CYP2C19 c.681G>A Is not in Complete Linkage Disequilibrium With c.332‐23A>G: Implications for Pharmacogenetic Testing

ABSTRACT

CYP2C19 plays an important role in the metabolism of many medications, including the antiplatelet agent clopidogrel, the antifungal agent voriconazole, selective serotonin reuptake inhibitors, select tricyclic antidepressants, and proton pump inhibitors. The Clinical Pharmacogenetics Implementation Consortium has published several guidelines emphasizing the importance of CYP2C19 genotype‐guided therapy to optimize patient outcomes. Formerly, the no function CYP2C19*2 allele was defined by three variants, c.332‐23A>G (splice defect), c.681G>A (splice defect), and c.991A>G (p.I331V). The discovery of two new haplotypes, one containing only a single variant, c.681G>A, and another with only c.681G>A and c.991A>G, challenged the assumption that c.681G>A always occurred together with c.332‐23A>G. PharmVar designated these new haplotypes as CYP2C19*2.018 and *2.019, respectively, prompting the revision of the CYP2C19*2 core allele to be defined solely by the single variant, c.681G>A. Although the variants interrogated to identify CYP2C19*2 and *35 remain the same, subjects who are heterozygous for c.332‐23A>G (present on most CYP2C19*2 and all *35 alleles) and c.681G>A (present on all CYP2C19*2 alleles) may, in rare cases, have a CYP2C19*2/*35 poor metabolizer diplotype (variants in trans) and not a CYP2C19*1/*2 intermediate metabolizer diplotype (variants in cis). CYP2C19*2 alleles with c.681G>A, but not c.332‐23A>G, were found in subjects across diverse populations and are estimated to have a frequency around 0.03%. These findings, along with the revision of the CYP2C19*2 core allele definition, have implications for variant testing, test interpretation and reporting.

Study Highlights

• What is the current knowledge on this topic?

  • ○
    The no function CYP2C19*2 and *35 alleles can be detected by genotyping two variants, c.332‐23A>G and c.681G>A, each of which causes aberrant splicing. It was assumed that CYP2C19*2 always has both splice variants, and consequently, individuals who are heterozygous for both variants are assigned a CYP2C19*1/*2 diplotype predicting intermediate metabolism. A CYP2C19*35 is called when c.332‐23A>G is present but not c.681G>A. Haplotypes with c.681G>A but not c.332‐23A>G have previously not been described.
• What question did this study address?

  • ○
    Are c.681G>A and c.332‐23A>G in complete linkage disequilibrium? If CYP2C19 haplotypes which only have c.681G>A exist as suggested by an initial case, how frequently do such haplotypes occur?
• What does this study add to our knowledge?

  • ○
    This study identified several individuals with c.681G>A but not c.332‐23A>G in the CYP2C19 gene. Long‐read sequencing data revealed two distinct novel haplotypes which were cataloged by PharmVar as CYP2C19*2 suballeles. All of Us population‐level data found these alleles across diverse ancestries at frequencies of up to 0.06%.
• How might this change clinical pharmacology or translational science?

  • ○
    Most clinical genotyping assays rely on c.681G>A to infer the presence of CYP2C19*2. However, if only c.681G>A is tested, CYP2C19*35 escapes detection. When testing both splice variants, patients who are heterozygous for both will most likely be assigned a CYP2C19*1/*2 diplotype (intermediate metabolizer, variants in cis), although a CYP2C19*2/*35 diplotype (poor metabolizer, variants in trans) may be present in rare cases. This ambiguity, which was previously not considered for diplotype calling and phenotype prediction, has implications for medication selection and dosing.

1. Introduction

Cytochrome P450 2C19 (CYP2C19) is an important drug‐metabolizing enzyme responsible for the biotransformation of numerous therapeutics, including antiplatelet agents such as clopidogrel, the antifungal agent voriconazole, selective serotonin reuptake inhibitors, select tricyclic antidepressants, and select proton pump inhibitors. The Clinical Pharmacogenetics Implementation Consortium (CPIC) and Dutch Pharmacogenetics Working Group (DPWG) have published guidelines that provide CYP2C19 genotype‐based recommendations for clinically prescribed drugs including clopidogrel [1, 2, 3, 4, 5, 6]. A guideline for clopidogrel and CYP2C19 testing has also been developed by the UK Centre of Excellence for regulatory science and innovation in pharmacogenomics [7].

CYP2C19 genetic variation [8] leads to a wide range of enzymatic activity. Metabolic capacity can affect drug response, treatment efficacy, and the risk of adverse events. CYP2C19 intermediate or poor metabolizers experience diminished activation of clopidogrel and altered pharmacokinetics of several other medications, underscoring the clinical utility of genotype‐guided therapy [2, 7].

The CYP2C19*2 allele is a commonly observed no function allele [8, 9, 10] which was defined by the Pharmacogene Variation Consortium (PharmVar) [11, 12] as having three core variants. While two of these variants, c.332‐23A>G (rs12769205) in intron 2 and c.681G>A (rs4244285) in exon 5 [13, 14, 15] cause aberrant splicing, the third variant, c.991A>G (p.I331V, rs3758580) is believed to not impact function (this variant is also present in many other CYP2C19 haplotypes). In contrast, the no function CYP2C19*35 allele is defined as having c.332‐23A>G and c.991A>G. Until now, there was no literature describing haplotypes having c.681G>A but not c.332‐23A>G; thus, it was assumed that these variants are in complete linkage disequilibrium (LD). Consequently, when c.332‐23A>G and c.681G>A were found heterozygous, both variants were assigned to the same chromosome (in cis) resulting in a CYP2C19*1/*2 genotype call predicting intermediate metabolism.

Clinical testing at St. Jude Children's Research Hospital (St. Jude) through the PG4KDS protocol [16] identified individuals harboring c.681G>A in the absence of c.332‐23A>G, indicating that these variants are not always occurring together. Additional cases were found among participants of the Genomic Answers for Kids (GA4K) initiative [17]. Individuals were also identified among All of Us Research Program (All of Us) participants [18]. This previously unrecognized haplotype diversity has important implications for pharmacogenetic testing and interpretation.

2. Methods

Clinical samples collected at St. Jude were tested by Right Patient Right Drug Diagnostics (RPRD) and genotyped using the PharmacoScan Assay Kit (Thermo Fisher Scientific, Waltham, MA). Gene translation, star allele assignment and diplotype calling were performed using the Axiom Analysis Suite v.4.0 with library files r6 or version v5.1.1.1 with library files r9 (Table S1). The initial observation of a potentially novel CYP2C19 haplotype occurred during routine clinical testing using r9 translation files, which flag variant combinations that do not match known star allele definitions as potential novel alleles. This finding prompted a retrospective review of prior CYP2C19 results generated with the earlier r6 translation files, which would have assigned these genotypes to established diplotypes and would not have identified them as potentially novel. Patients described in this study were enrolled in, and consented to, the ongoing IRB‐approved PG4KDS: Clinical pharmacogenomics implementation protocols at St. Jude [16].

PacBio HiFi long‐read whole genome sequence data (LRS) available through the IRB‐approved Children's Mercy GA4K repository program [17], were queried for individuals that are heterozygous for c.681G>A and lack c.332‐23A>G or are homozygous for c.681G>A and heterozygous for c.332‐23A>G. CYP2C19 variants were extracted from long‐read assemblies using bcftools and Variant Effect Predictor and visually inspected using the IGV Integrative Genomics Viewer [19, 20, 21]. Novel haplotypes were submitted to PharmVar for star allele designation.

Population‐level estimates for haplotypes containing c.681G>A without c.332‐23A>G were derived from the All of Us Researcher Workbench v7 short‐read sequencing (SRS) dataset (n = 245,394). Participants having at least one copy of c.681G>A but not c.332‐23A>G were identified after applying genotype quality thresholds (GQ ≥ 20). Genotypes were subsequently confirmed using All of Us genotyping array data (n = 312,945). Variant call rate, Hardy–Weinberg equilibrium, and LD metrics were computed using Hail v0.2.

3. Results

Among the 7939 St. Jude patients tested at RPRD, four individuals (0.05%) were identified having c.681G>A in the absence of c.332‐23A>G suggesting the presence of a previously unrecognized haplotype. The first two cases were identified prospectively through clinical testing using the r9 translation files (Table S1) which require the presence of both, c.681G>A and c.332‐23A>G, for calling CYP2C19*2. Because the observed variant combination did not correspond to a defined CYP2C19 star allele, these patients were flagged as likely having a novel allele and called as *1/unknown and *17/unknown, respectively. This observation prompted a retrospective review of CYP2C19 results previously analyzed with the older r6 translation files which used c.681G>A and c.276G>C, but not c.332‐23A>G, for calling CYP2C19*2 (Table S1). The review identified two additional patients with CYP2C19*1/*2A and *2A/*17 calls made in the absence of c.276G>C and c.332‐23A>G. The CYP2C19*1/*2A case was confirmed by Illumina SRS. All four patients self‐reported as non‐Hispanic White.

The patients genotyped as CYP2C19*1/*2A and *2A/*17 were predicted to be CYP2C19 intermediate metabolizers. A review of their electronic health records (EHR) revealed that neither was prescribed a CYP2C19‐related medication (citalopram, clopidogrel, dexlansoprazole, escitalopram, lansoprazole, omeprazole, pantoprazole, sertraline, or voriconazole). No phenotype was assigned to the two additional patients identified. Per EHR review, the patient genotyped as CYP2C19*1/unknown received sertraline for the management of anxiety and recurrent major depressive disorder. Sertraline was initiated at 25 mg/day and titrated up weekly to 75 mg orally daily. One year later, the sertraline dose was increased to 100 mg orally daily, then to 100 mg orally in the morning and 150 mg orally at night for management of symptoms of obsessive‐compulsive disorders. The patient tolerated sertraline therapy well without any reported adverse effects. The patient with the CYP2C19*17/unknown genotype had received voriconazole prophylaxis at a dose of 10 mg/kg orally every 12 h. The initial steady state voriconazole serum trough concentration was subtherapeutic at 0.8 mcg/mL (reference range 1–6 mcg/mL). Following an increase in the voriconazole dose to 12 mg/kg orally every 12 h, the patient achieved a trough concentration of 1.8 mcg/mL. No voriconazole‐associated adverse effects were reported or observed.

Four subjects with c.681G>A but not c.332–23A>G were identified in the GA4K database including a mother‐daughter dyad. PacBio HiFi LRS data allowed variant phase resolution over the entire gene including upstream and downstream regions. Figure 1a illustrates the diplotype found in three individuals including the mother‐daughter dyad. This novel CYP2C19*2.019 haplotype is characterized by a single coding variant, c.681G>A. The third unrelated individual has two novel haplotypes, CYP2C19*2.018, which is characterized by two core variants, c.681G>A and c.991A>G, and CYP2C19*12.002, which has an additional variant in the 3′UTR compared to the previously defined *12.001 (Figure 1b). The mother–child dyad self‐reported as Hispanic while no information regarding ethnicity was available for the other individuals.

FIGURE 1.

Graphical overview of the diplotypes containing novel CYP2C19*2 star alleles lacking c.332‐23A>G. Variants causing a splice defect or a function‐altering amino acid change are highlighted by red lines while all other variants are represented by black lines. Panel (a) depicts the diplotype in which the novel CYP2C19*2.019 suballele was found (one individual and a mother–child dyad). Panel (b) shows the novel CYP2C19*2.018 suballele; this individual also had a novel CYP2C19*12 suballele. Phase was ascertained for all subjects with long‐read next generation sequencing.

Population‐level analysis using the All of Us SRS data (n = 245,394) identified 112 participants having c.681G>A without c.332‐23A>G, corresponding to an approximate allele frequency of 0.023%. Concordance between SRS and array‐based genotyping for these variants was 100% among identified carriers. The frequency of the alleles within All of Us genetic ancestry groups was 0.05% in European (n = 57), 0.06% in Admixed American (n = 25), 0.04% in African (n = 19), 0.04% in East Asian (n = 2), and 0.05% in Other. Variant call rates for both loci were greater than 0.99 and Hardy–Weinberg equilibrium p‐values by ancestry were at least 0.04 in both sequencing and array datasets. LD between c.332‐23A>G and c.681G>A was 0.98, consistent with the rare occurrence of recombined haplotypes.

4. Discussion

Several individuals were identified to have c.681G>A but not c.332‐23A>G, presenting a challenge for CYP2C19*2 allele calling. Such cases were identified through clinical genotyping, with additional individuals identified in the GA4K repository. LRS data resolved the variant phase across the entire gene, revealing two distinct novel haplotypes, of which one contains two core variants, c.681G>A and c.991A>G, and one contains a single core variant, c.681G>A (Figure 1).

These haplotypes were designated as novel CYP2C19*2 suballeles, *2.018 and *2.019, based on c.681G>A causing aberrant splicing and preventing the formation of functional protein [13, 14]. Briefly, c.681G>A was previously described to cause a 40‐bp deletion in the transcript at the beginning of exon 5 that results in the deletion of amino acids p.215–227 and produces a premature stop codon [14, 15] rendering the resulting protein without a heme‐binding region. Human liver samples which were heterozygous for c.681G>A had about equal amounts of the canonical transcript and the variant transcript with the 40‐bp deletion while a homozygous liver sample only had transcript encoding the truncated protein. The impact of this variant on splicing was also demonstrated using an in vitro mini gene approach [13]. Based on available data, the PharmVar expert panel concluded that c.681G>A is likely obliterating activity. In vivo functional data will be exceedingly difficult to obtain as the new CYP2C19*2.018 or *2.019 alleles are rare and must be paired with a known no function allele to unequivocally assess their activity.

Hence, all haplotypes with c.681G>A are cataloged by PharmVar under the same star number, i.e., CYP2C19*2. Because both new haplotypes (*2.018 and *2.019) lack c.332‐23A>G, and *2.019 also lacks c.991A>G, all the haplotypes (i.e., suballeles) now listed under CYP2C19*2 only share c.681G>A rather than three core variants. Accordingly, PharmVar removed c.332‐23A>G and c.991A>G from the CYP2C19*2 core allele definition leaving c.681G>A as the sole defining variant. Notably, c.332‐23A>G has historically been used by some assays as a proxy for CYP2C19*2 identification due to its co‐occurrence with c.681G>A. However, our findings and the resulting updated CYP2C19*2 core allele definition underscore that c.332‐23A>G is not a reliable surrogate for detecting haplotypes with c.681G>A, i.e., CYP2C19*2 alleles.

The discovery of the novel CYP2C19*2.018 and *2.019 haplotypes and revised CYP2C19*2 core allele definition have important implications for pharmacogenetic testing, clinical reporting, and phenotype assignment. If a pharmacogenetic test interrogates c.681G>A and c.332‐23A>G, which is required to identify CYP2C19*2 and *35, subjects presenting heterozygous for both variants are typically reported as having a CYP2C19*1/*2 diplotype assuming these variants are in cis. However, our findings demonstrate that the variants can also be in trans meaning that the patient could also have a CYP2C19*2/*35 diplotype as illustrated in Figure 2. Individuals with a CYP2C19*1/*2 diplotype are predicted to be intermediate metabolizers while those with a CYP2C19*2/*35 diplotype are predicted to be poor metabolizers and consequently, misclassification of a patient as CYP2C19*1/*2 can directly affect selection or dosing of drugs metabolized by CYP2C19 including several SSRIs and antidepressants, voriconazole and proton pump inhibitors. Since there are no recommendations regarding the reporting of ambiguous diplotypes, it is common practice to report the most likely diplotype based on allele frequencies [22]. Because CYP2C19*2/*35 is expected to be rather rare based on our allele frequency estimates, many laboratories may not report CYP2C19*2/*35 as a possible alternate but ‘default’ to the more likely CYP2C19*1/*2. Moreover, since the CYP2C19*35 allele is predominantly found in individuals with African ancestry at an estimated frequency of 2%–3% (data derived from the UK Biobank and All of Us are available through ClinPGx [10]) the alternate CYP2C19*2/*35 diplotype may be predominantly found in people of this ancestral background.

FIGURE 2.

CYP2C19*1/*2 vs. *2/*35 diplotype ambiguity. Subjects who test heterozygous for c.332‐23A>G and c.681G>A without information regarding their phase can have a CYP2C19*1/*2 (c.332‐23A>G and c.681G>A are in cis) diplotype predicting intermediate metabolism or a CYP2C19*2/*35 (c.332‐23A>G and c.681G>A are in trans) diplotype predicting poor metabolism. c.991A>G is a ubiquitous variant that is typically not interrogated in panel‐based testing; this variant was removed from the CYP2C19*2 core definition along c.332‐23A>G because it is no longer found on all *2 suballeles.

Lastly, subjects testing heterozygous for c.681G>A and lacking c.332‐23A>G may have triggered an ‘unknown allele’ call in the past as the result was inconsistent with star allele definitions. As described in two of our clinical cases, a CYP2C19 phenotype could not be assigned to the patients. With the introduction of the new CYP2C19*2.018 and *2.019 suballeles and PharmVar updating their CYP2C19*2 core allele definition allows these diplotypes to be called and CYP2C19 phenotype assigned.

5. Conclusion

Two commonly observed CYP2C19 variants, c.681G>A and c.332‐23A>G are not in perfect LD as assumed in the past. Thus, testing strategies that rely on detecting only one of these splice‐defect causing variants fail to identify rare CYP2C19*2 alleles lacking c.332‐23A>G and/or identify CYP2C19*35. Furthermore, unless variant phase is known, patients who are heterozygous for c.681G>A and c.332‐23A>G most likely have a CYP2C19*1/*2 diplotype predicting intermediate metabolism but could conceivably have a CYP2C19*2/*35 diplotype predicting poor metabolism requiring different drug treatments. Clinical laboratories, implementers and prescribers should be aware of these limitations and interpretation challenges.

Author Contributions

A.G. and A.J.T. designed the research, analyzed the data, and wrote the manuscript. E.C.B., A.H., A.D.D., P.E.E., U.B. and C.E.H. performed the research, analyzed the data, and wrote the manuscript. U.B., and W.Y. performed the research and analyzed the data.

Funding

P.E.E. received grants OT2 OD026554 and OT2 OD002748 from the National Institutes of Health (NIH) Office of the Director (OD) supporting the All of Us Research Program and A.H. was supported by TL1 TR001858‐08, an NIH‐funded Clinical and Translational Science Fellowship. C.E.H. received funding from American Lebanese Syrian Associated Charities (ALSAC).

Conflicts of Interest

A.G. is the Director of the Pharmacogene Variation (PharmVar) Consortium. A.J.T., A.D.D., and U.B. are supported in part and/or hold equity in RPRD Diagnostics LLC. E.C.B., C.E.H., A.H., P.E.E. and W.Y. declared no competing interests for this work.

Supporting information

Table S1: CYP2C19 variants tested by the PharmacoScan platform.