Precision medicine: does ethnicity information complement genotype-based prescribing decisions?

Abstract

Inter-ethnic differences in drug response are all too well known. These are underpinned by a number of factors, including pharmacogenetic differences across various ethnic populations. Precision medicine relies on genotype-based prescribing decisions with the aim of maximizing efficacy and mitigating the risks. When there is no access to genotyping tests, ethnicity is frequently regarded as a proxy of the patient’s probable genotype on the basis of overall population-based frequency of genetic variations in the ethnic group the patient belongs to, with some variations being ethnicity-specific. However, ever-increasing transcontinental migration of populations and the resulting admixing of populations have undermined the utility of self-identified ethnicity in predicting the genetic ancestry, and therefore the genotype, of the patient. An example of the relevance of genetic ancestry of a patient is the inadequate performance of European-derived pharmacogenetic dosing algorithms of warfarin in African Americans, Brazilians and Caribbean Hispanics. Consequently, genotyping a patient potentially requires testing for all known clinically actionable variants that the patient may harbour, and new variants that are likely to be identified using state-of the art next-generation sequencing-based methods. Furthermore, self-identified ethnicity is associated with a number of ethnicity-related attributes and non-genetic factors that potentially influence the risk of phenoconversion (genotype–phenotype discordance), which may adversely impact the success of genotype-based prescribing decisions. Therefore, while genotype-based prescribing decisions are important in implementing precision medicine, ethnicity should not be disregarded.

Introduction

Inter-ethnic differences in drug response are all too well known and have been reviewed previously.^1–3 One of the earliest well-characterized examples is primaquine-induced haemolysis that is more frequent in Africans compared to Caucasians.⁴ Hepatotoxicity associated with ibufenac, a precursor of ibuprofen, led to its removal from the UK market, whereas it was rare in Japan, where it continued to remain on the market for much longer.⁵ More recent examples of inter-ethnic differences in drug safety include gefitinib-induced interstitial lung disease.⁶ This complication, with a mortality rate of 20–50%, is significantly more frequent in the Japanese population compared to non-Japanese populations.^6,7 Another example of inter-ethnic differences in susceptibility to adverse drug reactions (ADRs) is cutaneous reactions to drugs such as abacavir, allopurinol and carbamazepine. For details, the reader is referred to comprehensive reviews by Aihara,⁸ Borroni⁹ and Pan and colleagues.¹⁰ Briefly, however, these reactions are immunologically based and dependent on the presence of specific HLA antigens, the frequencies of which vary markedly between different ethnic groups. For example, a study in Han Chinese in Taiwan showed HLAB*15:02 to be strongly associated with carbamazepine-induced Steven–Johnson syndrome/toxic epidermal necrolysis (SJS/TEN).¹¹ This allele is present in about 8% of Chinese patients compared to only 1% of Caucasians; in Europe, the frequency of carbamazepine-induced SJS/TEN is about 5–6% compared to 25–33% in Taiwan.¹² Another striking example of inter-ethnic differences is BiDil, a fixed-dose combination of 20 mg isosorbide dinitrate and 37.5 mg hydralazine hydrochloride, indicated for the treatment of cardiac failure. The outcomes from three sequential large studies (V-HeFT-I, V-HeFT II and A-HeFT) showed this combination to be effective only in self-identified black patients compared to white Caucasians.^3,13 The third study (A-HeFT) was terminated prematurely because of a statistically significant benefit in the African patients treated with BiDil compared to placebo. Consequently, in June 2005, the US Food and Drug Administration (FDA) approved BiDil for use in the treatment of heart failure in self-identified black patients. BiDil represents the first drug ever to be approved for use in a specific ethnic group. Inter-ethnic differences in drug response are often reflected in geographical differences in the doses of drugs approved for local clinical use,^14–18 as well as in drug labels.¹⁹

Precision medicine is aimed at maximizing efficacy and mitigating the risks of pharmacotherapy by prescribing decisions based on a patient’s genotype, enabling the selection of the right drug at the right dose. This notion has a sound pharmacological basis since drug response in an individual patient is determined by dose–concentration and/or concentration–response relationships, both of which are driven by genetically determined expressions of drug metabolizing enzymes (DMEs) and responsiveness of pharmacological targets. Hitherto, the principal focus of precision medicine has been on genetic variations of DMEs such as CYP2C9, CYP2C19, CYP2D6, UGT1A1 and TPMT, although the attention is also increasingly turning to genetic variants of pharmacological targets and other determinants of drug disposition such as drug transporters.²⁰ Together, CYP2D6, CYP2C9 and CYP2C19 are responsible for phase I metabolism of about 40% of clinically used drugs.²¹ The number of variant alleles of the genes that encode for DMEs has grown steadily over the past decades and numerous no function and decreased function alleles have been defined by the Human Cytochrome P450 (CYP) Allele Nomenclature Database (www.cypalleles.ki.se), which has transitioned to the Pharmacogene Variation Consortium (PharmVar; www.PharmVar.org).²² Not surprisingly, the Clinical Pharmacogenetics Implementation Consortium (CPIC) has promulgated guidelines to facilitate genotype-based prescribing decisions for a number of actionable drug–gene pairs (details can be found at https://cpicpgx.org).

Available evidence, however, indicates that there are not only between-individual but also between-population differences in drug pharmacokinetics and pharmacodynamics that require characterization for global application of precision medicine. Therefore, whether prescribing decisions should be individualized or made precise on the basis of a patient’s genotype or ethnicity is an issue that deserves some attention. Following an evaluation of the importance of ethnicity, genotype, smoking, body weight, age and gender in determining CYP2C9 enzyme activity in Swedes and Koreans, it was concluded that together with genotype, ethnicity and environment factors need to be considered for population-specific dose optimization and global precision medicine.²³

While the terms ‘race’ and ‘ethnicity’ are frequently used interchangeably, each of these terms denotes distinctly different attributes which are widely discussed in academia. Broadly speaking, race refers to a person’s physical appearance, such as colours of skin, eyes and hair and structures of bone/jaw etc., whereas ethnicity refers to racial ancestry plus other factors such as nationality, culture, language and beliefs. Although there are three basic races (Caucasians, Negroids and Mongoloids), there are a large number of ethnic groups, even within one geographical area. In this review, we explore whether information on ethnicity can contribute to making genotype-based prescribing decisions and precision medicine more effective.

Ethnicity and variant alleles controlling drug metabolism

Early indications that different ethnic groups harbour different variants of DME genes came from a number of observations on CYP2D6-mediated metabolism of its substrate drugs. For example, Ghanaians, but not Caucasians, showed trimodality of debrisoquine hydroxylation,²⁴ and their ability to oxidize sparteine, a CYP2D6 substrate, was found to be independent of their capacity for debrisoquine oxidation.²⁵ Reports also emerged describing a dissociation of debrisoquine and metoprolol oxidations in Zambians²⁶ and Nigerians.²⁷ In another study, phenotyping of a black Zimbabwean population with debrisoquine and metoprolol revealed two poor metabolizers (PM) of debrisoquine, but five of metoprolol.²⁸ Interestingly, the two subjects who presented as PMs with both probe drugs were also genotypic PMs (each having two non-functional alleles); in contrast, no known sequence variations could explain the PM phenotype for metoprolol among the three remaining subjects. Notably, two of the three discordant subjects from whom DNA was available did not have a PM genotype. Droll and colleagues²⁹ reported inter-ethnic differences in the disposition of three CYP2D6 substrates (debrisoquine, dextromethorphan and sparteine) among Caucasian, Chinese and Ghanaian extensive metabolizers of CYP2D6 and their sensitivity to inhibition by quinidine. Similarly, Wennerholm and colleagues³⁰ found differences in the disposition of four CYP2D6 probe drugs in black Tanzanians compared with Swedes. Since these early observations, further evidence has accumulated, confirming that there are wide inter-ethnic variations in the frequencies of variant alleles of genes that determine DME activities, with some variants being ethnicity-specific. Nevertheless, it has been shown that as long as ethnicity-specific alleles are accounted for, the kinetics of drug metabolisms are influenced more generally by genetic variations in these DMEs than by ethnicity or geographic region. It is also well known that genetic diversity is greater in Africa than in other continental populations. Hence, it is not surprising that many CYP alleles of clinical relevance show a marked difference in distribution in Africa.^31,32 To highlight global differences, selected allele frequencies for CYP2C9, CYP2C19 and CYP2D6 genes, discussed in more detail in the following sections, are shown in Table 1. A more recent example is a study in Trinidadians and Tobagonians that revealed novel non-functional allelic variants in subjects with Indian heritage.³³ There is now abundant evidence showing that there is widely different prevalence of variant alleles across different ethnic groups, with some variants being confined primarily to one or the other ethnic group. When extrapolating data from one ethnic group to another, it is also self-evident that inter-ethnic differences (in drug pharmacokinetics or response) will be most apparent for those DME alleles that are not only functionally relevant but are also prevalent with widely different ethnic frequencies.

Table 1.

Selection of P450 allelic variation across populations, highlighting variability within and among populations.

Allelic variant	Function	Africa	African American	Caucasian	East Asia	Americas	Middle East	South Central Asia	Oceania
CYP2C9
*2	Decreased	0–9%	1–4%	8–16%	0–1%	0.3–14%	5–27%	2–26%	0–3%
*3	None	0–3%	0.5–2%	4–11%	1–5%	0–6%	2–19%	6–13%	1–4%
*5	Decreased	0–3%	0.7–2.5%	0%	0%	0–2%	0–0.1%	0%	n/a
*6	None	0–2%	0–1.3%	0%	0%	0–1%	0%	0%	n/a
*8	Decreased	2–8%	3–12%	0–1%	0%	0–2%	0–1%	0–1%	n/a
*11	Decreased	1–5%	1–2%	0–1%	0–0.2%	0–1%	0%	0–1%	n/a
CYP2C19
*2	None	4–22%	12–25%	8–27%	6–49%	2–31%	6–24%	9–51%	20–78%
*3	None	0–7%	0–1%	0–6.8%	0–21%	0–4%	0–20%	0–6%	2–33%
*4	None	0%	0%	0–1%	0–0.5%	0–0.2%	n/a	0%	0%
*17	Increased	10–18%	18–22%	11–33%	0–6.2%	1–25%	22–26%	12–18%	3–6%
CYP2D6
*4	None	1–7%	4–8%	8–33%	0–4%	0.2–43%	4–13%	3–18%	0–8%
*5	Gene deletion	1–17%	3–9%	0–9%	0–10%	0–5%	1–4%	0–16%	1–8%
*10	Decreased	3–19%	3–8%	0.4–15%	9–64%	0–12%	1–9%	4–55%	0–6%
*17	Decreased	9–34%	14–26%	0–2.2%	0–0.2%	0–18%	0–3%	0–1%	0–0.2%
*29	Decreased	4–20%	5–8%	0–0.3%	0%	0–11%	0–2%	0–0.2%	0%

Allele frequencies for CYP2C9, CYP2C19 and CYP2D6 are from tables compiled for the Clinical Pharmacology Implementation Consortium (CPIC) and available through PharmGKB. Frequencies are rounded and might slightly deviate from those posted as new literature is added.

n/a, no frequencies are available.

Ethnic variation of CYP2D6

Of the more than 100 alleles catalogued by PharmVar for CYP2D6, many show wide inter-ethnic variations in their prevalence.^34,35 Globally, the functional CYP2D6*1 and *2 alleles are relatively frequent across populations. Among the variants of clinical interest (i.e. those with increased, decreased or no activity), many range widely in their frequencies across different ethnic groups as exemplified in Table 1.^34,35 Consequently, the frequency of CYP2D6 PM individuals varies considerably across the globe, averaging from 0.4–0.5% among Asians to 5–10% in European Caucasians. We discuss CYP2D6*10 below in greater detail to highlight the implications of inter-ethnic differences in the frequencies of CYP2D6 alleles.

At a population level, the frequency of the decreased function CYP2D6*10 allele can be as high as 64% in Asians (averaging 42%), but is considerably lower (3–7%) in other populations, and is the lowest in white Europeans and Oceanians.³⁵ CYP2D6*10 is of particular interest in the context of tamoxifen metabolism and clinical outcomes in patients with breast cancer because CYP2D6 catalyses the bioactivation of tamoxifen to its therapeutically active metabolite, endoxifen.³⁶ CYP2D6*10 appears to be a severely decreased function allele for tamoxifen compromising the bioactivation step. Individuals with CYP2D6*10/*10 genotypes were observed more frequently in both Malaysian Malay (28.9%) and Chinese (57.1%) breast cancer patients,³⁷ and some studies have reported poorer clinical and survival outcomes in (Asian) breast cancer patients with this genotype.^38,39 Kiyotani and colleagues⁴⁰ have reported that tamoxifen dose requires adjustment in patients carrying the CYP2D6*10 allele to maintain therapeutically effective concentrations of endoxifen, and that dose adjustment does not significantly affect the incidence of adverse events. CYP2D6*10 also substantially decreases the metabolic elimination of other drugs. For example, the concentration of, and exposure to, active moieties of atomoxetine (atomoxetine + 4-hydroxy-atomoxetine) in the CYP2D6*10/*10 group are much higher than those in the CYP2D6*1/*1 group, with a potential for adverse effects,⁴¹ although an earlier study had reported that despite a higher mean exposure to atomoxetine in CYP2D6*10/*10 subjects, the adverse events reported by these subjects were indistinguishable from those of other Japanese participants and that comparison of the pharmacokinetics, safety, and tolerability of atomoxetine between Japanese and US subjects showed no clinically meaningful ethnic differences.⁴² Choi and colleagues⁴³ have reported that subjects with a CYP2D6*10/*10 genotype had significantly greater exposure to tamsulosin, potentially placing these subjects at an increased risk of developing drug-induced adverse events, such as hypotension, dizziness and fainting. A prospective study compared the consumption of tramadol in three CYP2D6*10-related genotypes in 70 Chinese gastric cancer patients recovering from gastrectomy. The total consumption of tramadol for 48 h in patients homozygous for CYP2D6*10 (n = 20) was significantly higher than that in patients without CYP2D6*10 (n = 17) and in patients heterozygous for CYP2D6*10 (n = 26), while the consumption did not differ between the last two groups.⁴⁴ CYP2D6*10 has also been shown to have a significant impact on the pharmacokinetics of tramadol in Chinese post-operative patients.⁴⁵

Ethnic variation in CYP2C9

Warfarin, a drug widely used for thromboembolic prophylaxis, is principally metabolized by CYP2C9 and has a narrow therapeutic window, with risks of bleeding or thrombosis when warfarin concentration and/or anticoagulation are outside therapeutic range. The efficacy of genotype-guided warfarin dosing was investigated in two key randomized controlled studies published in 2013, the European Pharmacogenetics of Anticoagulant Therapy (EU-PACT) and Clarification of Optimal Anticoagulation through Genetics (COAG) trials.^46,47 The two studies came to divergent conclusions and various explanations have been advanced to account for this.^48,49 One of the explanations focuses on ethnicity as the two studies were performed in different populations. Whereas COAG included 27% African Americans, only 1% in the EU-PACT study were Africans. Both studies genotyped patients for CYP2C9*2 and *3 and VKORC1-1639A. As reviewed by Pirmohamed and colleagues,⁴⁸ it is known that the CYP2C9*2 and CYP2C9*3 allele frequencies in African Americans (2% and 1% respectively) are much lower than in European Caucasians (10% and 6% respectively). Similarly, the VKORC1-1639A allele is considerably more common in Caucasians compared to African Americans (40% versus 11%). Furthermore, some variants including CYP2C9*8 and *11 are found in about 5% of African Americans, but are either absent or rather rare in Caucasians, and were not taken into account when dosing African American patients in COAG. The less common CYP2C9*5 and *6 alleles are also predominantly found in African Americans (and were not tested in COAG or EU-PACT). It is interesting to note that the CYP2C9*3 allelic frequency is the highest (36.2%) among Jahais, one of Malaysia’s aboriginal populations, making them the most frequent carriers of this allele thus far reported in any ethnic group globally.⁵⁰

Scott and colleagues⁵¹ have reported an African American male with a lower than predicted therapeutic warfarin dose who was initially genotyped as CYP2C9*1/*1, but revealed a CYP2C9*8/*8 genotype upon further analysis. Their study also demonstrated that CYP2C9*8 was the most frequent variant in African Americans, and a combined allele frequency of CYP2C9*2, *3, *5, *6, *8 and *11 of about 13%. Given that most warfarin pharmacogenetic dosing algorithms only recommend testing for CYP2C9*2 and CYP2C9*3, they reported that the inclusion of CYP2C9*8 alone reclassified the phenotypes of almost 10% of African Americans, or when combined with CYP2C9*5, *6 and *11, of more than 15%. Drozda and colleagues⁵² have also shown that genotype-guided warfarin dosing that did not include CYP2C9 variants that are important to African Americans can lead to significant dosing errors in this population. The CPIC have recently updated their guidelines on genotype-based warfarin dosing,⁵³ which now recommends against pharmacogenetic dosing of warfarin in black patients when only CYP2C9*2 and *3 genotype results are available. Based on the performance of nine algorithms in Chinese patients, Peng and colleagues⁵⁴ suggested that dosing algorithms should be devised for specific ethnic groups and preassigned-dose groups for better prediction of warfarin dose. In a study evaluating the performance of 16 previously published warfarin dosing algorithms in Korean patients,⁵⁵ the algorithms had comparable predictive abilities but showed limited accuracy depending on ethnicity, warfarin dose and VKORC1 genotype. These investigators concluded that further studies were necessary in Korean populations to develop genotype-guided warfarin dosing algorithms with greater accuracy.⁵⁵ A recent study in Puerto Ricans identified novel genetic variants that increased the predictability of stable warfarin dosing among Caribbean Hispanics, thus further corroborating the need for population-specific studies.⁵⁶ Taken together, it is not surprising that ethnicity-specific warfarin algorithms continue to appear with predictable regularity.

Although there is a paucity of hard data at present, the above limitations of genotype-guided warfarin therapy in different ethnic populations may be applicable to other narrow therapeutic index drugs undergoing metabolic elimination (phenytoin or oral hypoglycaemic agents) or bioactivation (losartan or cyclophosphamide) by CYP2C9. For example, the CYP2C9*3 allele in particular was reported to be associated with a high risk of phenytoin-induced neurological toxicity in Indian epileptic patients,⁵⁷ but the risk is likely to be associated with other CYP2C9 variants including CYP2C9*5, *6, *8 and *11 in Africans and their descendants. Indeed, CYP2C9*6 was first described in a female African American with phenytoin toxicity.⁵⁸ The risk is further aggravated in patients with compromised CYP2C19 activity, which also contributes to the metabolism of phenytoin. Patients who are carriers of decreased or non-functional alleles of both CYP2C19 and CYP2C9 are rare among Caucasians (about 1% of the population), but are more frequent in Asians (about 10%).⁵⁹

Ethnic variation in CYP2C19

The non-functional CYP2C19*2 and *3 alleles are the most commonly genotyped CYP2C19 alleles. However, CYP2C19*17 shows the greatest inter-ethnic diversity and is responsible for rapid and ultrarapid metabolizer phenotypes. Following a meta-analysis that included 52,181 healthy volunteers, Fricke-Galindo and colleagues⁶⁰ reported that CYP2C19*17 was 42- and 24-fold more frequent in Mediterranean–South Europeans and Middle Eastern populations, respectively, than in East Asians, in whom CYP2C19*2 and CYP2C19*3 alleles were more frequent (Table 1). The frequency of PMs of CYP2C19 varies widely across the globe, ranging from 0.87% in Hispanics to 70.8% in Vanuatu locals.^61–63

Clopidogrel, an antiplatelet agent, is a prodrug that requires bioactivation by CYP2C19 in the prevention of thrombotic episodes. Only 15% of the prodrug is available for transformation to the active moiety; the remaining 85% is hydrolysed by esterases to inactive forms. The impact of CYP2C19 polymorphism on clinical outcomes of patients treated with clopidogrel varies with ethnicity. One meta-analysis of CYP2C19 genotype effect stratified clopidogrel studies by predominant indication (percutaneous coronary intervention versus others) and by ethnicity of the population (Caucasian versus Asian).⁶⁴ The authors concluded that the reported association between CYP2C19 non-functional allele carriage and major cardiovascular outcomes differs based on the ethnicity of the study population, and to a lesser extent, clopidogrel indication, and considered their findings of potentially major importance, given that over 50% of Asians carry one or more such CYP2C19 alleles. In another meta-analysis of 36 studies involving 44,655 patients with coronary artery disease treated with clopidogrel, carriers of one and two decreased function alleles accounted for 42.5% and 10%, respectively, among Asians, while the corresponding frequencies were 25.5% and 2.4%, respectively, among Westerners. The effect of clopidogrel was decreased in only the carriers of two decreased function alleles among the Asians in contrast to the carriers of one or two of these alleles in the Western population. Furthermore, the decreased effect of clopidogrel was significant only after 30 days of treatment in Asians, but mainly within the first 30 days in the Western population.⁶⁵ In contrast, the increased function CYP2C19*17 allele, frequent in Mediterranean–South Europeans and Middle Eastern populations, has been reported as an independent risk factor for bleeding following clopidogrel therapy⁶⁶ and for loss of therapeutic response to the antifungal drug voriconazole.⁶⁷

Substrate-specificity of an allele and ethnicity

Not infrequently, the metabolic activity of an allele is substrate-dependent and this is likely to impact precision medicine. For example, CYP2D6*17, an allele first described in black Africans,⁶⁸ is generally considered as a decreased function allele. Individuals carrying this allele display decreased hydroxylation of debrisoquine,⁶⁸ but exhibit normal hydroxylation of risperidone to 9-hydroxy-risperidone.⁶⁹ Another example is an intronic SNP (IVS8–109T) that can occur on different CYP2C9 haplotypes. This SNP was associated with increased activity using losartan as the probe drug in Ecuadorian subjects,⁷⁰ but decreased activity, using a cocktail of probe drugs, in healthy Swedish individuals.⁷¹ This discordance may be explained by the intronic SNP being in incomplete linkage with other SNP(s) on those haplotypes impacting substrate-specific metabolism. To complicate the matter further, the same intronic SNP has now been reported to lead to decreased metabolic activity towards phenytoin in Mexican Mestizo patients with epilepsy.⁷² One study recently reported that a common variant of P450 Oxidoreductase gene (POR*28) was associated with altered CYP2C9 activity in Swedish, but not Korean, healthy subjects.⁷³ Its impact was determined to be only minimal in a Spanish population.⁷⁴ It is at present unknown whether there are other untested/unknown SNP(s) on the POR*28 allele in Swedes, or other mechanisms at play, that may explain these findings.

Ethnicity and variants of pharmacological targets

Although not studied as extensively as DMEs, genetic variations have also been described for a number of pharmacological targets. The role of pharmacological targets is well illustrated by the indications approved for many small-molecule tyrosine kinase inhibitors (TKI) since these are effective in only those patients who carry specific genetic variants. For example, somatic mutations within the kinase domain of the epidermal growth factor receptor (EGFR) are present in approximately 10% of patients with non-small-cell lung cancer (NSCLC). These mutations were identified in the tyrosine kinase domain of the EGFR gene in eight of nine patients with gefitinib-responsive lung cancer, as compared with none of the seven patients with no response.⁷⁵ The two most common mutations that account for greater than 85% of all EGFR gene mutations are in-frame deletions in exon 19 that occur around an LREA string of amino acids located between residues 747–750 of the EGFR protein and an amino acid substitution in exon 21 (L858R). Mutations in exon 18 are much less frequently observed and account for only about 4% of all EGFR gene mutations. Whereas about 60–70% of the patients harbouring the two common EGFR mutations respond to treatment with EGFR-inhibiting TKIs, only 10–20% of those without these mutations do so.^76–78 In one study, the mutation rate was significantly correlated with gender (women 73.3% versus men 20%) and ethnicity.⁷⁹ In terms of inter-ethnic differences, exon 19 deletions and the exon 21 L858R substitution are present in about 3% of patients from the Middle East, 10% of Caucasian patients and 20–50% of Asian patients with NSCLC.^79–82 African Americans are significantly less likely to harbour these mutations compared to white patients.^83,84 Not surprisingly, the efficacy of TKIs such as gefitinib in NSCLC is superior in females of East Asian descent compared to other populations.⁸⁵

In order to illustrate further the complex role of ethnicity, we briefly summarize below functionally relevant genetic variations in three important drug targets, namely serotonin transporter, cardiac ion channels and organic anion-transporting polypeptide (OATP) transporters.

Serotonin transporter

The serotonin transporter (5-HTT, also known as SERT) is encoded by the SLC6A4 gene. The promoter region contains a functional insertion/deletion polymorphism with two common alleles that have been designated the ‘short’ (S) and ‘long’ (L) alleles. For details of this complex polymorphism, the reader is referred to other reviews.^86–88 Briefly, the S-allele is associated with decreased 5-HTT expression and a number of serotonin-related psychiatric disorders such as bipolar disorder, depression and violent suicide. The majority of the studies published to-date have shown that L-allele carriers have a faster and better response to SSRI antidepressants if they are Caucasians,⁸⁹ in whom the S-allele is also known to be associated with either resistance to treatment or a delayed response. In South Indian patients, a significant association between the LL genotype and response to fluoxetine treatment has also been described.⁹⁰ Tomita and colleagues⁹¹ have reported that among patients with major depression who do not respond to paroxetine, a lower plasma concentration or a lower oral dose of paroxetine might be more effective in those with the SS genotype, and a higher plasma concentration might be more effective in those with the SL or LL genotype. The CPIC guideline on SSRIs acknowledges the accumulating evidence that variations in the genes encoding the 5-HTT and 5-HT_2A (a subtype of the 5-HT₂ receptor that belongs to the serotonin receptor family encoded by the HTR2A gene) receptors are associated with clinical response and adverse effects to SSRIs, and that as additional studies are published, dosing recommendations based on these genes may be warranted.⁹²

There are significant inter-ethnic differences in the prevalence of the 5-HTT S- and L-alleles.⁸⁷ Lotrich and colleagues⁹³ have shown that the L-allele is highly prevalent in Americans of African ancestry, ranging from 77 to 87%, compared to Americans of European ancestry (typically 56–60%). There is a much higher frequency of the S-allele in East Asian (79%) than in European (42%) populations.⁹⁴ Another study determined that the frequency of the S-allele was highest in Indians (42.9%) followed by Malays (23.5%), and was absent in the Chinese.⁹⁵ Following a study of white (n = 47) and Korean (n = 118) participants with major depressive disorder who were treated with escitalopram, Bousman and colleagues⁸⁹ have reported that among those with the LL, but not LS or SS, genotypes, white patients had greater reductions in depression score, response rates and remission rates compared with Koreans. In contrast, Poland and colleagues⁹⁶ have reported that in spite of a significantly greater frequency of the L-allele in African Americans, as compared to Caucasians, the clinical response between the two groups was comparable, with this polymorphism not being significantly associated with clinical response in either ethnic group, and suggested that this apparently contradictory result may be related to drug transporter kinetics,⁹⁷ which may also be impacted by polymorphic expression.

Cardiac ion channels

A large number of drugs across a wide range of therapeutic classes are known to cause prolongation of the QT interval of the surface electrocardiogram.⁹⁸ The duration of this interval is determined by a complex orchestration of currents mediated by cardiac ion channels encoded by KCNH2, KCNE2, KCNQ1, KCNE1 and SCN5A genes. Significant prolongation of QT intervals, with or without the resultant potentially fatal proarrhythmia known as torsade de pointes, is one of the two most common safety issues responsible for either the withdrawal of a drug from the market or prescribing restrictions on its use.^99,100 Available evidence suggests that a number of patients with drug-induced QT interval prolongation harbour variants in ion channel genes that determine the duration of the action potential and, therefore, the QT interval.^101–104 Furthermore, studies in first-degree relatives of patients with drug-induced prolongation of QT interval support a genetic predisposition to drug-induced long QT syndrome.¹⁰⁵ There are marked inter-ethnic differences in the frequency of variant alleles of KCNH2, KCNE2, KCNQ1, KCNE1 and SCN5A genes that encode for cardiac potassium and sodium channels.^106,107 For example, Ackerman and colleagues¹⁰⁶ identified 49 distinct amino acid-altering variants (36 novel): KCNQ1 (n = 16), KCNH2 (n = 25), KCNE1 (n = 5), and KCNE2 (n = 3). More than half of these variants (26/49) were found exclusively in black subjects. Excluding two common polymorphisms, 25% of black subjects had at least one nonsynonymous potassium channel variant compared with 14% of white subjects. Therefore, it is likely that ethnicity may be an additional important determinant of drug-induced QT prolongation. At present, there are hardly any data on prevalence of drug-induced torsade de pointes stratified by ethnicity, but there are some studies indicating that ethnicity may impact a patient’s QT-sensitivity to a drug,¹⁰⁸ although other studies have concluded otherwise.^109–111 In broad terms, the evidence against differential ethnic sensitivity to QT interval prolongation has been gathered with drugs of low QT-prolonging potency such as moxifloxacin, and it is worth appreciating that the greater the potency, the more evident will be the inter-ethnic differences. In one study of 58 patients (20 were African Americans) with atrial fibrillation, ethnicity had a significant impact on efficacy of ibutilide, a potent QT-prolonging drug, with increased rate of conversion sinus rhythm seen in the African American patients and increased non-conversions seen in the white patients. QT intervals were prolonged after drug administration, with a mean change of 24.6 ms for all patients; however, it was greater in African American patients than in white patients (27.4 versus 23.3 ms), and torsade de pointes occurred in four patients, three of whom were African American.¹¹²

Organic anion-transporting polypeptide transporters

OATP transporters are involved in the uptake of a number of drugs and their metabolites at most epithelial barriers. Polymorphisms in the genes (SLCOs) encoding these transporters determine inter-individual variability in drug disposition and response. SLCO1B1-encoded OATP1B1 has been shown to regulate the hepatic uptake of statins. The most common statin-related ADR is skeletal muscle toxicity. Indeed, one agent (cerivastatin) was removed from the market in 2001 as a result. Frequency of this toxicity varies depending on the diagnostic criteria; overall, statin-related myalgias are common, occurring in 1–5% of exposed subjects. A retrospective analysis of all 8610 drug-associated rhabdomyolysis cases reported to the FDA between January 2004 and December 2009 included 2164 with simvastatin, 1039 with atorvastatin and 742 with rosuvastatin.¹¹³

Several sequence variations have been discovered in the SLCO1B1 gene, of which 521T>C and 388A>G have been widely discussed in the literature.^114,115 These two SNPs have been demonstrated to be the most prevalent and functionally important for OATP1B1 transport function. The frequency of the low-function 521T>C variant varied markedly between populations. The lowest frequencies were observed in Oceania (0.0%) and sub-Saharan Africa (1.9%), and the highest frequencies in Europeans (18%) and American native populations (24%).^114,116

SLCO1B1 variants, particularly 521T>C, have been strongly associated with simvastatin-induced myopathies.^117,118 At present, there are hardly any data on inter-ethnic differences in the frequency of simvastatin-induced myopathy. However, because statin-induced myopathy is a concentration-dependent ADR and the SLCO1B1 521T>C SNP increases the risk of myopathy during treatment with simvastatin, it seems reasonable to speculate that there are likely to exist inter-ethnic differences. Furthermore, the fact that 521T>C only explained 60% of the myopathy cases in the SEARCH study¹¹⁸ suggests that other genetic and/or non-genetic factors (especially drug interactions) likely contribute to the susceptibility to statin-induced muscle toxicity.^119,120 Reduced function of OATP1B1 due to genetic variation and/or drug–drug interactions (especially with clopidogrel) appears to explain, at least in part, cerivastatin-induced rhabdomyolysis.^121,122

Regulatory aspects of global drug development, ethnicity and precision medicine

Recognizing that there may be important inter-ethnic differences in drug response, it is not surprising that regulatory authorities require (a) adequate participation of various demographic subgroups of patients by gender, ethnicity and age in clinical trials; and (b) analyses of safety and efficacy data in terms of these subgroups. Failure to provide these subgroup analyses could result in the authority refusing to file the application and ‘if evidence is available to support the safety and effectiveness of the drug only in selected subgroups of the larger population with a disease, the labeling shall describe the evidence and identify specific tests needed for selection or monitoring of patients who need the drug’. Regulatory aspects of ethnicity of study population have been reviewed previously.^2,3

With regard to ethnicity and precision medicine based on genotype-based prescribing decisions, three key regulatory guidelines are ‘Ethnic factors in the acceptability of foreign clinical data’,¹²³ ‘General principles for planning and design of multi-regional clinical trials’¹²⁴ and ‘Use of pharmacogenetic methodologies in the pharmacokinetic evaluation of medicinal products’.¹²⁵ This latter document clarifies the requirements related to the use of pharmacogenetics in the pharmacokinetic evaluation of medicinal products. It applies predominantly to small-molecule drugs as genetic effects on the pharmacokinetics of biological drugs today are much less understood. The FDA has long recognized the lack of easily accessible information about participation in drug trials and is now piloting a new transparency initiative called the ‘Drug Trials Snapshots’ since January 2015,¹²⁶ which aims to describe the demography of populations enrolled in clinical trials with each drug.

Genetic ancestry complicates reliability of self-identified ethnicity

There is no reason to believe that, all else being equal, two patients of different ethnicities sharing the same defined genotype will respond differently. However, when genotyping is not possible, a physician may have to rely on the ethnicity of the patient to guide individualized prescribing decisions. Therefore, in the context of precision medicine, when there is no access to genotyping tests, ethnicity is frequently regarded as a proxy of the patient’s probable genotype on the basis of overall population-based frequency of genetic variations in the ethnic group the patient belongs to¹²⁷; as summarized above, some of the variant alleles of DMEs are either ethnicity-specific or more frequent in one ethnic group compared to others.

The challenge for precision medicine is the reliability of self-identified ethnicity, an issue that has acquired great importance with increasing transnational and transcontinental migration of people and progressively increasing admixture of races and/or ethnicities. In the context of precision medicine, reliability of self-identified ethnicity has been well investigated in Brazil, where the population has evolved by extensive admixture of three different ancestral roots (Amerindians, Europeans and Africans) and where self-identified ethnicity and skin colour were found to be poor indicators of genetic ancestry.¹²⁸ The frequency distribution of allelic variants in many clinically relevant pharmacogenes varies considerably among Brazilians and is not captured by self-identification by race, ethnicity or skin colour.¹²⁹ Therefore, it is unwise to extrapolate data from one genetically (relatively) well-defined ethnic group such as European Caucasians to admixed populations. Durso and colleagues¹³⁰ studied the association of 15 SNPs, previously known to be linked with skin colour, in unrelated Brazilian individuals self-identified by their colour and reported that these SNPs could not predict self-assessed colour in Brazilians at an individual level.

It follows, therefore, that the other more important challenge to precision medicine arises from admixing of populations of different genetic ancestry. There are now calls for the development of race-specific or admixture-based algorithms that may facilitate improved genotype-guided warfarin dosing algorithms above and beyond that seen in individuals of European ancestry.¹³¹ For example, Suarez-Kurtz and Botton¹³² have shown that a warfarin dosing algorithm derived for an admixed cohort performed equally well in self-reported white and black patients, in marked contrast to the considerably poorer performance of other warfarin algorithms in patients of African descent compared to those of European ancestry. They attributed this discrepancy to the extensive European/African admixture among Brazilians. More recently, Duconge and colleagues¹³³ have reported an admixture-adjusted, genotype-guided warfarin dosing refinement algorithm developed in Caribbean Hispanics. This showed better predictability than a clinical algorithm that excluded genotypes and admixture, and outperformed two prior pharmacogenetic algorithms in predicting effective dose in this population.

Tanaka and colleagues¹³⁴ have reported that Japanese drug labels usually include information/data on Japanese patients as the major population. In contrast, US drug labels include data on white and non-white subpopulations, such as Asians and blacks, as these are major proportions of the US population structure. The authors also point out that ethnicity may not be the most informative classification, and that classification based on a patient’s genetic profile may be more important for predicting drug response. In the US, some drug labels have a special warning focused on ethnicity, whereas others recommend changes to the dose of the drug on the basis of genetic profiles (the basis of which is usually explained in the pharmacokinetic section of the label).^19,135 Suarez-Kurtz and Botton¹³⁶ have also suggested that accounting for ancestral genetic heterogeneity not only requires gathering information from trials at different population levels, but also calls for a critical appraisal of racial/ethnic labels that are commonly used in the clinical pharmacology literature, but do not accurately reflect genetic ancestry and population diversity. There are also calls for provision of country-/ethnicity-specific data in product labels so that they are useful to clinicians, and where data are not available the prevalence of genetic variation in the population of a country needs to be determined.¹³⁷

Impact of ethnicity-dependent non-genetic factors on precision medicine

Genotype-based precision medicine assumes that there is concordance between DME genotype and its metabolic phenotype, and that the responsiveness of the pharmacological target is uniform between individuals. Self-reported ethnicity is a complex combination of genetic and non-genetic factors that could be used by prescribing physicians as a predictor of drug response. Notwithstanding the superiority of prescribing decisions based on a patient’s genotype rather than his/her ethnicity, there are ethnicity-dependent non-genetic factors that could significantly impact the effectiveness of precision medicine. Foremost is the phenomenon of phenoconversion, which could have differential impact across different ethnicities. Phenoconversion implies discordance between genotype and predicted phenotype, whereby a genotype-predicted normal metabolizer, for example, presents with a PM phenotype. The two principal causes are comedications¹³⁸ and comorbidities,¹³⁹ the pattern of both of which differ between ethnicities or geographical regions. The presence of alternative, often compensatory and also polymorphic, metabolic pathway(s), however, further complicates prescribing decisions on the most appropriate dose based on the genotype that predicts metabolic activity for a single metabolic pathway.

With regard to comedication-induced phenoconversion, antidepressants such as paroxetine and fluoxetine are among the most potent inhibitors of CYP2D6 activity. Studies have demonstrated racial and ethnic disparities in the diagnosis and treatment of major depressive disorder. For a variety of reasons, African Americans, Hispanics and Asian Americans were shown significantly less likely to receive this diagnosis from a healthcare provider than were non-Hispanic whites. This is likely to result in greater use of these antidepressant agents in white populations. Schofield and colleagues¹⁴⁰ have reported that black and minority ethnic groups are up to four times less likely to be newly diagnosed with depression or pre-scribed antidepressants compared to white British patients. Similarly, Delaney and colleagues¹⁴¹ had earlier reported that in the US, Caucasian participants had the highest rate of antidepressant medication use (12%) compared with Asian (2%), African American (4%) and Hispanic (6%) participants. The use of complementary and alternative medicines, including certain herbal medicines that may modulate CYP450 enzyme activity and may induce phenoconversion, also varies widely between ethnic groups.¹⁴² Among the inflammatory comorbidities, infections such as HIV and leishmaniasis are known to downregulate the activities of a number of CYP450 enzymes. Since the prevalence of such comorbidities vary geographically, it seems reasonable to conclude that comorbidity-induced phenoconversion rates may also vary in a geographical manner.

Final notes and conclusions

Considering decreasing costs and increasing access to genetic testing, the need to draw any inferences on a patient’s likely genotype on the basis of his/her ethnicity is diminishing. Based on the six DME genes of Drs J. Craig Venter and James Watson, two Caucasian men whose genomes were sequenced, Ng and colleagues¹⁴³ argue that their genetic differences underscore the importance of personalized genomics over a race-based approach to medicine.

To accurately predict phenotype from genotype, a patient will need to be tested for variant alleles that are relevant for his/her ethnic background which, as already discussed, might not always be clear or accurate if self-identified, and/or the patient is living in a multi-ethnic society. However, the use of ethnicity as a proxy of potential difference between populations in drug response seems legitimate when, for a variety of reasons, including affordability and access, there is no information on a patient’s genotype, or a patient presents with a genotype that is not predictive (as indicated for warfarin, see CPIC guideline⁵³). In this regard, the key issues are the genetic ancestry of the patient and the reliability of ethnicity as self-identified by the patient.

Notwithstanding, as discussed in this review, there are inter-ethnic differences in non-genetic attributes or factors that impact adversely on the success of prescribing decisions based exclusively on genotype. Ng and colleagues¹⁴³ give the example of a patient who required at least 20 hospital visits over 5 years as his doctor tried to establish the correct dose of warfarin and who, during the fourth year of treatment, was discovered to be a PM based on the sequence of his CYP2C9 gene, and this, regardless of his skin colour, soon led to stabilization of his warfarin dose. It is worth emphasizing that non-genetic factors have a greater impact in the more prevalent intermediate or normal metabolizer cohorts and have little or no impact in the relatively far less prevalent genotypic PMs.^144–146 Ng and colleagues¹⁴³ acknowledge that there are also complex cases in which a variant does not act in the expected manner for a particular ethnic group, and cultural factors such as diet and environment can also influence drug response.

Overall, it is evident that self-identified ethnicity is often a poor marker of genetic ancestry but can serve as a marker of important non-genetic influences. Since the success of precision medicine is significantly impacted by genetic as well as non-genetic and other still unknown factors, consideration of ethnicity complements genotype-based prescribing decisions. Because of increasing admixing of populations, genotyping a patient potentially requires testing for all known clinically actionable variants that the patient may harbour, and new variants that are likely to be identified using state-of-the-art next-generation sequencing-based methods. This review reinforces the observation made a long time ago by Kalow¹⁴⁷ that ‘pharmacogenetics was broadened by the observation that multifactorial genetic influences, in conjunction with environmental factors, usually determine drug responses’, and that ‘it is wise to expect that, even after we have reached the goal to establish personalized medicine, we will not have eliminated all uncertainties’.