Association of Antidiabetic Drug Target Genes with Inflammatory Bowel Disease: A Mendelian Randomization Study

Introduction

Inflammatory bowel disease (IBD), mainly incorporating ulcerative colitis (UC) and Crohn’s disease (CD), is a progressive gastrointestinal disease that alternated between relapse and remission. The clinical manifestations include hematochezia, abdominal pain, and diarrhea,^1,2 possibly accompanied by extra-intestinal impairments, for instance, liver, joint and skin. IBD prevailed in 0.3% of Europeans,³ and a growing number of incidence of IBD in individuals over 60 years old, posing a challenge for the management of the disease.⁴

Type 2 diabetes (T2D) is a common metabolic disorder, and mainly occurs in middle-aged and elderly people.⁵ Chronic inflammation has been demonstrated in the pathogenesis of T2D.⁵ Observational studies have shown an increasing prevalence of T2D in chronic inflammatory diseases such as rheumatoid arthritis and psoriasis,⁶ so was IBD.^7,8 Accordingly, how to manage hypoglycemic drugs for patients with IBD is worth exploring and meaningful. Traditional pharmaceutical interventions, including biguanides (such as metformin), sulfonylureas (SUs), thiazolidinediones (TZDs), insulin, alongside innovative antidiabetic agents like dipeptidyl peptidase-4 inhibitors (DPP-IVi), sodium-glucose cotransporter 2 inhibitors (SGLT-2i), and glucagon-like peptide-1 receptor agonists (GLP-1RA), have proven efficacy in managing T2D.⁹ Observational studies have shown the effect of different classes of antidiabetic drugs on IBD, yet the results have been inconsistent. In terms of the possible association between IBD and T2D, clarifying the effect of antidiabetic drugs on IBD is necessary, challenging, and charming.

Randomized controlled trials (RCTs) are the criterion for determining the efficacy of drugs. Nevertheless, undertaking an RCT requires temporal demands, substantial financial requisites, and the augmented necessity for human involvement. Mendelian randomization, a statistical method, employs genetic variants as instrumental variables (IVs) to deduce causal links between exposures and outcomes. This method’s principle is the random assignment of parental genetic variants to offspring, influencing the offspring’s phenotype. In parallel, genetic variation preceding the disease can help prevent reverse causality. Thus, to some extent, it is similar to RCTs. This method is additionally employed to infer the effects of drug target genes on outcomes. The expression quantitative trait loci (eQTLs) encoding drugs can be applied as IVs to perform an MR analysis.¹⁰ Several studies have investigated the relationship between statins and COVID-19 and antihypertensive drugs and psychiatric disorders utilizing this method.^10,11

In this study, we performed an MR analysis to explore the effects of antidiabetic drugs on IBD.

Method

Study Review

Two methods, SMR and TSMR, were utilized to elucidate the effects of antidiabetic drugs on IBD. These methodologies necessarily rely on three core assumptions: i) strong association between genetic variation with exposure (relevance); ii) the absence of connection between genetic variation and confounding factors (independence); iii) the impact of genetic variations on outcomes is exclusively mediated by the exposure (exclusion restriction). The flow chart of the present study was shown in Figure 1.

Figure 1 Flow chart of Study. Abbreviations: eQTLs, expression quantitative trait loci; GWAS, genome-wide association study; FBG, fasting blood glucose; HbA1c, glycated hemoglobin A1c; T2D, type 2 diabetes; IBD, inflammatory bowel disease; UC, ulcerative colitis; CD, Crohn’s disease; SMR, summary-data-based Mendelian randomization, TSMR, two sample Mendelian randomization; HEIDI, heterogeneity in dependent instruments; IVW, inverse-variance weighted; WM, weighted median.

Effect of Antidiabetic Agents on IBD

Data Sources and Instrumental Variables Selection

SMR

We searched target genes of antidiabetic drugs DrugBank (https://go.drugbank.com/) and ChEMBL (https://www.ebi.ac.uk/chembl/); consistent genes in both databases were included in the analysis (Supplementary Table 1).

Given that the mechanisms of metformin were complicated and inconsistent in the two databases, it was not included. The expression quantitative trait loci (eQTLs) of the above drug target genes were obtained from the eQTLGen consortium (https://www.eqtlgen.org/), inclusive of blood samples from 31,684 subjects, of which the cis-eQTLs were selected to act as a proxy for these drugs. In addition, we sought the relevant genes from GTEx Consortium V8 (http://www.gtexportal.org/) if those were not available in the eQTLGen consortium. The cis-eQTLs with p < 5*10-08, minor allele frequency (MAF) > 0.01, and F value [calculate as (beta/SE)²] > 10 were selected as IVs for analysis as previous study.¹¹

The antidiabetic agents were mainly applied to treat T2D for lowering blood glucose and glycated hemoglobin, so we utilized GWAS summary data from the European population for FBG (281,416 cases), HbA1c (215,977 cases), and T2D (80,154/853,816 cases) as the positive control.^12,13 We included the genes with statistical significance in SMR or TSMR. The data for IBD derived from a meta-analysis encompassed patients with IBD (25,042 cases/34,915 controls), UC (12,366 cases/33,609 controls) and CD (12,194 cases/28,072 controls).¹⁴ Data sources were detailed in Table 1. The GWAS data were adjusted for sex and/or age and/or body mass index (BMI). Furthermore, these data underwent rigorous quality control to ensure data reliability.

Table 1 Data Sources for Phenotypes

TSMR

GWAS summary data of T2D, FBG, or HbA1c, whose SMR result was statistically significant with cis-eQTLs of target genes, were used as the proxy. SNPs with P < 5*10-08 and weakly linkage disequilibrium (r² < 0.1) with the target gene (located within ± 100kb region and MAF > 0.01) were selected as IVs.¹⁵ The GWAS data for IBD are the same as above.

Statistical Analysis

For SMR, IVs were cis-eQTLs; SMR software (20,220,322) was used to coordinate and analyze them. For TSMR, the primary method was IVW; when there was only an IV, the Wald ratio method was used. These analyses were conducted using RStudio (2023.06.0+421).

Sensitivity Analysis

For SMR, pleiotropy is due to an SNP associated with multiple genes, which was checked for the HEIDI test. The P-value of HEIDI < 0.01 indicated pleiotropy.¹⁰ Genes with pleiotropy were excluded in the further analysis as in previous studies.¹⁰

For TSMR, MR-Egger and weighted median (WM) were employed for sensitivity analysis if there were sufficient IVs. Heterogeneity was detected by Cochran’s Q test; the leave-one-out method tested whether the result was responsible for a single SNP.

We conducted co-localization analyses taking advantage of the cis-eQTL of the gene and GWAS (located within ± 1MB window) or glucose GWAS and IBD GWAS within the target gene region (located within ± 500kb window), which provided the possibility of two phenotypes sharing genetic variation.

Multiple testing was performed with Bonferroni corrections to adjust the threshold of significance. For SMR, P value <0.0028 (6 genes and 3 phenotypes) was of significance, while 0.0028 < P value < 0.05 meant evidence for the suggestion. For TSMR, P value < 0.008 (2 drugs and 3 phenotypes) stood for significance, while 0.008 < P value < 0.05 was on behalf of suggestion.

Effect of Blood Glucose on IBD

IVs Selection

Three criteria met as follows: ① P < 5*10E-8; ②r² < 0.001 and kb = 10,000 for removing linkage disequilibrium; ③ the value of F >10.

Statistical Analysis

The primary analysis method was IVW, the most efficient one, whose assumption was that all the IVs played a role. Additional methods were MR-Egger and WM. If the results of these three methods were not in the same direction, the p-value was tightened as previous study.¹⁶

Sensitivity Analysis

Cochran’s Q test was used for the heterogeneity; the MR-Egger intercept test determined whether there was horizontal pleiotropy; the MR-PRESSO test detected the outliers. Leave-one-out was conducted to evaluate the robustness of this research and detect outliers. After removing the outlier, we conducted a reanalysis.

Result

Gene Selection of Antidiabetic Drugs

Seven common hypoglycemic target genes were searched in 2 databases (Supplementary Table 1), of which ABCC8 cannot be queried in the eQTLGen consortium since we searched this gene expressed in the colon for data analysis in the GTEx Portal. Six genes were associated with type 2 diabetes, fasting blood glucose, or glycosylated hemoglobin, except GLP1R (Supplementary Table 2).

Relationship Between Cis-eQTL of Antidiabetic Drug Targets and IBD

The suggestive evidence showed that increasing SLC5A2 in the blood may increase the risk of UC (OR per 1-SD = 1.968 95% CI = 1.007–3.848, P = 0.048). It was suggestive evidence that increasing the expression of PPARG in the blood alleviated CD (OR per 1-SD = 0.861, 95% CI = 0.758–0.978, P = 0.021), not UC or IBD. It was the evidence for a recommendation that increasing KCNJ11 in the blood may reduce the risk of CD (OR per 1-SD = 0.728, 95% CI = 0.586–0.903, P = 0.004) or IBD (OR per 1-SD = 0.833, 95% CI = 0.706–0.983, P = 0.031) (Figure 2). DPP4 or INSR in the blood or ABCC8 in the colon did not affect IBD, UC, or CD (Supplementary Table 3).

Figure 2 Relationship between cis-eQTL of antidiabetic drug targets and IBD (Results based on SMR). Abbreviations: UC, ulcerative colitis; CD, Crohn’s disease; IBD, inflammatory bowel disease; OR, odds ratio; CI, confidence interval; eQTL, expression quantitative trait loci; SMR, summary-data-based Mendelian randomization. Notes: *P < 0.05, **P < 0.01.

Association Between T2D, HbA1c, or FBG -Mediated Antidiabetic Drug and IBD

We analyzed the cis-eQTL of different hypoglycemic target genes with T2D, HbA1c, or FBG as positive controls, and the positive results were used as proxy hypoglycemic agents (Supplementary Table 2). Only two target genes (KCNJ11 and PPARG) were found agents in the GWAS of T2D. ABCC8, another gene that codes for sulfonylureas, and KCNJ11 could not be discovered as a proxy in HbA1c GWAS. One SNP proxied for KCNJ11 by T2D was shown as Supplementary Table 4.

The results were consistent with SMR, and increasing expression of KCNJ11 reduced the risk of CD (OR = 0.475, 95% CI = 0.297–0.761, P = 0.002). The suggestive evidence suggested that increasing KCNJ11 expression mediated by T2D lowered the risk of increased IBD (OR = 0.663, 95% CI = 0.460–0.956, P = 0.028) (Figure 3).

Figure 3 Association between T2D, HbA1c, or FBG mediated hypoglycemic agents and IBD (Results based on TSMR). Abbreviations: UC, ulcerative colitis; CD, Crohn’s disease; IBD, inflammatory bowel disease; IBD, inflammatory bowel disease; HbA1c, glycated hemoglobin A1c; FBG, fasting blood glucose; TSMR, two sample Mendelian randomization; OR, odds ratio; CI, confidence interval. Notes: *P < 0.05, **P < 0.01.

Two SNPs were as IVs for T2D-mediated PPARG (Supplementary Table 5). T2D-mediated PPARG had no effect on inflammatory bowel disease (Figure 3). So was HbA1c-mediated PPARG (Supplementary Tables 6 and 7).

Sensitivity Analysis

In the SMR analysis, no pleiotropy between KCNJ11 and CD was detected in the light of results of HEIDI test (Supplementary Table 3). No substantial evidence existed to support the association between the cis-eQTL of KCNJ11 and CD within the gene region (posterior probability = 52.5%) (Supplementary Table 8).

For TSMR, there was only one SNP agent for KCNJ11, which was insufficient to conduct a heterogeneity test. For PPARG, no heterogeneity was detected on the basis of Cochran’s Q test (Supplementary Tables 9 and 10). Insufficient SNP was included to perform other sensitivity analyses. The results of colocalization analysis indicated that there was no substantial evidence of the correlation between the T2D and CD in the KCNJ11 region (posterior probability = 0.229%) (Supplementary Table 11).

Relationship Between T2D and CD

We conducted a Mendelian randomization study to elucidate the relationship between T2D and CD. One hundred and thirty-seven SNPs were proxies for T2D, and the result showed no causality between T2D and CD (Supplementary Tables 12 and 13, Supplementary Figure 1, and 2).

Discussion

Mendelian randomization is a common tool used to deduce causality only under certain assumptions (relevance, independence, and exclusion restriction). This study employed two methods to demonstrate the impact of antidiabetic drugs on individuals with IBD. This study found that reduction of the expression of KCNJ11 in the blood increased the risk of CD according to the results of SMR and TSMR. At the same time, no evidence showed that an association existed between ABCC8 in the colon and IBD, CD, or UC in terms of SMR, suggesting employing SUs (inhibitors of KCNJ11 and ABCC8) may increase the risk of CD. In parallel, the evidence pointed to no causality between T2D and CD, implying any association of KCNJ11 with CD is independent of its association with T2D. Suggestive evidence showed that TZDs (PPARG agonists) could reduce the risk of CD, while SGLT2 lowered the risk of UC according to the result of SMR. Whereas, TSMR did not support the former result, and we did not perform TSMR resulting from lacking a proxy for the latter. In addition, the association between PPARG or SLC5A2 and glycemic traits was contradictory with clinical practice; these results should be explained with prudence accordingly. Moreover, we found that DPP4 inhibitors (DPP4i) or INSR agonists did not affect IBD.

Previous studies have explored the potential relationship between antidiabetic drugs and IBD, but conflicting results existed. Wang et al and Radel et al found that DPP4i did not increase the risk of IBD,^17,18 and this result was consistent with our study. Devin et al reported that DPP4i increased the overall risk of IBD by 75% compared with other hypoglycemic agents, and this risk was significant at 1–2 years and 3–4 years after treatment with the drug.¹⁹ The activation of PPARγ, which is encoded by PPARG, inhibits NF-κB and MAPK signaling pathways to decrease the production of inflammatory factors and inflammatory cell infiltration.²⁰ TZDs, PPARγ (encoded by PPARG) agonists, have shown benefits for the intervention of UC in clinical studies. A double-blind RCT study illustrated that oral rosiglitazone (one of the TZDs) improved the rate of clinical response and clinical remission in patients with mild-to-moderate active UC. However, it did not improve the rate of endoscopic remission and was associated with edema and myalgia.²¹ Another two clinical studies have also shown the potential benefits of oral or enema rosiglitazone in treating active UC.^22,23 These small samples of RCTs confirmed the benefits of rosiglitazone on UC, but the benefit on CD remained unclear. Our study shows that TZDs may benefit CD according to the results of SMR, but not for UC. Of note, the effect of PPARG on glycemic traits was not accordant with clinical practice; therefore, the results should be interpreted cautiously.

Strengths and Limitations

Mendelian randomization can effectively avoid reverse causality and confounding factors. In the present study, we utilized two genetic tools for proxy antidiabetic drugs to estimate the effects of those drugs on IBD; the two classes of results could confirm each other. The IVs we selected were strongly associated with antidiabetic drugs in each method. Sensitivity analyses were also performed to account for the robustness of the results.

This work has several limitations. Firstly, colocalization analysis showed weak evidence of shared genetic variation in the KCNJ11 with CD. This was because colocalization generally provides more conservative results.²⁴ Secondly, the effects predicted by genetic variation are cumulative over a lifetime, and the impact of short-term administration using these drugs on IBD is not clear. Thirdly, our study used data from European subjects and cannot be generalized to other ancestry. Fourthly, no SNP was proxied to ABCC8 in HbA1c GWAS; hence, we did not conduct TSMR. However, the ABCC8 and KCNJ11 are closely located on the same chromosome; T2D-mediated SNPs are the same in the two gene regions.

Conclusion

In conclusion, it was found that genetically proxied sulfonylureas has side effects on CD, whereas other hypoglycemic drugs showed no impact on IBD consistent with the results of SMR and TSMR. This study could offer insights into the clinical medical management of hypoglycemic agents in IBD patients with T2D. However, the impact on public policy remains unclear, because clinical guidelines are unlikely to be changed on the basis of MR evidence alone. Further clinical studies are needed to confirm this work and explore the possible mechanisms.