The advantages and challenges of diversity in Pharmacogenomics: Can minority populations bring us closer to implementation?

Abstract

Health disparities exist among minorities in the United States, with differences seen in disease prevalence, mortality, and responses to medications. These differences are multifactorial with genetic variation explains a portion of this variability. Pharmacogenomics aims to find the effect of genetic variations on drug response, with the goal of optimizing drug therapy and development. Although genome-wide association studies have been useful in unbiasedly surveying the genome for genetic drivers of clinically relevant phenotypes, most of these studies have been conducted in mainly European and Asian descent participants, contributing to a growing health disparity in precision medicine. Diversity is important to pharmacogenomic studies, and there may be real advantages to the use of these complex genomes in pharmacogenomics. In this review we will outline some of the advantages and confounders of pharmacogenomics in minorities, the role of genetic variation in pharmacologic pathways and highlight a number of population-specific findings.

1. Disparities in precision medicine

Health disparities exist in a wide range of diseases among minorities in the United States, with differences seen in prevalence, mortality, and responses to medications. These differences are multifactorial and include socioeconomic factors such as access to care, environmental stressors, adequate prescription drug coverage, and institutional racism. But along with these social and environmental factors, genetic variation also explains a portion of this variability. Pharmacogenomics has aimed to find the effect of genetic variations on drug response, with the goal of optimizing drug therapy and development. Taking advantage of the technological advances in genomics, a growing list of clinical biomarkers of drug response and adverse drug reactions have been identified. Although these pharmacogenomically relevant markers have improved our understanding of the underlying mechanisms behind drug treatments, these are often identified in patients of European ancestry and do not always replicate in other populations (1–3), as the allelic frequency, linkage disequilibrium (LD), and confounding environmental factors differ across populations (4).

To date, most of the genome-wide association studies (GWAS) have been performed in people of European descent, resulting in an increased health disparity in precision medicine. A 2009 analysis on disparities in human genomics revealed that 96% of participants in GWAS were of European descent (5). Since then, the proportion of non-European individuals included in GWAS has increased to approximately 20%. Much of this increase is the result of studies conducted in populations of Asian ancestry, with relatively small increase in representation of African Americans, Hispanics or Native populations (6). In a recent scientometric review of GWAS, the authors found an explosion in the cohort sizes of GWAS, with all of the increase coming from very large European studies, albeit with a slight increase in diversity of the replication cohort, though this was again mostly in Asian populations (7). These reviews of ongoing GWAS efforts highlight the growing problem of diversity in genomics and the potential of missing critical population-specific SNPs that are important to the delivery of precision medicine to all. Inclusion of diverse populations in genomic studies is essential for evaluating the accuracy and wider relevance of findings that would help us understand the genetic heterogeneity in complex traits.

In this review we will outline some of the advantages and confounders to the use of minority populations in genomics in general, the role of genetic variation in pharmacologic pathways and highlight some specific examples describing population-specific findings in pharmacogenomics.

2. Potential confounders and advantages of genomics in minority populations

2.1. Ancestry versus race

Most studies of African Americans (usually referring to African Ancestry individuals in the continental US) used self-identified race as the criteria for inclusion in studies. While important social and cultural elements are encompassed by self-identified race, race is the sociological classification of human groups with shared biological characteristics (8). Ancestry, by comparison, can be ascertained by an individual’s genetic information (9). Human populations have migrated many times and mixed with other groups, potentially making the race of individuals differ from their genetic ancestry. Because ancestry varies in admixed populations, this additional source of variation may affect the association of SNPs to the phenotype of interest. Additionally, African ancestry group outside of the continental US use labels such as Black Brazilian, or Afro-Caribbean, as examples. These labels are also derived from a social context and distinct cultural traditions of this group. However, these labels do not always reflect the underlying genetic ancestry which may differ significantly from African Americans (10).

An infamous example of the conflation of race and ancestry can be seen in the clinical trials related to heart failure, which is more prevalent in African Americans with difference in therapeutic response (11). Effort to capitalize on these differences led to the development of the drug BiDil (hydralazine plus isosorbide dinitrate). BiDil showed minimal benefit over enalapril in heart failure patients in general, but it was more effective in the subset of African Americans clinical trial participants (12). This led to the approval and marketing of BiDil as an African American-specific therapy. However, this finding became controversial as no effort to identify the mechanism behind it was reported. This study used race as a proxy for ancestry and hence lost the unique opportunity to discover if ancestry-specific variables (or SNPs) contributed to the difference in response. Interestingly, while more effective in African Americans, there were also white patients that benefited from this medication, suggesting a more complex story than the dichotomous therapeutic effect first reported.

The problems related to ancestry versus race/ethnicity become even more complex in minority populations such as Hispanics as their genome is trichotomous, with ancestral contributions of indigenous American, European, and African ancestry (13). The proportion of genetic ancestry derived from each of these lineages varies substantially among and within each subgroup. As an example, Mexican Americans, are a subgroup of the US populations with origins in the country of Mexico. However, depending on whether their ancestors emigrated from European nations to Mexico or originated from the indigenous people of Mexico or a combination of the aforementioned geographical areas, there can be substantial differences in ancestry among Mexican Americans based on their ancestral origins and geographic locations within Central America. Similarly, Puerto Rican individuals may be comprised of European, Native American and African ancestries at varying amounts. However, population-based studies continue to lump Hispanics together, which negates the potential effect of these differing proportion of ancestry. Of note, the Pharmacogenomics Knowledgebase (PharmGKB) has recently recommended standardization of ancestry groups reported in the pharmacogenomics literature to the following seven geographically defined groups: American, Central/South Asian, East Asian, European, Near Eastern, Oceanian, and Sub-Saharan African, and two admixed groups: African American/Afro-Caribbean and Latino (14). Efforts such as these will help to better communicate the populations in which pharmacogenomic discoveries are made.

Concerted efforts to include ancestry proportions into genetic studies will help us to understand true population-based risk as well as the potential biological drivers, as seen in the recent work in Asthma response (15). This study identified population-specific associations to bronchodilator response in asthma that were linked to genetic regions of either African or European ancestry. These types of studies highlight that not only SNPs but the ancestral background of SNPs may affect their association to clinical phenotypes.

It should be noted that most clinical pharmacogenetic laboratories do not use genome-wide arrays, hence limiting the ability to infer genetic ancestry in patients. Interestingly,, investigators have shown that the DMET Plus array (Illumina) distinguish the continental populations (Asian, African and Europeans) and provide an estimate of ancestry (16), but it’s ability to assess the ancestry of admixed population has not been shown.

2.2. Discovery of new drug targets using minority populations

One of the goals of pharmacogenomics is aimed to identifying genetic variation to improve drug discovery and development. Many diseases such as obesity, asthma and hypertension show differences of prevalence between populations that are multifactorial. The reasons for these health disparities are not fully understood; however, population-specific genetic variants may explain some of the potential drivers of disease and also uncover new drug targets. The best-known example of how population-specific genetic variation was leveraged into intelligent drug target identification was the discovery of PCSK9 in the regulation of plasma low-density lipoprotein (LDL) cholesterol levels. Nonfunctional variants of PCSK9 were associated with low circulating levels of LDL cholesterol (17). About 3% of African Americans were carriers of these variants and had approximately 88% reduction in the incidence of coronary heart disease. These protein coding variants were carried at much lower frequencies in white populations (18). Thus, by leveraging the increased allele frequency of PCSK9 coding SNPs in African Americans, a new drug target was discovered. The identification of PCSK9 gene was only 15 years ago, however, it has been translated into an effective medical therapy in a relatively short period of time. This discovery has ushered in an exciting new era for atherosclerotic cardiovascular disease and drug treatments, highlighting how the use of a minority genome may benefit all populations (19).

2.3. Potential to fine-map causative variants

A unique aspect of genome biology in African ancestry populations is that they have considerably more genetic variation (i.e. SNPs), and lower extent of LD than European or Asian populations. LD is the nonrandom association of alleles at different loci within a chromosome and is the consequence of many factors, including population admixture, genetic drift, mutation, and natural selection (20). These differences in LD structure may result in the differences seen in GWAS findings between populations. This greater amount of genetic variation means that there are SNPs that are found in populations of African ancestry and therefore we would never be able to fully account for the contribution of these SNPs in genetic association by studying European populations alone. A study investigating the structure of genetic variation in major U.S. populations sequenced 3873 genes in 152 unrelated individuals from European (n = 40), Latino/Hispanic (n = 34), Asian (n = 38), and African American (n = 40) populations and observed that African Americans had the highest percentage of rare SNPs (64%), lowest percentage of common SNPs (36%) and most importantly, 45% of all SNPs in this population were unique (21). European genomes have extended LD blocks (22), and therefore when a genetic association is found, one can never be sure which SNP is the true causative allele. Because of the shorter blocks of LD seen in African Americans and hence fewer associated SNPs, investigators may be able to narrow down the region that harbors causative alleles. Also, marker SNPs that can be reliably predictive of pharmacogenomic phenotypes in Europeans will sometimes be poor predictors of the phenotypes in African Americans because the marker SNPs are poorly correlated with the true causal SNP in African populations. Additionally, the allele frequencies of SNPs vary across human ancestry groups (23). These differences may result in differences in statistical power to detect association of SNPs to clinical phenotypes.

Genetic admixture results from the addition of a new genetic lineage into a population. The results of this is a genome which contains a mosaic of both populations. For admixed populations, such as African Americans, genetic ancestry can vary substantially among individuals, with the proportion of African ancestry ranging from 20% to nearly 100% in self-identified African Americans (24). However, at any specific loci the genetic ancestry can vary drastically, even between individuals that have relatively similar proportions of ancestry. This more fine-scaled ancestry is dubbed, local ancestry. Figure 1 illustrates these differences on a chromosomal level. Pharmacogenomics is still at the early stages of how these local ancestry differences affect association studies. However, it is becoming clear that accounting for both proportion of ancestry (also known as global ancestry) and local ancestry is necessary to accurately assess genetic association in admixed populations (25).

Figure 1. Global Ancestry versus Local Ancestry .

The principal component (PC) analysis of 1000 genome ancestral populations (CEU - European, YRI - African and CHB - East Asian) and National Institute of General Medical Sciences (NIGMS) African American (AA) populations (dbGaP Study Accession: phs000211.v1.p1) shows separation of the parental populations, with the AA individuals falling along the axis between the CEU and YRI. The black dots show two individuals with the similar global ancestry. The local ancestry of these two individuals (inferred via RFMix v1.5.4) at each chromosome is shown. While these individuals have similar global ancestry, their local ancestry differs markedly at specific locations. This difference in local ancestry adds an additional variable in genetic association analysis.

3. Pharmacogenomics and minority populations

The differences in allele frequency, LD structure and ancestry outlined above can be seen in findings across pharmacogenomics. These differences manifest as population-specific genetic associations, differences in drug response related to drug metabolism or enzymatic conversion, and adverse events susceptibility. It follows that these complexities result in population-specific considerations when assessing the impact of specific genetic variants on clinically relevant phenotypes. Perhaps more importantly, they highlight the inadequacy of a single variant (or set of variants) to be sufficient for the accurate prediction of drug response in all people. For example, rs7200749a missense variant for VKORC1 gene, is associated with increased warfarin dose requirement (26), and varies widely in frequency between populations. Hence, it can only be identified as associated with drug response in populations in which the allele frequency is high enough to achieve adequate statistical power. The Exome Aggregation Consortium (ExAC) that contains exome sequencing data on individuals from six different ancestral populations (African [n = 5203], South Asian [n = 8256], East Asian [n = 4327], Finnish [n = 3307], Non-Finnish European [n = 33,370], and admixed American/Latino [n = 5789]) showed that half of all the functional variants in drug-related genes are unique to only one of the six populations and only 0.1% of functional variants occur with an allele frequency ≥ 0.1% across all populations (27).

3.1. Cross-population differences in drug response

An FDA review of drug approvals between 2008 and 2013 found that approximately one-fifth of new drugs demonstrated some differences in exposure and/or response across ancestry groups (28). However, identification of the specific variants responsible for these differences require dedicated trials in diverse populations. Many of the SNPs influencing major drug metabolism pathways show striking population differences in allele frequency and enzyme activityTable 1 shows allele frequency differences between global populations for Clinical Pharmacogenetics Implementation Consortium (CPIC) level A actionable recommendations found in current medications.

Table 1:

Allele frequency difference between global populations for Clinical Pharmacogenetics Implementation Consortium (CPIC) level A actionable recommendations (https://cpicpgx.org/genes-drugs/).

Drug	Gene	Disease	Variants	Population frequency				Clinical phenotypes
				EUR	AFR	EAS	AMR
capecitabine	DPYD	Neoplasma	rs75017182 G>C	0.024	0.001	0	0.006	Increased Drug Toxicity
			rs55886062 A>C	0.001	0	0	0	Increased Drug Toxicity
			rs67376798 T>A	0.007	0.001	0	0.003	Increased Drug Toxicity
			rs3918290 C>T	0.005	0.001	0	0.001	Increased Drug Toxicity
citalopram	CYP2C19	Depression	CYP2C19*2 (rs4244285 A>G)	0.145	0.17	0.312	0.105	Decreased metabolism of citalopram in people with Depressive Disorder
			CYP2C19*3 (rs4986893 G>A)	0	0.002	0.056	0	Increased exposure to citalopram or escitalopram
			CYP2C19*17 (rs12248560 C>T)	0.224	0.235	0.015	0.12	Increased median concentration/dose ratio and median parent drug/metabolite ratio
clopidogrel	CYP2C19	Stroke and heart disease	CYP2C19*2 (rs4244285 A>G)	0.145	0.17	0.312	0.105	Diminished platelet response to clopidogrel treatment and poorer cardiovascular outcomes
			CYP2C19*3 (rs4986893 G>A)	0	0.002	0.056	0	Enhance clopidogrel response and an increased bleeding risk in ACS/PCI patients
			CYP2C19*4 (rs28399504 A>G)	0.001	0	0.001	0.003	Poor metabolism of clopidogrel and increased risk for secondary cardiovascular events
			CYP2C19*17 (rs12248560 C>T)	0.224	0.235	0.015	0.12	Poor metabolism of clopidogrel
			rs11568732 T>G	0.069	0.07	0.095	0.036	Increased risk of hemorrhage inpatients on clopidogrel
codeine	CYP2D6	Pain and cough	CYP2D6*4 (rs3892097 C>T)	0.186	0.061	0.002	0.13	Decreased metabolism of codeine in people with Anemia, Sickle Cell
			CYP2D6*6 (rs5030655 A>deletion)	0.02	0.001	0	0.003	Decreased response to codeine in children with Anemia, Sickle Cell
			CYP2D6*17 (rs16947 A>G)	0.657	0.446	0.86	0.673	Decreased response to codeine in children with Anemia, Sickle Cell
			CYP2D6*40	0	0.0619	0	0.0108	Decreased response to codeine in children with Anemia, Sickle Cell
escitalopram	CYP2C19	Depression	CYP2C19*2 (rs4244285 A>G)	0.145	0.17	0.312	0.105	Increased exposure to citalopram or escitalopram
			CYP2C19*3 (rs4986893 G>A)	0	0.002	0.056	0	Increased exposure to citalopram or escitalopram
			CYP2C19*17 (rs12248560 C>T)	0.224	0.235	0.015	0.12	Decreased serum concentration of escitalopram
			rs12248560 C>T	0.224	0.235	0.015	0.12	Decreased serum concentration of escitalopram
fluorouracil	DPYD	Cancer	rs115232898 T>C	0	0.023	0	0.003	Decreased activity of DPYD
			rs3918290 C>T	0.005	0.001	0	0.001	Increased Drug Toxicity
			rs75017182 G>C	0.024	0.001	0	0.006	Increased risk of Drug Toxicity
			rs55886062 A>C	0.001	0	0	0	Decreased activity of DPYD
			rs67376798 T>A	0.007	0.001	0	0.003	Decreased activity of DPYD
fluvoxamine	CYP2D6	Obsessive-compulsive disorder	CYP2D6*4 (rs3892097 C>T)	0.186	0.061	0.002	0.13	Decreased dose of fluvoxamine
			CYP2D6*5	0.02829	0.0638	0.0517	0.0217	Increased fluvoxamine plasma concentrations
			CYP2D6*10 (rs1065852 G>A)	0.202	0.113	0.571	0.148	Increased fluvoxamine plasma concentrations
halothane	CACNA1S	Inhalation anesthetic	rs772226819 G>A	0	0	0.00012	5.96e-05	Increased sensitivity to halothane and isoflurane
irinotecan	UGT1A1	Colon Cancer	rs4148323 G>A	0.007	0.001	0.138	0.012	Increased Neutropenia
isoflurane	CACNA1S	Anesthetic	rs772226819 G>A	0	0	0.00012	5.96e-05	Increased sensitivity to halothane and isoflurane
nortriptyline	CYP2D6	Depression	CYP2D6*4 (rs3892097 C>T)	0.186	0.061	0.002	0.13	Increased risk of Side Effects
			CYP2D6*5	0.028	0.064	0.0517	0.022	Decreased metabolism of nortriptyline
			CYP2D6*10 (rs1065852 G>A)	0.202	0.113	0.571	0.148	decreased metabolism of nortriptyline
			CYP2D6*2xN	0.01119	0.01605	0.0379	0.01896	Decreased metabolism of nortriptyline
			CYP2C19*2 (rs4244825 A>G)	0	0.271	0	0.023	Decreased concentrations of nortriptyline
			CYP2C19*3 (rs4986893 G>A)	0	0.002	0.056	0	Decreased concentrations of nortriptyline
peginterferon alfa-2b	IFNL3	Hepatitis B or C	rs12979860 C>T	0.309	0.669	0.08	0.339	Increased response to peginterferon alfa-2b
			rs8099917 T>G	0.168	0.042	0.076	0.277	Increased response to peginterferon alfa-2a
			rs11881222 A>G	0.293	0.307	0.084	0.372	Increased response to peginterferon alfa-2a
tramadol	CYP2D6	Pain	CYP2D6*4 (rs3892097 C>T)	0.186	0.061	0.002	0.13	Increased risk of non-response and required higher dose
			CYP2D6*10 (rs1065852 G>A)	0.202	0.113	0.571	0.148	Decreased tramadol clearance
			CYP2D6*5	0.028	0.064	0.0517	0.022	Increased risk of non-response and required higher dose
			CYP2D6*3 (rs35742686 T>deletion)	0.019	0.002	0	0.006	Decreased metabolism and decreased response to tramadol
voriconazole	CYP2C19	Fungal infections	CYP2C19*2 (rs4244285 A>G)	0.145	0.17	0.312	0.105	Decreased metabolism of voriconazole=
			CYP2C19*3 (rs4986893 G>A)	0	0.002	0.056	0	Decreased metabolism of voriconazole=
			CYP2C19*17 (rs12248560 C>T)	0.224	0.235	0.015	0.12	Increased dose of voriconazole
warfarin	CYP2C9	Blood clots	rs28371685 C>T	0.002	0.024	0	0.001	Decrease warfarin dose requirement
			CYP2C9*3 (rs1057910 A>C)	0.073	0.002	0.034	0.037	Sensitive to the anticoagulant effect of warfarin, with increased risk of bleeding
			CYP2C9*2 (rs1799853 C>T)	0.124	0.008	0.001	0.099	Decreased metabolism of warfarin
			CYP2C9*8 (rs7900194 G>A)	0.002	0.053	0	0.001	Lowest warfarin maintenance requirement
			rs7089580 A>T	0.222	0.206	0.015	0.118	Lower dose of warfarin requirement
			rs28371686 C>G	0	0.017	0	0.001	Increased dose of warfarin requirement
			rs12777823 G>A	0.151	0.251	0.314	0.107	Decrease warfarin dose requirement

Pharmacogenomic investigations in the past few years have revealed substantial population-differences in the metabolism, efficacy and safety profiles of many clinically important drugs. It is important to note that most pharmacokinetic studies of drugs are conducted in European with little data available in other populations (29). Hence dosing recommendations are highly biased toward Eurocentric pharmacokinetics. A prime example of pharmacokinetic differences is tacrolimus, an immunosuppressive drug used to prevent and treat allograft rejection. Polymorphisms in CYP3A5 have been associated with pharmacokinetic variability and response to the drug. The CYP3A5*3 allele (rs776746), which encodes truncated protein is found at a high frequency (94%) in individuals of European ancestry but at a much lower allele frequency (18%) in African Americans. This difference in allele frequency results in a functional CYP3A5 enzyme in a majority of African Americans (30), while most Europeans do not have a functional enzyme. A recent study of tacrolimus dosing in 50 African Americans showed that the CYP3A5*1 allele is associated with normal function (also referred to as expressers), sub-therapeutic concentrations, and higher dose requirements for tacrolimus compared to non-expressers, i.e. patients with a CYP3A5*3/*3 diplotype (31).

Genetic variations in drug metabolizing enzymes, like SNPs and gene copy number variations (CNVs) are important determinants of drug response. A study on global and local differences in SNP profiles in 283 drug metabolizing enzymes and transporter genes across 62 ethnic groups showed that there is a positive selection on variation in genes encoding for drug metabolizing enzymes and that these genetic differences contribute to population heterogeneity in drug response (32). CYP2D6, contributing to the metabolism of about 25% of clinically used drugs is highly polymorphic with at least 100 SNPs and CNVs identified (33–36). There are large population differences in the frequencies of CYP2D6 alleles, among them the non-functional CYP2D6*3 (rs35742686), *4 (rs3892097), *5 (gene deletion), and *6 (rs5030655) alleles. Of the decreased function alleles, CYP2D6*10 (rs1065852) and *41 (rs28735595) are more common in Asians and Europeans, respectively, while CYP2D6*17 (rs28371706) and *29 (rs61736512) are found in people with African ancestry (37). CYP2D6 gene duplications CNVs (at least three copies) with the ultrarapid metabolizer phenotype are present in approximately 28% of North Africans, Ethiopians, and Arabs; 10% of Caucasians; 3% of African Americans; and 1% of Hispanics, Chinese, and Japanese (33, 37, 38). Therefore, drugs like codeine that are biotransformed by CYP2D6 to active metabolites can have a lack of efficacy in poor metabolizers and exaggerated effects in ultrarapid metabolizers (39, 40). Clinical cases have described toxicity in newborns as a result of breast-feeding mothers taking codeine who are also ultrarapid metabolizers (41–43). This has resulted in changes to FDA labels of codeine-containing medications to highlight this risk. However, recent restrictions on using codeine in children have repercussions for treating Sickle Cell Disease (SCD), which disproportionally impacts African American children (44). Tricyclic antidepressants, selective serotonin reuptake inhibitors (SSRI) or β-blockers that undergo CYP2D6-mediated biotransformation to inactive metabolites may result in adverse effects in poor metabolizers and a lack of efficacy in ultrarapid metabolizers (45–47). These allele frequency differences may explain the observation that blacks require lower doses of tricyclic antidepressants than white patients to attain similar treatment response to major depression (48).

CYP2C9 metabolizes over 20% of currently marketed drugs (49). CYP2C9 *2 (rs1799853) and *3 (rs1057910) are common in Caucasians, but less frequent in African Americans and East Asians, whereas CYP2C9*8 (rs7900194) and CYP2C9*11 (rs28371685) are found exclusively in people of African ancestry. The incorporation of CYP2C9*8 and CYP2C9*11 into genotyping panels has been shown to improve warfarin dose prediction in African Americans (50). Additionally, CYP2C19 metabolizes at least 10% of clinically used drugs including antiplatelet drug, clopidogrel, pain medications, and a number of SSRIs (51). Polymorphic expression of CYP2C19 influences the risk of bleeding as well as non-response as clopidogrel, which requires CYP2C19 mediated conversion to have a therapeutic effect (52). Previous studies have shown that the prevalence of nonfunctional alleles CYP2C19*2 (rs4244285) and CYP2C19*3 (rs4986893) and increased function allele CYP2C19*17 (rs12248560) vary by ancestry. The frequencies of the CYP2C19*2 and CYP2C19*3 alleles are higher in Asian populations. The CYP2C19*17 allele, which is found in African ancestry populations (24%), was associated with an increased 1-year mortality rate and an increased risk of bleeding in African Americans (53). A recent study has shown that the proportion of African ancestry may be associated with the hepatic expression on CYP2C19 (54). Even in genes as important as these, most GWAS of drug response have been conducted in European and Asian populations, leaving the effect of population-specific variants unknown.

3.2. Warfarin pharmacogenomics in diverse populations

Population-differences in drug dose requirement have been seen with the well-studied drug warfarin, used for the prevention and treatment of thrombotic events and venous thrombosis. Effective and therapeutic anticoagulation is difficult due to the dose requirements variability between patients. VKORC1 and CYP2C9 have been established as important contributors to warfarin dose variability for individuals of European or Asian descent, however, they do not fully explain the dose variability for individuals of African descent. Three main variants CYP2C9*2, CYP2C9*3, and VKORC1 −1639G>A (rs9923231) are typically used for estimating warfarin sensitivity. These variants have explained different proportions of the variability in warfarin dose in different populations, mostly due to allele frequency. The mean minor allele frequency (MAF) of rs9923231 for East Asian populations is 92% compared with 10% for African American populations indicating strong population differences in effect (55). A targeted study of African Americans uncovered a SNP (rs12777823) approximately 37Kb upstream of the CYP2C cluster that was associated with lower warfarin dose requirement in this population (3). Interestingly, this SNP, while present in other populations, was only associated to warfarin dose in African Americans. This difference in SNP association may be due to differences in LD structure. Figure 2 illustrates the LD structure differences between populations at this locus and the consequences on genetic associations.

Figure 2. LD structure differences between populations may affect SNP associations .

The SNP (rs12777823) in CYP2C cluster was shown to associate to warfarin dose but only in African Americans. A previous GWAS of clopidogrel response found rs12777823 to be the top association because of its high LD to CYP2C19*2 in the Amish population (92). The same conclusion could not be made with the African American warfarin GWAS. While both populations carry this allele, linkage disequilibrium (LD) differences are seen in between populations. (A) A diagram of the location of the associated SNPs, rs12777823 and rs4244285 (CYP2C19*2) in relation to the other CYP2C genes. (B) A detailed linkage disequilibrium plot showing the LD blocks in African American and European ancestry at this chromosomal position between these two SNPs, with a magnified portion to show the LD between rs12777823 and rs4244285. The rs12777823 SNP (red box) is found within the same block as several SNPs in the European genome including rs4424285 (blue box), while the rs12777823 is located in an LD block with only one other SNP in African American populations. Although the rs12777823 is in high LD (r² = 0.84) with rs4244285 in Europeans, it is at much lower LD (r² = 0.49) in African Americans. All coordinate positions are according to UCSC genomic build GRCh37/hg19. SNPs along this locus were selected from the HapMap database Haplotype blocks (triangular black shape) were obtained by haplotype analysis using Haploview (version 4.2). LD is displayed by standard color schemes, with bright red for very strong LD (LOD # 2, D’ = 1), blue for intermediate LD (LOD < 2, D’ = 1), and white for no LD (LOD < 2, D’ < 1). (LOD = Log of odds). This LD plot also illustrates the short blocks of LD (i.e. more LD blocks in African ancestry individuals as compared to European individuals at the same locus). (C) Table showing the allele frequency of rs12777823 in different populations and the association to warfarin dose requirement, with the clinical interpretation from CPIC (93).

The following studies highlight novel discoveries that can be made, even in relatively small minority cohorts, by leveraging the genomics of non-Europeans. A study of an Indian population showed the risk allele frequency of CYP4F2*3 (rs2108622) was higher in north Indians (30%−44%), as compared with African American (12%), Caucasian (34%) and Hispanic (23%), suggesting higher warfarin doses requirement for stable anticoagulation (56). A GWAS using an extreme phenotype strategy in African Americans found a novel association between a genetic variant (rs7856096) in FPGS, a gene known to affect folate homeostasis, and lower warfarin dose requirements among African Americans (56). Again, this is a SNP that is only found in African ancestry individuals and hence would play no role in the dose requirements of other populations. Furthermore, a study on Iranians showed high frequencies of the APOE E3 allele (60%) in patients and lower dose requirement of warfarin (57). Additionally, a study by Duconge et. al. on Puerto Ricans described the no function variant, NQO1*2 (rs1800566), which was significantly associated with resistance to warfarin in this population (58).

Bleeding is the most severe complication of warfarin therapy (59). Although effectiveness and safety are routinely monitored by the international normalized ratio (INR), most cases of fatal bleedings occur at a therapeutic INR (60, 61). The frequency of bleeding from warfarin is 15%−20% per year, with life-threatening bleeds accounting for as much as 1%−3% (62). However, these values are based on studies conducted predominantly in patients of European origin and do not account for differences in responsiveness to warfarin across different ancestral groups (63). Patients of African descent are at significantly higher risk of major bleeding from warfarin which contributes to adverse clinical outcomes and heath disparities in anticoagulation therapy in this population (64). Warfarin associated bleeding is known to depend on several patient-related risk factors including demographics, clinical conditions and genetic variants (65). Previous studies reported polymorphisms in warfarin target enzyme, (VKORC1) and warfarin metabolizing enzyme (CYP2C9) associated with increased risk of bleeding from the drug (66, 67). However, these commonly studied genetic variants are not predictive of warfarin-associated bleeding in African Americans. A recent GWAS specifically focused on African Americans identified 4 SNPs in LD (rs115112393, rs16871327, rs78132896, and rs114504854) associated with warfarin-related bleeding at INR < 4, and found within the regulatory region of the gene EPHA7 (68). These variants were shown to effect transcription, and hence may play a role in the regulation of EPHA7 and platelet function. More importantly, these are found only in people of African ancestry and would not have been discovered in previous studies of European ancestry patients. Given the lack of comprehensive genomics studies in African Americans, this discovery, i.e. the identification of a novel gene related to vascular homeostasis, is highly impactful and an important step forward towards pharmacogenetics-guided anticoagulation therapy.

3.3. Antihypertensive Pharmacogenomics in diverse populations

Hypertension is the most common chronic condition with life-long medications. Antihypertensive therapy reduces cardiovascular morbidity and mortality due to stroke, heart failure and ischemic heart disease (69). African Americans have the highest prevalence of hypertension in the world, with age-adjusted prevalence of 44.9% in men and 46.1% in women respectively (70). Furthermore, hypertension has an earlier onset, greater severity, and higher rate of target organ damage, contributing to decreased longevity in African Americans as compared with European Americans (71). A study using data of the blood pressure lowering arm of the Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT-BPLA) (72) showed clinically significant ancestral differences in blood pressure lowering response to both first- and second- line antihypertensive drugs (73). Patients of African ancestry were significantly less responsive to atenolol monotherapy compared to patients of European origin. In contrast, response to amlodipine monotherapy did not differ significantly between patients of European, African and South-Asian origin. Addition of a diuretic to atenolol had similar effect in all three populations. However, upon addition of perindopril to amlodipine monotherapy, patients with Africans ancestry had a lesser response and patients with South Asians ancestry had a greater blood pressure lowering response.

African Americans are more sensitive to sodium intake and thus are more likely to present with increased blood pressure. A study by Tu et. al. showed that increase in aldosterone concentration and decrease in renin activity in plasma were associated with blood pressure in African Americans, which was not seen in European ancestry individuals (74). They also exhibit a significantly poor blood pressure lowering response to beta-blockers and angiotensin-converting enzyme (ACE) inhibitors and ACE receptor blockers compared to European Americans (53). Genetic testing prior to drug prescription has great potential to advance personalized antihypertensive therapy and prevent drug-associated adverse events, though currently unavailable.

It is recommended that the first-line antihypertensive drugs (a calcium channel blocker or thiazide-type diuretics) should be used differently between African-American and non-African American hypertensive individuals (75). Hydrochlorothiazide (HCTZ), the most commonly prescribed thiazide diuretic and antihypertensive agent in the United States, has been associated with a series of adverse effects, including an increased risk of developing diabetes, hypertriglyceridemia and hyperuricemia (76, 77). A GWAS investigating HCTZ-induced changes in uric acid identified five unique gene regions associated with HCTZ-induced uric acid elevations in African Americans (LUC7L2, COX18/ANKRD17, FTO, PADI4, and PARD3B), and one region associated with these elevations in Caucasians (GRIN3A) (78). In another GWAS focused on hypertriglyceridemia, a known adverse effects of thiazide diuretics identified two SNPs, rs12279250 and rs4319515, located in the NELL1 gene, associated with change in fasting plasma triglycerides in African Americans, where each copy of the risk allele was associated with a 28 mg dl⁻¹ increase in the change in triglycerides (77). NELL1 represses adipogenic differentiation, and the authors speculate that HCTZ modulates adipocyte differentiation through NELL1 leading to accumulation of plasma triglycerides in patients carrying the risk allele. These studies demonstrate that including diverse populations in genomic research can uncover novel mechanisms underlying drug effects and adverse events.

3.4. Asthma pharmacogenomics in diverse populations

Asthma is the most common chronic inflammatory disease of the airways among children, which is incurable but can be managed by the regular use of asthma-controlling medications. The prevalence of asthma shows population differences, with prevalence of 18.4% in Puerto Ricans, 13.0% in African Americans, 8.2% in Whites and 4.8% in Mexicans (79). The asthma death rate of Puerto Ricans and African Americans is four-fold higher compare to that of Mexican Americans (80). In addition, drug response to asthma therapies has marked population-differences. It has been demonstrated that Puerto Rican and African American children with asthma were significantly less responsive to albuterol than European American children, which also associated with increased mortality among African-American subjects (81). Although multiple socioeconomic and environmental factors are associated with such disparities in asthma, some of the differences may be due to variations in genetic susceptibility.

A meta-analysis of GWAS of asthma in diverse populations, including individuals of European American, African American, and Latino/Hispanic ancestry, identified a new asthma susceptibility locus at PYHIN1 (rs1102000) in subjects of African descent but not in Latinos/Hispanic or European Americans (p = 3.9 ×10⁻⁹) (82). A GWAS in 2 independent African ancestry populations identified 3 SNPs (rs10515807, rs6052761, rs1435879, mapped to the genes ADRA1B, PRNP, and DPP10 respectively) relevant to asthma and allergic disease, however, none of them were replicated in European ancestry studies (83). Again, the differences in LD and allele frequency may explain these contrasting findings. A whole-genome sequencing pharmacogenetics study identified genetic variants important for albuterol drug response in racially diverse children. In this study, the investigators found rs28450894 (located within NFKB1 and quantitative trait locus for SLC39A8), which is significantly associated with drug response, was predominantly found in African ancestry populations (8.8–28.7%), versus European populations (3.7–7.6%) and Puerto Ricans (6.2%), and Mexicans (1.5%) (80). An admixture mapping meta-analysis of genetic ancestry identified a new locus at 18q21 that contributed to asthma susceptibility and therapies in Latinos (84) as well as a recent paper that identified two novel loci in 8p23 and 8q24 specific to asthma risk in African populations (85). These novel loci found through admixture mapping and population-specific cohort may yield novel drug targets and biomarkers that may be useful if replicated, for patient stratification.

4. The challenge of ancestry diversity in clinical pharmacology and therapeutics

With the knowledge gained from the human genomics, researchers are learning how inherited variations in genes affect individual’s response to medications. These genetic differences can be used to predict the effective medications for a person and to help prevent ADRs. However, the lack of minority studies has limited the applicability of clinical recommendations that can be made from these discoveries in a diverse population such as the US.

The lessons from the warfarin pharmacogenomics clinical trials speak to this issue. In 2013, the EU-PACT trial reported that the anticoagulation control was improved when drug dosing incorporated genetic factors to calculate a patient’s warfarin dose. Meanwhile, the COAG trial reported that the similar genotype guided dosing did not make a difference among patients. Of note, the COAG trial included African Americans while the EU-PACT was conducted in a European ancestry cohort. While the COAG trial showed no improvement in warfarin dosing in the genotype guided arm, it showed lower mean percentage of time in the therapeutic range (TTR) for INR in the genotype guided arm when restricted to only the African American participants. In both these trials, only SNPs found to associate with warfarin dose in European or Asian were used to calculate warfarin dose in the genotype guided arm. Thus, by excluding ancestry-specific SNPs, a real possibility exists for misclassification of African Americans as requiring a normal dose of drug, when they may in reality carry other SNPs that are highly relevant to drug dose. These trials highlight that a “comprehensive” SNP panel must recognize the difference in predictive SNPs between populations and eschew the impulse to use only the information gathered for in the largest discovery cohorts, which bias our implementation studies toward discoveries in Europeans.

The main roadblock to the implementation of a more inclusive precision medicine is well-done large-scale studies such as those seen in whites. These types of studies provide the opportunity for replication of previous findings and, with larger cohort sizes, the ability to find SNPs with smaller effects sizes. Notable cohorts such as the TopMed and MESA have recently been made publicly available with a conscience effort to have a larger proportion of non-European participants (61.4% underrepresented minorities) (86). New efforts have also begun. The National Institute on Minority Health and Health Disparities (NIMHD) began funding a group of five Transdisciplinary Collaborative Centers (TCC) focused on minority genomics, with the goal of incorporating underrepresented US minorities in precision medicine. One of these centers, ACCOuNT (African American Cardiovascular Pharmacogenetic Consortium), was tasked with the discovery of novel genetic variants in African-American related cardiovascular phenotypes and incorporation of African-American specific genetic variations into clinical recommendations that could be delivered back to patient and physicians at the point of care (87). These efforts along with larger national efforts, such as the All of Us, will start to provide the needed data to move the needle towards equality in precision medicine. Importantly, these initiatives have made good-faith efforts to engage minority communities and patient populations in study design and recruitment. These are important steps for minority populations that may be weary of both science and medicine. In addition, these interactions will also help future efforts by developing communication tools needed to engage these communities in research.

Challenges continue in the ability of the scientific communities to adequately evaluate and analyze the genomes of non-European populations within studies. Many of the commercial arrays that are used to genotype participants were developed to capture common genetic variants in populations of European ancestry and enabled efficient imputation of variants from European-descent reference panels. Recent efforts within the human genomics community have identified almost 300,000,000 bp of sequence in the African ancestry genomes that were not represented in the human reference genome (88). This suggests that a more inclusive approach is needed to accurately map the human genome for all populations. The Population Architecture using Genomics and Epidemiology, Phase II (PAGE II) was initiated by the National Human Genome Research Institute to expand our understanding of complex trait loci in diverse ancestral populations. Towards this effort, the Multi-Ethnic Genotyping Array (MEGA) was designed to substantially increase variant coverage across multiple ethnicities that would improve fine-mapping and functional discovery of clinically relevant mutations and uncover novel, population-specific, disease associations (89). Genome-wide SNP panels have continued to improve their coverage of genetic variation in populations outside of Europe. This effort was specifically highlighted in the work of H3Africa and the Welcome Trust, which developed a pan-African genotyping array (90). However, most clinical laboratories do not use genome-wide panels and instead rely on targeted genotyping panels containing clinically relevant SNPs. These SNPs are selected due to their discovery and replication in the scientific literature and thus are biased toward SNPs discovered in European Ancestry subjects. This perpetuates the growing disparity in our ability to deliver pharmacogenomics to non-European populations.

However, with these new efforts, more efficient methods on assessing the translatability of new findings is needed. The impracticality of a clinical trial for every new genetic variant discovered is obvious. Real-world innovative trial designs for implementation such as those seen in the U-PGx trial across Europe (91) may be needed on a national level to better understand the benefits and potential risk of pharmacogenomics implementation.

In conclusion, we have outlined the potential confounders that may result from the addition of admixed minority populations to genomics analysis. Many advantages remain, such as the discovery of population-specific SNPs, the ability to uncover novel biological mechanisms through the discovery of new genes within a biological or disease pathway, and the improvement in the prediction of clinical phenotypes in these understudied populations. Clearly, more work remains as new pharmacogenomics phenotypes are discovered. But with large national and international efforts, these discoveries may be more quickly translated into all populations.