Next-Generation Sequencing of CYP2C19 in Stent Thrombosis: Implications for Clopidogrel Pharmacogenomics

Abstract

Purpose

Describe CYP2C19 sequencing results in the largest series of clopidogrel-treated cases with stent thrombosis (ST), the closest clinical phenotype to clopidogrel resistance. Evaluate the impact of CYP2C19 genetic variation detected by next-generation sequencing (NGS) with comprehensive annotation and functional studies.

Methods

Seventy ST cases on clopidogrel identified from the PLATO trial (n = 58) and Mayo Clinic biorepository (n = 12) were matched 1:1 with controls for age, race, sex, diabetes mellitus, presentation, and stent type. NGS was performed to cover the entire CYP2C19 gene. Assessment of exonic variants involved measuring in vitro protein expression levels. Intronic variants were evaluated for potential splicing motif variations.

Results

Poor metabolizers (n = 4) and rare CYP2C19*8, CYP2C19*15, and CYP2C19*11 alleles were identified only in ST cases. CYP2C19*17 heterozygote carriers were observed more frequently in cases (n = 29) than controls (n = 18). Functional studies of CYP2C19 exonic variants (n = 11) revealed 3 cases and only 1 control carrying a deleterious variant as determined by in vitro protein expression studies. Greater intronic variation unique to ST cases (n = 169) compared with controls (n = 84) was observed with predictions revealing 13 allele candidates that may lead to a potential disruption of splicing and a loss-of-function effect of CYP2C19 in ST cases.

Conclusion

NGS detected CYP2C19 poor metabolizers and paradoxically greater number of so-called rapid metabolizers in ST cases. Rare deleterious exonic variation occurs in 4%, and potentially disruptive intronic alleles occur in 16% of ST cases. Additional studies are required to evaluate the role of these variants in platelet aggregation and clopidogrel metabolism.

Introduction

The occurrence of stent thrombosis (ST) after percutaneous coronary intervention (PCI) is associated with a very high mortality rate [1]. Certain patient characteristics such as diabetes mellitus, as well as certain procedural issues such as stent under-expansion or fracture, have been implicated in ST [2]. Dual antiplatelet therapy (DAPT) is the cornerstone of medical treatment after PCI to prevent recurrent ischemic events and clopidogrel with aspirin is the most widely used antiplatelet therapy [3]. Clopidogrel is a pro-drug that is converted into an active metabolite primarily by the cytochrome P450 enzyme CYP2C19. In the last decade, a variable response to clopidogrel after PCI has been attributed to genetic variants primarily in the pharmacokinetic pathway of clopidogrel metabolism. The most significant of these are CYP2C19 loss-of-function (LoF) alleles, primarily CYP2C19*2 and *3, which are associated with decreased clopidogrel active metabolite levels and high on-treatment platelet reactivity [4–6]. Other LoF alleles (CYP2C19*4, *5, *6, *7, *8) have been identified in CYP2C19 and have been shown to be associated with ST in patients taking clopidogrel after PCI [2, 7].

Up to 70% of variation in clopidogrel response as measured by residual adenosine diphosphate (ADP)–mediated platelet aggregation is hypothesized to be heritable [8]. However, in the Platelet Inhibition and Patient Outcomes (PLATO) trial, one of the largest DAPT randomized trials, 62.5% of clopidogrel-treated patients with definite ST did not have any known CYP2C19 LoF alleles (*2 or *3) [9]. The role of genetic variation other than known LoF alleles in CYP2C19 using next-generation sequencing (NGS) techniques in a ST cohort has not been comprehensively studied.

Our study objective was to use complete CYP2C19 gene sequencing and extensive annotation strategies with functional validation to describe variants that may be associated with genetically mediated clopidogrel resistance in ST that may not have been identified by genotyping in an attempt to further define the pharmacogenetics of clopidogrel-CYP2C19 inter-actions in the pathophysiology of ST.

Methods

Cases and controls were identified from Mayo Clinic (MC) Biobank [10] and PLATO [11] trial database (Fig. 1). Subjects gave informed consent for both studies. We identified patients diagnosed with ST at the time of cardiac catheterization from January 1, 2001 to March 31, 2014 (n = 382) at Mayo Clinic, assessing electronic medical records and the Rochester Epidemiology Project (REP) [12, 13]. Inclusion criteria were as follows: (1) Patients age > 18 years undergoing PCI with stent implantation, (2) taking clopidogrel with aspirin after PCI with evidence of definite ST during 1 year follow-up for drug-eluting stents and 1 month follow-up for bare metal stents, (3) compliance and continuation of DAPT after index PCI documented in the medical record, and (4) adequate DNA available for analysis. From MC Biobank, we identified 68 patients with ST and stored Biobank DNA sample. Fifty-six patients were excluded because diagnosis was not confirmatory on evaluation of coronary angiogram, patients had late ST (> 1 year after stent placement), or patients were not taking clopidogrel at the time of ST. Only 12 MC Biobank patients were included. The PLATO trial (www.ClinicalTrials.gov NCT00391872) randomized patients with acute coronary syndrome to ticagrelor versus clopidogrel. All patients with adjudicated definite ST event on clopidogrel after PCI, with DNA samples and fulfilled inclusion criteria were included in the present study (n = 58). These two sources combined results in the 70 patients with definite ST (Fig. 1). All cases were matched 1:1 (12 matching controls from MC Biobank and 58 from PLATO) with controls for age, race, sex, diabetes, chronic renal failure, stent type, and acute coronary syndrome presentation. Controls were on clopidogrel and did not have ST or MACE (major adverse cardiovascular events) during the guideline-directed timeline after PCI. In PLATO, given the ACS (acute coronary syndrome) indication, all patients were on DAPT for 12 months irrespective of type of stent. The protocol was reviewed by the IRB and approval was obtained for this study.

Fig. 1.

Selection criteria of the study population

Exome Sequencing

Whole exome sequencing (WES) was performed using Agilent’s SureSelect Human All Exon v5 + UTRs 75 MB kit protocol augmented with custom primers to cover the entire CYP2C19 gene. Paired-end libraries were prepared following manufacturer’s protocol (Illumina and Agilent) using the Bravo liquid handler from Agilent. One microgram of genomic DNA was fragmented to 150–200 bp using Covaris E210 sonicator. The ends were repaired, and an “A” base was added to 3′ ends. Paired-end Index DNA adaptors (Agilent) with a single “T” base overhang at the 3′ end were ligated. We purified resulting constructs using AMPure SPRI beads (Agencourt). Adapter-modified DNA fragments were enriched by 4 cycles of PCR using SureSelect forward and SureSelect ILM Pre-Capture Indexing reverse (Agilent) primers. Concentration and size distribution of libraries was determined with Agilent Bioanalyzer DNA 1000 chip. Seven hundred and fifty nanograms of prepped library was incubated with whole exon biotinylated RNA capture baits supplied for 24 h at 65 °C. Captured DNA:RNA hybrids were recovered using Dynabeads MyOne Streptavidin T1 (Dynal). Eluted DNA from beads was purified using Ampure XP beads (Agencourt). Purified capture products were amplified using SureSelect Post-Capture Indexing forward and Index PCR reverse primers (Agilent) for 12 cycles. Exome libraries were loaded one sample per lane onto Illumina TruSeq v3 paired end flow cells with 9pM concentrations to generate cluster densities of 600,000–800,000/mm² following standard protocol using the Illumina cBot and TruSeq Rapid Paired end cluster kit version 3. Flow cells were sequenced as 101-base paired end reads on an Illumina HiSeq 2000/2500 using TruSeq SBS sequencing kit version 3 and HiSeq data collection software v2.0.12.0. Base calling was performed using Illumina’s RTA version 1.17.21.3. For SS v5 + UTRs 75 MB need 6 Gb of sequencing so 3–4 samples per lane should give 60–75 X coverage. Sequencing was performed at Medical Genome facility Genome Analysis Core, Mayo Clinic, Rochester, MN.

Bioinformatics Analysis

Reads were mapped to GrCh37/hg19 using GenomeGPS (in-house analysis pipeline) which performs alignment using Novoalign (V2.07.13); realignment and recalibration using Genome Analysis Tool Kit (GATK, v3.3); germline variant calling (SNV and indels) using GATK HaplotypeCaller; and Variant Quality Score Recalibration (VQSR), following GATK best practices, version 3 [14–16]. Variants with pass filter and first tranche were selected for further analysis. Transcript annotations are based on RefSeq transcript NM_000769.1 and Pharmacogene Variation (PharmVar) guidelines to determine list of known PGx variants associated with clopidogrel metabolism.

Phasing/Imputation Approach

Haplotypes downloaded from PharmVar 3.3 for CYP2C19 alleles were grouped into families by the primary numeral in the name (e.g., *1A and *1B are in the same *1 family). We defined tag SNPs/indels and obligates SNPs/indels as follows: Tag SNPs/indels were SNPs/indels present in at most 2 families, and Obligate SNPs/indels were tag SNPs/indels that were also functional variants (non-synonymous, frameshift, indels in splice regions, stop gain, stop loss) present in at most 2 families or unique to a specific family (e.g., *17 promoter SNPs). Per sample, we identified which tag SNPS/indels or obligates SNPs/indels were variable in the sample and combined all possible pairs of the haplotypes containing those tag SNPs/indels and obligates SNPs/indels to generate predicted genotypes for each pair of *alleles. We sorted solutions by number of mismatching alleles in the genotypes of obligate SNPs/indels, tag SNPs/indels, and other SNPs/indels (excluding those that were non-functional ubiquitous polymorphisms). All non-perfectly matching solutions were manually reviewed for the mismatch scores or samples with multiple identical scoring solutions in different families. We identified samples with novel functional SNPs/indels if not defined in Pharmvar 3.3 haplotypes or known star alleles.

LoF alleles were defined in this study as CYP2C19*2 (splicing defect c.681G>A), CYP2C19*3 (c.636G>A), and CYP2C19*8 (c.358T>C). Rapid metabolizers (RM) were defined as containing one gain of function (GoF) CYP2C19*17 allele (c.−3402C>T and c.−806C>T); ultra-rapid metabolizers (UM) are samples with two *17 GoF alleles. Intermediate metabolizers (IM) meant the sample contained a single LoF allele, and poor metabolizers (PM) contained two LoF alleles.

DMS Method

Engineered “landing pad” HEK 293T (human embryonic kidney) cells were used as a platform to integrate pooled variant libraries, resulting in one variant per cell [17, 18]. CYP2C19 ORF was fused to GFP and mCherry as transfection control. Individual variants with different GFP/mCherry were used to indicate different levels of protein expression. Pooled variant libraries were generated by nicking mutagenesis; recombinase integrated the libraries into landing pad cells, one per cell [17, 19].

Multiplexed functional selection performed with flow cytometry (FACS) sorted cells into different bins. Amplicon sequencing of DNA collected in each bin used the Illumina HiSeq4000 Sequencing System, followed by variant calling. The fastq files were aligned with CYP2C19 reference using BWA mem aligner version 0.7.15. Samtools mpileup version 1.5 was used with a custom python script for SNV calling. A base quality score cutoff of 20 and a mapping quality score cutoff of 20 were applied for SNV calling. The frequency of variants appearing in each bin was used to determine protein expression levels (Supplemental Fig. 1).

Western blot analyses were performed using BFP⁻/mCherry⁺ cells integrating individual variant of CYP2C19. Proteins were lysed and separated by SDS-PAGE prior to transfer to PVDF membranes. Membranes were incubated with rabbit polyclonal CYP2C19 antibody (Abcam, Cat. No. 137015) at 1:2000 dilution. β-Actin protein was measured using rabbit β-Actin (D6A8) monoclonal antibody (Cell Signaling, Cat. No. 8457), mCherry protein was measured using mouse monoclonal mCherry antibody (Sigma, Cat. No. SAB2702291), and their expression was used as loading controls. Proteins were detected with SuperSignal™ West Dura Extended Duration Substrate (Thermo Scientific, Rockford, IL). Radiographic images were captured on X-ray films or with ChemiDoc™ Touch Image System (Bio-Rad, Hercules, CA). Protein density quantification was performed with NIH ImageJ software (https://imagej.nih.gov/ij/download.html).

Annotation of Intronic Variants

Variants identified only in cases, not present in controls and vice versa, were analyzed with three tools to evaluate potential splicing motif variations that may lead to a LoF effect. Each variant identified only in cases or controls was manually reviewed using Alamut Software ® v.2.11 –Qt v.5.5.1 (https://www.interactive-biosoftware.com/). Alamut provided visualization of splicing predictions from SpliceSite Finder-like [20, 21], MaxEntScan [22], GeneSplicer [23], and NNSPLICE [24] algorithms and high confidence branchpoints previously described [25]. We also used Human Splicing Finder (HSF) v.3.1 (accessible on: http://www.umd.be/HSF3/index.html) combines 12 algorithms [26]. Finally, regSNP-intron [27], a random forest classifier that integrates both splicing signals and conservation metrics, outputs a splicing disruption probability score (accessible on: https://regsnps-intron.ccbb.iupui.edu/). We combined predictions into 5 categories. “Very strong prediction” (VSP) was assigned when all three tools suggested a deleterious effect on splicing. “Strong prediction” (SP) was assigned when 2 tools predicted the creation or alteration of a splicing motif. “Moderate prediction” (MP) suggests one tool predicted a splicing alteration. “Weak or no effect prediction” (WNP) refers to instances where only splicing enhancers or silencer motifs alterations were observed or no alterations were predicted. Estimates of both donor and acceptor site changes at a single location, were deemed as “conflicting prediction” (CP). Minor allele frequencies (MAF) were obtained from 1000 Genomes and gnomAD database. RegulomeDB [28] was used to capture potential sites which may cause regulatory changes, via disruption of binding sites for transcription factors. Deep intronic variants in this study are defined as beyond 100 base pairs away from exon-intron junctions. “Novel” variant in this study means the genomic alteration has not been reported in association with pharmacogenomics studies or any publication to the best of our knowledge.

Statistical Analysis

The numerical data are presented as mean ± standard deviation. Comparisons between groups were executed with the two-tailed t test. Categorical variable concerning distribution of genotypes or CYP2C19 phenotype summary was expressed with frequencies and/or percentages and evaluated with conditional logistic regression (coxph in R) accounting for matched variables. Protein expression data were analyzed using GraphPad Prism7 software (GraphPad Software, La Jolla, CA). Protein expression was analyzed using ANOVA followed by appropriate post hoc tests for multiple comparisons. p < 0.05 was considered statistically significant.

Results

Study Population

Table 1 shows baseline characteristics of the cohort. Mean age of population was 62.7 ± 11.9 years. There were no significant differences in age and sex distribution, prevalence of diabetes, and chronic renal failure between the groups. The median time to ST (from index PCI) was 8.5 days (IQR 4–49 days).

Table 1.

Baseline characteristics of the study population

Variables	Cases	Controls	p value
Age	62.5 ± 12.0	62.9 ± 12.0	0.86
Females	28.5%	28.5%	1.0
Diabetes	30%	30%	1.0
Chronic renal failure	8.5%	2.8%	0.14
Drug-eluting stent	25.7%	25.7%	1.0
Acute coronary syndrome	100%	100%	1.0

DNA Sequencing

Overall, an average 71.2% mapped reads aligned to captured region, and 98.1% of the complete CYP2C19 gene region was covered at minimum 10× read depth (Supplemental Tables 1 and 2). Overview of genetic variation seen in cases and controls is shown in Supplemental Table 3. A total of 501 variants (SNV and indels) were identified in the 140 samples, of which 413 were present in cases and 327 were present in controls (Supplemental Table 3). The majority of variation was observed in intronic regions (n = 489). There were 169 variants only observed among cases and 84 only present in controls, reflecting a greater allelic heterogeneity in ST cases.

CYP2C19 Haplotypes, Metabolizer Status, and Stent Thrombosis

A schematic representation of allelic frequencies of CYP2C19 star alleles identified is presented in Supplemental Fig. 2. CYP2C19*2 alleles were more commonly observed in cases (19.2%) as compared with controls (15.0%). Paradoxically, the CYP2C19*17 allele was more often present in cases (27%) than controls (17.5%). A rare CYP2C19*8 allele (c.358T>C, rs41291556), known to disrupt enzymatic function [29], was identified in a single case. We also identified two rare coding alleles present in cases only, CYP2C19*15 (c.55A>T, rs17882687) and CYP2C19*11 (c.449G>A, rs58973490), previously described to not cause significant change in enzymatic function [30]. We compared the estimated to observed frequency of CYP2C19*17 and CYP2C19*2 haplotypes in our study population and found it to be in Hardy-Weinberg equilibrium demonstrating no bias in the selection of controls.

To better understand diploid CYP2C19 status, Table 2 shows the frequencies in cases and controls after our imputation method to estimate phase and individual metabolizer phenotype. We observed an overall association of metabolizer status and ST in our cohort (p = 0.02). PM (*2/*2 and *2/*8 alleles) were only observed in cases with ST. IM were present in 28.6% of cases and 31.4% of controls, with no significant association to ST (p = 0.301). RM carrying one CYP2C19*17 allele were more frequent in cases (41.4%), compared with controls (25.7%), and thus associated with ST (p = 0.014, OR = 2.94 95% C.I. 1.24–6.98). However, there were equal numbers of UM (2.9%) in both cases and controls.

Table 2.

Observed frequencies of genotype combinations and CYP2C19 metabolizer phenotype based on Pharmvar 3.3 definitions. PM - Poor metabolizers, IM - Intermediate metabolizer, RM - Rapid metabolizer, UM - Ultra-rapid metabolizers and EM - Extensive “normal” metabolizers

Phenotype summary	Diplotypes	Cases (%) N = 70	Controls (%) N = 70	Univariate OR (95% C.I.)	p value
PM	*2/*2	3 (4.3%)	0	n.a.	0.998
	*2/*8	1 (1.4%)	0
IM	*2/*17	2 (2.9%)	3 (4.3%)	1.63 (0.64—4.1–5)	0.301
	*1/*2	17 (24.3%)	19 (27.1%)
	*2/*11	1 (1.4%)	0
RM	*1/*17	28 (40%)	18 (25.7%)	2.94 (1.24–6.9–7)	0.014*
	*15/*17	1 (1.4%)	0
UM	*17/*17	2 (2.9%)	2 (2.9%)	1.67 (0.20–13.−43)	0.628
EM	*1/*1	14 (20.0%)	28 (40%)	n.a.	n.a.
	*1/*15	1 (1.4%)	0

n.a. not applicable

^*Statistically significant

Exonic Genetic Variation in CYP2C19

Secondly, we evaluated exonic variation for LoF effect by in silico analysis and by using an in vitro cell system. A total of eleven exonic variants were identified and outlined in Table 3. No indels were identified overlapping coding regions. There were 8 out of 11 exonic variants that were known star allele haplotypes. Our study identified one novel variant in a case on exon 7. This is a very rare synonymous variant, c.1137C>T, with a high CADD score (35.0), but predicted neutral by PredictSNP2 and not reported in population databases. The 2 rare exonic variants (rs370051475 and rs149072229) found only in control samples were predicted not to affect protein function by in silico analyses.

Table 3.

Exonic variants called by next-generation sequencing of CYP2C19 from ST cases and controls. NM_000769.4 transcript was used for cDNA nomenclature. MAF: minor allele frequency. For SIFT, and Polyphen-2 and PredictSNP2 predictions: T—tolerated, B—benign prediction, PD—probably damaging, D—deleterious or damaging and N - neutral. CADD Phred-like score was interpreted as deleterious if > 20. Splicing alterations: WNP - Weak or no effect prediction and VSP - Very strong predictions. FACS: Flow cytometry multiplexed assay for protein expression estimates. n.a. - not applicable, n.c. - not computed based on variation type and n.d. - not detected.

dbSNP ID (Haplotype)	cDNA & Protein change	Cases (%) N = 70	Control (%) N = 70	MAF	In-silico Tools: SIFT | Polyphen-2 | CADD | PredictSNP2 | Splicing-related	Functional Data
						FACS Estimate	Western Blot
rs771120274 (novel)	c.1137C>T p.(Tyr379=)	1 (1.4%)	0 (0%)	n.a.	n/c | n/c | 35.0 | N | WNP	~25–50%	Decreased
rs17882687 (CYP2C19*15)	c.55A>C p.(Ile19Leu)	1 (1.4%)	0 (0%)	0.21%	T | B | 0.02 | N | WNP	~25–50%	Decreased
rs41291556 (CYP2C19*8)	c.358T>C p.(Trp120Arg)	1 (1.4%)	0 (0%)	0.15%	D | PD | 24.3 | D | WNP	<25%	n.d.
rs58973490 (CYP2C19*11)	c.449G>A p.(Arg150His)	2 (2.9%)	0 (0%)	0.26%	T | B | 3.5 | N | WNP	~ wt	Increased
rs370051475 (novel)	c.1077C>T p.(Ile359=)	0 (0%)	1 (1.4%)	0.015%	n/c | n.c. | 2.9 | N | WNP	~25–50%	Decreased
rs149072229 (novel)	c.241G>A p.(Glu81Lys)	0 (0%)	1 (1.4%)	0.037%	T | B | 15.4 | N | WNP	~ wt	~wt
rs4244285 (CYP2C19*2)	c.681G>A p.(Pro227=)	24 (34.3%)	22 (31.4%)	17.59%	n.c. | n.c. | 0.07 | N | VSP - New Acceptor Site	~wt	~wt
rs3758580 (CYP2C19*2)	c.990C>T p.(Val330=)	21 (30.0%)	23 (32.8%)	17.71%	n.c. | n.c. | 0.05 | N | WNP	25–50%	Decreased
rs17878459 (CYP2C19*2)	c.276G>C p.(Glu92Asp)	3 (4.2%)	3 (4.2%)	2.26%	T | B | 6.29 | N | WNP	~wt	~wt
rs17885098 (multiple)	c.99C>T p.(Pro33=)	15 (21.4%)	13 (18.5%)	7.60%	n.c. | n.c. | 5.37 | N | WNP	~wt	~wt
rs3758581 (multiple)	c.991A>G p.(Ile331Val)	11 (15.7%)	13 (18.5%)	6.19%	T | B | 0.01 | N | WNP	~wt	~wt

In vitro Functional Studies of Exonic Variants and Comparison with In silico Analyses

We evaluated the exonic variants effect on protein expression via HEK293 cells transfection with constructs containing the variants (n = 11). In Fig. 2a, we illustrate the fluorescent ratio in quartiles demonstrating an estimate of protein expression levels. To confirm these findings, CYP2C19 protein concentration of cell lysates was quantitatively assessed with Western blots (Fig. 2b) and image quantitation (Fig. 2c) in triplicates. There was concordance between in silico predictions and in vitro functional validation for 54% (6/11) of the exonic variants as shown in Table 3. The known LoF missense p.(Trp120Arg) or CYP2C19*8 showed less than 25% protein expression via FACS, and no protein was detected in Western blots (p < 0.001). Common missense variants c.991G>A (rs3758581) and c.276G>C (rs17878459) and synonymous variants c.99T>C (rs17885098), c.681G>A (rs4244285), and the rare missense c.241G>A (rs149072229) resulted in no difference in protein expression levels as compared with WT, concordant with non-deleterious in silico predictions.

Fig. 2.

Deep mutational scanning results for exonic variants identified in cohort. a Bar plots depicting the percentage of variant appearing in each bin from variant calling, which indicates protein expression. P1: < 25%, P2: 25–50%, P3: 50–75%, P4: > 75% of protein expression. b Western blots of lysate from BFP⁻/mCherry⁺ cells. c Quantitative western blot analysis. Bars depict standard deviation. (*p ≤ 0.05, **p < 0.01, ***p < 0.001 vs WT)

There were 5 exonic variants in which there was discordance between in silico prediction and in vitro functional studies. CYP2C19*11 (c.449G>A, rs58973490) was expected to not affect protein function; however, protein expression was increased (p < 0.001) as determined by Western blot. The CYP2C19*15 construct showed decreased protein expression levels compared with WT (p < 0.05). Similarly, a rare novel c.1137C>T synonymous variant resulted in 50% reduced protein expression levels (p < 0.01), predicted by most in silico methods to be benign. There were two other synonymous variants, with benign in silico predictions: c.990C>T (rs3758580) and the rare c.1077C>T (rs370051475) that resulted in significant protein expression reduction in vitro. There were 3 cases (4%) versus only 1 control that carried potentially deleterious rare exon-locating variants as determined by Western blots.

Intronic Genetic Variation in CYP2C19 and Stent Thrombosis

The exonic variation only accounts for 3% of the variation detected. There were 169 intronic alterations observed only in cases, twofold the number unique to controls (n = 84). To delineate the significance of these variants, annotation of potential splicing-mediated LoF effect was performed using in silico algorithms. Figure 3a illustrates the case-specific bias of increased allelic variation found in introns 1, 5, 6, 7, and 8 as compared with controls. Intronic variants were observed in 74% of cases and 64% of controls. Splicing prediction resulted in 8% of variants in cases and 6% of variants in controls being categorized as VSP or SP as shown in Fig. 3b. The majority of unique intronic variants in both cases (72%) and controls (78%) had weak or no predictions. There were less than 2% of intronic variants that had conflicting interpretations (Fig. 3b).

Fig. 3.

Splicing predictions annotation of intronic variants unique to cases and controls. a Distribution of intronic number of SNVs and/or indels only observed on cases and not in controls; and vice versa across the CYP2C19 gene (unscaled). b In silico prediction summary of potential splicing motif alterations per variants uniquely identified in cases and controls. c Intronic variants with VSP across the CYP2C19 gene (scaled) with imputed metabolic status and type of splicing motif alteration predicted

We summarize the VSP and SP predictions, imputed metabolizer status and their genomic localization in Fig. 3c. There are 11 cases (16%) that had intronic variants with VSP and SP for creation or alteration of a splicing motif compared with only 6 controls (9%). The majority of VSP and SP intronic variants in controls resulted in a new donor site prediction with only one new potential acceptor site prediction. In contrast, 10 of 13 (77%) intronic variants with VSP and SP that were unique to cases create or activate a cryptic intronic acceptor site. The location of these potentially deleterious intronic variants in cases, as compared with controls, shows a clustering of these variants in cases at the first and last introns of the CYP2C19 gene.

Discussion

This study describes findings of comprehensive sequencing of CYP2C19 resulting in the identification of 160 SNVs and 9 indels unique to clopidogrel-treated patients who experienced ST. We demonstrate all poor metabolizers in this cohort were in ST cases and not in controls confirming their well-known association with clopidogrel resistance. Furthermore, we detected the rare star alleles CYP2C19*8 (rs41291556, MAF 0.15%), CYP2C19*11 (rs58973490, MAF 0.002%), and CYP2C19*15 (rs17882687, MAF 0.21%) alleles in patients with ST but not in controls. Functional validation of these alleles with an in vitro cell system suggested 2 of these variants to be deleterious (CYP2C19*8 and *15). Typically, these alleles are not routinely assayed in current genotyping tests and may be missed when assessing clopidogrel resistance. In our study, NGS yielded greater intronic variation in cases compared with controls that was regionally specific. Annotation of the intronic variation observed with in silico algorithms demonstrated a larger number of potentially deleterious candidates (affecting splicing) in ST cases that may represent novel LoF CYP2C19 alleles associated with clopidogrel resistance. The heterogeneity of genetic variation observed in CYP2C19 sequencing adds to the complexity of interpreting genetic clopidogrel resistance and presents a challenge when sequencing as opposed to genotyping for known actionable variants in “pharmacogenes.”

We elected to study ST because it is the closest drug phenotype associated with clopidogrel failure or resistance. DAPT is recommended after PCI, and a majority of the patients in the USA receive clopidogrel with aspirin [1]. A meta-analysis of 9 studies (n = 9685 patients) reported the presence of 1 or 2 CYP2C19*2 and or *3 LOF alleles that was associated with a significantly increased risk of ST and the composite endpoint of cardiovascular death, MI, or stroke, compared with non-carriers [7]. In 2010, the U.S. Food and Drug Administration (FDA) issued a “boxed warning” on clopidogrel that CYP2C19 testing could be “used as an aid in determining therapeutic strategy.” However, in a review of the data supporting this suggested approach, ACC/AHA guidelines [31] pointed there were no prospective randomized clinical trials conducted demonstrating that changing DAPT based on identification of these genetic variants improved clinical outcomes. This unanswered question is being explored in the recently completely Tailored Antiplatelet Therapy Following PCI (TAILOR PCI; ClinicalTrials.gov identifier: NCT01742117) clinical trial, which will compare prospective genotype-directed antiplatelet therapy with standard antiplatelet therapy [32, 33].

Despite the lack of prospective evidence, laboratories offer genotyping platforms that screen for CYP2C19 LOF alleles and embed DAPT recommendations in the EHR based on the CYP2C19 genetic profile of patients [34]. Patients are frequently genotyped for only common and known CYP2C19 LOF alleles without assessing contribution of rare or other non-coding genetic variation. Although NGS cost and sensitivity for detecting variants has improved, difficulty continues in interpreting the significance of these variants [35]. We attempt to overcome this problem, by performing comprehensive annotation and functional studies to discern the potential effect of rare exonic and intronic variations detected during CYP2C19 sequencing.

NGS in 70 cases with ST and 70 matched controls detected not only common LoF alleles (CYP2C19*2) but also rare LoF alleles like CYP2C*8. As previously described in other studies, poor metabolizers were observed in cases but not in controls of our cohort. Interestingly, heterozygote carriers of the CYP2C19*17 allele (RM) were present at a significantly higher frequency in ST cases when compared with controls. CYP2C19*17 allele increases transcription and protein expression, but its role in clopidogrel platelet activity inhibition is controversial, and current recommendations do not endorse altering DAPT based on the presence of this allele [36]. In one study, carriers of CYP2C*17 allele when treated with clopidogrel were found to have greater inhibition of platelet aggregation as compared with non-carriers and were at an increased bleeding risk at 30 days [37]. However, this study did not genotype for CYP2C19 LoF alleles hence its effect in attenuating these outcomes is unknown. Other studies have demonstrated conflicting evidence, regarding the effect of CYP2C*17 on platelet aggregation and bleeding risk [6, 38–40], which may be attributable to the variant not having an independent effect by virtue of it being in linkage disequilibrium with the CYP2C19*2 loss-of-function variant [41].

We also highlight the potential importance of CYP2C*15 in clopidogrel resistance not only by describing its presence in ST cases but by also demonstrating its previously unknown effect on protein expression in vitro. We also detected a novel, very rare synonymous variant c.1077C>T in a ST case with a deleterious effect on protein expression in vitro. Two cases were carriers of the CY2C19*11 allele (rs58973490, p.Arg150His), and functional validation revealed a significant increase in protein expression as a result of this rare variant as opposed to previous observations that indicate that the CY2C19*11 does not cause significant increase in CY2C19 activity [29, 42]. The effect of CYP2C19 variants on protein expression generally correlates with enzymatic activity, but exceptions have been reported [43].

Two very rare exonic variants were present only in controls—a missense variant c.241G>A was found not to result in decreased CYP2C19 enzyme levels and a synonymous variant c.1077C>T resulted in decreased protein expression. The potentially deleterious genetic variant in controls demonstrates the concept of variable penetrance in pharmacogenomics. In summary, functional validation of the exonic variation (n = 11) detected with a high throughput assay and Western blots was concordant with 54% (n = 6) of our in silico predictions. The variant c.681G>A is essential to the CYP2C19*2 allele and was not expected to show changes in protein expression in our assay as it would result in a splicing alteration and our constructs only contained the ORF. Deleterious synonymous variants described may be affecting mRNA structure and stability, thus affecting the rate of translation [44] and/or protein folding or degradation [45].

This study identified 13 intronic variants (with MAF < 1%) uniquely present in cases, which may represent novel CYP2C19 LoF alleles. The majority of these candidates were present in introns 1 and 6. Many of these intronic variants appear to create or activate cryptic acceptor regions. Intronic variants are increasingly being recognized for playing an important role in the regulation of alternative splicing, gene expression, and possibly mRNA transport [46]. Intronic variation has been described previously for CYP2C19 at intron 2, 23 nucleotides from the exon-intron boundary (referred as CYP2C19*35) [47] resulting in inclusion of the intron (exon 2B) and altering canonical transcript levels. The intronic SNVs identified in this cohort are located more than 100 base pairs away from exon-intron junctions. To the best of our knowledge, no additional potentially deleterious deep intronic CYP2C19 variant candidates have been described to date.

MAF differences in potential LoF intronic variants could have important implications for clopidogrel resistance in different populations. For example, rs74152367, predicted to create a cryptic donor site on intron 3, is not present in non-Finnish Europeans (NFE), but present in 4.74% of Africans (AFR) in gnomAD, with all homozygotes reported in this ethnic group. Similarly, rs111315047 is present in 0.026% of NFE but seen in 1.69% of AFR populations with similar differences identified for rs186254704 and rs77923111 variants in intron 6. These significant MAF differences are not observed for any of the intronic SNVs with SP and VSP that were unique to controls.

Our study has several limitations. Functional validation will be required to determine the biological significance of deep intronic variation and the role they may play in drugs metabolized by CYP2C19. The overall sample size limits the power to detect significant association of the role of both homozygous and heterozygous states of common CYP2C19 genetic variants due to a lower effect size of these variants on ST. For example, for CYP2C19*1/*2, i.e., intermediate metabolizer status, we would require double the sample size to reach adequate power (~ 80%) to detect a significant association with ST. However, this is the largest study of in-depth sequencing of CYP2C19 in patients with ST in which strict criteria for inclusion and appropriately adjusted controls were used. With the increased availability of NGS and decreased cost of NGS, large cohorts may help identify rare genetic variants that may play a role in clopidogrel resistance and confirm the findings of this study. There may also be other “omic” factors that play a role in ST and adverse cardiovascular events, and an integrated approach incorporating these technologies may be required to fulfill the promise of precision medicine [48].

In summary, in this comprehensive sequencing study of CYP2C19 in clopidogrel-treated patients with ST, PM were only found in ST cases, with a significantly higher frequency of so-called RM. In addition, rare CYP2C19 haplotypes such as *8, *15, and 3 protein-disruptive exonic variants including a novel synonymous variant (c.1137C>T) were present in cases and confirmed by functional in vitro genomic studies to be deleterious. A higher burden of variation within intronic regions of CYP2C19 was also detected in ST cases. Annotation of intronic variation identified potential candidates contributing to the underlying CYP2C19 metabolizer status, but these need to be functionally validated. In conclusion, CYP2C19 sequencing identifies deleterious rare exonic and potentially disruptive intronic variants, which are not currently covered by commercial genotyping platforms, in a small minority of patients with ST. The association of some of these rare variants with platelet aggregation and clopidogrel metabolism needs to be further studied before recommendations can be made to routinely genotype or sequence for these variants in clinical practice.