Introduction

Inter-individual variability in drug response that leads to either adverse drug reactions (ADRs) or low drug efficacy is commonly observed in clinical practice and poses significant burden to patient health and health care system. Insufficient drug efficacy and drug-related toxicity occur in up to 50% of the clinical treatments and ADRs accounts for 6–9% of all hospital admission, of which up to 40% are life threatening [1,2,3]. Importantly, it is estimated that 20–30% of the variability in drug response can be explained by genetic polymorphisms that are primarily localized in genes involved in drug absorption, distribution, metabolism and excretion (ADME), as well as in drug target genes and immune-related genes [4].

The human cytochrome P450 (CYP) superfamily, comprises 57 functional genes and constitutes the largest family of enzymes involved in phase I drug metabolism [5]. Among them, CYP2C9 is the most abundantly expressed CYP2C isoform in the liver and accounts for around 20% of the hepatic CYP proteins quantified by mass spectrometry [6]. It metabolizes a variety of commonly prescribed drugs, including coumarin anticoagulants, NSAIDs and sulfonylureas, as well as endogenous substrates, such as arachidonic acid [7, 8]. Genetic polymorphisms in CYP2C9 have long been recognized as a determinant of inter-individual CYP2C9 variability. The most well-studied CYP2C9 alleles are CYP2C9*2 (NC_000010.11:g.94942290C > T, p.R144C, rs1799853) and *3 (NC_000010.11:g.94981296A > C, p.I359L, rs1057910). In vitro, CYP2C9*2 reduces enzyme activity by 50–70% whereas CYP2C9*3 almost completely abrogates enzyme function (reduction of 75–99%) [9, 10]. Both alleles have also been associated with decreased metabolism of many CYP2C9 substrates in vivo, including S-warfarin and phenytoin [11]. Besides *2 and *3, multiple other variant alleles can affect CYP2C9 activity, including the decreased function alleles CYP2C9*5 (NC_000010.11:g.94981301C > G, p.D360E, rs28371686), *8 (NC_000010.11:g.94942309G > A, p.R150H, rs7900194), *11 (NC_000010.11:g.94981224C > T, p.R335W, rs28371685), *14 (NC_000010.11:g.94942234G > A, p.R125H, rs72558189) and the loss-of-function (LOF) alleles *6 (NC_000010.11:g.94949283del, p.Lys273fs, rs9332131) and *13 (NC_000010.11:g.94941958 T > C, p.L90P, rs72558187) [12]. Due to the significant impact of CYP2C9 variations, the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) include CYP2C9 genotyping in the drug labels or summary of product characteristics of 19 drugs. Specifically, testing is required for the sphingosine-1-phosphate receptor modulator siponimod in multiple sclerosis and CYP2C9 genotype is also considered as actionable information for dosage of warfarin, phenytoin and several non-steroidal anti-inflammatory drugs (NSAIDs) [13].

Considerable variation in CYP2C9 allele frequencies across different populations and ethnicities has been observed. Previous studies reported that CYP2C9*2 was most abundant in European populations (minor allele frequency, MAF between 11.1 and 14.4%), whereas the highest frequency of CYP2C9*3 was found in Asians (MAF up to 13%) [14,15,16,17]. However, frequencies were commonly extrapolated from subpopulations within the same geographic group and the generalizability of these results remains questionable [14]. In addition, evaluation of CYP2C9 allele frequencies with higher resolution is required to account for the complex patterns of ethnogeographic variability that are not reflected when only aggregated populations are considered. To better understand the global distribution of CYP2C9 variability, we here systematically analyzed CYP2C9 allele frequency data from the literature, covering in total 81,662 unrelated individuals across 70 countries and 40 ethnogeographic groups. In addition, we translate these allele frequencies into functional metabolic consequences, thus providing the first comprehensive overview of genetic and inferred functional variability at a global scale.

Results

Geographic distribution of functionally important CYP2C9 alleles

Analysis of the frequencies of functionally relevant CYP alleles across 70 countries showed that CYP2C9*2 was most abundant in Europe and across the Middle East (Fig. 1; Table 1). The prevalence was overall highest in Iran (minor allele frequency; MAF = 18.1%), followed by Croatia (MAF = 16.5%), Lebanon (MAF = 15.4%) and France (MAF = 15%). In contrast, CYP2C9*2 was absent in East Asian populations and low in South Asia with frequencies pivoting around 5%. In Africa, CYP2C9*2 was generally absent in Sub-Saharan Africa, but relatively high in North African populations (up to 12%). Notably, data about the genetic variability of CYP2C9 is still lacking for many African countries, suggesting that further exploration in these ethnogeographic groups remains to be important. In the Americas, high CYP2C9*2 frequencies were observed in Brazil (10.7%), but not in Ecuador (0.5%), Mexico (3.7%) and Peru (3.8%).

Fig. 1
figure 1

Global distribution of CYP2C9*2 and *3 alleles. Frequencies of 65 countries were color-coded with the highest frequency in red, the average frequency across all countries in yellow, and the lowest frequency in green. Countries with no frequency information available are colored white

Full size image
Table 1 National allele frequencies of CYP2C9*2 and *3
Full size table

Global frequency distributions of CYP2C9*3 align with patterns of CYP2C9*2. European and Middle Eastern populations feature high CYP2C9*3 frequencies particularly in Spain (10.1%) and Turkey (9.8%), whereas the allele was absent or rare in Sub-Saharan Africa and East Asia. A recent study furthermore revealed very high frequencies of CYP2C9*3 in the United Arab Emirates (21.3%), which is in stark contrast to other Middle Eastern populations where CYP2C9*3 frequencies pivot around 6%. In contrast to CYP2C9*2 however, CYP2C9*3 was very common in South Asia with frequencies as high as 11.9% in Pakistan and 11.6% in Bangladesh. In South America, CYP2C9*3 frequencies are relatively higher in Uruguay (7.6%), Columbia (6.8%) and Brazil (6%) but lower than 5% in all other reported countries. Notably, frequency data for Australia was mostly derived from a pan-ethnic Australian population (n = 2,509), which results in an overall close alignment with data from European populations. In contrast, the indigenous Australian Tiwi population exhibited very high frequency of CYP2C9*3 (36%) whereas CYP2C9*2 was absent.

Besides the variants defining CYP2C9*2 and *3, more than 700 additional CYP2C9 variant alleles have been described, of which approximately 40% have been estimated to impact gene function [18]. While the vast majority of these variants are very rare and their ethnogeographic distribution has not been investigated, the frequencies of six additional functionally relevant CYP2C9 variant alleles (*5, *6, *8, *11, *13 and *14) have been investigated in multiple populations (Additional file 1: Table S1). Overall, the six alleles were observed in 16 populations, of which *8 and *11 were prevalent in African and South American populations with highest frequency found in Mozambican (14.6%) and Guarani populations (4.4%), respectively. *13 was identified in East Asia and African Americans with frequencies between 0.4 and 1.5%, whereas *5 and *6 were most abundant in the United Arab Emirates (7.8%) and Sudan (2%), respectively.

Frequencies of CYP2C9*2 and CYP2C9*3 across ethnic groups

In addition to geographic patterns, we analyzed CYP2C9*2 and CYP2C9*3 frequencies across 40 ethnic groups (Table 2). CYP2C9*2 was high in Sephardi Jews (MAF = 19.4%), a Jewish diaspora population originating from the Iberian Peninsula, as well as in Ashkenazim (MAF = 13.5%) who are of Middle Eastern origin with evidence of European admixture [19]. CYP2C9*2 prevalence was very high in Kosovars (MAF = 17.5%), whereas frequencies in neighboring Serbian (MAF = 12.3%) and North Macedonian (12.4%) populations were considerably lower. While CYP2C9*2 is mostly absent or rare in South and East Asia, specific subpopulations, such as Uyghurs (MAF = 7.8%) from Northwest China and the Kannadiga ethnic group (MAF = 6%) from southwest India, feature considerably high frequencies.

Table 2 Allele frequencies of CYP2C9*2 and *3 across 35 ethnogeographic groups
Full size table

For CYP2C9*3, highest frequencies (MAF = 36.2%) were found in the Jahai people, an indigenous population living in Malaysia. This prevalence is higher than in any other population analyzed and is in stark contrast to national CYP2C9*3 frequencies in Malaysia (2.9%) and other Southeast Asian countries, such as Indonesia (2.4%), Vietnam (3%) and Thailand (5.3%), demonstrating the importance to consider ethnic backgrounds in addition to geographic factors for pharmacogenomic mapping studies.

CYP2C9 phenotype distribution across different countries and regions

Next, we extrapolated the functional consequences of the observed genetic differences by inferring CYP2C9 metabolic phenotypes based on the frequencies of functionally important CYP2C9 alleles. To this end, we considered CYP2C9*2 and *3, as well as the less common variant alleles *5, *6, *8, *11, *13 and *14 that decrease or abolish enzyme function. Globally, the prevalence of CYP2C9 poor metabolizers (PMs) is relatively low, ranging from 3–4% in Southern Europe and the Eastern Mediterranean coast to < 1% in Asian and African populations, with the exception of Emiratis (11.1%) due to high frequencies of CYP2C9*3 and *5 (Table 3; Fig. 2). Similar patterns were found for CYP2C9 intermediate metabolizers (IMs), which were overall most common in the United Arab Emirates (48.7%), Croatia (41.2%) and Iran (40.3%) where almost every second individual was estimated to exhibit reduced CYP2C9 metabolism. In Africa where the prevalence of CYP2C9 IM is generally low, we found that around 35.5% were IMs in Mozambique due to the exceptionally high frequency of the decreased function allele CYP2C9*8 in this population (MAF = 14.6%). In contrast, reduced CYP2C9 metabolism was virtually absent in Ghana and Nigeria (Table 3).

Table 3 Frequencies of inferred CYP2C9 metabolizer phenotype
Full size table
Fig. 2
figure 2

CYP2C9 metabolizer phenotype across different countries and regions. Pie charts illustrate the percentage of normal metabolizer (NM, in green), intermediate metabolizer (IM, in orange) and poor metabolizer (NM, in red) for representative countries

Full size image

Discussion

Genetic variations in CYP2C9 are major determinants of impaired CYP2C9 enzyme activity. In this study, we presented country-specific frequencies of functionally important CYP2C9 alleles at a global scale by integrating available genotype data from 108 studies comprising > 81,000 individuals in total. Our analysis showed that CYP2C9*2 is most prevalent in Middle Eastern populations (up to 18.1%), followed by South European populations (up to 16.5%), whereas CYP2C9*3 is most abundant in Emiratis (21.3%) and South Asian populations (up to 11.9%), followed by South European populations (up to 10.1%). These results are overall in accordance with previous studies that analyzed frequency data aggregated by continent or major ethnic groups [14, 15].

Our data indicate extensive variation within superpopulations, as evidenced by frequencies of CYP2C9*3 in the Jahai people that were up to tenfold higher than in geographically adjacent populations in Malaysia, Thailand and Vietnam. This population is considered to be among the earliest settlers entering the Malaysian peninsula from Africa over 50,000 years ago and endogamy results in a distinct genetic profile and high frequency founder mutations [20]. Similar results were observed for the Uygur population from Xinjiang, China, that harbors high frequencies of CYP2C9*2 (7.8%), an allele which is otherwise very rare in East Asia (< 0.1%). Previous studies showed that the Uygurs are a highly admixed population with admixture mapping suggesting a European ancestry contribution of 47% [21]. In South America, we observed large frequency differences of CYP2C9*2 (between 0.5% in Ecuador to 6% in Chile) but not *3 across different mestizo populations (Table 2). This is not surprising given the well-documented fluidity between Amerindian and European ancestral contributions [22] and similar results in different mestizo populations in Mexico [23].

Besides differences between ethnic groups, we also observed heterogeneity between allele frequencies of countries within the same macrogeographical region. For example, the frequency of CYP2C9*2 in Turkey (10.5%) were considerably lower than in its neighboring countries Bulgaria (12.5%), Greece (12.9%), Lebanon (15.4%) and Iran (18.1%). Similarly, CYP2C9*3 prevalence was graded across Europe with overall lower frequencies in North and Central Europe (4.7%-7.2%) compared to South and East Europe (7.9%-10.1%) in agreement with previous reports [24]. These results were corroborated by a large aggregate study in Scandinavia, which confirmed *3 frequencies of 6% in a large sample of 3,503 individuals from Norway, Sweden, Denmark and Finland [25]. Most pronounced differences were observed between the indigenous Tiwi population and groups of European ancestry in Australia with frequencies of CYP2C9*2 (12.8% and 0% in Europeans and Tiwi, respectively) and CYP2C9*3 (6.9% and 36%, respectively) differing by more than fivefold [26]. Genetic variability profiles of Tiwi were moreover drastically different from indigenous Polynesian and Maori populations. Substantial variability in prevalence of CYP2C9*2 (0.5–13%) and CYP2C9*3 (2.3–7.6%) was also observed across South America likely due to differing admixture between Amerindians, Europeans and Sub-Saharan Africans [27]. Therefore, genetic information of small geographically defined groups cannot provide accurate estimations of national allele frequencies particularly in countries with high population diversity [28]. Furthermore, our analyses conclude that there remains a pronounced underrepresentation of ethnogeographic groups, particularly in Central and Southeast Asia, Oceania and parts of Africa. Combined, these results demonstrate that pharmacogenomic studies require high ethnogeographic resolution to maximize its socioeconomic benefits, particularly for indigenous or founder populations with distinct genetic profiles.

Genetic profiles constitute important factors to infer metabolic phenotypes and CYP2C9 genotypes were sufficient to correctly identify the majority of individuals with impaired CYP2C9 activity [29, 30]. Estimation of metabolizer status at the population-scale, requires the aggregation of prevalence data from as many as possible functionally relevant alleles. While distribution patterns of CYP2C9*2 and *3 have been extensively profiled, considerably less information is available about frequencies of alleles considered to be population-specific, such as, *5, *6, *8, *11 for Africans and *13, *14 for Asians. Consequently, the calculated IM and PM frequencies could be underestimated for countries with missing data. Besides star alleles, also rare variants without functional information can contribute to altered CYP2C9 metabolism and have been associated with hypersensitivity to CYP2C9 substrates [31]. Overall, rare variants that have not been assigned star alleles have been estimated to account for 3.4% of the genetically encoded functional variability in CYP2C9 and could thus be an additional, but minor source of underestimation [18]. In addition, co-medication with inhibitors or inducers of CYP2C9 or impaired liver function can also affect CYP2C9 metabolic phenotypes, indicating that prescribers must integrate both genetic and non-genetic factors to guide treatment decisions.

CYP2C9 functionality is clinically relevant for a variety of commonly prescribed drugs. CYP2C9*2 and *3 have been associated with mean weekly warfarin dose [32] and carrier status of reduced function CYP2C9 alleles impacts the risk of warfarin-induced bleeding complications [33,34,35]. Consequently, guidelines recommend the use of pharmacogenetically guided dosing algorithms that consider CYP2C9*2 and *3, with adjustments for patients with African ancestry based on carrier status of CYP2C9*5, *6, *8 and *11 [36]. Reduced CYP2C9 metabolizer status is furthermore linked to phenytoin related neurotoxicity [37] and our data indicate that up to 40% of patients in Southern Europe and the Middle East might benefit from maintenance dose reductions of 25–50% [38]. Furthermore, reduced CYP2C9 activity is associated with higher drug exposure and increased risk of gastrointestinal bleeding, hypertension and myocardial infarction upon treatment with multiple NSAIDs, including oxicams, celecoxib, flurbiprofen and ibuprofen [39,40,41,42]. The current treatment recommendations are to reduce starting doses in poor CYP2C9 metabolizers by 50–75% and to titrate upwards with caution after steady-state concentrations are reached NSAIDs [43]. Based on global CYP2C9 allele distributions, these recommendations apply to more than 3% of patients in Croatia, Italy, Spain, France and Israel.

In order to implement CYP2C9 genotyping in a clinical setting, it is essential to consider the cost-effectiveness of preemptive CYP2C9 genotyping followed by treatment adjustment for decreased function allele carriers. For warfarin, previous cost-effectiveness analyses indicated that dosing guided by both CYP2C9 and VKORC1 genotypes was likely to be cost-effective compared to conventional dosing regimens [44, 45]. We are not aware of studies that evaluate the cost-effectiveness of preemptive CYP2C9 genotyping to guide therapy with phenytoin or NSAIDs. However, previous studies have shown that allele frequencies are one of the most important determinants of the cost-effectiveness of preemptive pharmacogenomic testing at the national level [46]. The data presented here might thus be useful for policy makers to evaluate whether CYP2C9 genotyping, most likely in a panel together with variants in other genes, might provide added value for national healthcare systems.

Conclusions

In summary, this study presents the worldwide distribution of CYP2C9 alleles and inferred metabolizer phenotypes with high ethnogeographic resolution. The results reveal global patterns as well as unexpected disparities of CYP2C9 genotype variability. Reduced CYP2C9 activity is most prevalent in South Europe and the Middle East, as well as in specific founder populations in Southeast Asia. The data presented here can serve as a valuable resource for population-specific CYP2C9 allele and phenotype frequencies that can provide important information for the guidance of personalized drug therapy and inform precision public healthcare at the global scale.

Methods

Data sources

We performed a systematic literature search in PubMed database covering publications before February 2023. All studies reporting frequencies of CYP2C9*2 (rs1799853) and CYP2C9*3 (rs1057910) in defined populations with cohort sizes ≥ 50 were included. We also included studies reporting frequencies of the functional alleles *5 (rs28371686), *6 (rs9332131), *8 (rs7900194), *11 (rs28371685), *13 (rs72558187) and *14 (rs72558189). In addition to published studies, we included population frequency data from the Genome Aggregation Database [47] and the 1000 Genomes Project [48]. As a result, we identified a total of 108 original articles reporting studied CYP2C9 allele frequencies from 81,662 unrelated individuals using a variety of methods for variant detection (Additional file 1: Table S1 and Additional file 2: Table S2). Frequency data for countries and ethnogeographic groups were aggregated using a weighted average approach using the cohort sizes as weighting factors.

Phenotype analyses

CYP2C9 allele function was defined based on the PharmVar consensus classifications [16]. CYP2C9 metabolizer phenotypes were defined according to the CPIC guideline [43]. Phenotype assignment of diplotypes is provided in Additional file 3: Table S3. Based on these definitions, frequencies of analyzed CYP2C9 alleles, i.e., decreased function allele *2, *5, *8, *11, *14 and LOF allele *3, *6 and *13, were used to calculate phenotype frequencies based on the Hardy–Weinberg equation. Frequencies of the CYP2C9 reference allele (*1) were calculated as f*1 = 1 − Σi fi, with fi being the frequency of each analyzed variant allele i.