Calcium‐channel blockers: Clinical outcome associations with reported pharmacogenetics variants in 32 000 patients

Aims Pharmacogenetic variants impact dihydropyridine calcium‐channel blockers (dCCBs; e.g., amlodipine) treatment efficacy, yet evidence on clinical outcomes in routine primary care is limited. Reported associations in pharmacogenomics knowledge base PharmGKB have weak supporting evidence. We aimed to estimate associations between reported pharmacogenetic variants and incident adverse events in a community‐based cohort prescribed dCCB. Methods We analysed up to 32 360 UK Biobank participants prescribed dCCB in primary care (from UK general practices, 1990–2017). We investigated 23 genetic variants. Outcomes were incident diagnosis of coronary heart disease, heart failure (HF), chronic kidney disease, oedema and switching antihypertensive medication. Results Participants were aged 40–79 years at first dCCB prescription. Carriers of rs877087 T allele in RYR3 had increased risk of hazard ratio (HF 1.13: 95% confidence interval 1.02 to 1.25, P = .02). Although nonsignificant after multiple testing correction, the association is consistent with prior evidence. We estimated that if rs877087 T allele could experience the same treatment effect as noncarriers, the incidence of HF in patients prescribed dCCB would reduce by 9.2% (95% confidence interval 3.1 to 15.4). In patients with a history of heart disease prior to dCCB (n = 2296), rs877087 homozygotes had increased risk of new coronary heart disease or HF compared to CC variant. rs10898815 in NUMA1 and rs776746 in CYP3A5 increased likelihood of switching to an alternative antihypertensive. The remaining variants were not strongly or consistently associated with studied outcomes. Conclusion Patients with common genetic variants in NUMA1, CYP3A5 and RYR3 had increased adverse clinical outcomes. Work is needed to establish whether outcomes of dCCB prescribing could be improved by prior knowledge of pharmacogenetics variants supported by clinical evidence of association with adverse events.


| INTRODUCTION
High blood pressure-hypertension-is a key modifiable risk factor for cardiovascular morbidity and mortality. While reducing raised blood pressures is the goal, only 1/3 of hypertensive patients treated with antihypertensive medications are estimated to reach target blood pressures. 1 The reasons for failure to control raised blood pressure are complex, but genetic factors are proposed to play a role, either directly on blood pressure or indirectly by influencing antihypertensive medication response, adverse events or medication adherence. 2 Calcium-channel blockers (CCBs) are the first line recommended antihypertensive for most adults with hypertension, and their use is widespread across the world. [3][4][5] There are 2 subgroups of CCBs, the most common being dihydropyridines (dCCBs), which are regarded as relatively safe and cost-effective. 6 Oedema is a common dCCB adverse effect, with incidence rates of 22%, [7][8][9] that affects the quality of life of patients and can lead to discontinuation of treatment. 10,11 The presence of oedema can result in additional prescribing, which in turn can cause additional adverse outcomes including falls, over diuresis, acute kidney injury and polypharmacy. 11,12 Genetic factors can predispose to side effects, as well as further complications.
The pharmacogenomics knowledge base (PharmGKB) documents genetic variants reported to influence dCCB effectiveness or adverse events. 2,[13][14][15] The levels of supporting evidence for each variant is variable, with many only having limited clinical evidence. Such evidence includes reported genes containing single-nucleotide polymorphisms (SNPs) include those encoding calcium channel subunits themselves, such as the voltage-gated calcium channels α1C (CAC-NA1C). SNPs in other ion channels are reported to alter dCCB responses (including PICALM, TANC2, NUMA1, APCDD1, GNB3, SLC14A2, ADRA1A, ADRB2 and CYP3A4). SNPs in ATP-binding cassette subfamily B member 1 (ABCB1) and in cytochrome p450 3A5 (CYP3A5) are reported to affect the clearance of dCCBs, and SNPs in cytochrome p450 ox reductase (POR) reportedly influence the plasma concentration of medicines. SNPs in nitric oxide synthase 1 adaptor protein (NOS1AP) increase risk of cardiovascular death, and SNPs in ryanodine receptor 3 (RYR3) 16 and in atrial natriuretic precursor A (NPPA) are reported to increase the risk of cardiovascular disease. In particular, RYR3 (in intracellular calcium channels) was found to be associated with heart failure (HF) and there is a need to examine its effect on stroke, and effects on heart disease in risky groups as it is unknown. 17 Evidence of impact on clinical outcomes, especially in routine primary care (rather than in acute hospital care settings) is currently limited for most pharmacogenetics variants reported to affect dCCBs.
Here we analyse the UK Biobank (UKB) community volunteer cohort with linked genetic and medical records. We aimed to determine the extent to which 23 commonly occurring (minor allele frequency >3%) pharmacogenetic variants in 16 genes reported to affect dCCB effectiveness or rates of adverse events are associated with clinical outcomes.  18  We also identified antihypertensive prescriptions apart from dCCB; diuretics, β-blockers, α-blockers, angiotensin converting enzyme inhibitors and other antihypertensives using the Read 2 codes and BNF codes (see Table S1 for details).

| Disease ascertainment
Primary and secondary care health records were used to examine the dCCB-related adverse events. Peripheral oedema diagnoses were ascertained from ICD-10 and ICD-9 codes: and converted to Read codes used in UK primary care records using UKB-provided diagnostic code maps. Cardiovascular events from hospital admissions records were available up to 14 years follow-up after baseline assessment (HES in England up to 30 September 2020: data from Scotland and Wales censored to 31 August 2020 and 28 February 2018, respectively), covering the entire period up to the date of censoring of primary care prescribing data. Diagnosis of myocardial infarction (MI)/ angina, stroke, chronic kidney disease (CKD), HF and ischemic stroke were ascertained using ICD-10 codes (see Supporting Information for further details).

| Genetic variants
We utilized genotype data from UKB, as described previously 20 (see Supporting Information for details). Our analysis included 451 367 participants (93%) identified as genetically European (identified by genetic clustering, as described previously 21 ): unfortunately, sample sizes from other ancestry groups were too small to analyse separately.
We analysed the genetic variants with documented effects on dCCBs effectiveness in the literature 22 Table S1 for details). Genotype status for 23 variants could be ascertained from the available UKB imputed data (release version 3) and minor allele frequencies were common enough to study (frequency varying from 3 to 46%-see Table S2 for details) in the UKB cohort: results for all 23 studied SNPs are reported. We calculated correlation coefficient to check linkage disequilibrium. We also investigated the association between genotype and likelihood of switching dCCB for an alternative antihypertensive prescription using Cox's proportional hazards regression models, with adjustment for age at first prescription, sex and genotyping principal components of ancestry 1-10 in patients. See Supporting Information for further details.

| Primary analysis
To adjust for multiple statistical testing and control the false discovery rate, we applied Benjamini-Hochberg correction to P values for the associations between 23 SNPs and each outcome (using R function p.adjust()).

| Secondary analysis in patients with heart disease diagnosis prior to dCCB treatment
We only included patients who had any heart diseases prior to the dCCB treatment for MI/angina/HF outcomes models as secondary analysis, as worsening angina and acute MI are reported as a caution for patients with coronary artery disease by the Food and Drug Administration in the prescribing information. 9 We also tested associations for stroke and the RYR3 calcium channel gene variant in patients on dCCBs, to examine the treatment effect, as stroke was associated with RYR3 in a genome-wide association study (GWAS) regardless of use of dCCBs. 24

| Prescribed additional antihypertensives
We identified the BNF and Read 2 codes for the antihypertensive medication classes (β-blockers, α-blockers, diuretics, angiotensin receptor blockers, angiotensin converting enzyme inhibitors and vasodilators) and determined whether patients receiving dCCB prescriptions also received another antihypertensive within the dCCB prescribing time period. We then included this variable as a covariate in analyses.

| Amlodipine and other dCCBs
We performed sensitivity analysis of our primary results splitting the dCCBs into 2 categories: in just those patients prescribed amlodipine (by far the most common dCCB) and other dCCB only and repeated the analyses described in the previous sections. Further splitting of nonamlodipine dCCBs was not feasible due to low numbers.

| Analysis of unrelated participants only
We identified participants related to the third degree or closer using KING kinship analysis. 25 We then repeated our primary results only in unrelated participants of European descent by randomly excluding 1 of each pair of related to the third degree or closer.

| Characteristics of the sample
There were 32 360 (45.6% female) patients who were prescribed dCCB in primary care. The mean age was 61.3 years (standard deviation [SD] 7.7). The number of prescriptions in a year varied from 1 to 25, with a mean of 9.2 (SD 4.6) and a median of 7.9 (interquartile range 6.3 to 13). The mean prescription period was 5.9 (SD 5.2) years, the median was 4.4 (interquartile range 1.6 to 9.1) (see Table 1 for details). The allele frequencies for the 23 studied genetic variants range from 3 to 50% (for details, see Table S3). We found no pairs of variants in high (R 2 < .8) linkage disequilibrium (see Table S4).

| Associations with prior evidence
We investigated 23 genetic variants with reported pharmacogenetic effects on dCCB effectiveness or adverse events. We found supporting evidence in the UKB for 5 of the 23 reported dCCB pharmacogenetic associations ( Table 2). Details of the 5 genes reported below, including secondary analysis of other adverse outcomes.

| RYR3
The RYR3 rs877087 T allele prevalence in people on dCCB treatment in UKB was 46%, and TT homozygotes was 21.3%.
Of the 32 360 patients prescribed dCCBs, 2292 developed HF during the follow-up period. Diagnoses were more common in RYR3 rs877087 TT homozygotes (n = 404, 6.1% of 6607) and CT heterozygotes (n = 943, 6.1% of 15 377) compared to common CC homozygotes (n = 491, 5.4% of 9090; Figure 1 and Table 3; see Table S5 for details  Table S6). Therefore, the overall incidence of CKD in patients prescribed dCCBs could be reduced by 8.6% if rs877087 T allele carriers received the same treatment benefit as noncarriers (95% CI 3.2 to 14.0).

| CYP3A5
Patients with CYP3A5 rs776746 TT (CYP3A5*3) genotype (0.47% of patients), a variant previously linked to kidney related outcomes 26 had increased risk of CKD (HR 2.12: 95% CI 1.34 to 3.38, P = .002) compared to CC homozygotes ( Figure 1 and Table 3). The association was still significant after Benjamini-Hochberg adjustment for multiple statistical testing (adjusted P = .03). When we repeated the analysis for patients who were on dCCB but had no CKD history, 12.3% of CYP3A5 rs776746 TT homozygotes without prevalent CKD were diagnosed with incident CKD compared to 6.6% of heterozygotes and 6.8% homozygotes for CC (HR 2.09, 95% CI 1.29 to 3.37, P = .003; see Table S9 for details). We estimated that if rs776746 TT homozygotes could experience the same treatment effect as CC homozygotes 11 CKD diagnoses could have been avoided (95% CI 4 to 18; see Table S6).
Therefore, the overall incidence of CKD in patients prescribed dCCBs could be reduced by 0.5% (95% CI 0.2 to 0.9) if rs776746 TT homozygotes received the same treatment benefit as CC homozygotes.
Of the patients on dCCB prescription, 5565 (14.2%) changed treatment from dCCB CCBs to other antihypertensives. CYP3A5 rs776746 TT homozygotes (n = 27/152) were also more likely to change treatments compared to common homozygotes; HR 1.59, 95% CI 1.09 to 2.32, P = .02, respectively (see Figure 1 and Table 3; see Table S10 for details). Incident MI/angina was less likely to occur in patients heterozygous for CYP3A5 rs776746 compared to CC homozygotes (P = .01).

| ADRA1A
Adrenoceptor  Figure 1 and Table 3). We estimated that if rs1048101 AA homozygotes could experience the same treatment effect as noncarriers (e.g., were prescribed an alternative antihypertensive medication unaffected by this genotype) 86 CKD diagnoses could have been avoided (95% CI 13 to 138; see Table S6). Therefore, the overall incidence of CKD in patients prescribed dCCBs T A B L E 4 Associations between hospital-diagnosed coronary heart disease (CHD) and stroke and variants of RYR3 in patients with CHD diagnosed prior to dihydropyridine calcium-channel blocker prescription 1.07, 95% CI 1 to 1.13, P = .04, FDR P = .28 for AC; Figure 1 and Table 3; see Table S11 for details). TWIST analysis showed that rs564991 was not associated with CHD in individuals never prescribed dCCBs (GMTE0 estimate P > .05; Table S4).

| SENSITIVITY ANALYSES
In total, 23 971 (61%) patients were also on another antihypertensive medication at some point during dCCB prescription time period. In the sensitivity analysis adjusted for receiving another antihypertensives during the dCCB prescribing period, significant associations with outcomes from the main analysis remained consistent. See Table S13 for the details.
The sensitivity analysis of patients on amlodipine (n = 31 357) and other dCCBs (n = 6854) separately were consisted with the primary analysis, is presented in Table S14.
After excluding one of patients related 27 042 patients remained.
Many associations remained significant such as APCDD1 and CHD, RYR3 and CKD, NUMA1 and treatment switch whilst some associations were not significant, the effect sizes were consistent to the whole cohort (see Table S15).

| DISCUSSION
CCBs, especially dihydropyridines (dCCBs) such as amlodipine, are commonly prescribed to reduce blood pressure. Many pharmacogenetic variants are reported to impact dCCB responses, with evidence from laboratory studies, randomized trials, or acute hospital settings.
However, data on clinical impact in routine care in the community is RYR3 mediates Ca 2+ release from ryanodine-sensitive stores, triggering cardiac and skeletal muscles. 27 A common variant in RYR3 (rs877087) increased risk of HF in a study of 2516 people randomized to amlodipine or to other antihypertensives. 17 We support and extend this literature in a substantially larger sample using longitudinal analysis methods: we report increased HF risk in both TT homozygotes (n HFs = 404 in 6607 genotypes; HR 1.15) and CT heterozygotes (n HFs = 943 in 15 377 genotypes; HR 1.12), compared to CC homozygotes (n = 9090). We used TWIST, 23 a novel pharmacogenetic causal inference framework, to estimate the population average GMTE on HF if all RYR3 T allele carriers could experience the same treatment effect as common CC homozygotes (e.g., they were prescribed an alternative medication): we estimate that HF risk would reduce by 9.2%, corresponding to 170 avoidable HF diagnoses in the studied patients. Further work is needed to determine the optimum strategy to reduce the risk of T allele carriers, for example, by prescribing an alternative treatment or with increased monitoring of patients. Furthermore, rs877087 has been associated with stroke in a GWAS, but the genotype effect in patients on treatment is unknown. 24 Our findings suggest that rs877087 CT had an increased risk of hospital diagnosed stroke (n = 203/15381, HR = 1.11, P = .02) compared to common CC homozygotes, but we found effect in TT homozygotes.
Observational studies of drug effects often suffer from indication and other biases: As doctors aim to prescribe each medication based on the patients' clinical state, statistically separating the effects of the medication from the effects of underlying disease is challenging, especially as data to correct potential confounders are seldom complete or entirely accurate. However, genotypes are inherited at conception and stay fixed, meaning that they predate receipt of the studied medications. In our study, we found that genotype was not associated with treatment initiation. Associations between genotypes and outcomes provide less confounded evidence than conventional observational associations, particularly because the participants and GPs were not told given genotype information by the UKB study. Because genetic variants are largely independent of traditional confounding, 28 and GPs and patients are unaware of these genotypes when making prescribing decisions and diagnosing outcomes, we can therefore assume that the difference between genotype carriers is due to the modifying effect of the genotype on medication (hence our naming in the GMTE -genetically modified treatment effect). This assumption is common to such Mendelian randomization studies. Therefore, our finding that variants in 2 genes (NUMA1 and CYP3A5) with switching treatment may result directly from dCCB pharmacokinetics or/and pharmacodynamics effects. Further work (both replication and experimental validation) is required to confirm the precise biological mechanisms involved.
NUMA1 rs10898815 was previously identified in a GWAS of blood pressure 29 but we found no reports on switching antihypertensive treatment. We found that AA homozygotes had increased likelihood of switching treatment, which was significant after multiple testing adjustment. CYP3A5 is a cytochrome p450 enzyme and metabolizer of dCCB; CYP3A5*3 is the most common nonfunctional allele (rs776746-T; with a prevalence of 6.6% in the UKB European cohort), which results in increased clearance of dCCB, 30 resulting in less successful treatment. Our findings support this: CYP3A5*3 homozygotes had increased risk of CKD, for which high blood pressure is a risk factor, and were 59% more likely to change treatments compared to common homozygotes.
In a pathway-focused GWAS, 31 genes in the ADRA1 pathway ultimately affect intracellular calcium release (which dCCBs block) and blood pressure. The isoforms (e.g., ADRA1A) was associated with hypertension in patients. In a study in mice, it also mediated renal vasoconstriction in hypertension. 32 34 Additionally, a previous review about pharmacogenomics of hypertension medications reported 4 variants associated with dCCB response in a small Japanese sample. 22 Of the 23 dCCB variants, we found evidence for an effect on outcomes/adverse events for only 10 variants-even fewer after adjustment for multiple statistical testing-suggesting that these few specific variants should be a priority for future study. The possible reasons for the lack of consistency include interethnic differences in studied populations, heterogeneity in exact phenotype studied, lack of adherence to medication or variability in medication history between patients, but may also include publication biases, in which false positive statistical associations (type 1 errors) tend to be overrepresented, especially from small studies.
However, we here add substantially to the evidence base for these variants due in part to the large sample size studied, but also the strengths of analysing real-world primary care prescribing and the novel pharmacogenetic analysis approach triangulating evidence from multiple analysis methods (TWIST 35 ). Using these data for pharmacogenetics analysis means that we are able to look at more adverse reactions over longer periods and have therefore increased confidence for the relevance to routine clinical care of hypertension of variants where significant effects on outcomes are identified.
In conclusion, our analysis of longer-term prescribing in real-world primary care data support the hypothesis that use of genetic information in antihypertensive prescribing might optimize treatment selection for specific patients to maximize efficacy and reduce incidence of adverse events. The variants identified as associated with adverse clinical outcomes are good candidates for studies to test whether dCCB treatment outcomes can be improved with pharmacogenetic guided prescribing.
Department of Health and Social Care. The funders had no input in the study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the article for publication. The researchers acted independently from the study sponsors in all aspects of this study.

COMPETING INTERESTS
All authors declare no support from any organization for the submit-

TRANSPARENCY
We affirm that this manuscript is an honest, accurate and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained. The full methods are available in the Supporting Information.

DATA AVAILABILITY STATEMENT
The genetic and phenotypic UKB data are available upon application to the UKB (www.ukbiobank.ac.uk/register-apply). The derived data fields used in our analysis will be available via the UKB, searching for application number 14631-we are not able to share these directly.