Genetic Determinants of Thiazide‐Induced Hyperuricemia, Hyperglycemia, and Urinary Electrolyte Disturbances – A Genome‐Wide Evaluation of the UK Biobank

Thiazide diuretics, widely used in hypertension, cause a variety of adverse reactions, including hyperglycemia, hyperuricemia, and electrolyte abnormalities. In this study, we aimed to identify genetic variants that interact with thiazide‐use to increase the risk of these adverse reactions. Using UK Biobank data, we first performed genomewide variance quantitative trait locus (vQTL) analysis of ~ 6.2 million SNPs on 95,493 unrelated hypertensive White British participants (24,313 on self‐reported bendroflumethiazide treatment at recruitment) for 2 blood (glucose and urate) and 2 urine (potassium and sodium) biomarkers. Second, we conducted direct gene–environment interaction (GEI) tests on the significant (P < 2.5 × 10−9) vQTLs, included a second UK Biobank cohort comprising 13,647 unrelated hypertensive White British participants (3,478 on thiazides other than bendroflumethiazide) and set significance at P = 0.05 divided by the number of vQTL SNPs tested for GEIs. The vQTL analysis identified eight statistically significant SNPs for blood glucose (5 SNPs) and serum urate (3 SNPs), with none being identified for the urinary biomarkers. Two of the SNPs (1 glucose SNP: CDKAL1 intron rs35612982, GEI P = 6.24 × 10−3; and 1 serum urate SNP: SLC2A9 intron rs938564, GEI P = 4.51 × 10−4) demonstrated significant GEI effects in the first, but not the second, cohort. Both genes are biologically plausible candidates, with the SLC2A9‐mediated interaction having been previously reported. In conclusion, we used a two‐stage approach to detect two biologically plausible genetic loci that can interact with thiazides to increase the risk of thiazide‐associated biochemical abnormalities. Understanding how environmental exposures (including medications such as thiazides) and genetics interact, is an important step toward precision medicine and improved patient outcomes.

disease, respectively. 4There are several international and national hypertension treatment guidelines, and although they differ in some aspects, there is consensus on non-pharmacological and pharmacological approaches to management of hypertension. 2our major drug classes are generally recommended, including the renin-angiotensin-aldosterone system inhibitors (e.g., angiotensin-converting enzyme inhibitors and angiotensin receptor blockers), beta-blockers, calcium channel blockers, and thiazide-type/thiazide-like diuretics. 2,5ince their introduction in the late 1950s, 6 thiazide diuretics have remained the cornerstone of antihypertensive treatment, being among the most widely used, and least costly antihypertensives. 2,7A meta-analysis of 12 randomized controlled trials (48,898  patients) confirmed their effectiveness in reducing all adverse CVD clinical outcomes, including stroke, coronary heart disease, heart failure, cardiovascular death, and all-cause mortality. 8In the United Kingdom, both thiazide-type diuretics (e.g., bendroflumethiazide and hydrochlorothiazide) and thiazide-like diuretics (e.g., chlorthalidone and indapamide) are available for use alone or as fixed dose combinations with other antihypertensives.However, bendroflumethiazide remains the most commonly prescribed thiazide. 9,10hiazide diuretics differ in their molecular structure and pharmacokinetic/pharmacodynamic profiles but have similar pharmacological actions, with diuresis mainly resulting from the inhibition of the thiazide-sensitive sodium chloride (Na + /Cl − ) co-transporter (NCC). 9,11The NCC, located in the renal distal convoluted tubule, is responsible for ~ 5-7% of renal sodium re-absorption. 9,11lthough the mechanism of thiazide-mediated blood pressure reduction is not fully understood, it may partly result from natriuresis and reduced blood volume that occurs when the NCC is inhibited. 12Partly as an extension of their pharmacological actions, thiazides have also been associated with biochemical/electrolyte abnormalities, including hyponatremia, hypokalemia, hyperuricemia, and hyperglycemia/impaired glucose tolerance, which vary in severity, occasionally resulting in mortality. 7,9,13lood concentrations of glucose and uric acid are partly determined by genetic factors.A UK Biobank study (n = 363,228 participants) that evaluated the genetic basis of 35 blood and urine biomarkers identified 363, 121, 15, and 1 independent loci associated with serum urate, blood glucose, urine sodium, and urine potassium, respectively, 14 results that are consistent with other studies. 15ene-environment interactions are also important to consider in terms of their impact on resultant concentrations of electrolytes. 16In addressing this, previous evaluations of this have mostly used a candidate gene approach, 17,18 partly due to the large sample size requirements for undertaking genotype-by-environment interactions (GEIs). 19,20One solution to the sample size challenge is to use electronic health records that are linked to large biobanks, such as the UK Biobank. 21,22Another solution, unique to studies undertaking genomewide GEIs, is to first perform a "screening step" which involves undertaking a variance heterogeneity analysis, also known as a genomewide variance quantitative trait locus (vQTL) analysis. 19,20Because SNPs with GEIs are likely to be vQTLs, this screening step (which does not require knowledge of the environmental risk factor) prioritizes the SNPs that can be directly tested for GEIs improving statistical power (few vQTLs, as opposed to all genomewide SNPs, are directly tested for GEIs). 19,20Using an approach similar to Wang and colleagues who investigated GEIs in relation to five environmental factors/covariates (age, sex, physical activity level, sedentary behavior, and smoking), 19 the aim of the present study was to identify, at genomewide level, genetic variants that interact with thiazide-use to increase the risk of biochemical and electrolyte abnormalities.

METHODS
This study adheres to the STrengthening the Reporting Of Pharmacogenetic Studies (STROPS) guideline 23 (Table S1) and the REporting of studies Conducted using Observational Routinely collected health Data (RECORD) statement 24 (Table S2).

Study design and setting
The UK Biobank, a large population-based prospective cohort, recruited > 500,000 participants aged between 37 and 73 years at the time of recruitment, which occurred between 2006 and 2010 (22 assessment centers in England, Scotland, and Wales). 21,22Phenotypic, health-related, lifestyle, and genomewide genotypic information were collected at baseline and/or follow-up assessments using several methods, including physical measurements, touchscreen questionnaires, biochemical assays, biological measurements, and multi-modal imaging.Additional follow-up information is provided through linkage with electronic health records, including death registrations, cancer registrations, hospital inpatient and outpatient episodes, and primary care.The UK Biobank obtained ethics approval from the North-West Multicentre Research Ethics Committee (approval number: 11/NW/0382) and all participants provided written informed consent before data collection.The analyses included in this study were approved by the UK Biobank (approved application number: 56653).

Study participants
We included UK Biobank participants who had self-reported hypertension at baseline and were of White British ancestry (based on selfidentification and principal component analysis of genetic data).We tested two cohorts: (i) bendroflumethiazide-treated participants, and (ii) participants taking other thiazides.Bendroflumethiazide-treated participants were about 7 times the number of participants taking other thiazides, and so we randomly split the hypertensive participants not taking any thiazides in a ratio of 7:1 and included these as controls in the 2 cohorts, respectively.Participants who withdrew consent were excluded from all analyses.Details of the codes used to identify hypertensive participants and those on self-reported thiazide treatment are available at https:// github.com/ iasii mwe/ ukb_ bfz, where UK Biobank data fields and codings for all study variables are also provided.

Variables
Exposure variables included all genotyped and imputed SNPs.The primary outcomes included two blood (glucose and urate) and two urine (potassium and sodium) biomarkers.Age, sex, and the first 10 principal components of genetic ancestry were considered as covariates during the primary analysis.

Data sources/measurement
Blood glucose (hexokinase analysis, analytical range 0.6-45 mmol/L), serum urate (uricase PAP analysis, analytical range 89-1,785 μmol/L), urine sodium (ion selective electrode analysis, analytical range 10-400 mmol/L), and urine potassium (ion selective electrode analysis, analytical range 2-200 mmol/L) were all measured using a Beckman Coulter AU5800. 25,26Sample collection, processing, and the quality performance checks have been detailed previously. 27enotyping and imputation in the UK Biobank have also been described previously. 21Briefly, DNA samples obtained from the UK Biobank study participants were genotyped using two similar (about 95% shared marker content) genomewide genotyping arrays: the Applied Biosystems UK Biobank Lung Exome Variant Evaluation (UK BiLEVE) Axiom Array by Affymetrix (49,950 participants and 807,411 markers) and the Applied Biosystems UK Biobank Axiom Array (438,427 participants and 825,927 markers).Per-marker (e.g., inclusion of markers present on both arrays, with a missing rate ≤ 5%, and a minor allele frequency (MAF) ≥ 0.0001) and per-sample (e.g., removing outliers based on missing rate and heterozygosity adjusted for population structure or those with mismatches between selfreported and marker-inferred sex) evaluation resulted in a dataset of 487,442 samples and 670,739 autosomal markers from both arrays.Efficient phasing and genotype imputation increased the number of markers (autosomal SNPs, short indels, and large structural variants) to 93,095,623 in the 487,442 individuals (version 3 imputed genotypes used in the present analysis). 21In addition to the quality control procedures centrally conducted by the UK Biobank, we applied additional variant filters and excluded variants with MAF <0.05 in the analyzed participants, missing genotype rate > 0.05, Hardy-Weinberg equilibrium (HWE) test P value < 10 −5 and imputation quality (INFO score) < 0.3 for consistency with the Wang et al. study. 19Per-participant quality control, on the other hand, included filtering out patients with: non-White British ancestry, discordant self-reported and genetic sex, sex chromosome aneuploidy, and outlying heterozygosity and missing rates.Based on a kinship threshold of 0.0884, we obtained the maximal set of unrelated participants as detailed at https:// github.com/ iasii mwe/ ukb_ bfz.

Sample size calculations
We evaluated sample size using functions within the R package "genpwr" 28 and standard deviations of the four outcomes in the UK Biobank participants (https:// bioba nk.ctsu.ox.ac.uk/ cryst al/ label.cgi? id= 717).For genomewide vQTL and QTL analysis, we used the "ss.calc.linear"function ("Function to Calculate Sample Size in Linear Models").Assuming an explained variability in outcome or a target coefficient of determination (R 2 ) of 1%, MAFs ranging from 5 to 20%, 80% power, and a similar true/test mode of inheritance (either additive, dominant or recessive), the minimum sample size requirement was 4,606 participants at a genomewide significance threshold of 2.5 × 10 −9 (see "Statistical methods" section on why this threshold was chosen).For the direct GEI tests, we used the "ss_envir.calc.linear_outcome" function ("Function to Calculate Power for Linear Models with logistic environment interaction") and similar assumptions (R 2 of 1% for genetic, environmental, and genetic/environment interaction values, MAFs ranging from 5 to 20%, 80% power, and a similar true/test mode of inheritance).We additionally estimated that 25% of hypertensive UK Biobank participants are prescribed thiazides, and this resulted in a minimum sample size requirement of 766 participants for a significance threshold of 0.05 (only one GEI test conducted).Considering Bonferroni-correction, this sample size requirement respectively increases to 1,298, 1,822, and 2,339 participants when 10, 100, and 1,000 GEI tests are conducted.

Statistical methods
Outcome transformation.Similar to Wang et al.'s study, 19 we used raw outcomes pre-adjusted for covariates without any nonlinear transformations.Specifically, the raw outcomes were linearly regressed on age and the first 10 principal components (in a sensitivity analysis conducted for the blood biomarkers, we additionally adjusted for the variables listed in the "Sensitivity analyses" subsection), and the resulting residuals used in subsequent analysis.Next, values that were more than five standard deviations from the mean were excluded before a standardization to z scores (mean 0, variance 1) was made for each sex grouping.
Predictor handling.Quantitative predictor variables were neither transformed nor categorized.Nonlinearity between these variables and the outcomes was assessed using restricted cubic splines in the rms R package. 29Splines were prepared with knot placement based on the percentile distributions of these variables (5 knots placed at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles were used). 30ssing data.Missing genotype data were imputed as previously described. 21Participants with missing outcome data were excluded from the corresponding analyses.In the primary analysis, no participant was excluded on the basis of missing covariate data because age, sex, and principal component analysis data was available for all participants, whereas in the sensitivity analyses, the covariates body mass index (BMI; 0.4% cases missing data), smoking status (0.4% missing), alcohol status (0.1% missing), degree of physical activity (0.8% missing), and Townsend index (0.1% missing) were singly imputed using the multivariate imputation by chained equations ("mice") R package. 31nomewide vQTL and genomewide QTL analysis.Genomewide vQTL analysis was conducted using the OmicS-data-based Complex trait Analysis (OSCA; https:// yangl ab.westl ake.edu.cn/ softw are/ osca/# vQTLA nalysis) software tool (−-vqtl-mtd option 2 ("Levene's test with median"), no genetic mode of inheritance assumed), 32 whereas genomewide QTL analysis (commonly referred to as a genomewide association study (GWAS)) was conducted using PLINK2 (−-glm option, additive genetic mode of inheritance assumed). 33he genomewide significance threshold was set to 2.5 × 10 −9 which represents a Bonferroni correction (4 investigated outcomes) for a stringent genomewide significance threshold of 1 × 10 −8 . 19,34,35fter performing genomewide vQTL/QTL analysis, the number of independent significant SNPs was determined using linkage disequilibrium (LD) clumping for each outcome and this was implemented in PLINK2 using parameters similar to Wang et al.'s (--clump-p1 2.5 × 10 −9 , --clump-p2 2.5 × 10 −9 , --clump-r2 0.01, and --clump-kb 5,000). 19Results were graphically presented using Manhattan and Quantile-Quantile plots (qqman R package 36 ).
GEI tests.Using the most significant vQTL SNPs, the pre-adjusted outcomes and an assumption of additive genetic mode of inheritance, we performed GEI tests (interaction term of SNP*thiazide status included in the model) using standard analysis of variance in R (version 4.2.2).We used a Bonferroni-adjusted significance threshold of 0.05 divided by the number of vQTL SNPs tested for GEIs.To visualize the effects of the SNPs significant during the GEI analysis, the outcomes (both raw and pre-adjusted) were classified according to genotype and thiazide status.Additionally, we used an extreme discordant phenotype strategy, 37 plotting the 10%, 20%, 30%, 40%, and 50% of patients exhibiting extreme phenotypes -this facilitated visualization of small effect sizes.
Sensitivity analyses.We conducted two sensitivity analyses for the blood biomarkers.First, we additionally considered BMI, smoking status, alcohol status, degree of physical activity, Townsend index (which reflects socioeconomic status), the genotyping array, antihypertensive medications (beta-blockers, calcium channel blockers, losartan, and non-losartan renin-angiotensin-aldosterone system inhibitors), antidiabetic drugs (for blood glucose), allopurinol status (serum urate), and an ARTICLE additional 30 principal components of genetic ancestry.In the second sensitivity analysis, we pooled and analyzed the two cohorts together.

RESULTS Participants
Taking into account those who had not withdrawn consent as of March 10, 2023, there were 109,140 unrelated hypertensive White British ancestry participants that were included in the analysis following the quality control steps detailed in Figure 1.They were split into 2 cohorts of 95,493 (the "first cohort") and 13,647 (the "second cohort") participants, with ~ 25% of the participants being on self-reported bendroflumethiazide (first cohort) and other thiazides (second cohort).Figures S1 and S2 show the baseline characteristics of these 2 cohorts, whereas Figure S3 shows that a linearity assumption between the continuous covariates and the outcomes could generally be relied upon.

Single nucleotide polymorphisms
Following quality control procedures, a total of 6,198,661 SNPs were analyzed.

Genomewide vQTL analysis
Genomewide vQTL analyses performed in the first cohort using the median-based Levene's test produced significant (P < 2.5 × 10 −9 ) SNPs for the blood (Figure 2) but not urine (Figure S4) biomarkers, with limited evidence of genomic inflation (Figure S5, highest genomic inflation factor = 1.062).LD clumping applied to the significant vQTL blood biomarker SNPs produced 5 (representing 147 SNPs) and 3 (representing 109 SNPs) independent SNPs for blood glucose and serum urate, respectively (Table S3), whose details are shown in Table 1 (Table S4 shows details for all the genomewide significant vQTL SNPs).which means vQTL-based screening (compared with using standard GWAS) narrowed the genomic search and reduced the multiple testing burden in addition to discovering new loci (5 of the 8 independent vQTL SNPs did not have effects on phenotypic mean using a significance threshold of 2.5 × 10 −9 ; Table 1).For the three that had both QTL and vQTL effects, the effects were in the same directions (Tables S4, S5), which means that  ) in the first cohort  S3 (number of genomewide significant and independent QTL SNPs), and Table S5 (tabular results for all genomewide significant QTL SNPs).

GEI tests
To determine if the associations between the eight independent significant vQTL SNPs and phenotypic variance could be explained by GEIs, we conducted interaction tests.In the first cohort, only two (one blood glucose SNP: cyclin-dependent kinase 5 regulatory subunit associated protein 1 like 1 (CDKAL1) intron rs35612982, P = 6.24 × 10 −3 ; and one serum urate SNP: solute carrier family 2 member 9 (SLC2A9) intron rs938564, P = 4.51 × 10 −4 ) SNPs passed the Bonferroni-adjusted significance threshold (P < 6.25 × 10 −3 ).None of these SNPs were significant in the second cohort.The interaction P values for all 8 analyzed SNPs are shown in Table 1, that also includes the effect sizes in thiazide-vs.non-thiazide-treated participants, vQTL and QTL analysis P values, and performance in the second cohort.
Figures S8 and S9 show the blood glucose and serum urate concentrations classified according to the genotypes of the top SNPs and bendroflumethiazide status.

Sensitivity analysis
To assess the impact of only adjusting for age, sex, and 10 principal components of genetic ancestry during primary analysis, we performed a sensitivity analysis in which we adjusted for additional covariates (BMI, smoking status, alcohol status, degree of physical activity, Townsend index, genotyping array, antihypertensive medications, antidiabetic medications, allopurinol status, and additional 30 principal components of genetic ancestry).The results are shown in Figures S10 and S11 (Manhattan plots), Table S3 (independent SNPs), Tables S6 and S7 (linear regression outputs for model excluding SNPs and principal components of genetic ancestry), Table S8 (genomewide significant vQTL SNPs), and Table S9 (genomewide significant QTL SNPs).One key finding was that except for genotyping array and Townsend index, the additional non-genetic covariates influenced the blood biomarker levels, as demonstrated by their P values in the linear regression model shown in Tables S6 and S7, and that their effect size directions agreed with the literature.Most importantly and for the vQTL analysis that informed the GEI tests, the genes identified during the sensitivity analysis (blood glucose: TCF7L2 and HLA-DQB1-AS1; serum urate: ABCG2 and SLC2A9) had been identified during the primary analysis.These results were similar to those of the second sensitivity analysis, in which pooling the two cohorts did not result in additional loci (vQTL plots in Figure S12).

DISCUSSION
We have conducted a genomewide GEI study in the UK Biobank using a two-stage approach similar to that used by Wang et al. 19 Our analysis has identified two loci/genes: (i) cyclin-dependent kinase 5 (CDK5) regulatory subunit associated protein 1 like 1 (CDKAL1) which may predispose to thiazide-induced hyperglycemia; and (ii) solute carrier family 2 member 9 (SLC2A9) which may predispose to thiazide-induced hyperuricemia.
The association between blood glucose and CDKAL1 has been widely reported with 63 unique publications (as of March  11, 2023) deposited in the GWAS Catalog (https:// www.ebi.ac.uk/ gwas/ ). 15However, only one of these studies reported GEIs 38 : this was undertaken in the UK Biobank (350,016 unrelated participants) and explored multi-ancestry genomewide vQTL analysis for 20 serum cardiometabolic biomarkers before conducting GEIs for 2,380 exposures, that included selfreported thiazide use.However, unlike the present study, none of the 847 significant GEIs were related to random blood glucose or serum urate.This difference can be explained by the fact that different populations were studied: Westerman et al. 38 analyzed a subset of 336,298 European-ancestry participants, whereas our study analyzed 109,140 hypertensive White British ancestry participants.Differences in the covariates adjusted for during phenotype preprocessing may also have contributed to the differences.To the best of our knowledge, no published study to date has reported the CDKAL1:thiazide influence on blood glucose.2][43] Based on the plots (Figure S8) of blood glucose concentrations classified according to rs35612982 genotype and bendroflumethiazide status, wild-type homozygotes not taking thiazides have the lowest median blood glucose-levels, and display the most substantial increase in blood glucose levels when treated with thiazides.Functional genomic studies will be needed to understand the mechanism of this interaction.
With regard to SLC2A9 and serum urate, 59 publications have been deposited in the GWAS Catalog up until March 11, 2023, with two studies 44,45 reporting on GEIs (interaction effects between genes and serum urate on bone mineral density and type 2 diabetes mellitus).In the literature, 17,[46][47][48][49] several studies have also reported/reviewed SLC2A9 interactions involving thiazides/diuretics.One of the studies, a primary study, 17 reported that the SLC2A9 intron rs13129697 SNP significantly (P = 0.010) interacted with thiazides/loop diuretic use and selfreported incident gout in 3,524 hypertensive participants.This SNP had a QTL P ~ 0 (6.30× 10 −694 ) but did not pass our stringent vQTL significance threshold despite having a low P value of 9.05 × 10 −09 .A post hoc GEI analysis, nevertheless, showed a significant effect (P = 2.1 × 10 −4 ).Of note is that, despite not including this SNP in our GEI analysis, we were still able to replicate the role of SLC2A9 as we detected many other significant SNPs at the same locus.The SLC2A9 variant rs938564 G allele reduced serum urate levels (raw phenotype) by 22.9 μmol/L in non-thiazide treated-participants (Figure S9), a protective effect which became more pronounced (25.4 μmol/L) in thiazidetreated participants.The difference in effect sizes (2.5 μmol/L) between the non-thiazide and thiazide treated groups may appear small but this is not unexpected at a population level.For example, using the extreme discordant phenotype strategy to identify ARTICLE individuals most likely to benefit from genotyping, 37 this difference becomes even more pronounced in the 10% of participants with the most extreme phenotypes (reductions of 85.8 vs. 98.8 or a difference of 13 μmol/L).According to the Genotype-Tissue Expression portal (https:// www.gtexp ortal.org/ home/ snp/ rs938564), rs938564 decreases the expression of SLC2A9 in adipose tissue (P = 2.7 × 10 −5 , normalized effect size = −0.17),which is important because adipose tissue can secrete uric acid. 50dditionally, the SLC2A9 SNP rs12498742 may have an additional effect beyond serum urate, with an interaction also being reported for systolic blood pressure (GEI P = 9.0 × 10 −4 ). 51We did not observe any significant GEIs in the second cohort that comprised patients taking non-bendroflumethiazide thiazides, but this could be explained by the different effects diuretics have on the renal handling of urate, 52 and/or the much smaller sample size of the second cohort.
We acknowledge some limitations to the study.We only analyzed White British ancestry participants and the UK Biobank recruited participants aged between 37 and 73 years, which may mean our results may not be generalizable to other populations or age ranges.Other limitations include using self-reported hypertension status and medications, which may be impacted by recall bias.However, these were the variables available in the UK Biobank and thus we were limited to their use.Moreover, self-reported variables have been widely used by other studies (more than 40 "hypertension" studies as of December 2022, https:// www.ukbio bank.ac.uk/ enabl e-your-resea rch/ publi cations), thus validating the approach.As we only relied on data collected at baseline, we were unable to quantitatively evaluate the worsening of the biomarker levels due to thiazide administration or to what extent the duration of antihypertensive drug administration affected the biomarker levels; this was, however, outside the scope of the current analysis.We also used urine sodium and potassium levels as biomarkers for plasma levels (i.e., increased urinary excretion may correspond to low plasma concentrations).Although not using the blood levels may have contributed to the observed results (no identified vQTL SNPs) for the sodium and potassium outcomes, we could not investigate the blood levels, as the UK Biobank did not collect blood for these two outcomes.Following Wang et al.'s approach, 19 we used a stringent genomewide significance threshold of 2.5 × 10 −9 .Visual inspection of the vQTL association plots shows that decreasing this threshold to the common 5 × 10 −8 threshold would not result in additional loci.Last, during the GEI analysis, we used outcomes pre-adjusted for age, sex, and the first 10 principal components.Although there is a chance that unaccounted covariates like hypertension severity might have impacted the results, the biological plausibility of the identified loci decrease this possibility.The strengths of this study include using a relatively large population-based study and using a powerful screening tool to improve statistical power to detect the small GEI effect sizes.
In conclusion, we have used a two-stage approach (vQTL analysis followed by direct GEI tests) to increase power to detect GEIs using the UK Biobank and identified two biologically plausible loci (CDKAL1 for blood glucose and SLC2A9 for serum urate).The SLC2A9mediated interaction has been previously reported in the literature, which means our study adds to existing clinical evidence to further support the notion that genetic variants can interact with thiazides to increase the risk of thiazide-associated hyperuricemia.Additional functional and/ or larger observational studies are required to further clarify these pharmacogenetic associations.Understanding how genetic variants alter the effect of environmental exposures, including medications such as thiazides, is an important step toward precision medicine, which is very relevant in the field of cardiovascular medicine given the high prevalence and mortality associated with CVDs.

Figure 1
Figure 1 Flow chart for included participants.Bold values represent the total number of participants at each stage.BFZ, Bendroflumethiazide; HTN, hypertension; PC, principal components.

Table 1 P
values for the independent vQTL SNPs significant (2.5 x 10 −9