³² More recently, HLA genetic markers were also found to be associated with antiepileptic (for example, lamotrogine) drug-induced Stevens–Johnson syndrome.^33,34

Rare, yet critical, adverse effects may also occur with psychotropic drugs. Most notably, clozapine is the only antipsychotic with demonstrated superiority for treatment-resistant schizophrenia,^35,36 yet it remains clinically underutilized in part because of its association with agranulocytosis, a potentially fatal blood dyscrasia observed in < 1% of patients.³⁷ The concomitant burden of routine blood monitoring could potentially be lifted by the identification of a PGx biomarker enabling clinicians to ascertain risk for agranulocytosis a priori. A recent candidate gene study demonstrated some promise in this regard, detecting a replicated association of an allele at the HLA-DQB1 locus with the risk of agranulocytosis in two small clozapine-treated cohorts.³⁸ The ORs were extremely high (OR = 16.86), especially compared with those typically reported in PGx studies of clinical efficacy/symptom response; nearly 90% of allele carriers developed agranulocytosis (see Figure 2). Unfortunately, the overall sensitivity of the marker was only 21%, indicating that a majority of individuals who develop agranulocytosis are not carriers of the allele, and presumably have other genetic risk factors. Thus, a more comprehensive risk profile would be necessary in order to obviate the need for invasive monitoring.

Figure 2.

Proportion of clozapine-induced agranulocytosis cases and controls in patients with and without the HLA-DQB1 marker (data extracted from Athanasiou et al.³⁸).

A more common adverse effect, with serious consequences for morbidity and mortality, is anti-psychotic drug-induced weight gain. Despite a substantial body of research on this side effect, recent data suggest that its severity has been consistently underestimated by studies in chronically treated adult populations. For example, such studies typically reveal an acute (≤12-week) body mass index increase of less than one unit (kg m^–2) for risperidone.^39,40 Pediatric studies, although not nearly as common in the literature, demonstrate consistently greater effect sizes given similar methodologies.^41,42 In a recent paper,⁴³ we reported on the weight and metabolic effects of second-generation antipsychotic drugs in a unique cohort of 272 antipsychotic drug-naive (defined as ≤1 week of prior treatment) pediatric patients beginning initial treatment with one of four second-generation antipsychotic drugs (olanzapine, risperidone, quetiapine or aripiprazole). The amount of drug-induced weight gain was dramatic—patients gained significant weight on each of the second-generation antipsychotic drugs with an overall mean weight gain of > 10 lb at 12 weeks of treatment (Figure 3). The amount of weight gain was similar across the age range (adjusted for height), and was not affected by pubertal status, ethnicity or sex of the subjects. To demonstrate the degree of underestimation in the general literature, it is noteworthy that the weight gain attributed to risperidone, for example, was more than four times greater in our sample than in the CATIE report.⁴⁰

Figure 3.

Weight gain observed after 12 weeks of exposure to second-generation antipsychotics in drug-naive youth (adapted from Correll et al.⁴³).

These data suggest that pharmacogenetic studies of weight gain that include previously treated patients are not optimally designed. Variable histories of prior drug exposure between subjects may confound attempts to identify subtle genetic effects on a complex phenotype such as weight gain. Consistent with this hypothesis, Reynolds et al.⁴⁴ studied 123 antipsychotic drug-naive schizophrenic Chinese patients and found that a promoter region polymorphism—759 C/T in the 5-HT_2C receptor gene—significantly influenced weight gain following antipsychotic treatment. Study participants with the T allele at this locus gained significantly less weight than subjects with the C allele at 6 and 10 weeks of treatment. This effect was observed in patients receiving risperidone or chlorpromazine, regardless of gender, and remained significant after exclusion of subjects who were either underweight or obese at baseline. Moreover, none of the 27 subjects carrying the T allele met criteria for severe weight gain ( > 7% increase from baseline body weight) after 6 weeks of treatment, compared with 28% of the 96 subjects without the T allele. Templeman et al.⁴⁵ reported similar results in a small first-episode cohort treated with a mix of antipsychotic medications including olanzapine. As expected, subsequent studies in previously treated patients have been less robust,² although meta-analysis confirms that the effect is not likely to be false positive.⁴⁶

Similar results were recently reported for the functional promoter region variant in DRD2 in a study of antipsychotic-induced weight gain in first-episode patients with minimal prior antipsychotic exposure.⁴⁷ Carriers of a single-nucleotide deletion in the DRD2 promoter (–141C Ins/Del) demonstrated substantially more weight gain than noncarriers after 6 weeks of treatment, regardless of medication (risperidone or olanzapine). Again, the effect sizes were relatively large; mean weight gain in deletion carriers at 6 weeks was ~6 lb higher than in noncarriers.

Adherence in pharmacogenetic studies

It is well recognized by clinical researchers that treatment adherence is poor among psychiatric patients. A recent review demonstrated that the rates of nonadherence are 28 to 52% in patients with major depressive disorder, 20 to 50% in patients with bipolar disorder and 20 to 72% in those with schizophrenia.⁴⁸ Typical methods of assessing adherence include patients’ self-report, family report, pill count, blood levels and pharmacy refill records, all of which may still underestimate the scope of the problem. In the CATIE trial, 74% of patients stopped their initially assigned medications within 18 months, and almost 30% stopped medication because of ‘patient’s decision’.⁴⁰

Not only can medication nonadherence result in symptom relapse clinically, but it can also lead to weakened signals in research studies of adverse events. However, this important issue has often been overlooked in PGx research, and the effect of nonadherence on statistical power has not been quantified. Intuitively, if a substantial proportion of subjects are nonadherent with treatment, it would be very difficult to detect a significant genotype–phenotype (that is, response to an administered drug) relationship regardless of the strength of the effect of the genetic marker. This could be particularly important for studies of side effects such as weight gain, as almost all subjects with documented adherence to second-generation antipsychotics gain some degree of weight;⁴³ a PGx study that failed to formally assess medication adherence would misclassify nonadherent subjects as impervious to weight gain.

To quantify the effects of medication nonadherence on statistical power in PGx studies, we conducted a Monte Carlo simulation study using the R statistical programming language.⁴⁹ In this simulation, modeled after a recently published PGx weight gain study by our group,⁴⁷ we assumed that a specific genetic marker increased one’s susceptibility to weight gain during 16 weeks of antipsychotic drug treatment. The primary analysis compares two genotypic groups (for example, marker carriers vs noncarriers). Medication nonadherent subjects are assumed to gain no weight in average,⁴³ and are randomly distributed between both genotype groups. After setting appropriate statistical parameters (that is, mean and s.d. of weight gain in each group, sample size, frequency of rare genotype, percentage of nonadherence in the sample and so on), the R program randomly generated 200 000 samples and performed a t-test on the difference between means of the two genotype groups in each sample. Power was calculated as the percentage of samples with a significant t-test (P < 0.05).

Figure 4 shows the power curves for different sample sizes as a function of nonadherence rate in the sample, with a 20% frequency of risk genotype and a moderate true effect size (that is, Cohen’s d = 0.50), which could represent a typical scenario in PGx studies. As the nonadherence rate increases, statistical power for detecting a significant signal drops down rapidly. Even with an N of 400, which would be a large sample size in pharmacogenetic studies, power drops below 0.70 when the nonadherence rate reaches 50%, which is not uncommon in a clinical trial.⁵⁰ However, by reducing the nonadherence rate from 50 to 10%, a sample size of 200 is adequately powered (~0.80). Therefore, by maximizing medication adherence, total sample sizes can be cut in half with maintenance of adequate study power.

Figure 4.

Effect of medication nonadherence rate on statistical power as a function of sample size in a simulated pharmacogenetic study of antipsychotic-induced weight gain, assuming risk genotype frequency of 20% and true effect size of 0.50.

Empirical support for this hypothesis is provided by recent studies of antipsychotic-induced adverse events, in which observed effect sizes were significantly greater than commonly reported. For example, in the clozapine-induced agranulocytosis study,³⁸ treated subjects were required to undergo weekly venipuncture for white blood cell counts; subjects who did not comply with venipuncture were restricted from receiving clozapine and therefore not included in the PGx analysis. In a recently completed PGx study of antipsychotic-induced weight gain (AK Malhotra et al., unpublished data), plasma antipsychotic drug plasma levels were collected at each study visit and subjects with undetectable levels were excluded from the genetic analysis. With this analytic plan, a sample of 139 subjects was sufficient to detect a genome-wide significant result, subsequently replicated in two additional samples with monitored adherence. These data suggest that careful attention to adherence could markedly enhance study power at lower overall cost than additional recruitment of larger sample sizes. Future PGx studies may consider use of ingestible event markers, a new technology in which digestible sensors are incorporated into oral medications. After activation by stomach fluids, the sensor can confirm ingestion of medication and report a number of physiological parameters to a receiver worn as a skin patch (www.proteusbiomed.com). Although the introduction of this technology into psychiatric research clearly poses practical challenges, it would represent a gold standard of accuracy for measurement of adherence.

Unique aspects of pharmacogenetics in psychiatry

The preceding sections present several strategies to enhance power in PGx studies of psychotropic medications, including use of reliable phenotypic measurements, a focus on first-episode subjects and careful monitoring of adherence. Nevertheless, it could be argued that these strategies would remain futile in the face of very small effect sizes and economically constrained sample sizes. For example, GWASs of psychiatric diagnostic phenotypes have suggested that effect sizes for individual common alleles are extremely small. The few risk alleles that have been replicably identified in such studies demonstrate ORs of ≤1.2, with each single-nucleotide polymorphism accounting for < 1% of the heritability of disorders such as schizophrenia, bipolar disorder and major depressive disorder.⁵¹ Similarly, small effect sizes have been the norm in GWASs for susceptibility to many complex diseases, including diabetes and cardiovascular disease, as well as quantitative traits such as height and weight, despite high heritability of these phenotypes.⁵² Consequently, GWASs for both disease susceptibility and normally distributed traits in the general population are now thought to require tens of thousands of subjects to account for even a small portion of the heritability, and such sample sizes are out of reach for PGx clinical trials.

However, it does not necessarily follow that pharmacogenetic effect sizes will mirror those observed in disease susceptibility studies. As noted previously, the ORs for specific HLA alleles implicated in rare but serious adverse reactions range from ~17 for clozapine-induced agranulocytosis³⁸ to ≥80 for other medication reactions.³¹ Loss-of-function pharmacokinetic genotypes can lead to a quadrupling of risk of bleeding events for patients treated with warfarin⁵³ or stent thrombosis in patients treated with clopidogrel.⁵⁴ More broadly, effect sizes (for ORs) greater than two are not uncommon in pharmacogenetic studies, including those in psychiatry. For example, the effects of a particular promoter region variant (rs3813929, or −759 T/C) in the gene encoding serotonin_2C receptors (HTR2C) have been robust; a meta-analysis of eight studies demonstrated a doubling of risk for clinically significant ( > 7%) weight gain from baseline associated with the C allele at this single-nucleotide polymorphism.⁵⁵

These effect sizes are possible because PGx studies are fundamentally based upon a time-limited, within-subject measurement: the change of a particular phenotype during a relatively brief clinical trial. Pharmacogenetic studies therefore control the vast array of unmeasured environmental factors working over variable amounts of time that can influence phenotypes measured in a cross-sectional GWAS (for example, of body mass index) conducted in the general population. Similarly, although there may be a multiplicity of subtle genetic and environmental pathways for development of psychiatric illness, it is reasonable that a few key genes in relevant pharmacokinetic and pharmacodynamic pathways can exert relatively large effect sizes in a clinical trials context. Thus, pharmacogenetic GWASs may be sufficiently powered with sample sizes measured in the hundreds, rather than in the tens of thousands.

Moreover, pharmacologic effects may also be independent of diagnosis or other historical and clinical characteristics of patients, permitting the pooling of studies of specific pharmacologic agents across multiple diagnostic groups. For example, the phenotype of antipsychotic-induced weight gain does not appear to differ in patients with schizophrenia, bipolar disorder or nonpsychotic behavioral disturbances,^41,43 and initial pharmacogenetic data suggest no difference across diagnostic groups as well (AK Malhotra et al., unpublished data). This consideration may mitigate, to some extent, the difficulties in recruiting and following well-characterized clinical trial samples.

Finally, for PGx to be clinically useful, large PGx clinical trials of head-to-head drug comparisons are needed to validate the strategy of selecting and dosing drugs based on genetic testing. Genetic markers need to be shown as true moderators of differential treatment and clinical outcome to be capable of guiding drug selection.⁵⁶ CYP2D6 is a good example. If a schizophrenic patient is a poor metabolizer, the clinician may choose quetiapine or ziprasidone, instead of risperidone or aripiprazole, which are metabolized primarily by CYP2D6. However, there have been no prospective data to test alternative treatment based on genetic markers. Furthermore, it is even more challenging in the case of selecting antipsychotic drugs based on genetic variants of dopamine receptors. The Del allele of −141C Ins/Del in DRD2 is associated with poor response to anti-psychotic drugs,⁹ and also increases the liability of weight gain induced by antipsychotics drugs.⁴⁷ However, all available antipsychotics to date exert their effect by D2 blockade. Even if a patient has the Del allele, there is no fundamentally alternative drug treatment. Future research should focus on developing new effective drugs without D2 antagonism, and thus provide more options in clinical management when a patient is a poor responder because of variants in the DRD2 gene.

Optimizing PGx studies

Unlike other branches of medicine, prospective PGx studies are lacking in psychiatry. A rare example is a recent study of treating severe alcohol abuse with ondansetron, in which patients were randomized by genotypes of the serotonin transporter gene (SLC6A4) and the treatment effect was significantly greater in the L/L genotype group as compared with the L/S and S/S groups.⁵⁷ This study highlighted an essential characteristic of a prospective pharmacogenetic clinical trial,⁵⁸ that is, randomization by genotype. However, to best position PGx testing for clinical practice, studies that match individuals with a particular genotype with a specific effective treatment, that is, testing the genotype × treatment interaction,⁵⁶ would be ideal. The drawback of this approach is that multiple medication arms will be required, increasing sample size requirements and study cost and therefore diminishing feasibility.

To optimize a PGx clinical trial, several key issues should be considered. Table 1 summarizes the pros and cons and alternative options associated with each key issue, many of which have already been discussed in previous sections. In the ideal scenario, we attempt to design a PGx study to examine the efficacy of alternative treatments associated with a hypothetical genetic marker. It is hypothesized that medication treatment A is efficacious for patients with a particular genotype of the marker, but not for patients without the genotype. In contrast, medication treatment B is efficacious for patients without the particular genotype, but not for patients with the genotype. Patients would preferably be in their first episode of illness with minimal prior medication exposure, and could be genotyped at baseline. Minimal turnaround time for genotyping will be needed, as many patients will require rapid initiation of treatment and randomization cannot be delayed for more than 24–48 h in many cases. Patients are then randomized into either A or B treatment, stratified on genotypes. Hence, this is a 2 × 2 randomized factorial design. The primary outcome is efficacy, that is, symptom reduction, which could be rated by a centralized rating system to increase assessment reliability. Medication adherence should be carefully monitored by plasma drug levels, or perhaps by novel techniques such as digestive event markers.

Table 1.

Major decision points in designing a pharmacogenetic/pharmacogenomic (PGx) clinical trial

Issues	Options	Pros	Cons
Sampling	First-episode/drug-naive patients	Eliminate confounding by prior drug exposure ↑ Effect size and power	Difficult to recruit Potential loss of generalizability
	Chronic patients	Easy to recruit	Confound by prior drug history, substance abuse, psychosocial issues Structural/functional brain changes associated with long-term treatment and illness Patients may be resistant to treatment
Choosing a phenotype	Drug efficacy/effectiveness	Most clinically relevant	Symptom rating scales are subjective to bias and reliability issues
	Adverse drug response	Relatively easy to assess with good reliability	Not major focus of clinical intervention
Genotyping	Commercial lab	Professional management and availability	Slower turnaround time (with some exceptions) Difficult to get customized genotyping
	In-house genotyping	Able to customize genotyping Fast turnaround time	More expensive for rapid testing just for the study
Designing intervention	One arm, one drug	Maximized sample size ↑ Homogeneity	Unable to test moderator effect of a genetic marker
	Multiple arms, multiple drugs	Able to test moderator effect of a genetic marker Most relevant in answering clinical questions	Smaller sample size for each drug ↑ Heterogeneity
Adherence monitoring	Blood drug level	Easy to obtain	Post hoc ascertainment of adherence leading to smaller sample size
	Digestible event marker	Real-time ascertainment of adherence enabling intervention to improve adherence and ↑ sample size	More expensive Patients may have difficulty accepting the technology
Minimizing placebo response	Centralized rating system	↑ Reliability Able to access more sites	Technology required Potential loss of confidentiality
	Placebo lead-in phase	Minimize true placebo response	Loss of treatment time ↑ Trial length

It should be recognized, of course, that many practical and logistical problems prevent a single study to achieve the perfect design. Nevertheless, the issues listed in Table 1 should be carefully considered in designing a PGx clinical trial. Short of a perfect PGx study, there are still many opportunities in psychiatric PGx that could be conducted to provide informative data for clinical application. For example, given that the Del allele of −141C Ins/Del in DRD2 is associated with poor response to antipsychotic drugs in schizophrenia,⁹ a clinical trial may be conducted to examine whether patients carrying the Del allele can benefit from early clozapine treatment, instead of going through multiple unsuccessful trials of non-clozapine antipsychotics. Taking advantage of the finding that the −759 C/T polymorphism in HTR2C is associated with antipsychotic-induced weight gain,⁴⁶ a study could be designed such that risk allele carriers are randomly assigned to regular antipsychotic drug treatment plus a weight loss drug such as metformin⁵⁹ versus antipsychotic drug plus placebo. Finally, a prospective, randomized clinical trial is needed to demonstrate the clinical utility of CYP2D6 genotyping. For CYP2D6 poor metabolizers, dosing risperidone at lower levels and/or slower titrations schedules, or choosing alternative drugs that are not metabolized by the CYP2D6 pathway, may be able to increase efficacy, minimize side effects and potentially shorten inpatient stays, resulting in decreased costs that could offset any costs associated with PGx testing.

Conclusions

It is important to consider what goals are reasonably expected of PGx research. Individually tailored therapies remain a somewhat distant goal, if for no other reason that our armamentarium is mechanistically limited. As noted above, as all antipsychotics have effects at the D2 receptor, it may not be possible to recommend a different form of treatment for individuals with DRD2 genotypes indicative of poor response and increased weight gain liability.^9,47 However, such individuals might be given priority for additional clinical attention, given increasing constraints on physician time and hospital length of stay. Similarly, individuals at risk for increased weight gain or other adverse events can be given lower doses, adjunctive therapies (psychosocial or pharmacologic) and/or increased monitoring. An analogy can again be made to the pharmacogenetics of warfarin,⁶⁰ in which it was recently demonstrated that the use of genetic markers to guide dosing cut the rate of adverse events in half. Although pharmacogenetic markers are unlikely to attain perfect sensitivity and specificity in the foreseeable future, they can still usefully inform clinical decision making, clarify prognosis and ultimately help guide the development of novel medication strategies.