Translational Pharmacogenomics: Discovery, Evidence synthesis and Delivery of Race-conscious Medicine

Abstract

Response to medications, the principal treatment modality for acute and chronic diseases, is highly variable, with 40–70% of patients exhibiting lack of efficacy or adverse drug reactions. With approximately 15–30% of this variability explained by genetic variants, pharmacogenomics has become a valuable tool in our armamentarium for optimizing treatments and is poised to play an increasing role in clinical care. This review presents the progress made towards elucidating genetic underpinnings of drug response including discovery of race/ancestry-specific pharmacogenetic variants and discusses the current evidence and evidence framework for actionability. The review is framed in the context of changing demographics and evolving views related to race and ancestry. Finally, it highlights the vital role played by cohort studies in elucidating genetic differences in drug response across race and ancestry and the informal collaborations that have enabled the field to bridge the “bench to bedside” translational gap.

In the United States, 60% of adults have a chronic disease and 42% have two or more, contributing to death and disability and accounting for >90% of the $3.3 trillion annual healthcare costs.^1,2 A similar global pattern of morbidity and mortality poses considerable challenges to health systems and economies not equipped to care for complex and expensive conditions.³

Medication therapy is the mainstay for chronic disease management, accounts for over $330 billion of annual US healthcare costs.^1,2 This does not include the additional costs related to variability in drug response. Variability in drug response, both lack of efficacy and adverse drug reactions (ADRs), is common. Many intrinsic (e.g., body weight, organ function) and extrinsic (e.g., concomitant medications, diet, lifestyle) factors can influence drug response. Genetic factors influence drug pharmacokinetics and pharmacodynamics⁴ and contribute to population differences in both response and susceptibility to ADRs.⁵ Pharmacogenomics, the study of genetic variation influencing response to drugs and tailoring medications to a patients’ genetic makeup is emerging as a valuable tool in our armamentarium for optimizing drug therapy.^4,6

Although racial disparities in chronic disease prevalence, morbidity and mortality are widely appreciated, response to medications is generally assumed to be similar by race. The clear and persistent under-representation of patients from diverse racial groups in randomized clinical trials (RCTs) does not support race-stratified analyses to identify the differential (if any) effect of clinical and genetic factors on drug response.⁷ Therefore, much of our current understanding of racial differences in drug response has been generated through cohort studies. In this review, we present the progress made towards elucidating genetic underpinnings of drug response including discovery of race/ancestry-specific pharmacogenetic variants and discuss the current evidence and evidence framework for actionability. We frame this review in the context of changing national and global demographics; discuss evolving views related to race and ancestry, the vital role played by cohort studies in representing a diverse population, and the collaborations that have bridged the “bench to bedside” translational gap.

We want draw the reader’s attention to use of the term ‘race’ in this review. As RCTs and cohort study participants self-identified their race (Black, White, Asian etc.), we retain this term in reviewing the evidence. Self-reported race categories have served as a starting point for pharmacogenomic research and represent a person’s self-identity. Interrogation of the genome has shed light on another facet of race; namely “biogeographical ancestry,” allowing a more precise determination of genomic variation such as minor allele frequencies (MAF) of actionable genetic markers. For these discussions, we refer to “race/ancestry.” Finally, we bring our current conversations on race: identity versus ancestry in the context of a changing demographic to discuss the challenges ahead.

A brief history and review of the current state of evidence:

Recognition of genetic underpinnings of variable drug response dates back to 510 BC, when Pythagoras observed the occurrence of hemolytic anemia after fava bean ingestion. The attribution of this observation to Glucose 6 phosphate dehydrogenase deficiency (G6PD) laid the foundation for a new field. By the 1950’s deficiency in pseudocholinesterase was associated with prolonged paralysis among patients receiving succinylcholine. This marked the beginnings of a new field; pharmacogenetics (pharmacology + genetics), first coined by Friedrich Vogel in 1959. Early studies in identical twins were followed by case-reports and case-series of exaggerated response to conventional doses of medications. By the 1990’s the field began to burgeon, now emerging as the cornerstone for precision medicine.^8,9

The substantial progress made (Figure 1) demonstrates the scope, span and longitudinal nature of discovery, evidence synthesis and guideline development, and implementation efforts. These achievements are underpinned by a cohesive national/ international effort, supported and guided by the National Institutes of Health (NIH). Much of the momentum was sparked by the convening of the first pharmacogenetic research workgroup (1998), the creation of the Pharmacogenetic Research Network (PGRN)¹⁰ and Pharmacogenomics Knowledge Base (PharmGKB; 2000)¹¹ and the technological advances brought to bear by the completion of the human genome project (2003).¹²

Figure 1: The timeline of pharmacogenomic discovery, the emergence of team science and the development of gene-drug clinical guidelines.

Figure 1a. Pharmacogenetic and pharmacogenomic publications over the last 30 years presented in the context of major developments (https://pubmed.ncbi.nlm.nih.gov/)

• Inset shows the increasing trend in publications by race. Note the trend lines are not to scale with total number of publications for each year

Figure 1b. The 4,547 gene-drug response PharmGKB associations document evidence of the influence of 1096 genes on 820 drugs¹¹

Figure 1c. Expert evidentiary review by the Clinical Pharmacogenetics Implementation Consortium documents the influence of 58 genes on 125 medications with 24 genes and 78 medications meeting Level A or B threshold for actionability¹⁴

Figure 1d. Actionability of the 457 pharmacogenomic biomarkers found in the Food and Drug Administration package labeling¹⁰⁵

Figure 1e. Level of evidence for the 147 pharmacogenetic associations in the Food and Drug Administration tables of pharmacogenetic associations¹⁰⁶

• Section 1: Pharmacogenetic associations for which the Data Support Therapeutic Management Recommendations
• Section 2: Pharmacogenetic Associations for which the Data Indicate a Potential Impact on Safety or Response
• Section 3: Pharmacogenetic Associations for which the Data Demonstrate a Potential Impact on Pharmacokinetic Properties Only

Initial investigations interrogated variation in a small number of genes based on a priori knowledge of pharmacology (absorption, distribution, metabolism and excretion and drug receptors). These gene-centric approaches leveraged well-curated data from patients exhibiting unexpected or exaggerated drug response, led to the identification of genes with large effects. With advances in genome sciences and technology, evolution of pharmacogenetics to pharmacogenomics, shifted focus from gene-centric studies to agnostic genome-wide association studies (GWAS) in large patient cohorts and assessed the entire spectrum of drug response to identify novel variants, inform new biology, and explain the genomic component of variable response. Rigorous individual and team-based reports documented the effects in diverse race groups, fueled discovery of novel gene variants, and provided robust estimates of their impact on drug response (Figure 1a).

To enable these discoveries, investigators stepped outside the conventional paradigm of lab-based research efforts to build consortia, collaborating across laboratories, departments, institutions, countries, and continents to build large, well phenotyped, racially diverse cohorts.¹³ Garnering support and direction from the NIH, networks leveraged established (and new) collaborations to emerge as the primary sources of information on pharmacogenomics. Building on early evidence, predominantly derived from patients of European ancestry, research in non-European populations (Figure 1a inset, not to scale) is increasing. The cohort studies have facilitated the identification of race/ancestry-specific variants, highlighted differences in linkage disequilibrium (LD) and MAF and elucidated their differential impact on drug response by race/ancestry.

The discovery efforts and translation of genomic variation influencing therapeutic effects and adverse drug reactions are catalogued by the PGRN (https://www.pgrn.org/) and genotype-phenotype relationships curated for dissemination by PharmGKB (https://www.pharmgkb.org). PharmGKB also hosts genotype-guided drug selection and dosing guidelines published by the Clinical Pharmacogenetics Implementation Consortium (CPIC), the Royal Dutch Association for the Advancement of Pharmacy - Pharmacogenetics Working Group, the Canadian Pharmacogenomics Network for Drug Safety and other professional societies. The PharmGKB along with a comprehensive catalogue of allelic variation of genes influencing drug response provided by the Pharmacogene Variation Consortium (https://www.pharmvar.org/ CYP Allele Nomenclature until 2018) provides the foundation for expert review and development of gene/drug clinical guidelines by CPIC (https://cpicpgx.org/).¹⁴ The Food and Drug Administration (FDA) catalogues package inserts updated with pharmacogenomic information (https://www.fda.gov/drugs/science-and-research-drugs/table-pharmacogenomic-biomarkers-drug-labeling), and in 2020, released guidance on gene/drug associations with evidence to suggest a genetic contribution to altered drug response (https://www.fda.gov/medical-devices/precision-medicine/table-pharmacogenetic-associations). FDA recognized gene/drug associations are classified into three sections, Section 1: data supports therapeutic management, Section 2: data indicates potential impact on safety/response, and Section 3: data indicates potential impact on pharmacokinetics only.

Investigations have assessed differences in the metabolism (dosing), efficacy, and safety profiles of commonly used drugs, with a majority of these efforts based on gene-centric studies. Among gene-drug associations annotated in PharmGKB, there are currently 4,751 clinical annotations and 715 drugs annotations. (Figure 1b). A majority of these associations (93.2%) are designated as “low-evidence” (PharmGKB level 3–4). However, this designation should not be equated with unimportance. The limited data, lack of replication or conflicting results indicate the need for continued research to support the evidence generation.

Associations where robust data are available undergo an expert review by CPIC. Currently CPIC reviews have been completed for 58 genes and 125 medications (level A through D). Evidence supporting the influence of 24 genes on 78 medications meets the CPIC level A/B evidence threshold and are considered clinically actionable and 25 of the gene-drug associations have published peer-reviewed guidelines (Figure 1c) and support the FDA package insert updates (Figure 1d). Among gene/drug associations acknowledged by the FDA (Figure 1e), the majority (47.7%) are considered to have sufficient evidence to support a pharmacogenetic association, and approximately 35% are expected to influence pharmacokinetic parameters only.

Both CPIC and FDA evidentiary reviews recognize the influence of pharmacogenetic variation on drug dosing, metabolism /pharmacokinetics, safety and efficacy (Figure 2a). Of the 83 drugs significantly impacted by pharmacogenes, 35 (Figure 2b) are commonly used and have consistently ranked in the most commonly used medications in the US by prescription volumes.¹⁵ This highlights the scope of pharmacogenomic actionability across clinical disciplines including oncology, psychiatry, cardiology, neurology, gastroenterology, rheumatology, anesthesiology and pain management, and transplant medicine.

Figure 2: The influence of pharmacogenes on a) drug metabolism, dosage, safety and efficacy b) frequently used medications.

Figure 2 provides an overview of broad impact of pharmacogenes. Both CPIC (left panel Figure 2a)¹⁴ and FDA (right panel Figure 2a)¹⁰⁶ evidentiary reviews categorize this influence as affecting drug pharmacokinetics/ metabolism and dosage to altered safety and efficacy profiles.

b) Commonly used medications influenced by pharmacogenes (bar chart indicates number of prescriptions and rank by prescription volume in the US) and number of medications influence acorss clinical disciplines (pie chart)¹⁵

Representation of race/ ancestry groups in pharmacogenomic discovery:

Among the 30,214 study parameters associated with PharmGKB variant annotations, (Figure 3a; study parameters),¹¹ the majority with known ancestry have included Whites/Europeans (34.6%) and Asians (33.4%). Among Asians, East Asians have accounted for 93.9%, whereas Central/ South Asians have accounted for approximately 6.5%. Fewer studies have focused solely on Blacks/Africans (4.9%) and Hispanics (1.5%). However, 25.5% of studies have included participants of multiple race/ancestry groups and among these, Blacks/Africans (29.9%) and Hispanics (12.2%) have had better representation. Although this provides a general overview of diversity at the variant annotation level, it is important to note that this does not represent the number of unique participants or studies, as studies that identified multiple variants are included for each variant. Among study parameters associated with PharmGKB variant annotations which identified a significant association (p≤0.05) and reported the biogeographical group of participants (Figure 3a; significant associations), the majority of significant findings have been in Whites/ Europeans (36.4%) and Asians (30.8%). Similar to results for all study parameters, fewer significant associations have been identified in studies conducted in Black/ Africans (4.2%) and Hispanics (1.5%); although studies which have included participants of multiple ancestries have had increased Black/ African (30.5%) and Hispanic (15.2%) representation. Despite increased representation in studies, at the individual level (Figure 3a; individual level), only approximately 5% of individuals have been of Black/ African (0.7%), Asian (3.9%), or Hispanic (0.2%) ancestry. Among individuals in studies which included multiple ancestries (16.2%), White/ Europeans (52.4%) and Asians (37.0%) have comprised the majority of individuals. However, Black/ Africans (6.1%) and Hispanics (4.5%) have had increased representation at the individual level when participants of multiple ancestries were included.

Figure 3: Race/ ancestry representation in pharmacogenomic discovery.

Figure 3a. Representation among study parameters associated with PharmGKB variant annotations overall, based on the number of significant associations, and among significant associations at the individual-level.

Data obtained from PharmGKB variant annotation study parameters [https://api.pharmgkb.org/v1/download/file/data/variantAnnotations.zip]. Studies conducted in animals/cells lines or biogeographical groups other than those reported or where the ancestry of participants was unknown or not reported were not included in the assessment. Reported race/ ancestry based on PharmGKB biogeographical groups. Black/ African includes African American/Afro-Carbibbean and Sub-Saharan African, Hispanic/ Latino includes American and Latino groups. This does not represent the number of unique studies and/or participants. Multiple indicates that participants of different ancestries were included in the study. Significant associations included those p≤0.05

Figure 3b. Individual-level representation among drug response related GWAS discovery and replication efforts, including CPIC level A associations only and all GWAS evaluating drug response [https://www.ebi.ac.uk/gwas/].

GWAS based efforts account for 4.3% of associations listed in PharmGKB, and GWAS evaluating drug response account for <1% of total reported GWAS associations. Among drug response GWAS (Figure 3b), chemotherapeutic agents (drug class) and warfarin (individual medication) have been the most extensively studied. Overall GWAS discovery efforts have been Eurocentric with Whites/Europeans accounting for 90% of individuals; while Blacks/Africans, Asians, and Hispanics comprise approximately 3% each. However, replication efforts have better Black/African (8.4%) and Asian (6%) representation. Hispanics remain underrepresented.¹⁶

Pharmacogenomic actionability – implications by race/ ancestry:

To appreciate the impact of actionable guidance to tailor medication regimens, variants meeting the CPIC level-A evidentiary threshold are presented in Table S1 and Figure 4. MAF’s are presented to allow the reader to appreciate the scope of actionability based on racial/ ancestral diversity in their clinical practice. As expected, the majority of clinically impactful variation is attributed to the CYP supergene family.⁶

Figure 4: Distribution of minor allele frequencies for common actionable pharmacogenomic variants by race/ biogeographic ancestry.

Frequencies were obtained from CPIC allele frequency tables for each gene, and with the exception of, G6PD, HLA-A/B, IFNL3, and VKORC1, are based on PharmGKB biogeographic groups [https://www.pharmgkb.org/page/biogeographicalGroups]. African American/Afro-Caribbean: reflect the extensive admixture between African, European, and Indigenous ancestries and, as such, display a unique genetic profile compared to individuals from each of those regions alone. American: includes populations from both North and South America with ancestors predating European colonization, including American Indian, Alaska Native, First Nations, Inuit, and Métis in Canada, and Indigenous peoples of Central and South America. Central/South Asian: includes populations from Pakistan, Sri Lanka, Bangladesh, India, and ranges from Afghanistan to the western border of China. East Asian: includes populations from Japan, Korea, and China, and stretches from mainland Southeast Asia through the islands of Southeast Asia. Also includes portions of central Asia and Russia east of the Ural Mountains. European: includes populations of primarily European descent, including European Americans. European region defined as extending west from the Ural Mountains and south to the Turkish and Bulgarian border. Latino: not defined by an exclusive geographic region, but includes individuals of Mestizo descent, individuals from Latin America, and self-identified Latino individuals in the United States. Individuals reflect mixed Native and Indigenous American, European, and African ancestry. Near Eastern: encompasses populations from northern Africa, the Middle East, and the Caucasus. It includes Turkey and African nations north of the Saharan Desert. Oceanian: includes pre-colonial populations of the Pacific Islands, including Hawaii, Australia, New Zealand and Papua New Guinea.

HLA-A/B and VKORC1 race/ethnic designations based on Human Genome Diversity Project. G6PD frequencies are reported for the following groups: Caucasian, South American, African, and Asian. The Mediterranean variant is also known as Dallas, Panama, Sassari, Cagliari, Birmingham. IFNL3 frequencies are reported based on Genomes Aggregation Database (gnomAD) v3 data in the following biogeographical groups: African/African American, Latino, East Asian, South Asian, and non-Finnish European. Alleles with zero frequency across groups are not reported.

Actionability in Blacks/ Africans:

Among Black/ African individuals, the most actionable pharmacogene is CYP3A5. CYP3A4/5 is responsible for the metabolism of approximately 30% of clinically used medications.⁶ The CYP3A5 non-functional allele (*3) is common in Whites/Europeans, whereas the majority of Blacks/Africans have the CYP3A5 normal function allele (*1). Additionally, Blacks/Africans (and less commonly Hispanics) have a higher prevalence of the *6 and *7 non-functional alleles. The differences in MAF renders CYP3A5 highly actionable in Blacks/African individuals.

Currently, the only CYP3A5 CPIC level A guideline is for tacrolimus, the most commonly used immunosuppressant for patients with organ transplants. Genotype-guided tacrolimus dosing achieves therapeutic immunosuppression more rapidly than a conventional dosing approach.¹⁷ This has implications for the 118,195 patients (29% Black, 20.2% Hispanic, 40% White) currently waiting for an organ transplant in the US (https://optn.transplant.hrsa.gov/data/view-data-reports/national-data/#). Moreover, numerous drugs/drug classes (opioids, statins, anticoagulants, antipsychotics, chemotherapy, antiretrovirals, benzodiazepines, calcium channel blockers, and antiepileptics) have lower-level evidence CYP3A5 genetic associations.¹¹ Given that many pharmacogenomic investigations have been conducted in Whites/Europeans, who primarily lack CYP3A5 function, CYP3A5 has the potential to be more actionable in Black/African individuals and have a broad impact on medications across therapeutic areas.

Similarly, actionable variants in CYP2C9 and CYP2B6, encoding enzymes responsible for the metabolism of approximately 13% and 7% of drugs, respectively,⁶ are more common among Black/African individuals. For CYP2C9, the reduced (*2) and non-functional (*3) alleles are found among all race/ancestry groups. However, Black/African individuals (and rarely other groups) harbor decreased function (*5, *8, and *11) and non-functional (*6) alleles at higher frequencies than individuals of other ancestries. Additionally, CYP2B6*6 (decreased function) is more common among Black/African individuals, and the CYP2B6 non-functional *18 allele is almost exclusively observed in Black/African individuals (~8%), and to a lesser extent in Hispanics (<1%). Although CYP2B6 is responsible for the metabolism of fewer medications, due to its expression in the brain, variants in this gene can contribute to the neurologic adverse effects associated with certain medications,⁶ and may be more applicable to Black/African individuals. Excluding these variants in CYP2C9 and CYP2B6 can result in incorrect metabolizer phenotype assignment and inappropriate therapeutic management decisions.

Differences also exist for Black/African individuals with respect to variation in G6PD, IFNL3, and DPYD. Variation in G6PD, associated with drug-induced hemolytic anemia, is more common among Black/African individuals. While currently only rasburicase and tafenoquine meet the CPIC level A evidence threshold for G6PD, numerous medications including antibiotics (e.g., ciprofloxacin, sulfamethoxazole/trimethoprim), antidiabetics (e.g., glipizide, glimepiride), and commonly used over the counter medications (e.g., aspirin, vitamin c) have CPIC level A/B to C evidence¹⁴ and may become more significant as evidence accumulates. While difficulties in phenotype assignment arise due to the location of G6PD on the X-chromosome, variation still has important implications for Black/African men and women (homozygosity).

Although variation in DPYD, associated with fluoropyrimidine toxicity, is relatively rare, the rs115232898 variant is primarily observed in Black/African individuals, further demonstrating the importance of genotyping for ancestry-specific variants. In addition, variation in IFNL3 influences response to peginterferon alfa and ribavirin therapy when used for hepatitis C treatment,⁴ and Blacks/Africans have the highest frequency of the unfavorable IFNL3 rs12979860 genotype across race/ancestry groups. Although newer agents are preferentially recommended for hepatitis C treatment, peginterferon alfa-2b has been investigated as a potential COVID-19 therapy.¹⁸ This highlights the importance of understanding pharmacogenomic associations that may influence response and/or susceptibility to adverse effects as medications are repurposed for new indications.

Actionability in Asians:

Among pharmacogenomic associations that are more actionable in Asians, the best characterized are CYP2C19 and HLA. CYP2C19 is responsible for the metabolism of a large number of therapeutic drug classes.⁶ Of CPIC level A gene-drug associations, eight have CYP2C19-related recommendations, and the majority of these are among the top 300 most prescribed drugs (antidepressants, proton-pump inhibitors, and antiplatelets) in the US.¹⁵ CYP2C19*2 and *3 are the most common loss of-function alleles in Asians. Up to 12% of Asians are homozygous for loss-of function and over 55% harbor at least one loss-of-function (LoF) allele.¹⁹

Clopidogrel, the most commonly used antiplatelet,¹⁵ requires bioactivation through CYP2C19 to its active form.²⁰ Clopidogrel-treated patients harboring LoF alleles have a higher risk of stent thrombosis and major adverse cardiovascular events (MACE). The higher frequency of LoF alleles in Asians, resulting in reduced or diminished activation of clopidogrel, renders actionability of CYP2C19 vital for 56% of Asians. This clopidogrel-CYP2C19 association is highlighted in Figure 6.

Figure 6:

Highlights pharmacogenomic discoveries in relation to the approval of the antiplatelet agents, and provides a breakdown of representation among the trials leading to the approval of these agents compared to pharmacogenomic efforts.

Figure 6a: Clopidogrel-related publications over the past 30 years and pharmacogenomic discoveries and collaborative efforts leading to major pharmacogenomic trials (Error! Hyperlink reference not valid. 6b: Racial/ancestral breakdown of participants included in the clinical trials leading to the approval of the antiplatelets, clopidogrel, prasugrel, and ticagrelor, and pharmacogenomic sub-analysis conducted within these trials.^{86,91,115–120}

Figure 6c: Racial/ancestral breakdown of pharmacogenomic efforts evaluating the association between CYP2C19 and antiplatelet therapy^{93,95,121–124}

Figure 6d: CPIC guidelines for CYP2C19-guided antiplatelet therapy and breakdown of actionable CYP2C19 metabolizer phenotypes by race/ancestry⁹²

ASA: aspirin; GWAS: genome-wide association study; HPR: high on treatment platelet reactivity; ACS/PCI: acute coronary syndrome/percutaneous coronary intervention; EM/UM: extensive metabolizer/ultrarapid metabolizer; IM: intermediate metabolizer; PM: poor metabolizer; LoF: loss of function; MACE: major adverse cardiac events

Asians have the highest probability of severe dermatological toxicity to medications, including antiepileptic drugs (AEDs), non-steroidal anti-inflammatories, and antibiotics.²¹ The assessment of exquisitely curated cases of Stevens-Johnson syndrome (SJS)²² led to the identification of a 2,500 fold-risk of SJS associated with human leukocyte antigen HLA–B*1502 among Han Chinese patients on carbamazepine therapy. Although SJS is rare, the strength of the association, and the sensitivity and specificity of the gene-outcome association²³ led to an ethnicity-specific black-box warning for carbamazepine by the FDA.²⁴ The HLA-cutaneous toxicity association extends to other medications, including abacavir and allopurinol.

Actionability in Hispanics:

Hispanics have the highest frequency of the NUDT15*2 no function allele, rare in both Whites/Europeans and Blacks/Africans. Patients harboring NUDT15*2 or other NUDT15 no function alleles require lower doses of thiopurine agents (mercaptopurine, azathioprine, and thioguanine), due to increased risk of severe myelosuppression.²⁵ Hispanics also have the highest frequency of the CYP2B6*4 increased function allele. While the guidelines for efavirenz, currently the only CPIC level A drug influenced by CYP2B6,¹⁴ recommend standard dosing in patients with the *4 allele, other drugs with lower-level evidence may be influenced by CYP2B6*4. For example, patients harboring CYP2B6*4 may experience decreased response to mirtazapine or may need higher methadone and bupropion doses.¹¹ As evidence emerges, the higher frequency of the CYP2B6*4 allele in Hispanics will have implications for actionability in this group. It is important to note that in the FDA guidance surrounding the collecting of race and ethnicity data in clinical trials, Hispanic is reported as an ethnicity and not a race.²⁶ This has important implications when evaluating drug response and for pharmacogenetic studies in Hispanics. Compared to other major populations, Hispanics typically have the highest composition of Native American genetic ancestry, and may report “other” race when demographic data is collected in this manner.^27,28 This may limit the identification of genetic influences that may be unique to Hispanics due to higher Native American ancestry, and this should be considered when conducting future studies.

Translational pharmacogenomics: weaving discovery, evidence synthesis and clinical implementation into one seamless framework:

The integration of research findings into practice is enabled by systematic developments across a continuum. For pharmacogenomics, the stages of this continuum²⁹ include genome-based discovery and development of genetic tests /interventions (T1 discovery), synthesis of the collective knowledge into evidence-based guidelines (T2 translation), dissemination and implementation of evidence-based guidelines into clinical practice (T3) and research to evaluate the “real world” health outcomes of pharmacogenomic application in practice (T4).

Three figures present pharmacogenomic efforts across the translational continuum, providing an appreciation for individual and team efforts, sustained focus needed for robust science and the unique contribution of an array of study designs including RCTs, observational cohort studies and pragmatic implementation trials. These examples cover intermediate outcomes (e.g., dose), adverse clinical (safety, efficacy) outcomes, and severe life-threatening (Stevens Johnson syndrome) outcomes. They highlight early Euro-centric discoveries, lack of racial diversity in clinical trials, and the importance of team science in studying diverse populations. Most importantly, the diversity of the investigative team and patient cohorts enabled discovery of race/ancestry-specific variants and demonstrated the importance of MAF in actionability at a population level.

Warfarin: the tale of two genes over the last 30 years:

Since its discovery (1945) and commercial approval (US approval; 1954), warfarin remains the most widely used oral anticoagulant with over 14.7 million prescriptions in the US.¹⁵ The capricious nature of the anticoagulation response to any given warfarin dose demands sophisticated and resource-intensive methods to determine the optimal dose for therapeutic anticoagulation. The intricacies of warfarin metabolism, the complexities of the coagulation system, and the multitude of clinical and genetic factors do not make this a straightforward task.

Three characteristics: commonly used, easily perturbed and variable response, and variability associated with poor outcomes (thromboembolism, hemorrhage) rendered warfarin a robust candidate for pharmacogenomic investigation. Although investigations have identified the influence of several genes on warfarin response (including APOE, GGCX, CALU, CYP4F2),^30–35 the bulk of the evidence supports the influence of polymorphisms in two genes: the cytochrome P450 (CYP) 2C9 (CYP2C9; the principal metabolic pathway) and the vitamin K epoxide reductase complex 1 (VKORC1; the target for warfarin). A majority of the pharmacogenomic reports (Figure 5a) have focused on identifying genetic underpinnings of variability in warfarin dose.

Figure 5.

illustrates the trajectory of discoveries in warfarin pharmacogenomics across races/ancestries, and demonstrates how increased racial representation in pharmacogenomic studies allowed for the identification of race/ancestry-specific variants, culminating in improved pharmacogenomic risk stratification based on self-identified ancestry.

Figure 5a: Warfarin-related publications over the past 30 years, highlighting major discoveries and advances in the understanding of warfarin pharmacogenomics across races/ancestries (https://pubmed.ncbi.nlm.nih.gov/).

Figure 5b: Racial/ancestral breakdown of participants included in warfarin-associated clinical trials. Trials include those leading to warfarin approval, the warfarin arm of trials leading to the approval of direct-acting oral anticoagulants, and pharmacogenomic sub-analyses of these trials.^107–114

Figure 5c: Racial/ancestral breakdown of participants included pharmacogenomic research efforts^{34,46,48,51–54}

Figure 5d: Evolution of CPIC warfarin dosing guidelines from a single-marker recommendation (2011)⁵⁵ to a race-stratified algorithm incorporating multiple genes/variants (2017)⁵⁶

Over 100 drugs are metabolized by CYP2C9 (https://www.pharmvar.org/gene/CYP2C9), including drugs with a narrow therapeutic index (e.g., warfarin, phenytoin), and other commonly prescribed drugs (e.g., tolbutamide, losartan, glipizide, and some nonsteroidal anti-inflammatory drugs. Recognition of the central role of CYP2C9 in warfarin metabolism provided a starting point for gene-centric studies, leading to the identification of CYP2C9*2 (R144C; 1994)³⁶ and CYP2C9*3 (A1075C; 1996),³⁷ both associated with lower warfarin dose (1999) and risk of hemorrhage (2002) for Whites/Europeans.^38,39 The recognition of racial differences in dose requirements,⁴⁰ followed by the identification of CYP2C9 variants unique to Japanese (CYP2C9*4) and African (CYP2C9*5,*6) ancestries⁴¹ set the trajectory for publications (Figure 5a, inset) in diverse race/ancestry groups.

Fifty years after the commercial introduction of warfarin, VKORC1, encoding the target protein for warfarin, was discovered (2004),³⁵ explaining a significant portion of the variability in warfarin dose.^42–44 GWAS analysis confirmed the influence of common variants in CYP2C9, VKORC1 and CYP4F2 among Whites/Europeans.^45,46

The burgeoning effort coalesced to form the International Warfarin Pharmacogenetics Consortium (IWPC) an investigative network across 25 countries. Data on over 5,000 warfarin users was brought to bear to evaluate and quantify the effect of VKORC1 and CYP2C9 in European, African and Asian patients.⁴⁷ This dataset also provided the opportunity to demonstrate that a single variant (−1639 G>A or 1173C>T) was predictive of dose among Asians, Blacks/Africans and Whites/ Europeans, assess differential prediction by race and show that the differences in prediction metrics were explained by differences in MAF for the VKORC1 variant.⁴⁸ The IWPC also provided the foundation for identification of a Black/African-specific novel variant through the conduct of the first GWAS (rs12777823)⁴⁹ and exome sequencing (rs7856096)⁵⁰ approaches. This enabled systematic assessment of individual gene-variants (e.g., CYP4F2)³⁴ and quantified the effect of all known gene-variants among Blacks/Africans.⁵¹

More importantly this international effort provided the rationale for two clinical trials, the European Pharmacogenetics of Anticoagulant Therapy (EU-PACT)⁵² and the Clarification of Oral Anticoagulation through Genetics (COAG),⁵³ which evaluated whether genotype-guided dosing improved anticoagulation control. In EU-PACT,⁵² genotype-guided dosing (compared to standard dosing), improved anticoagulation control (60.3% vs 67.4%, p<0.001) while COAG showed genotype-guided dosing (vs. clinically guided dosing) did not (45.4% vs 45.2%, p=0.91). The differences in diversity among EU-PACT (98.6% European 0.9% African, 0.5% other) and COAG (67% European, 27% African, 6% Hispanic) highlighted the role of race/ ancestry in the discordant findings.

In both trials, variants included in the genotype intervention (CYP2C9*2,*3 VKORC1 rs9923231) and the dosing algorithm were based on findings from Eurocentric cohorts. Although the effect of these variants was at least partially confirmed in other race/ancestry groups, the lack of inclusion of race/ancestry-specific variants (e.g., CYP2C9*5,*6*8*11, rs12777823) and race-adjusted rather than race-stratified dosing algorithms resulted in poor predictive performance of genotype-guided dosing algorithms in Blacks/Africans. Collectively these results demonstrate that the effect of predictors varies by race/ancestry and that the effects are dependent on both, predictor prevalence and its biological effect. This highlights the need for adequate racial representation to facilitate identification of the differential impact of a predictor by race/ancestry and enable adaptive algorithms that can incorporate race-specific effect size to personalize therapy.⁵⁴ The evidence is reflected in the warfarin FDA 2010 label update and 2011 CPIC guidelines⁵⁵ supporting the use of CYP2C9*2,*3 and VKORC1 rs9923231 variants for all patients. The 2017 CPIC guidelines⁵⁶ for adult and pediatric patients recommend against the use of genotype-guided warfarin dosing unless data on additional African-specific variants (CYP2C9*5, *6, *8, *11, rs12777823) is available for self-reported Blacks/Africans (Figure 5d).

Early gene-centric investigations showed increased risk of warfarin-related hemorrhage among patients harboring CYP2C9, and more recently, EPHA7 variants.^39,57,58 However, the differences in outcome definition and incongruity in assessment of confounders render the interpretation of gene-hemorrhage associations tenuous. This explains why, despite the strength of gene-dose association, clinicians and payers have been hesitant to endorse the use of genotype-guided therapy, as it has not been shown to reduce warfarin-related hemorrhage, the leading cause of adverse-drug-event related hospitalizations.⁵⁹

While RCTs are considered the gold standard for research, it is important to recognize that much of our understanding of genomic underpinnings of warfarin response by race/ancestry has been elucidated by well phenotyped cohort studies (Figure 5b, 5c). Race groups were under-represented or un-acknowledged in the Stroke Prevention in Atrial Fibrillation trials (SPAF I-III), the results of which garnered FDA approval for the use of warfarin in atrial fibrillation. The recent direct-acting oral anticoagulant (DOAC) trials recruited >18,000 patients globally, with improved Asian representation. However, Blacks/Africans remain under-represented.

Figure 5 helps illustrates four key issues. First, that the same gene-variant (e.g., VKORC1) can have different impact (effect-size) in dosing algorithms by race/ancestry at the population level, influenced by the MAF in the race group.⁴⁸ Second, failure to include race/ancestry-specific variants and basing decisions on Eurocentric studies can do harm. This has led to the development of race/ancestry specific recommendations for warfarin dosing. Third, gene-outcome associations are phenotype dependent (dose vs. hemorrhage) and synthesis of gene-effects should account for incongruity in assessment of confounders. Finally, evidence synthesis from varied sources including RCTs (large homogenous data) and cohorts (real world data in diverse population), and sustained team science are necessary to understand and address racial differences/ knowledge gaps. In this regard, cohort studies have played a vital role in understanding genomic predictors of medication response in diverse race/ ancestral groups.

Human leukocyte antigen (HLA): the untapped hotspot for pharmacogenomic actionability?

Variation in human leukocyte antigen (HLA) genes is increasingly recognized as a significant predictor of severe drug-induced hypersensitivity reactions (HSR), including Stevens-Johnson syndrome and toxic epidermal necrolysis (SJS/TEN).⁶⁰ HLA variation has been linked to severe cutaneous reactions related to the use of AEDs (HLA-B*15:02, HLA-A*31:01), abacavir (HLA-B*57:01, antiviral), and allopurinol (HLA-B*58:01)

HLA-B*15:02, HLA-A*31:01, and antiepileptic drugs (AEDs):

Epilepsy is among the leading causes of disability, globally.³ For several of the AEDs used to treat epilepsy (and other conditions); variation in HLA genes is associated with increased risk of severe and life-threatening cutaneous reactions. These include phenytoin (US approval; 1953), carbamazepine (US approval; 1968), fosphenytoin (US approval 1996), and oxcarbazepine (US approval; 2000). Although the higher risk of severe and life-threatening cutaneous reaction in patients of Asian descent was recognized, the genetic underpinnings linking HLA-B*15:02 with an over 2,500-fold increased risk of carbamazepine-related SJS/TEN in Han Chinese patients were not discovered until 2004.²²

Replication of the carbamazepine-HLA-B*15:02 association, and 10 to 100-fold higher post-marketing reporting rates of SJS/TEN in Asian countries, led to a FDA clinical review.²⁴ This resulted in the carbamazepine label update with a new black-box warning for HLA-B*15:02, providing the first official update requiring testing based on race/ancestry. Since this update, the FDA labels of phenytoin, fosphenytoin, and oxcarbazepine have also been updated to reflect information about increased risk of toxicity in HLA-B*15:02 carriers. Although these associations are actionable, only carbamazepine carries a black-box warning and a testing required designation for individuals of Asian ancestry.¹¹

While increased risk of carbamazepine-induced toxicity has also been ascribed to HLA-A*31:01, this allele is associated with a broader range of hypersensitivity reactions, including maculopapular exanthema (MPE) and drug reaction with eosinophilia and systemic symptoms (DRESS), in addition to SJS/TEN.⁶¹ The HLA-A*31:01-carbamazepine association, reported in Han Chinese patients, was replicated in Koreans, and further confirmed in independent GWASs in Japanese and Europeans.¹⁶ The carbamazepine FDA package label was updated to reflect this increased risk of toxicity in patients positive for HLA-A*31:01. However, in contrast to HLA-B*15:02, the update only warned that HLA-A*31:01 was actionable and did not provide a recommendation or requirement for testing. Since this update, the HLA-A*31:01 association has been confirmed in Indians with carbamazepine-related SJS/TEN.⁶² Similarly, North Africans with carbamazepine-associated DRESS were found to have higher frequencies of HLA-A*31:01 compared to carbamazepine-tolerant controls.⁶³ Updated CPIC guidance regarding the use of carbamazepine, oxcarbazepine, phenytoin, and fosphenytoin in the context of HLA-A and HLA-B genotypes is available, with recommendations to select alternative AEDs in carriers of HLA-B*15:02 and/or HLA-A*31:01.^61,64

HLA-B*57:01 and abacavir:

Abacavir, approved in the US in 1998, is an antiretroviral agent used to treat HIV-1 infection as a single drug or as part of combination antiretroviral therapy (ART) products (Trizivir, Epzicom, and Triumeq). Despite increased availability of ART, the global burden of HIV/AIDS remains high; being the second leading cause of disability among adults aged 25–49.³ The association of HLA-B*57:01 with increased risk of abacavir-related hypersensitivity reactions (HSR), including SJS/TEN, is not unique to one race/ancestry group and was first reported in cohorts composed predominately (>70%) of White/Europeans.⁶⁰

PREDICT-1 (2008), a large multi-national, double blind study, demonstrated that screening for HLA-B*57:01 was effective in preventing abacavir-related HSR, with a negative predictive value of 100% and positive predictive value of 47.9%.⁶⁵ In SHAPE, a retrospective case-control study, the abacavir-HLA-B*57:01 association was replicated in Black/African and White/European patients. Furthermore, this study also demonstrated the importance of appropriate phenotype ascertainment in pharmacogenomic studies.⁶⁶ While all Black/African and White/European patients with an immunologically confirmed patch test for abacavir-related HSR were HLA-B*57:01 carriers, when all clinically suspected cases were included, only 14% of Black/African patients and 44% of White/European patients were HLA-B*57:01 positive. Results of these studies led to an update of the abacavir FDA black-box warning against use in patients positive for HLA-B*57:01 due to increased risk of HSR, recommendations which are also reflected in the abacavir CPIC guideline.⁶⁷ Recent studies have evaluated the prevalence of HLA-B*57:01 in Nigeria,⁶⁸ and HIV positive patients in the US.⁶⁹ No Nigerian patients were positive for HLA-B*57:01, and among US patients positive for HLA-B*57:01, approximately 85% were White/European and 15% were Black/African. These efforts provide additional evidence to inform implementation efforts.

HLA-B*58:01 and allopurinol:

Since its introduction (US 1966), allopurinol, indicated for the treatment of gout and hyperuricemia, remains the predominant hypouricemic medication,¹⁵ with over 16 million prescriptions in the US, annually. Patients positive for HLA-B*58:01 are at increased risk of severe cutaneous adverse reactions (SCAR) related to allopurinol treatment. This association was first reported in Han Chinese patients,⁶⁰ and further replicated in studies in Japan, Korea, Thailand, and France, including a 2011 GWAS conducted in Japanese patients.^16,70 Due to mounting evidence supporting this association, the 2012 American College of Rheumatology (ACR) clinical practice guidelines for gout recommended HLA-B*58:01 screening for high-risk patients. This included patients of Han Chinese and Thai descent regardless of renal function, and patients of Korean descent with ≥ stage 3 chronic kidney disease.⁶⁰ Additionally, the 2013 CPIC guideline reinforced this recommendation, stating that allopurinol was contraindicated in patients positive for HLA-B*58:01.⁷⁰ A 2016 study of US hospitalizations related to SJS/TEN, found that Asians and Blacks/Africans were overrepresented compared to the national population. Compared to Whites/Europeans, SJS/TEN-related hospitalization rate ratios for Asians and Blacks were 11.9 and 5.0, respectively, which correlated with US HLA-B*58:01 frequencies.⁷¹ In 2018, the allopurinol drug label was updated to include pharmacogenetic guidance, however despite the evidence supporting the association, replication across different races/ancestries, and guidance from medical societies, testing is recommended rather than required. Since this update, studies have also confirmed the HLA-B*58:01-allopurinol association in Vietnamese,⁷² Brazilian,⁷³ and Malaysian patients.⁷⁴ Notably, the 2020 ACR gout treatment guideline update added an additional recommendation for HLA-B*58:01 screening in African Americans.⁷⁵

While RCTs conducted for drug approvals have largely lacked diversity, inclusion in pharmacogenomic studies allows for the identification of associations related to underlying genetic architecture. These examples highlight the evolution of pharmacogenomic knowledge over time and demonstrate how lower-level associations may become more actionable as evidence emerges, particularly for patients of races/ancestries who were not included in discovery efforts. It also reinforces the importance of replicating associations in diverse patient populations. For example, although Asians have a higher risk of drug induced SJS/TEN, to identify this risk, it is necessary to study populations where both the risk allele and drug exposure are prevalent. Prospective trials screening for HLA*58:01 (allopurinol) and HLA-B*15:02 (carbamazepine) conducted in Taiwan, demonstrate that screening for these high-risk alleles prior to initiating therapy can be deployed at a population-level to prevent treatment-related SJS/TEN.^76,77 It is important to note that a recent survey identified a knowledge gap among pediatric neurologists in recognition of racial/ancestral differences in carbamazepine related SJS/TEN risk.⁷⁸ Similarly, a survey among members of the Asia Pacific Association of Allergy, Asthma and Clinical Immunology societies revealed that only 25% of responders stated that HLA testing was mandatory before prescribing carbamazepine, and only 8.3% prior to prescribing allopurinol.⁷⁹ Despite strong evidence and recommendations reinforcing the increased risk of severe HSR in Asians, there is still a lack of clinical uptake and knowledge surrounding this risk.

In addition to the outlined CPIC level A associations, variation in other HLA genes, including HLA-DRB1, HLA-C, HLA-DPB1, HLA-DQA1, has been implicated in drug-induced toxicities.¹¹ While these associations currently have lower level evidence, based on evidence surrounding HLA-mediated drug induced toxicities, race/ancestral differences may influence susceptibility. Given the strong evidence base supporting the influence of HLA variation on drug-induced HSR, the highly polymorphic nature of the HLA region, and the influence of HLA variation on increased disease susceptibility risk,⁶⁰ HLA variation may contribute to immune-mediated HSR across drug classes and/or disease states. Due to complexities associated with the high degree of variation within the HLA region, methods for appropriate genotyping must be considered (e.g., direct sequencing etc.) prior to HLA testing.^60,64

Frequencies of the HLA high-risk alleles associated with drug-induced toxicity vary widely based on geographic distribution. Due to this, there are differences in the impact of implementing these associations in clinical practice. While allele frequencies make associations related to HLA-B*15:02 and AEDs highly actionable in predominately Asian countries, they are less actionable in other geographic locations. Additionally, while the negative predictive values for HLA-B*15:02, HLA-B*57:01, and HLA-B*58:01 are 100%, the positive predictive values range from 1.5% (allopurinol) to 55% (abacavir), which must be considered for implementation efforts^60,64 However, these high-risk alleles are still present in patients in the US and Mexico, although at lower frequencies.

Clopidogrel: the aggregate evidence

Clopidogrel (US approval; 1997) remains the most widely prescribed P2Y12 (purinergic receptor P2Y12) inhibitor (Figure 6a). With over 20 million prescriptions it ranks 39^th among the most frequently prescribed medications in the US.¹⁵ Clopidogrel is a prodrug requiring bioactivation by CYP2C19.²⁰ Indicated for use in patients with acute coronary syndromes (ACS), myocardial infarction, stroke, peripheral artery disease and coronary artery disease, the bioactive thiol metabolite exerts its therapeutic effect by irreversibly blocking the ADP P2Y12 receptor thereby inhibiting of platelet activation and aggregation.

CYP2C19 is a highly polymorphic enzyme with 38 variants currently identified (https://www.pharmvar.org/gene/CYP2C19), the first LoF variant, CYP2C19*2, was identified in 1994.⁸⁰ CYP2C19 is responsible for metabolism of over 10% of drugs in clinical use, including clopidogrel, proton pump inhibitors (e.g., omeprazole), and anxiolytics (e.g., diazepam).

Recognition of the central role of CYP2C19 in clopidogrel bioactivation,²⁰ stimulated research with reports first focusing on clopidogrel pharmacokinetics and pharmacodynamics (Figure 6a). Patients possessing CYP2C19 LoF alleles had lower active metabolite levels and higher residual platelet reactivity (HPR).⁸¹ This association was also confirmed by GWAS analysis.⁸² Attempts to reduce HPR by increasing clopidogrel dose demonstrated some success among patients with one- LoF allele (HPR 52% at 75mg to 10% at 225 or 300 mg) but did not overcome HPR in patients with two-LoF.⁸³ Although HPR has been used as surrogate for MACE, its predictive and discriminatory power is modest.⁸⁴ HPR-guided therapy has not demonstrated improved clinical outcomes and is not endorsed by the cardiology community for routine use.⁸⁵

The influence of CYP2C19 LoF on MACE outcomes has been extensively reported (Figure 6a). The collective evidence was summarized by two important meta-analyses with discordant conclusions.^86,87 Due to the potential bias introduced by inclusion of patients with varying acuity (e.g., acute coronary syndrome, and stable coronary disease), the lack of prospective genotype-based treatment, and despite the FDA’s 2010 black-box warning, the cardiovascular community has not endorsed a genotype-guided strategy.⁸⁸ Along with the US approval of prasugrel (2009) and ticagrelor (2011), innovations in genotyping technology ushered the development of CYP2C19 point-of-care tests.⁸⁹ Compared to clopidogrel, both prasugrel and ticagrelor demonstrated superior efficacy, independent of CYP2C19 genotype, albeit with an increased bleeding risk.^90,91

The availability of CYP2C19 point-of-care tests, alternate P2Y12 inhibitors, and CPIC (2013)⁹² and FDA guidance (Figure 6d), created the starting point for pharmacogenomic implementation efforts. The use of prasugrel or ticagrelor is strongly recommended for poor metabolizers (PMs; harboring 2- LoF alleles) with a moderate recommendation for intermediate metabolizers (IMs; harboring 1- LoF allele).⁹² Institutions implementing genotype-guided P2Y12 selection collaborated with the Implementation of Genomics In pracTicE (IGNITE) pharmacogenomics working group to share best practices. Generating real-world evidence, this multi-institutional national effort has demonstrated that integration of genotype-guided care in the routine care of cardiac patients is feasible and scalable, that ~30% of patients harbor ≥1 CYP2C19 LoF allele, that genotype-guided P2Y12 selection reduces the risk of MACE by 2-fold (clinical utility)⁹³ and is the most cost-effective strategy.⁹⁴

With clopidogrel remaining the dominant P2Y12 inhibitor, its similar efficacy among patients without LoF variants, along with lower bleeding risk and substantially lower costs, two trials recently evaluated whether the use of genotype-guided P2Y12 selection is beneficial. The POPular Genetics trial randomized 2,488 patients undergoing PCI to standard (prasugrel or ticagrelor) or genotype-guided treatment (prasugrel or ticagrelor for patients with CYP2C19 LoF and clopidogrel for patients with no-LoF) across ten European sites. Genotype–guided strategy was noninferior to standard treatment and resulted in significantly lower risk of bleeding.⁹⁵ The Tailor-PCI⁹⁶ trial randomized 5,302 patients undergoing PCI across 40 sites (US, Canada, Mexico and South Korea) to receive point-of-care genotyping (with prescription of ticagrelor for LoF carriers and clopidogrel for non-LoF carriers) versus standard clopidogrel. Patients in the genotype group had a 34% lower risk of the composite endpoint (cardiovascular death, myocardial infarction, stroke, stent thrombosis, and severe recurrent ischemia). However, this failed to meet the statistical significance threshold (p=0.057). There were no differences in the risk of bleeding.

In the most recent synthesis of the evidence, data on 15,949 patients from seven randomized controlled trials were pooled (98% presented with ACS and 77% underwent PCI) with secondary analysis including another 2,859 patients from nonrandomized controlled trials. The risk of ischemic events was 30% lower (RR 0.7; 95%CI 0.59–0.83) among LoF carriers on prasugrel or ticagrelor (7.0%) compared to LoF carries receiving clopidogrel (10.0%). Treatment (prasugrel or ticagrelor vs. clopidogrel) did not influence the rate among patients with no-LoF variants.⁹⁷

The selective use of prasugrel or ticagrelor among CYP2C19 LoF carriers combined with the selective use of clopidogrel among patient with non-LoF alleles has the potential to reduce bleeding complications related to the broad use of more potent P2Y12 inhibitors while avoiding the increased ischemic risk associated with unselected use of clopidogrel. The available evidence confirms the use of genotype-guided P2Y12 inhibitor therapy in poor metabolizers and now supports expanding the FDA black-box warning to include intermediate metabolizers.⁹⁸

Figure 6 helps reinforce four key issues. The first, documentation and adjustment for the baseline phenotype (stable vs. acute disease) and confounders is vitally important. Second, technological advancements can enable point-of-care-genotyping allowing a feasible and scalable implementation of pharmacogenomics to improve patient outcomes. Third, as shown in Figure 6b, diverse race groups were under-represented in RCTs conducted to garner FDA approval, a void that is being addressed by pharmacogenomic studies (Figure 6c) including the recently completed Tailor-PCI trial.⁹⁶ Finally, the strength and consistency of gene-outcome associations are similar across race/ancestry with population level actionability driven by MAF. High MAF prevalence has also supported assessment of CYP2C19-outcome associations in Asian stroke patients treated with clopidogrel. This evidence along with the population prevalence of actionability played a crucial role in the building the recent class action lawsuit by the state of Hawaii against the makers of clopidogrel.⁹⁹

National and Global Implications of Pharmacogenomics:

The collective evidence is presented in juxtaposition to a changing and increasingly diverse population to highlight three key themes, which can be viewed from a regional, national and international perspective. First, to provide an appreciation of therapeutic implications including differences in disease prevalence, frequency of medication use, and actionability based on differences in variant MAF, which may lead to differences in pharmacogenetic actionability among different biogeographic groups. Second, to demonstrate the importance of identifying genetic variants that may be unique to a race/ancestral group. Third, and most importantly, to highlight the potential national and global impact of pharmacogenomics and genomic medicine. To translate this promise into impact, research participation must reflect the biogeographic diversity of our world.

Although the 2020 US census results are awaited, projections indicate an increasingly diverse US population (Figure 7a). Of note, the doubling of the multi-racial population (2.6% to 6.2% in the next 40 years) has special implications for race/ancestry discussed later in the review. Figure 7b displays states with the highest proportion of self-reported African, Asian and Hispanic populations in the US. Asians comprise ~6% of the US population and ~60% of the global population (Figure 7c) and will continue to represent the majority of the global population for the foreseeable future. The African population has grown from 9% in 1950 to 17% in 2020 and will comprise 25% of the global population by 2050. Europe’s contribution towards the world population was 22% in 1950, 9.6% in 2020 and is projected to be 7% by 2050.

Figure 7: National and Global opportunities for pharmacogenomics.

Figure 7a: Racial demographics of the US from 2000 to 2060, based on recorded Census data and future projections¹²⁵

Figure 7b: States with the highest representation of Blacks/Africans, Asians and Hispanic in the US¹²⁶

Figure 7c: Race representation in US pharmacogenomic research and implementation efforts^100,101

Figure 7d: Demographics of the global population¹²⁷

Recent US efforts assessing the prevalence of use of medications among adult and pediatric patients with pharmacogenomic actionability demonstrate abundant opportunity to guide prescribing in pediatric and adult populations and a readiness for implement pharmacogenomics among diverse populations (Figure 7d) in the US.^100,101

Capturing racial identity and ancestry:

The debate around the classification and use of race in clinical practice and biomedical research has been a contentious issue for centuries.¹⁰² The use of race in clinical algorithms,¹⁰³ the stark racial differences in COVID-related outcomes driven by differences in access to care and chronic disease burden, and the increasing attacks on Blacks and more recently Asians, has brought this debate to the forefront. The collective visceral response has rallied global support around the “Black Lives Matter” and “Asian Lives Matter” movements. These events have called upon us to act, reigniting a more thoughtful conversation on recalibrating the use of race in medicine and research.¹⁰⁴

We must recognize that our collective thoughts on race have been shaped, and have shaped the politics of the times. Race, introduced in US medical curricula in 1790, is now entrenched as an essential biologic variable and an independent risk factor for disease rather than a mediator of structural inequalities. Eradicating racism is a moral imperative. However, the call to abolish race from research and clinical practice is far too simplistic and too extreme a stance. We must appreciate the strong correlation between a person’s continent of ancestral origin and self-identified race, recognize situations where self-reported race, despite being an imperfect surrogate, is useful and where the exclusion of race as a variable, could cause harm, especially for the most disadvantaged populations. Lack of racial/ancestral diversity during discovery can result in associations that fail to replicate in patients of other ancestries. Applying this information uniformly without validation in non-Europeans can lead to differential benefit or even harm in patients of non-European ancestry. Figure 5 illustrates this. Warfarin dosing algorithms, where the effect-size (dose adjustment) of variants are based on predominantly Eurocentric data, were tested in the COAG trial.⁵³ The stark differences in dose prediction by self-identified race, and the poorer anticoagulation control achieved among Blacks (vs. Whites)⁵³ highlights that other factors (socio-demographical, clinical, genetic, etc.) need to be considered. Given that the correlation of self-reported Black race with African ancestry for warfarin dose is high (85%±9% discovery; 86%±8% replication), inclusion of novel ancestry-specific markers⁴⁹ can help explain some of the disparity. However, much of the variability remains unexplained,^54,57 leaving room to evaluate whether inclusion of other variables (diet, exercise, social determinants) can bridge the differences. To this end, we must also conduct additional research to understand when differences in drug response are due to population differences in the distribution of the underlying factors. We must acknowledge that our current measure of race does not capture its many facets and constructs. These constructs of identity, ancestry and social determinants of health will become increasingly important as the population becomes more admixed (see Figure 7b; multi-racial population). Therefore, we must develop accurate and precise measurements of racial identity (including social, economic, political and power constructs of race) and integrate genetic granularity of ancestry enabled by advances in genomics.

Ongoing and future research studies should consider capturing the many facets of race and ancestry to disentangle the influence of social, power, economic and biological/genomic components on drug response.

The progress made by genomic medicine has created tangible benefits for science and society. Pharmacogenomic research and implementation efforts have leveraged this opportunity to engage an increasingly diverse population and create a roadmap for integrating genomic medicine into practice. To ensure that all individuals benefit from these advances, we must recognize that our ideas about race and ancestry, entrenched over centuries, will take a sustained (often-uncomfortable) dialogue in order to re-examine, revise, refine and reform them. We must commit to this first step in our quest to deliver race-conscious medicine.