Mendelian Randomization Analysis of Genetic Proxies of Thiazide Diuretics and the Reduction of Kidney Stone Risk

Key Points Question Are thiazide diuretics associated with reduced risk of kidney stones? Findings In this genetic association study of up to 1 079 657 adults, genetic proxies of thiazide diuretics were associated with a statistically significant 15% lower odds of kidney stones. Meaning These results suggest that genetic proxies of thiazide diuretics estimate long-term drug effects; this finding supports the use of thiazide diuretics for kidney stone prevention.


Introduction
Although kidney stones affect nearly 10% of the population worldwide, therapeutics are limited. 1,2iazide diuretics reduce the excretion of urinary calcium and are recommended by multiple clinical guidelines as a medical intervention for the prevention of calcium kidney stones. 3,46][7] However, the largest trial-the NOSTONE trial (Efficacy of Standard and Low Dose Hydrochlorothiazide Treatment in the Prevention of Recurrent Nephrolithiasis)-did not show a difference in composite symptomatic and radiographic stone recurrence among 416 patients with recurrent calcium-containing stones receiving various hydrochlorothiazide doses compared with placebo. 8,9The study's ability to detect stone events may have been limited due to its short duration of follow-up.[12] However, a more extensive and long-term clinical trial would require significant resources.
Alternative approaches that use existing data sets are needed to understand the potential efficacy of thiazide diuretics, including large-scale data sets linked to genetic data suitable for analyzing drug effects.One such approach is mendelian randomization, which can investigate the association between naturally occurring genetic variation in drug targets and disease risk. 13By virtue of the random allocation of genetic variation at conception, this approach is less subject to unmeasured confounding or reverse causation biases that can occur in observational studies. 14though it does not supersede the evidence of a randomized clinical trial, mendelian randomization provides an additional level of evidence for the preventive potential of thiazide diuretics in kidney stone disease.
In this study, genetic proxies of thiazide diuretics were derived from naturally occurring genetic variation in the thiazide-sensitive sodium chloride cotransporter gene.Kidney stone outcomes were derived from 3 biobanks.The objective was to use mendelian randomization to assess the association of genetic proxies of thiazide diuretics with the risk of kidney stones.

Overview
Mendelian randomization is a method for investigating potential causal relationships between exposures and disease outcomes. 15The first step is to identify genetic variants as instrumental variables that robustly associate with an exposure.7][18] The second step is to assess the association between the instrumental variables and an outcome.This requires several assumptions, including that the genetic variants serving as instrumental variables are only associated with the exposure, have no common cause with the outcome, and only affect the outcome via the exposure. 19is study derived exposures and outcomes from genome-wide association studies (GWAS).
Genetic proxies of thiazide diuretics and negative controls were derived from a GWAS of systolic blood pressure from the International Consortium for Blood Pressure (ICBP).Kidney stone outcomes were derived from GWAS in the Million Veteran Program (MVP), UK Biobank (UKB), and FinnGen study (FinnGen).The mendelian randomization association of genetic proxies of thiazide diuretics with kidney stones was estimated in each of these cohorts individually, and then combined in a random-effects meta-analysis (eFigure in Supplement 1).

This research adhered to the Strengthening the Reporting of Observational Studies in
Epidemiology Using Mendelian Randomization (STROBE-MR) documentation. 51All analyses were conducted using GWAS summary statistics, which previously obtained ethical review board

Instrumental Variable Selection
To proxy thiazide diuretics, we identified genetic variants associated with systolic blood pressure at the SLC12A3 gene and its 48 enhancer and promoter regions.The SLC12A3 gene encodes the thiazide-sensitive sodium-chloride cotransporter located in the distal convoluted tubule of the kidney. 20This protein plays a critical role in sodium excretion and the maintenance of salt homeostasis.Thus, the sodium-chloride cotransporter function affects the regulation of blood pressure. 21Genetic proxies of β-blockers and systolic blood pressure were chosen as negative controls because they have no known effect on urinary calcium excretion or kidney stone risk and were used as negative controls in previous observational studies of kidney stone disease. 22To proxy β-blockers, we identified genetic variants associated with systolic blood pressure at the ADRB1 gene and its 67 enhancer and promoter regions.The ADRB1 gene encodes β1-adrenergic receptor, which is present in heart and kidney tissues and plays a role in cardiac contractility and renin release.To proxy the total effect of systolic blood pressure, we selected genetic variants associated with systolic blood pressure across the genome.
Genetic proxies for thiazides, β-blockers, and systolic blood pressure were selected from a GWAS of systolic blood pressure from the ICBP. 23Briefly, this global collaboration involved a fixedeffects inverse variance-weighted meta-analysis of several large-scale studies to identify common and rare genetic variants associated with blood pressure traits (757 601 total study participants).
Details for the 77 independent studies and study-level genomic controls are provided in supplementary tables published by Evangelou et al. 23 Summary statistics were downloaded from the OpenGWAS application programming interface (API). 24,25We identified genetic variants associated with systolic blood pressure from within drug target genes and their enhancer and promoter regions as determined from the GeneHancer integrated database. 26Significant associations had a Bonferroni level of significance based on the total number of genetic variants tested (ie, α < .05divided by variants tested) (eTable 1 in Supplement 1).
We performed linkage disequilibrium clumping on significant genetic variants to select uncorrelated instrumental variables based on the 1000 Genomes Reference Panel using the European ancestry superpopulation as reference.This approach utilizes the PLINK clumping method, which prunes genetic variants in linkage disequilibrium within a specified 10 000 kb window, ultimately retaining the genetic variant with the lowest P value. 27The main analysis, negative controls, sensitivity analyses, heterogeneity tests, and pleiotropy tests used instrumental variables with a clumping threshold of r 2 = 0.4. 18To enhance the robustness of our findings, we used additional sensitivity analyses of the main analysis using instrumental variables with varying clumping thresholds from r 2 equaling 0.2, 0.1, 0.05, and 0.01, as done in previous drug-target mendelian randomization studies. 28The strength of instrumental variables was assessed using F-statistics, which are related to the variance in the phenotype explained by the genetic variants.
F-statistics were estimated by the formula F = (β 2 / standard error 2 ). 29An F value above 10 was considered high instrument strength. 30

Main Analysis
Kidney stone outcomes were derived from GWAS summary statistics from 3 large biobanks-the MVP, UKB, and FinnGen.Kidney stones were defined using administrative and diagnosis codes.
Kidney stones are a binary outcome based on the presence or absence of such codes and were not transformed.All 3 GWAS were minimally adjusted for age, sex, and 10 principal components of ancestry.These cross-sectional designs do not report a duration of follow-up.Data analysis was performed in May 2023.
The MVP GWAS was created as part of the genome-wide PheWAS project, a collaboration between the US Department of Veterans Affairs and the US Department of Energy 31 (eMethods in Supplement 1).The case definition was urinary calculus as defined by phecode 594, and results were derived from European ancestry individuals defined by HARE (harmonized ancestry and race and ethnicity) (39 955 cases and 400 379 controls). 32Cases had at least 2 occurrences of the phecode on different days and controls had no occurrence of the phecode.The GWAS of urinary calculus was adjusted for age, sex, and first 10 principal components using a mixed model approach in SAIGE (Scalable and Accurate Implementation of Generalized mixed model). 33Results were filtered to remove variants with poor imputation quality (r 2 < 0.3) or that were very rare (minor allele count below 30).
The UKB is a biomedical database and research resource containing genetic, lifestyle, and health information from approximately half a million participants aged between 40 and 69 years in the UK, as described elsewhere. 34We obtained UKB data for individuals with European ancestry on the Pan-UKBB project phenotype trait "Calculus of Kidney Stone or Ureter" defined by International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code N20 (5530 cases, 415 001 controls).Summary statistics were downloaded from the Pan-UKBB project portal. 35e FinnGen includes analyses of the genome and health registry data of approximately half a million Finnish individuals, encompassing low frequency and high impact variants, as described elsewhere. 36We obtained FinnGen data from the IEU OpenGWAS project phenotype trait "Urolithiasis" defined by ICD-10 codes N20, N21, and N23 (5347 cases, 213 445 controls).This reflects data freeze 5 (spring 2020), consisting of 218 792 individuals.Summary statistics were downloaded from the OpenGWAS API. 24,25

Secondary Outcomes
The secondary outcomes were serum laboratory values relevant to the treatment of kidney stones, including calcium, phosphorus, vitamin D, urate, albumin, alkaline phosphatase, potassium, cholesterol, glucose, and hemoglobin A 1c .In clinical practice, these measurements are often used in the evaluation or follow-up of patients who receive thiazide diuretics for kidney stones. 37Thiazide diuretics reduce calcium excretion in the urine, which causes serum calcium levels to rise.They also cause extracellular volume contraction and consequently increase uric acid reabsorption in the proximal tubule of the kidney. 38Thiazides also affect insulin sensitivity and glucose metabolism leading to hyperglycemia.Hypercholesterolemia is affected by thiazide use, but the mechanism is not clearly understood. 39Finally, potassium levels in the serum fall from the diuretic-induced increased sodium delivery to the distal nephron as well as an indirect effect on aldosterone-mediated actions of the sodium-potassium pump in the collecting tubule of the kidney. 40e secondary outcomes were downloaded from the OpenGWAS API. 24,25The calcium, phosphorus, vitamin D, urate, albumin, alkaline phosphatase, potassium, cholesterol, glucose, and hemoglobin A 1c outcomes were derived from the Neale lab analysis of UKB round 2 GWAS.The GWAS was adjusted for age, sex, age squared, age × sex, age × sex squared, and the first 20 principal components, and outcomes were inverse rank-normal transformed.The potassium outcome was not available from that resource, so it was obtained from the BioBank Japan. 41,42The GWAS was adjusted for age, sex, and the first 10 principal components, and outcomes were also inverse rank-normal transformed.

Statistical Analysis Data Harmonization and Mendelian Randomization Analyses
The analyses were performed using the TwoSampleMR package in R. The associations of instrumental variables were extracted from the outcome GWAS using the outcome_data() function.
For instrumental variables not present in the outcome data, a proxy in high linkage disequilibrium (r 2 > 0.8) would be substituted based on the 1000 genomes reference panel European population.
Harmonization of genetic variants in the exposure and outcome data sets was performed using the harmonize_data() function.Ambiguous variants, where alleles do not correspond for the same genetic variant, were removed.Palindromic variants, where the alleles on the forward strand are equivalent to the reverse strand, were also removed unless the minor allele frequencies allowed us to infer which alleles were on the forward strand.The mendelian randomization inverse variance weighted effect estimate of drug proxies on outcomes was performed using the mr() function.The inverse-variance weighted effect estimate of drug proxies on kidney stones was calculated separately for MVP, UKB, and FinnGen, and then combined in a random-effects meta-analysis. 43The effect of drug proxies on all outcomes was inverted to reflect the negative relationship between blood pressure medications and blood pressure.An association was considered significant for 2-sided P < .05.

Sensitivity Analyses
Sensitivity analyses were performed using weighted median, weighted mode, and multiplicative random-effects inverse-variance weighted estimates at multiple clumping thresholds.The weighted median analysis estimates a causal effect while allowing for up to half of the single nucleotide polymorphisms to be invalid instrumental variables. 44The weighted mode determines causal effects assuming that the plurality of single nucleotide polymorphisms are valid instrumental variables, and has a low likelihood of inflating type 1 error. 45Lastly, examining the inverse-variance weighted at multiple clumping thresholds (with r 2 equaling 0.4, 0.2, 0.1, 0.05, and 0.01) allows for maximization of power at higher thresholds and maximizing validity at lower thresholds. 28The same sensitivity analyses were done for main outcome and the negative controls.

Heterogeneity and Pleiotropy Tests
The presence of heterogeneity or horizontal pleiotropy indicates potential violations of modeling assumptions, or that certain genetic variants are exerting a direct effect on the outcome not through the exposure. 46Heterogeneity was tested with Cochran Q and MR (mendelian randomization)-Egger Q statistics, which are expected to conform to a χ 2 distribution with degrees of freedom equal to the number of genetic variants minus 1. 47 Horizontal pleiotropy was tested with the MR-Egger intercept test, which suggests directional pleiotropy when the intercept of the MR-MR-Egger model is significantly different from zero, and the MR-PRESSO global test. 48,49If heterogeneity or pleiotropy were detected, the MR-PRESSO outlier test would be used to repeat the analyses after outlier removal. 50The same heterogeneity and pleiotropy tests were done for the main outcome and the negative controls.

Computing System
Analysis was performed using R version 3.

Study Population
The study included up to

Association of Genetic Proxies of Thiazide Diuretics With Kidney Stone Risk
Genetic proxies of thiazide diuretics, β-blockers, and systolic blood pressure were identified based on their association with systolic blood pressure in the ICBP (eTables 2-4 in Supplement 1).All instrumental variables demonstrated strong validity (F-statistic greater than 10) (eTable 5 in Supplement 1).The mendelian randomization effect of genetic proxies of thiazide diuretics on kidney stones was estimated in the MVP, UKB, and FinnGen individually (eTables 6-9 in Supplement 1) and then combined in a random-effects meta-analysis (eTable 10 in Supplement 1).

Discussion
For over 3 decades, thiazide drugs have been the standard of care for the prevention of kidney stone recurrence. 3,4,52This practice was built upon observational data sets and multiple small clinical trials.
However, the largest randomized clinical trial failed to find a protective effect of thiazides on kidney stone recurrence. 9Our mendelian randomization study provides further evidence that the action of thiazides may be sufficient to prevent kidney stone formation.
The use of mendelian randomization in this context offers several benefits.Genetic proxies may overcome potential confounding or reverse causation bias that may arise in studies of drug effects.
For example, salt intake can increase urinary sodium and blunt the hypocalciuric effect of thiazide diuretics, but dietary habits are difficult to control in clinical studies. 53Mendelian randomization, on the other hand, is based on genetic variation assigned at birth and unaffected by environmental factors such as diet.Additionally, mendelian randomization provides a rapid and cost-effective means of investigating drug targets, whereas a large clinical trial would require immense effort and cost.Genetic proxies can approximate the effect of thiazide drugs over the course of a lifetime, as opposed to the confined duration of clinical trials.
The finding that genetic proxies of thiazide diuretics increase serum calcium while also reducing the risk of kidney stones is important.First, it supports the robustness of our instrumental variables, which appear to mimic the expected change in calcium seen in clinical practice.Prior reports estimate an increase of 0.8 mg/dL of serum calcium among thiazide users. 54Second, it supports the theory that modulation of calcium excretion through thiazide diuretics is a relevant mechanism for the reduced risk of kidney stones.While the exact mechanism is not completely understood, thiazides do appear to affect calcium absorption in the kidney and modulate uptake in bone.

Limitations
It is important to acknowledge the limitations of the methods and data sources.events and treatments in other cohorts. 55This suggests that our outcome corresponds to clinical events rather than asymptomatic imaging findings or inconspicuous stone formation.The MVP cohort is a unique population consisting primarily of older, male veterans who may have different risk factors for urinary calculus than the general population.Additionally, the drug proxies obtained from the ICBP, the largest genetic study of blood pressure traits available, was derived from European ancestry individuals, which limits the generalizability of our results.The consortium also includes some overlap with the UKB, which may introduce bias into the estimate.Lastly, mendelian randomization relies on multiple assumptions, including the validity and independence of instrumental variables.While we demonstrate that the instrumental variables were related to the exposure based on robust F-statistics, we cannot definitively show that they were independent of the exposure-outcome relationship or the outcome itself.However, the results from the sensitivity analyses were consistent with our main analysis.

Conclusions
In this genetic association study, genetic proxies of thiazide diuretics were associated with reduced kidney stone risk.In light of clinical trials that have challenged their efficacy, these findings may support the role of thiazide diuretics in the prevention of kidney stones.
3.3 (R Project in Statistical Computing) in the VanderbiltAdvanced Computing Center for Research and Education (ACCRE) computing environment.The TwoSampleMR package was used for the main analysis, secondary outcomes, sensitivity analyses, and heterogeneity and pleiotropy tests using the packages described in Methods.The ieugwasr package was used for clumping using the packages described in Methods.The MR-PRESSO package was used for MR-PRESSO heterogeneity and pleiotropy tests.The meta package was used for metaanalyses.

Figure 1 .
Figure 1.Association of Genetic Proxies of Thiazide Diuretics, β-Blockers, and Systolic Blood Pressure With Risk of Kidney Stones

Table .
Characteristics of the Study Population Genetic Proxies of Thiazide Diuretics and the Reduction of Kidney Stone Risk P for MR-PRESSO global test < .001).However, no outliers were detected in the MR-PRESSO outlier test (eTable 12 in Supplement 1).The association of genetic proxies of thiazide diuretics was tested against laboratory values related to the treatment of kidney stones (Figure2).Genetic proxies of thiazide diuretics were JAMA Network Open.2023;6(11):e2343290. doi:10.1001/jamanetworkopen.2023.43290(Reprinted) November 14, 2023 6/12 Downloaded from jamanetwork.comby guest on 11/17/2023 intercept = .81, The kidney stone phenotype was based on diagnosis codes that were not validated in the cohorts studied.However, analogous diagnosis codes have demonstrated a high positive predictive value for kidney stone Figure 2. Association of Genetic Proxies of Thiazide Diuretics With Serum Laboratory Values Criteria for Instrumental Variables Selected From the International Consortium for Blood Pressure eTable 2. Instrumental Variables for Thiazide Diuretics From the International Consortium for Blood Pressure eTable 3. Instrumental Variables for Beta Blockers From the International Consortium for Blood Pressure eTable 4. Instrumental Variables for Systolic Blood Pressure From the International Consortium for Blood Pressure eTable 5. F-Statistics for Instrumental Variables eTable 6. Harmonization of Genetic Proxies of Thiazide Diuretics With Kidney Stone Risk in the Million Veteran Program, UK Biobank, and FinnGen Study eTable 7. Harmonization of Genetic Proxies of Beta Blockers With Kidney Stone Risk in the Million Veteran Program, UK Biobank, and FinnGen Study eTable 8. Harmonization of Genetic Proxies of Systolic Blood Pressure With Kidney Stone Risk in the Million Veteran Program, UK Biobank, and FinnGen Study eTable 9.The Inverse Variance-Weighted Effect of Genetic Proxies of Thiazide Diuretics, Beta Blockers, and Systolic Blood Pressure on Kidney Stones in the Million Veteran Program, UK Biobank, and FinnGen Study eTable 10.The Combined Random-Effects Model Meta-Analysis of Kidney Stone Risk in the Million Veteran Program, UK Biobank, and FinnGen Study eTable 11.Heterogeneity and Pleiotropy Testing for the Main Analysis and Negative Controls (Cochran's Q, MR-Egger Q, MR-Egger Intercept, and MR-PRESSO Tests) eTable 12. Sensitivity Analyses for the Main Analysis and Negative Controls (Weighted Median, Weighted Mode, and Multiplicative Random-Effects Inverse Variance Weighted Effect Estimates at Multiple Clumping Thresholds) eTable 13.The Inverse Variance-Weighted Effect of Genetic Proxies of Thiazide Diuretics on Serum Laboratory Values eMethods.Genome-Wide PheWAS Methods From Million Veteran Program eReferences.