Carbamazepine, an effective anticonvulsant drug, also used to treat bipolar disorder and to relieve pain in trigeminal neuralgia, is known to trigger hypersensitivity reactions that typically affect the skin. While some of these reactions are mild, other conditions such as Stevens-Johnson syndrome (SJS), toxic epidermal necrolysis (TEN), and drug reactions with eosinophilia and systemic symptoms (DRESS) are potentially life-threatening. There is strong evidence of genetic predisposition for these idiosyncratic reactions, linked to specific human leukocyte antigen (HLA) types, especially HLA-B*15:02 in Southeast Asian populations and HLA-A*31:01 in more widespread regions (Dean et al., 2015; Yip and Pirmohamed, 2017). Accordingly, the Clinical Pharmacogenetics Implementation Consortium (CPIC) and the Dutch Pharmacogenetics Working Group (DWPG) dosing guidelines for carbamazepine recommend the choice of an alternative drug for carbamazepine-naive patients carrying at least one copy of either HLA-B*15:02 or HLA-A*31:01 (Phillips et al., 2018; Dutch Pharmacogenetic Working Group, 2021). Recommendations for genetic testing for HLA-B*15:02 and/or HLA-A*31:01 for carbamazepine-naive patients before they start treatment have also been issued by the Canadian Pharmacogenomics Network for Drug Safety (Amstutz et al., 2014) and the United States Food and Drug Administration (Food and Drug Administration, 2021).

The fact that “gold-standard” methods for HLA haplotyping are time-consuming and relatively expensive, prompted efforts to validate single nucleotide variants (SNVs), as reliable tags for HLAs linked to idiosyncratic carbamazepine reactions. Two SNVs, namely rs1061235 (GRCh38.p13 chr 6, NC_000006.12:g.29945521A>T) and rs17179220 (GRCh38.p13 chr 6, NC_000006.12:g.29853458G>A) have been proposed for tagging the HLA-A*31:01 haplotype: de Bakker et al. (2006) first reported complete linkage disequilibrium (LD) between rs1061235 and HLA-A*31:01 in the HapMap European (CEU) sample. Subsequent work revealed that rs1061235 is also linked to HLA-A*33 haplotypes, which impacts the predictive performance of rs1061235 to identify carriers of HLA-A*31:01 (Amstutz et al., 2013; Thorstensen et al., 2014; Buchner et al., 2021). High LD of rs17179220 with HLA-A*31:01 was initially detected in Han Chinese (Zhou et al., 2016), and subsequently verified in a large set of ancestrally diverse populations (Erlichster et al., 2019).

The present study explores the reliability of rs1061235 and rs17179220 for tagging HLA-A*31:01 in populations with Native American ancestry. Two previous studies included individuals with Native American ancestry: Amstutz et al. (2013) enrolled five Canadian Aboriginal children whereas Erlichster et al. (2019) examined a subset of samples from the Admixed American superpopulation of the 1000 Genomes Project [1 KG_AMR; (1000 Genomes Project Consortium et al., 2015)]. Our study covers the overall 1 KG_AMR cohort, as well as Native American individuals from the Human Genome Diversity Project (HGDP) (Cavalli-Sforza, 2005) and from three groups living in indigenous reservation areas of Brazil, that differ largely in frequency of HLA haplotypes (Belich et al., 1992; Petzl-Erler et al., 1993; Parham et al., 1997). We did not extend our study to tags for HLA-B*15:02, the other major HLA haplotype linked to carbamazepine-induced idiosyncratic reactions, because this haplotype is rare or absent in our study cohorts.

This is the first study of the predictive performance of previously identified SNP tags for HLA-A*31:01 in indigenous populations of the Americas, represented in the HGDP project and recruited from three reservations areas in Brazil. We extended our analysis to the 1 KG_AMR superpopulation—comprised of individuals with Native American, European and African admixed ancestry—and to a subcohort of the 1 KG_AMR, denoted 1 KG_NAT, of individuals with predominant (> 70%) Native ancestry.

The diversity of our study cohorts is reflected in the frequency of the HLA-A*31:01 haplotype, which ranged from 0.02 in 1 KG_NAT to 0.65 in the KRC (Tables 1, 2). Particularly impressive is the extent to which the indigenous groups from Brazil differ in HLA-A*31:01 frequency, first reported by Petzl-Erler and colleagues (Petzl-Erler et al., 1993; Parham et al., 1997). Of notice, the GRC (Guarani) and KRC (Kaingang), in the extreme of the HLA-A*31:01 frequency range (0.04 and 0.65, respectively) live in the same indigenous reservation area in southern Brazil (Rio das Cobras) and despite their proximity, remain linguistically and culturally distinct and are predominantly endogamous (Petzl-Erler et al., 1993; Parham et al., 1997). Large differences in HLA-A*31:01 frequency were also observed between Karitiana (0.54) and Suruí (HLA-A*31:01 absent), two tribes from the Brazilian Amazon represented in the HGDP cohort. However, the small number of individuals of the distinct HGDP American groups, especially Suruí (n = 8), are an obvious caveat to intergroup comparisons.

The diversity of the study cohorts is also evident in the MAF of rs1061235 and rs17179220, which varied 3-5 fold across cohorts, and in the predictive performance of these SNVs as tags for HLA-A*31:01. We highlight especially NPV, the predictive metric of primary interest for pharmacogenetic-informed carbamazepine prescription. Both SNVs showed perfect NPV (NPV = 1.0) in 1 KG_AMR and 1 KG_NAT, while in HGDP NPV was perfect for rs17179220, but not for rs1061235 (NPV = 0.91). In the Native populations from Brazil, genotyped only for rs17179220, perfect NPV was observed in GWK, but not GRC (NPV = 0.88) or KRC (NPV = 0.80). Of notice, rs17179220 was not present in the HLA-A*31:01 haplotype in GRC, such that the PPV and specificity were null. Data from previous genotyping of the Kaingang and Guarani samples for HLA-A, B and C haplotypes (Petzl-Erler et al., 1993; Parham et al., 1997) reveal that in KRC, all the eight different HLA-A*31-Bx-Cx haplotypes included the rs17179220 A allele, while the A*24 and the A*02-B*35 and A*02-B*39 haplotypes as well as a few non-HLA-A*31 haplotypes introduced by gene flow from European or African were rs17179220*G. However, of 10 A*02:12-B*51:01:01-C*01:02 KRC haplotypes, eight included the G and two the A allele. In GKW, all the HLA-A*31 haplotypes but also all the eight HLA A*02:01-B*40:04-C*03:04 haplotypes carried the rs17179220 A allele.

Prompted by a Reviewer’s suggestion, we extended our analysis of the predictive performance of rs1061235 and rs17179220 as HLA-A*31:01 tags to other admixed populations of the Americas, with predominant European or SubSaharan African, rather than Native ancestry (Table 3). This exercise was applied to a previously defined subcohort [denoted 1 KG_AMR (European)] comprising the 80 1 KG_AMR samples with the largest proportion of European ancestry (Suarez-Kurtz et al., 2020) and to the 80 1 KG_ACB (African Caribbean) samples with the largest proportion of African ancestry (Debortoli et al., 2021). HLA-A*31:01 frequency ranged from 0.01 in the 1 KG_ACB subcohort to 0.06 in 1 KG_AMR (European). Sensitivity and NPV were perfect for both rs1061235 and rs17179220 in 1 KG_AMR (European) and for rs1061235, but not rs17179220, in the 1 KG_ACB subcohort. We emphasize that the small number of individuals carrying the HLA-A*31:01 haplotype in both these subcohorts, especially the 1 KG_ACB (n = 2) is a major caveat for quantification of the predictive metrics. Of notice, the rs1061235 SNV was in complete linkage disequilibrium with the HLA-A*33 in the 1 KG_ACB subcohort, and in strong disequilibrium in 1 KG_AMR (European), where four of the five HLA-A*33 carriers had also the rs1061235 variant allele.

We acknowledge that the low frequency of HLA-A*31:01 in the GRC sample and the low number of individuals of the distinct groups in the HGDP American cohort pose limitations to our study. Practical and ethical difficulties are commonly encountered in recruiting participants from Native American populations, and the pivotal study linking HLA-A*31:01 to carbamazepine-induced cutaneous reaction in Native Americans included only 5 Aboriginal Canadian children (Amstutz et al., 2013). Another potential limitation relates to the use of different methodologies for characterization of HLA haplotypes in the indigenous populations from Brazil compared to the 1 KG and HGDP samples, which may have affected the results.

In conclusion, we present the first assessment of the predictive performance of rs1061235 and rs17179220 for tagging HLA-A*31:01 in cohorts of Native American ancestry. NPV, the metric of primary interest for pharmacogenetic-informed carbamazepine prescription, ranged from 0.80 to 1.0, which supports the notion that these SNPs are not optimal proxies for HLA-A*31:01 across populations with Native American ancestry.