CYP2C9, an enzyme of the cytochrome P450 superfamily, provides the major pathway for metabolism of several commonly-prescribed therapeutic medicines, such as anticoagulants (e.g., warfarin), anticonvulsants (phenytoin), antidepressants (fluoxetine), angiotensin receptor blockers (losartan), diuretics (torsemide) and nonsteroidal anti-inflammatory drugs (NSAIDs). The clinical relevance of the CYP2C9 pathway is reflected in the CPIC (Clinical Pharmacogenetics Implementation Consortium) guidelines: three of the 26 guidelines currently available have therapeutic recommendations based on CYP2C9 metabolic phenotypes, predicted from CYP2C9 genotypes (Johnson et al., 2017; Theken et al., 2020; Karnes et al., 2021). The CYP2C9 gene is highly polymorphic, with 85 star alleles (haplotypes) currently defined in the Pharmacogene Variation Consortium (https://www.pharmvar.org/gene/CYP2C9). The frequency distribution of functionally relevant CYP2C9 alleles listed in the CPIC guidelines shows considerable variability across geographical regions and ethnicities: CYP2C9*2 is most prevalent in Europe and the Middle East, CYP2C9*3 shows the highest frequency among South Asians, while CYP2C9*5,*8 and *11 are typically observed in populations of African descent (Sistonen et al., 2009; Céspedes-Garro et al., 2015; Zhou et al., 2023; https://gnomad.broadinstitute.org). Variation in frequency of functional CYP2C9 alleles is also observed among the heterogeneous and admixed populations of Latin America, which are the target of the present study. Thus, significant trends for increasing frequency of CYP2C9*2 and CYP2C9*3 as the individual proportion of European ancestry increases, and for increasing frequency of CYP2C9*5 and CYP2C9*11 as African ancestry increases were first reported in an admixed Brazilian cohort (Suarez-Kurtz et al., 2012). Accordingly, both CYP2C9*2 and CYP2C9*3 were reported at lower frequencies in native Tepehuanos from Mexico than in admixed Mexicans (Dorado et al., 2011). Rodrigues-Soares et al. (2020) expanded the study of the distribution of CYP2C9 variants and predicted phenotypes to 33 populations from Latin America, and concluded that European admixture accounts for the presence of CYP2C9*2 in Native American populations, while CYP2C9*3 was already present in the pre-Columbian Americas.

The present study explores the pharmacogenomic implications of the distinct distribution of clinically relevant CYP2C9 variants across Latin American populations, represented by Brazilians and the four cohorts of the One Thousand Genomes Project Admixed American (1KG_AMR) superpopulation. The results show that the proportion of individuals at high risk for adverse effects of drugs predominantly metabolized by CYP2C9, the proportion of patients with CPIC therapeutic recomendation for adjustment of doses of NSAIDs and the number of patients that need to be genotyped to prevent NSAIDs-associated adverse effects vary markedly among Latin American populations.

The present study analysed the distribution of CYP2C9 functional variants and predicted CYP2C9 metabolic phenotypes among Latin American populations and explored its phamacogenomic implications. Latin America comprises 20 countries and six territories distributed over 20 million km² across South, Central and North America, and the Caribbean. The extent and dynamics of admixture among the three major parental populations, namely, autochthonous (Native) Amerindians, Europeans and sub-Saharan Africans, resulted in the kaleidoscopic diversity of the present-day Latin American population, in excess of 650 million people (∼9% of global population). Importantly, the relative proportions of these major ancestral roots vary widely across and within nations in the American continent (Pena et al., 2011; Bonifaz-Peña et al., 2014; Ruiz-Linares et al., 2014; Belbin et al., 2018; Suarez-Kurtz and Parra, 2018), and the resulting heterogeneity represents a major caveat against lumping Latin Americans into a single racial/ethnic category, such as Latino/Hispanic (Huddart et al., 2019).

A total of 1,404 samples from the South American countries Brazil, Colombia and Peru, from Puerto Rico in the Caribbean and from individuals with Mexican ancestry living in North America were analysed in this study. The results showed that the distribution of CYP2C9 alleles and diplotypes, and diplotype-predicted CYP2C9 phenotypes vary significanly across the distinct cohorts, as well as among self-identified as White, Brown or Black Brazilians. Differences in average proportions of biogeographical ancestry across the study groups (Supplementary Table S4), especially Native American and African ancestry, are the likely explanation for these results. Previous studies of Native Americans and admixed cohorts of predominant Native ancestry revealed absence or rarity of all variant CYP2C9 alleles analysed here. Examples are the absence of CYP2C9*3, *5, *8 and *11, and rarity of *2 (frequency 0.01) in Native American samples of the Human Genome Diversity Project (Cavali-Sforza, 2005); absence of CYP2C9*3, *5 and *11, and rarity of CYP2C9*2 (frequency 0.01) in Guarani from an indigenous reservation area in Brazil (Vargens et al., 2012); absence of CYP2C9*2 and *3 in nine and seven out of fifteen Ibero-American populations with estimated average Native ancestry >85% (Rodrigues-Soares et al., 2020). Accordingly, PEL, with the highest proportion of Native ancestry (0.84) among this study cohorts, displayed the lowest frequency of alleles CYP2C9*2 (0.024) and *3 (0.012), and absence of alleles *5, *8 and *11. By contrast, the latter three alleles, which associate with African descent were detected only in Brazilians and PUR, the two study cohorts with the highest average proportions of African ancestry. Among Brazilian subcohorts, the frequency of CYP2C9*8 increased from self-reported White, to Brown to Black iindividuals, in parallel with the average proportions of African ancestry (Supplementary Table S4).

A previous study of genetic divergence across Latin American populations, using the fixation index statistics (F _ST ), reported little pharmacogenetic differentiation for the CYP2C9 variant alleles examined in the present analysis (Suarez-Kurtz and de Araújo, 2022). In addition, the Cramér´s V test disclosed small magnitude of the effects of association between the study groups (cohorts and Brazilian subcohorts) and AS´s or metabolic phenotypes. Nevertheless, the distinct distribution of genotype-predicted CYP2C9 phenotypes across the study groups resulted in potentially relevant pharmacogenomic implications. First, the proportion of individuals at high risk for “an adverse response to medications that are affected by CYP2C9”, according to the PharmGKB EHR Priority Result Notation (https://www.pharmgkb.org/page/cyp2c9RefMaterials), differed significantly across study cohorts and subcohorts. Second, the proportion on individuals with CPIC therapeutic recommendations for initial dosing and/or choice of NSAIDs (Theken et al., 2020) ranged 5.4-fold across cohorts and 2.4-fold in the Brazilian subcohorts. Importantly, these CPIC recommendations are classified as Moderate or, in the case of tenoxicam, Optional. There are no Strong recommendations for dose adjustment or choice of a different drug in the NSAIDs CPIC guideline, although a Strong recommendation for dose adjustments in CYP2C9 PMs is made in the CPIC guideline for phenytoin (Karnes et al., 2021); such recommendation applies to 1% – 2% of Brazilians, CLM and PUR, but not to MXL or PEL, due to the absence of PMs in the latter two cohorts. Of notice, the CPIC recommendation for phenytoin dosing rely also on the presence/absence of the HLA-B*15:02 haplotype. Thirdly, the NNG in order to prevent one additional gastrointestinal bleeding event in patients exposed to NSAIDs ranged 4.3-fold (230–989 individuals) across cohorts and 1.6-fold (250–402) in Brazilian subcohorts. Collectively, these findings are likely to impact the perceived benefits, cost-effectiveness and clinical adoption of pharmacogenetic screening for NSAIDs and other drugs that are CYP2C9 substrates. Indeed, the probability that a patient will have a clinically relevant pharmacogenetic variant is a key parameter in cost-effectiveness analyses of routine PGx screening (Veenstra, 2016; Hughes, 2018).

Although a thorough discussion of cost-effectiveness is out of scope here, an estimate of the costs of CYP2C9 genotyping to prevent one additional NSAID-associated gastrointestinal bleeding event in the cohorts might be useful to put the NNG into context. Using the estimated NNGs and the current US$ vs. Brazilian Real exchange rate, the purchase cost of reagents and disposables for DNA extraction and genotyping the five target CYP2C9 variants, using Taqman allele discrimination assays, ranges from US$ 5,131 (PUR) to 22,031 (PEL), when 10 DNA samples are assayed concomitantly. Among Brazilians the corresponding costs range from US$ 5,571 (White) to 8,945 (Black).

Limitations of this study include: First, CYP2C9 genotyping was restricted to the most common and decreased function alleles listed in the CPIC guidelines for CYP2C9 substrates, and the reference *1 allele was assigned by default (Johnson et al., 2017; Theken et al., 2020; Karnes et al., 2021). Second, the cohorts investigated do not fully represent the diversity and heterogeneity of Latin American populations, and caution must be exerted in extrapolating the present findings across Latin America. The evidence (Lewontin, 1973) that the largest proportion of genetic variation (>80%) among human populations is found within local geographic groups is especially relevant in highly admixed populations, such as Latin Americans. Accordingly, the absence of genetic ancestry analysis adds to the limitations of this study. Finally, the NNG estimates were based on odd ratios and frequency of adverse events in trials carried out in distinct populations and clinical settings, which may not be reproduced in the cohorts examined in this study.

In conclusion, the present analyses show that the differential distribution of CYP2C9 functional alleles across Latin American populations has pharmacogenomic implications that are likely to impact the perceived benefits, cost-effectiveness and clinical adoption of pharmacogenomic screening for drugs that are predominantly metabolized by CYP2C9. The diversity and heterogeneity of Latin Americans should caution against their lumping into a single racial/ethnic category, such as Latino/Hispanic, as is often done in pharmacogenomic trials.