Exploration of the association of antidiabetic drugs with urolithiasis: A drug-targeted Mendelian randomization study

Abstract

Objective

Type 2 diabetes mellitus has previously been reported to be potentially associated with urolithiasis. We conducted a Mendelian randomization (MR) study to explore whether there is a causal relationship between genetic susceptibility to common antidiabetic drugs and urolithiasis risk.

Methods

We used genetic variants from two different sources as instruments to proxy the exposure to antidiabetic drugs for our MR research design. The variants included loci regulating expression traits of the target genes, and genetic variants associated with blood glucose nearby or within antidiabetic drug target genes from genome-wide association studies. We ultimately calculated estimates using inverse-variance weighted MR (IVW-MR) and summary-data-based MR methods.

Results

The Bonferroni-corrected IVW results suggested potassium inwardly rectifying channel subfamily J member 11 (KCNJ11)-mediated blood glucose was associated with a lower risk of urolithiasis (odds ratio [OR]: 0.15; 95% confidence interval [CI]: 0.06–0.39; p=1.19×10⁻⁴). Similarly, we also observed a higher expression of KCNJ11 was linked to a decreased risk of urolithiasis in the summary-data-based MR analysis (OR: 0.81 per 1 mmol/L decrement in blood glucose; 95% CI: 0.70–0.95; p=0.008). We found suggestive evidence of the positive relationship between insulin receptor expression and urolithiasis (OR: 5.67; 95% CI: 1.01–31.97; p=0.049), which was not supported when using cis-expression quantitative trait locus as an instrument.

Conclusion

This study provided evidence for a potential causal link between KCNJ11-mimicked sulfonylureas and the reduced risk of urolithiasis. Given the limitations of this study, it is essential to investigate further using the latest data from large-scale genetic studies and relevant clinical data to validate our findings from the MR study.

1. Introduction

Urolithiasis is a prevalent disorder that impacts people of all age groups worldwide [1]. The renal parenchyma serves as a deposition site for both organic and inorganic crystalline constituents, initiating the formation of urolithiasis [2]. The most prevalent form of urolithiasis is calcium oxalate stones, though other types include calcium phosphate stones, uric acid stones, and cystine stones, with the distribution of stone types varying by region [3,4]. Approximately 10% of the global population is affected by urinary stone disease. The incidence of urolithiasis is rising worldwide, and the recurrence rate is alarmingly high, with a 10-year recurrence rate reaching up to 50% [5]. According to the National Health and Nutrition Examination Survey in the United States, the self-reported prevalence of urolithiasis increased from 3.2% in 1976–1980 to 8.8% in 2014 [5], causing a great public health problem. The development of urolithiasis is influenced by various factors, including dietary habits, genetic predispositions, and chronic kidney disease.

The etiology of urolithiasis is complex, involving metabolic disorders, structural abnormalities, infections, and dietary factors [6]. A previous study highlighted a significant link between abnormal lipid metabolism and the formation of urinary tract stones [6]. Kirejczyk et al. [6] discovered a strong connection between urinary tract stone formation and metabolic syndrome, including lipid metabolism disorders, hyperglycemia, hypertension, obesity, and insulin resistance. Glucose metabolism disorders have been identified as separate risk factors for the formation of urinary tract stones [7]. Type 2 diabetes mellitus (T2DM) is one of the most common diseases in the elderly population. The study of Lien et al. [8] showed that abnormal fasting blood glucose, impaired glucose tolerance, diabetes, dyslipidemia, and impaired renal function may all influence the incidence of urinary stones, with elevated blood glucose levels potentially increasing the risk of developing urinary stones. Diabetes, particularly T2DM, has been identified as an independent risk factor for uric acid stones. Compared to a 6% incidence rate of urinary stones in non-diabetic individuals, the incidence of uric acid stones in T2DM patients is 29%, representing a statistically significant difference between the two groups (p<0.05) [9]. In the same way, several epidemiological studies also suggest that patients with T2DM are at an increased risk for stone formation [[10], [11], [12]]. Hypoglycemic agents such as sulfonylureas and glucagon-like peptide 1 (GLP-1) receptor agonists can lower blood glucose levels through multiple mechanisms. These drugs play a crucial role in delaying the progression of diabetes and improving outcomes, and are widely used globally [13]. In addition to promoting insulin secretion and improving insulin resistance, these medications may also influence the production and excretion of metabolic byproducts such as uric acid and oxalate [14,15]. Some observational studies suggest that treating and managing insulin resistance may improve or mitigate the occurrence and recurrence of kidney stones [15,16]. However, the findings from observational studies examining the relationship between hypoglycemic agents and kidney stones remain controversial and even contradictory. Moreover, observational studies provide generally unreliable estimates of causal relationships [17].

The Mendelian randomisation (MR) analysis can address the limitations of observational studies in reliably estimating causal relationships. MR is a genetic epidemiological method that identifies the causal relationship between genetic variants, as instrumental variables (IVs), in the exposure and the risk of the outcome [18,19]. MR methods can address concealed confounding effects and reverse causality in the relationship between the exposure and the outcome. Genetic variations, especially those present in genes that encode proteins targeted by medications, eventually mimic mechanism of drug action [20].

Hence, we gathered the published data from extensive genetic studies and performed an MR analysis in this study to test the effects of genetic variations in antidiabetic drug targets on the urolithiasis risk.

2. Materials and methods

2.1. Study design

We performed the present study utilizing a two-sample MR and the summary-data-based MR (SMR) design, based on open-access data from the genome-wide association study (GWAS) and expression quantitative trait locus (eQTL) consortia. Generally, cis-eQTLs are defined as genetic variants regulating proximal gene expression within 1 Mb. Please refer to Supplementary Table 1 for detailed information on all datasets in this study. The present study followed the STROBE-MR statement [17].

We used two kinds of genetic variants associated with antidiabetic drug targets as instruments for these drug exposures in our MR study design. The genetic variants used were categorized into two sources: synonymous variations within genes encoding the relevant drug target proteins, and single nucleotide polymorphisms (SNPs) from GWAS summary data, serving as proxies for antidiabetic drug use. There are three core assumptions for the MR analysis. (a) The selected SNPs must be significantly related to the exposure [21]. We used the F-statistic to assess the strength of the instrument-exposure association [22]. The expression for F is given by R²(n–k–1)/[k(1–R²)]. R² represents the cumulative explained variance of the selected SNPs on circulating glucose levels; k denotes the number of selected SNPs; and n represents the sample size. If the value of the F-statistic is greater than 10, the result suggests no evidence of weak instrument bias. (b) The included genetic variants must be associated with the risk of the outcome solely through the exposure, not via confounding factors [23]. MR-Egger regression was employed to identify any horizontal pleiotropy between the genetic variant and the outcome. (c) The SNPs should be independent of confounders. A flowchart of the study design is depicted in Fig. 1.

Figure 1.

A flowchart illustrating the study design process of Mendelian randomization. GLP-1, glucagon-like peptide 1; GLP1R, glucagon-like peptide 1 receptor; INSR, insulin receptor; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; PPARG, peroxisome proliferator-activated receptor-gamma; eQTL, expression quantitative trait locus; MAF, minor allele frequency; HEIDI, heterogeneity in dependent instruments; GWAS, genome-wide association study; SNP, single nucleotide polymorphism.

2.2. Selection of genetic instruments

We first identified seven primary classes of glucose-lowering drugs, including metformin, dipeptidyl peptidase-4 inhibitors, sodium-glucose co-transporter 2 inhibitors, insulin/insulin analogues, GLP-1 analogues, sulfonylureas, and thiazolidinediones (TZDs) [24]. Additionally, we retrieved information on pharmacologically active protein targets and relevant encoding genes from the DrugBank and ChEMBL databases [25,26]. Due to the uncertainty of potential molecular mechanisms underlying physiological effects, metformin was excluded from further analysis. Due to the insufficient number of SNPs extracted from the exposure data, which precluded further analysis, dipeptidyl peptidase-4 inhibitors and sodium-glucose co-transporter 2 inhibitors were also excluded from the study. At the same time, we identified nearly 14 million SNPs for blood glucose from the GWAS summary statistics, comprising over 300 000 individuals. Additional information regarding the dataset can be found in the study with PMID 24816252 [27]. To extract more IVs, we selected genetic variants significantly linked to the glucose level at genome-wide significance (p<1×10⁻⁵) as the proxies for the four glucose-lowering drug classes. We also set the following filtering criteria to ensure the reliability of the tool: eligible SNPs (minor allele frequency >1%) were required within ±100 kb windows of the target gene region of each drug class and in a state of linkage disequilibrium (r²<0.30).

To decrease potential pleiotropic effects, we thoroughly searched for potentially relevant traits (secondary phenotypes) associated with each SNP using the LDtrait tool retrieved from the GWAS Catalog (https://ldlink.nih.gov/?tab=home) until March 2024 [28,29] to evaluate whether IVs were related to common urolithiasis confounding factors. A p-value less than 1×10⁻⁵ was considered indicative of no confounding factors. Ultimately, we determined that the uric acid level was the sole confounding factor associated with the IVs. After accounting for the effects of confounding SNPs, the reanalysis results indicated that the estimates remained statistically significant.

For the outcome dataset, SNPs related to urolithiasis outcomes were obtained from GWAS. The corresponding GWAS ID is ebi-a-GCST90018935, comprising 6223 cases and 482 123 controls [30]. The diagnosis of urolithiasis in the inpatient registry was defined by the presence of stone(s) within the urinary tract. The following ICD-10 codes were used to identify relevant cases across different analytical models: N20, N21 (kidney and lower urinary tract stones) for the CNV U-shape, deletion-only, duplication-only, and mirror models; and N20 (kidney and ureter stones) and N21.0 (bladder stones) for the standard and gene-based burden analyses. All participants in both the exposure and outcome were of European ancestry. Supplementary Table 1 and Supplementary Fig. 1 provided details on the GWAS information from the included studies.

The GLP-1 agonists, the insulin analogues, and the peroxisome proliferator-activated receptor-gamma (PPARG) stimulants in this study are all antidiabetic drugs approved by the U.S. Food and Drug Administration. A previous study proved that these four drugs can effectively reduce blood glucose levels through different mechanisms [24]. Therefore, we selected blood glucose as the biomarker in the next analysis. To confirm the causal relationships derived from GWAS-based genetic instruments, we extracted cis-eQTLs for the four drug-target genes as proxies for antidiabetic drug use. We acquired all the eQTL summary-level data for the four genes from the eQTLGen Consortium (https://www.eqtlgen.org/). The details of the included eQTLs are provided in Supplementary Table 1. We identified cis-eQTL variants linked to the expression of GLP1R, INSR, KCNJ11, or PPARG (p<1×10⁻⁵) in blood, with minor allele frequency >0.01 and within a 1 MB window near the genes [31].

2.3. Statistical analysis

The two-sample MR method effectively avoided sample overlap. When using the genetic variations associated with blood glucose levels as IVs, the two-sample MR analysis was conducted to assess whether there was a causal relationship between genetic susceptibility to blood glucose and the urolithiasis risk. The assessments were conducted through standard MR methods, namely the inverse-variance weighted (IVW), MR-Egger regression, weighted median, and weighted mode [23,[32], [33], [34]]. IVW was adopted as the primary MR analytical method to infer causality since it required all selected SNPs to be valid IVs [32]. In addition, the SMR method was employed when eQTLs served as instruments, since it could explore the relationship between gene expression and outcomes by using aggregate data from GWAS and eQTL studies [35]. Therefore, we performed the analysis via the SMR software (version 1.3.1, https://cnsgenomics.com/software/smr/#Overview). The significance level for each comparison was calculated with the Bonferroni correction formula: α=0.012 (0.05/4) for four drug classes [36]. For the genetically instrumented drug class identified to be associated with urolithiasis risk, we also investigated the colocalization between blood glucose and urolithiasis risk within the target gene region. To assess whether the genetic associations of GLP1R, INSR, KCNJ11, and PPARG with urolithiasis risk were driven by shared causal variants, we performed a Bayesian colocalization analysis using the COLOC R package (version 5.1.0) [31]. This method evaluates the posterior probability that the same variant explains both the exposure (gene expression or protein levels) and outcome (stone disease) associations.

All statistical analyses were conducted using R software (version 4.2.1, https://www.r-project.org) with the following key packages: TwoSampleMR (version 0.5.7), LDlinkR (version 2.0.0), MR-PRESSO (version 1.0), and COLOC (version 5.2.0). Unless otherwise noted, we set the statistical significance at a two-sided p-value of less than 0.05.

2.4. Sensitivity analysis

To verify the reliability of the IVs used in the study, we conducted a positive-control analysis utilizing GWAS-derived genetic instruments for blood glucose to assess their causal associations with three established cardiometabolic phenotypes: hypertension, body mass index, and hip circumference (HIP), since in previous clinical studies, glucose-lowering drugs were associated with the above-mentioned outcome traits. Given the established role of antidiabetic drugs in lowering blood glucose, we examined the association between the tool variables from eQTL and glucose levels. Please refer to Supplementary Table 1 for details of all datasets used in the positive control analysis.

For the SMR analysis, we searched for data of target genes related to antidiabetic drugs for sensitivity analysis of genes near the top SNP. We used the heterogeneity in dependent instruments (HEIDI) test to determine whether the causality between the four drug-target gene expression and urolithiasis risk was due to linkage disequilibrium. The HEIDI test was performed in the SMR software, a tool developed to test for the pleiotropic association between gene expression and complex traits using summary-level data from GWAS and eQTL [37]. The p-value of the HEIDI test that was less than 0.05 was regarded as indicating that the observed association resulted from linkage disequilibrium [38].

For the IVW-MR method, the MR Pleiotropy RESidual Sum and Outlier (MR-PRESSO) method could automatically detect outliers in the IVW linear regression and remove them to correct MR estimation [39]. We performed MR-PRESSO to identify outliers, and the MR analysis was repeated after eliminating heterogeneous SNPs. The MR-PRESSO Global test with p<0.05 indicated the evidence of horizontal pleiotropy. We also used the Cochran Q test to assess heterogeneity [40]. As a test for directional horizontal pleiotropy, a non-significant MR-Egger intercept test (p>0.05) suggested the absence of such pleiotropic bias among the IVs [41].

2.5. Data availability and ethics statement

The analyses presented herein are based on aggregated, de-identified GWAS summary statistics which are publicly available under open licenses (e.g., CC BY). All source datasets (GWAS Catalog, UK Biobank, Meta-Analyses of Glucose and Insulin-related traits Consortium [MAGIC], DIAbetes Genetics Replication And Meta-analysis Consortium [DIAGRAM], Within-Family GWAS Consortium) permit secondary analysis for research purposes without requiring separate ethical approval or specific permission, provided their terms of use are followed. This study complies fully with these conditions.

3. Results

3.1. Primary results

In the primary IVW-MR analysis, we identified 4, 2, 4, and 1 qualified IVs near the GLP1R, INSR, KCNJ11, and PPARG genes (p<1×10⁻⁵) from a summary dataset with GWAS ID ukb-d-30740_irnt on blood glucose levels from the UK Biobank (Fig. 2). The F-statistics of all IVs exceeded 10, and this was deemed indicative of no bias caused by the weak SNPs (Supplementary Table 2). The IVW-MR analysis revealed a significant association between INSR-regulated glycemic traits (per 1 mmol/L increment in genetically predicted blood glucose) and urolithiasis susceptibility (OR: 5.67; 95% CI: 1.01–31.97; p=0.049). On the other hand, the IVW-MR results provided stronger evidence for the relationship between KCNJ11-mediated blood glucose levels (equivalent to per 1 mmol/L decrement in blood glucose) and the risk of urolithiasis (OR: 0.15; 95% CI: 0.06–0.39; p=1.19×10⁻⁴). Similarly, the estimates for sulfonylureas in the weighted median method (OR: 0.21; 95% CI: 0.06–0.70; p=0.012) and MR-Egger regression (OR: 0.11; 95% CI: 0.002–5.69; p=0.4) also validated a decreased risk of urolithiasis. Nevertheless, the standard error of MR-Egger regression was larger than that of the traditional IVW-MR method since its 95% CI is relatively wider. However, the present IVW-MR analysis provided no evidence that GLP1R-mediated blood glucose and PPARG-mediated blood glucose were associated with urolithiasis outcomes. All p-values of the IVW Cochran's Q test and MR-Egger intercept test exceeded 0.05, suggesting an absence of heterogeneity and pleiotropy among the SNPs (Table 1).

Figure 2.

Associations between the blood glucose levels mediated by the genes GLP1R, INSR, KCNJ11, or PPARG and the urolithiasis risk. OR, odds ratio; SNP, single nucleotide polymorphism; CI, confidence interval; GLP-1, glucagon-like peptide 1; GLP1R, glucagon-like peptide 1 receptor; INSR, insulin receptor; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; PPARG, peroxisome proliferator-activated receptor gamma.

Table 1.

Associations between the blood glucose levels mediated by the genes GLP1R, INSR, KCNJ11, and PPARG and the urolithiasis risk.

Exposure gene	Outcome	SNP, n	Beta	SE	p-Value for IVW-MR association	p-Value for Cochran Q test	p-Value for MR-Egger intercept	p-Value for the MR-PRESSO Global test
GLP1R	Urolithiasis	4	−0.576	0.685	0.4	0.5	0.5	0.6
INSR	Urolithiasis	2	1.735	0.883	0.049	0.9	NA	NA
KCNJ11	Urolithiasis	4	−1.925	0.500	0.0001	0.9	0.6	0.6
PPARG	Urolithiasis	1	−0.874	1.387	0.5	NA	NA	NA

SNP, single nucleotide polymorphism; MR-PRESSO, Mendelian Randomization Pleiotropy RESidual Sum and Outlier; NA, not applicable; GLP1R, glucagon-like peptide 1 receptor; INSR, insulin receptor; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; PPARG, peroxisome proliferator-activated receptor gamma; SE, standard error.

3.2. SMR effects of instrumental variables for antidiabetic drugs on urolithiasis risk

The results from the SMR analysis are shown in Fig. 3. In the section of SMR analysis, we identified 227, 156, 448, and 1488 cis-eQTLs from the eQTLGen Consortium, corresponding to the drug target genes GLP1R, INSR, KCNJ11, and PPARG, respectively. We selected the most significant cis-eQTL SNP as the genetic variation for each drug target gene. No weak IVs were found among the identified SNPs after the harmonization process (Supplementary Table 3). Eventually, the genetic variation in sulfonylurea targets was found to meet the 0.05 screening threshold through the SMR analysis (OR: 0.81 per 1 mmol/L decrement in blood glucose; 95% CI: 0.70–0.95; p=0.008). The association was consistent in the IVW-MR analysis of KCNJ11, further confirming the protective effect of sulfonylureas on urinary stones. Furthermore, we established meaningful connections between medication exposure and blood glucose levels using genetic tools based on eQTL (Supplementary Table 4). The results presented additional confirmation and confidence in the effectiveness of the chosen genetic instruments in our MR study. The HEIDI test indicated that all detected associations, with the sole exception of the link between PPARG expression and urolithiasis (p=0.043), were not influenced by linkage disequilibrium (p>0.01) (Supplementary Table 3).

Figure 3.

SMR association between expression of the genes GLP1R, INSR, KCNJ11, or PPARG and the risk of urolithiasis. OR, odds ratio; SNP, single nucleotide polymorphism; CI, confidence interval; GLP-1, glucagon-like peptide 1; GLP1R, glucagon-like peptide 1 receptor; INSR, insulin receptor; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; PPARG, peroxisome proliferator-activated receptor gamma; SMR, summary-data-based Mendelian randomization.

3.3. Positive control investigation

The positive control analyses reproduced the known associations of each drug's genetic proxies with T2DM and related diseases, as intended (Fig. 4). The present MR study showed that increased expression of the GLP1R, INSR, and KCNJ11 genes in blood exhibited statistically significant negative associations with T2DM risk. For PPARG, the association was non-significant (p>0.05) but showed a directionally consistent negative trend. Collectively, these patterns indicated a protective effect of gene-mediated glucose levels against T2DM (Fig. 4A). Genetic variants in the target genes of sulfonylureas and GLP-1 analogues were indicative of increased insulin secretion, whereas those for insulin/insulin analogues and TZDs were linked to reduced insulin resistance, which is consistent with their known pharmacological mechanisms (Fig. 4B–C). Further studies demonstrated that INSR-mediated blood glucose levels were associated with a decreased risk of hypertension, whereas sulfonylurea-targeted genetic variants were associated with an increased risk of hypertension (Fig. 4D). Furthermore, the estimates for insulin analogues and sulfonylureas suggested a decrement in both body mass index and HIP (Fig. 4E–F). Genetic variations in the targets of TZDs were associated only with decreased HIP. The results of the positive control analysis demonstrated broad alignment with previous clinical experience [8,16].

Figure 4.

Pharmacogenetic associations between glucose-lowering drug-targeted genetic variants and T2DM-related traits in positive control analyses. (A) With T2DM; (B) With insulin secretion; (C) With insulin resistance; (D) With hypertension; (E) With BMI; (F) With HIP. OR, odds ratio; SNP, single nucleotide polymorphism; CI, confidence interval; GLP-1, glucagon-like peptide 1; GLP1R, glucagon-like peptide 1 receptor; INSR, insulin receptor; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; PPARG, peroxisome proliferator-activated receptor gamma; BMI, body mass index; HIP, hip circumference; T2DM, type 2 diabetes mellitus.

3.4. The Bayesian colocalization analysis for antidiabetic drugs

We conducted the Bayesian colocalization analysis using the COLOC R package to further explore whether the association of identified drug proxy genes with urolithiasis was driven by loci within the corresponding genomic regions. Four antidiabetic drugs corresponded to the target-encoding genes (±1000 base pairs of GLP1R for GLP-1 analogues, INSR for insulin/insulin analogues, KCNJ11 for sulfonylureas, and PPARG for TZDs) (Fig. 5). From the eQTL results of the colocalization analysis (Supplementary Table 5), no solid evidence was found to confirm the colocalization between blood glucose and urolithiasis within the PPARG gene region. We observed suggestive evidence for colocalization from the results of sulfonylureas (the combined prevalence of heterozygous pathogenic/likely pathogenic variants and heterozygous variants of uncertain significance was 41.9%; SNP=rs214927) and insulin/insulin analogues. There were supporting data showing that GLP-1 analogues and urinary calculi were significantly associated with SNPs within the region, although they did not share the same causal variants.

Figure 5.

Genetic associations between blood glucose regulation and urolithiasis susceptibility (based on EBI GCST90018935 GWAS data). (A) GLP1R locus (±1000 bp); (B) INSR locus (±1000 bp); (C) KCNJ11 locus (±1000 bp); (D) PPARG locus (±1000 bp). GLP1R, glucagon-like peptide 1 receptor; INSR, insulin receptor; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; PPARG, peroxisome proliferator-activated receptor gamma. GWAS, genome-wide association study; eQTL, expression quantitative trait locus; chr, chromosome.

4. Discussion

The current MR study investigated the causal effect of antidiabetic agents’ targets on urolithiasis risk through comprehensive genetic datasets, including blood glucose levels (with over 300 000 participants) and urolithiasis (6223 stone disease cases and 482 123 controls). We ultimately found significant evidence for the association of KCNJ11 expression and KCNJ11-mediated glucose levels with a lower risk of urolithiasis (p=0.008). It indicated that KCNJ11-mediated blood glucose levels may reduce the incidence of urolithiasis by approximately 20%; therefore, an increase in KCNJ11 gene expression could significantly decrease stone formation. The current MR study preliminarily demonstrated the role of sulfonylureas, modelled by variants of the KCNJ11 gene, in the prevention of kidney stones. This finding further consolidates and expands upon previous research results [24].

There is plenty of epidemiological evidence to support a relationship between T2DM and nephrolithiasis [42,43]. Renal calculi occur in at least one in 10 patients with T2DM. Similarly, a retrospective cross-sectional study demonstrated that T2DM severity was a significant risk factor for kidney stone disease, as determined by glycemic control [44]. Poor blood glucose control can lead to a decreased renal glomerular filtration rate, increased overabsorption of urine by the renal tubules, resulting in increased urine calcium and reduced phosphate content, and eventually the formation of kidney stones [9,45]. Although the adjustment of lifestyle and dietary habits is the main means of preventing urinary tract stones [46], individuals with diabetes may also need to coordinate medication use.

Therefore, the medication for urolithiasis in patients with diabetes should balance the therapeutic effectiveness and associated risks of both diseases. In the current clinical scenario, few appropriate medications have been developed for the treatment or prevention of urolithiasis, which, to some extent, reflects the value of our MR study.

It is more cost-effective to rediscover and use old drugs from a completely new perspective than to develop new drugs. In the early 21st century, due to the emergence of other new classes of glucose-lowering drugs, the clinical utilization of sulfonylureas has declined [47]. Sulfonylureas are commonly prescribed for managing T2DM [48]. However, in recent years, with the mitigation of adverse reactions such as hypoglycemia susceptibility, the new generation of sulfonylureas is expected to be widely used in clinical therapy [47,49].

The KCNJ11 gene encodes for the Kir6.2 subunit of the ATP-dependent potassium channel (KATP), which is an important gene that regulates insulin secretion of pancreatic β-cells [50]. Sulfonylureas can bind to the channel's sulfonylurea receptor (SUR), causing it to close and then promote insulin secretion. It is worth noting that Ho et al. [51] and Wang [52] reported the presence of KATP channel activity in the kidneys, while Chutkow et al. [53] demonstrated the expression of the SUR2 gene and protein throughout the mouse kidney. Furthermore, another pharmacological study confirmed that, in addition to being present in β-cells, SURs are predominantly expressed in the kidneys, particularly in renal tubular cells [48]. The reabsorption of calcium by renal tubules is a crucial mechanism for maintaining calcium homeostasis in the body [54]. In the proximal convoluted tubule, most filtered calcium is actively reabsorbed, a process that relies on the sodium/calcium exchanger [54]. The sodium/calcium exchanger utilizes the transmembrane sodium gradient to transport calcium ions from the filtrate back into the bloodstream [55]. Crucially, this essential sodium gradient is established and maintained by the Na⁺/K⁺-ATPase (sodium-potassium pump) on the basolateral membrane of tubular cells. [56]. We hypothesize that genetic variation or drug-mediated modulation of the renal KATP/SUR complex may indirectly influence tubular ion transport (potentially affecting intracellular ATP/ADP ratios, membrane potential, or other secondary pathways), thereby creating a permissive environment for altered calcium handling and stone formation. This potential crosstalk provides a biological rationale for our genetic findings linking sulfonylurea targets to kidney stone risk. This may promote the sodium/calcium exchange at the luminal side of the tubule, ultimately reducing the concentration of calcium in the filtrate. This mechanism could potentially decrease the incidence of calcium stone formation in patients using sulfonylureas. Several observational studies indicate that addressing and managing insulin resistance could help reduce or prevent the formation and recurrence of kidney stones [6,15].

In addition, metabolic syndrome, characterized by insulin resistance, is one of the important pathogenic factors for urolithiasis. Sulfonylureas may reduce the incidence of urinary tract stones by improving insulin resistance. This MR study targeting the drug's genetic proxy preliminarily demonstrated the role of sulfonylureas, modelled by variants of the KCNJ11 gene, in the prevention of kidney stones. This finding effectively consolidated and extended the results of previous studies [24,42].

Our MR study used variants on the KCNJ11 gene to proxy sulfonylurea medications, and the results showed that genetically mimicking the sulfonylurea drug class was associated with a lower risk of urolithiasis. Similarly, a review that integrated multiple large-scale cohort studies concluded that modern sulfonylurea drugs are effective hypoglycemic agents [57]. In addition to their efficacy in achieving blood glucose control, they have also been proven to be renally safe and contribute to preventing diseases of the urinary system [57,58]. However, causal inference is not permitted owing to the retrospective characteristics of observational studies. The MR findings in the present study provided more reliable evidence for causal inference and addressed the limitations of observational studies, such as the relatively small sample size and reverse causality.

Our MR approach has the following strengths: the primary strength of this study is the application of genetic variants to proxy different drug classes, which can effectively eliminate potential confounding factors and reverse causation. Additionally, we employed two different genetic instruments to proxy the drugs under study, allowing for mutual validation of effect estimates. Thirdly, we conducted multiple sensitivity analyses to test the validity of the genetic instruments and the assumptions of the MR study. MR-PRESSO and MR-Egger regression intercept term tests and HEIDI tests were used to examine and resolve horizontal pleiotropy.

Concurrently, we acknowledge the limitations of this study. Firstly, we found no available qualified cis-eQTLs of the studied genes in blood from other databases for the SMR analysis. Therefore, the cis-eQTL database we selected had a relatively limited data source, which may affect the reliability of the results. Secondly, all the datasets included in this study are from populations of European ancestry, so the findings may not be generalizable to other ethnic ancestries. Different populations may harbor different genetic variants due to variations in genetic drift, selection pressures, and migration patterns over time. Certain alleles that confer risk or protection might be rare or even absent in non-European populations, leading to potential differences in disease susceptibility. Additionally, genetic susceptibility to urolithiasis and diabetes may differ significantly across populations. Certain risk alleles for diabetes might also show differing effects depending on ancestry, highlighting the need for broader inclusion of diverse populations in genetic studies to identify population-specific risk factors. Personalized medicine efforts should account for ethnic differences by using genetic risk models that are trained on diverse populations. This will help improve the accuracy of risk prediction tools for individuals from different ethnic backgrounds. It is necessary to collect data from non-European populations to further validate the findings. Thirdly, in the positive control study, genetic variation in the targets of sulfonylureas and insulin analogues was associated with a higher risk of T2DM, inconsistent with the drugs' mechanism of action. It might require more compelling evidence for the existing association. Furthermore, it remains challenging to translate the genotype-phenotype associations into a specific and feasible solution for treatment. The MR study can only represent a prediction on the on-target effects of antidiabetic agents since we only included the well-characterized protein targets in our analysis.

The Bayesian colocalization analysis suggested no suggestive evidence of colocalization for PPARG. The first possible reason for the lack of colocalization is a distinct genetic architecture for the two traits. The disease is influenced by genetic variants affecting other biological processes that do not impact gene expression [31]. The lack of colocalization suggested that the two traits do not share a common causal variant, implying that gene expression is not directly driving the disease risk in a specific locus. The genetic variants affecting the disease may be independent from those regulating gene expression at that locus, thus undermining direct causal inference from the genetic variants alone. The second possible reason for the lack of colocalization is complex genetic interactions [59].

To translate the MR findings into clinical practice, particularly regarding the potential protective effect of sulfonylureas on the risk of urolithiasis, it is important to consider how these insights could be implemented, as well as the necessary steps for further validation. For patients with a known family history of urolithiasis, previous stone formation, or other risk factors (e.g., obesity, hypercalciuria, or hyperoxaluria), sulfonylureas might be considered as a part of an early intervention strategy to prevent the development of kidney stones.

However, before sulfonylureas can be repurposed for urolithiasis prevention, several important considerations need to be addressed. While the MR analysis can provide suggestive evidence of causality, it is based on genetic associations, which may not directly reflect the full biological mechanism [60]. The underlying molecular mechanisms by which sulfonylureas could prevent urolithiasis need to be better understood. Experimental studies or clinical trials will be necessary to clarify these mechanisms.

Furthermore, the MR findings, while promising, are based on observational genetic data and cannot replace clinical trials. Randomized controlled trials would be required to definitively establish whether sulfonylureas have a protective effect against urolithiasis. Safety and side effects should also be taken into consideration. Although sulfonylureas are generally well tolerated, they come with known side effects such as hypoglycemia, weight gain, and, in some cases, cardiovascular risks. In repurposing these medications for urolithiasis prevention, it is critical to assess whether the benefit-risk ratio is favourable, especially for non-diabetic patients who might not experience the glycemic benefits [61]. Additionally, the long-term safety in preventing kidney stones needs to be examined. If these studies are successful, sulfonylureas could become a part of a broader strategy to manage the kidney stone risk, potentially improving outcomes for a diverse group of patients at high risk for urolithiasis.

In summary, we simultaneously used genetic variants related to KCNJ11 expression or KCNJ11-mediated glucose as instruments to proxy the exposure of sulfonylurea drugs. Both results showed a significant association of sulfonylurea drugs with a lower risk of urolithiasis. This suggests that taking sulfonylureas might have preventive and/or therapeutic effects on T2DM patients with urinary calculi.

5. Conclusion

We brought together several large-scale GWAS and eQTL datasets to investigate the causal relationship between antidiabetic drugs and urolithiasis through a systematic MR analysis. The final results of the MR study revealed associations between antidiabetic drugs and the risk of urolithiasis, which might provide new ideas for developing drugs for the treatment of stone diseases. The underlying mechanisms should be elucidated in further research, and the role of antidiabetic drugs in urolithiasis risk could be evaluated in basic and clinical trials.

Author contributions

Study concept and design: Xidong Wang, Qiancheng Mao.

Data acquisition: Yingying Yang, Ming Liu.

Data analysis: Xidong Wang.

Investigation: Tianqi Wang.

Drafting of the manuscript: Xidong Wang.

Critical revision of the manuscript: Xidong Wang, Qiancheng Mao, Jitao Wu.

Conflicts of interest

The authors declare no conflict of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article.