Introduction

Type 2 diabetes mellitus (T2DM) is a chronic metabolic disorder characterized by insulin resistance and impaired pancreatic β-cell function, leading to persistent hyperglycemia1,2. In recent years, the prevalence of T2DM has risen sharply. In 2021, an estimated 536.6 million people worldwide were living with T2DM, a figure projected to increase to 783.2 million by 20453. T2DM not only imposes significant physical and psychological burdens on individuals but also contributes to a rapidly growing global economic burden4. According to the American Diabetes Association, T2DM patients face a high risk of chronic complications, including neuropathy, retinopathy, nephropathy, and cardiovascular diseases, resulting in an estimated economic burden of $327 billion in 20175.

T2DM is a lifelong condition with no known cure6. Current treatment strategies include lifestyle interventions, dietary modifications, physical activity, oral hypoglycemic agents, and insulin therapy. While metformin is widely recommended as the first-line pharmacological treatment, it is also associated with adverse effects, including gastrointestinal discomfort, lactic acidosis, and vitamin B12 deficiency7,8. As insulin secretion progressively declines, most individuals with T2DM will eventually require insulin therapy9. However, long-term insulin use is associated with an increased risk of hypoglycemia, cancer, and mortality10. Given these challenges, there is an urgent need for novel pharmacotherapies for T2DM.

Although randomized clinical trials (RCTs) are the gold standard for evaluating therapeutic strategies, they are often time-consuming, expensive, and logistically challenging11,12. Integrating genetic evidence into drug development is emerging as an effective approach to accelerate the identification of therapeutic targets. Previous studies have shown that drugs with genetic support are twice as likely to succeed in clinical trials compared to those without13. Recently, Mendelian randomization (MR), which leverages genetic variants as instrumental variables to infer causal relationships between exposures and outcomes, has been widely used to prioritize candidate drug targets.

In this study, we aimed to identify circulating blood proteins as potential therapeutic targets for T2DM using Mendelian randomization (MR). We integrated cis-expression quantitative trait loci (cis-eQTL) data for circulating proteins from the eQTLGen Consortium (31,684 individuals) with T2DM summary statistics from the Integrative Epidemiology Unit (IEU) Open Genome-Wide Association Studies (GWAS) Project (61,714 cases, 593,952 controls). Identifying promising drug targets can help streamline drug development and improve its efficiency.

Materials and methods

Study design

This study expanded the application of MR in drug development from individual target to the whole genome. First, we employed blood druggable gene expression profiles as the exposure and T2DM as the outcome in an MR framework. Next, we assessed whether the candidate causal genes identified through MR were colocalized with T2DM. To further evaluate the therapeutic potential of these genes, we conducted phenome-wide association studies (PheWAS), drug target prediction, and molecular docking. Finally, we quantified the expression levels of these target proteins in T2DM patients using enzyme-linked immunosorbent assay (ELISA). The overall study framework is illustrated in Fig. 1.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Study design for identifying and validating the druggable targets of T2DM.

Blood druggable gene expression quantitative trait loci dataset

We used blood druggable gene expression profiles as the exposure. Blood cis-eQTL data was obtained from the eQTLGen Consortium (https://eqtlgen.org/), as previously described14. Briefly, the eQTLGen dataset includes 31,684 blood samples integrated from 37 datasets, all preprocessed in a standardized manner. The druggable gene list comprises 4,302 autosomal protein-coding genes, identified by integrating data from multiple genome-wide association studies and linking them to known drugs15. Among these, 1,375 proteins serve as therapeutic targets in clinical development, 646 are associated with drug targets and compounds, and 2,281 belong to key drug target families. To construct genetic instruments for the 4,302 druggable targets, we selected cis-eQTLs within ± 100 kb of each gene’s genomic position, identifying 2,113 druggable genes with cis-eQTLs.

Source of outcome dataset

The GWAS summary statistics (ebi-a-GCST006867) for T2DM, included 61,714 cases and 593,952 controls of European ancestry, was retrieved from the IEU Open GWAS Project (https://gwas.mrcieu.ac.uk/)16. Odd ratios (ORs) were calculated to evaluate the association between single nucleotide polymorphisms (SNPs) and T2DM.

MR analysis

We assessed the causal relationship between the blood druggable gene and T2DM using MR analysis with the TwoSampleMR package17. Multiple quality control methods were applied. The SNPs were selected as genetic instruments based on the following criteria: (i) being associated with the druggable gene at the genome-wide significance threshold (P ≤ 5 × 10−8); (ii) being strong instruments with F-statistic > 10; (iii) having no linkage disequilibrium at the threshold of r² < 0.1 within 1000 kb, assessed using the European reference panel from the 1000 Genomes Project18.

The effect of blood druggable genes on T2DM was estimated using the Wald ratio method when only a single instrumental variable was available. For multiple instrumental variables, causal effects were estimated using the inverse variance weighted method. To account for false positives due to multiple testing, we calculated the false discovery rate (FDR) to determine adjusted significance thresholds, considering FDR < 0.05 as statistically significant. To assess potential pleiotropy, we performed the MR-Egger intercept test, which provided an estimate of horizontal pleiotropy. Cochran’s Q test was used to evaluate heterogeneity among instrumental variables. Steiger filtering was applied to verify the directionality of causality, ensuring that the genetic instruments were more strongly associated with the exposure (gene expression) than with the outcome (T2DM).

Colocalization analysis

Colocalization analysis serves as a key approach to determine whether pairs of traits share the same causal variant within a specific region, thereby strengthening the causal evidence established through MR. The analysis enumerates all variant-level hypotheses, which correspond to five mutually exclusive hypotheses: H0: no association with either trait in the region; H1: association with only trait 1; H2: association with only trait 2; H3: associations with both traits, yet, having different single causal variants; H4: associations with both traits, and sharing the same single causal variants. Posterior probabilities of each hypothesis were calculated using a Bayesian approach. The significance threshold for H4 was set at PP.H4 > 0.50 (posterior probability), and colocalization analysis was performed using utilized coloc package.

PheWAS

PheWAS, which could identify cross-phenotype associations, was designed to examine the causal association of SNPs or other attributes with a wide range of traits. PheWAS has been instrumental in uncovering shared biological mechanisms across human diseases. After identifying the candidate drug targets, we performed a PheWAS to comprehensively assess the association between genetic variants in these drug targets and disease outcomes. The analysis was performed on the AstraZeneca PheWAS Portal (https://azphewas.com/), leveraging exome sequencing data from approximately 450,000 individuals in the UK Biobank, encompassing approximately 15,500 binary and 1,500 continuous phenotypes. More details referred to the published paper19.

Candidate drug prediction

Evaluating protein-drug interactions is crucial for determining the viability of target proteins as drug candidates. We utilized the Drug Signatures database (DSigDB) within Enrichr (https://maayanlab.cloud/Enrichr/) to predict candidate drug. DSigDB currently comprises 22,527 gene sets, encompassing 17,389 unique compounds and 19,531 genes. This database links gene expression with drugs or compounds, providing direction for drug repurposing and translational research20.

Molecular docking

Molecular docking is a computational technique used to predict interaction between small molecules (ligands) and target proteins (receptors). This technique assesses the binding affinity and interaction stability of potential drug candidates with their target proteins, facilitating drug discovery and validation. To comprehensively evaluate the binding energy and interaction patterns between candidate drugs and their receptor proteins, we employed atomic-level molecular docking. This method predicts ligand-receptor binding affinity and interaction dynamics, aiding in the prioritization of drug targets and providing a foundation for further experimental validation of potential drug candidates. The chemical structures of drug compounds were obtained from the PubChem (https://pubchem.ncbi.nlm.nih.gov/), while structure of protein encoded by druggable gene was obtained from the Protein Data Bank (http://www.rcsb.org/). Corresponding IDs for target proteins were retrieved from the EMBL-EBI database (https://www.ebi.ac.uk/). Molecular docking simulations were performed using AutoDock 4 (https://autodock.scripps.edu/)21, and results were visualized using PyMOL software.

Validation of the target proteins

We performed a matched case-control study to validate the identified drug target protein. A total of 44 pairs of patients with T2DM and sex- and age-matched healthy controls were recruited from Yijishan Hospital of Wannan Medical College in 2024. Blood specimens were collected, and serum levels of target proteins were measured using ELISA according to protocol. Protein concentrations (ng/mL) were quantified based on the standard curve. This study was approved by Institutional Review Board of Wannan Medical College (WNMC-AWE- 2024007).

Statistical analysis

Protein concentrations were logarithmically transformed, and expressed as mean ± standard deviation. The fold change was calculated by dividing the mean value in T2DM group by the mean value in control group. Paired t-tests were used to compare the differences between patients with T2DM and healthy controls. Statistical significance was determined by a two-sided P value < 0.05. All statistical analyses were conducted using R software (version 4.3.0).

Results

Screening T2DM-related drug targets

Initially, we characterized 4,302 unique human protein-coding genes as potential therapeutic targets. After linking protein-coding genes with cis-eQTL in blood, 2,113 druggable genes remained. The causal associations of 2,113 druggable genes with T2DM were evaluated by utilizing two-sample MR methods. After FDR correction, we identified 42 genes that were potential drug targets for T2DM (Fig. 2). Of them, 24 genes were positively associated with the risk of T2DM, and 18 genes were negatively associated with the risk of T2DM at the level of FDR < 0.05 (Fig. 3). MR-Egger intercept test indicated that none of the identified T2DM-related genes exhibited significant pleiotropy (P > 0.05, Supplementary Table S1), suggesting that pleiotropy is unlikely to bias our causal estimates. Therefore, MR-PRESSO was not applied. Cochran’s Q test suggested that, except for BMP8 A, CAMKK2, and C4 A, the remaining T2DM-related genes showed no significant heterogeneity (P > 0.05, Supplementary Table S2), indicating that the instrumental variables were consistent and valid. The results of Steiger filtering confirmed that all selected genetic instruments explained a greater proportion of variance in gene expression than in T2DM (P < 0.05, Supplementary Table S3), indicating that gene expression influences T2DM rather than the reverse. Given that T2DM was unlikely to causally influence gene expression, bidirectional MR was not conducted.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Manhattan plot from druggable genome-wide association of T2DM using MR analysis*. * The dashed line indicated a significant threshold of FDR < 0.05.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Forest plot displaying the association of 42 druggable genes with the risk of T2DM at level of FDR < 0.05.

Colocalization analysis between druggable gene and T2DM

We conducted a colocalization analysis to identify loci where the same causal variant influences both druggable genes and T2DM. Six genes (CLSTN1, DLD, KCNJ11, MLX, RELA, and ULK1) were found to colocalize with T2DM (Table 1), indicating that these druggable genes and T2DM share the same causal genetic variants. These findings highlight the genetic overlap between these six druggable genes and T2DM.

Table 1 Colocalization results between 6 druggable gene and T2DM.
Full size table

PheWAS analysis of identifying common mechanisms between T2DM and other diseases

The PheWAS was conducted to estimate the associations between genetic variants in six druggable genes and the phenome. The results showed that CLSTN1 was associated with age-related macular degeneration (AMD), indicating that drugs targeting the CLSTN1 gene might impact eye diseases (Table 2).

Table 2 The association between CLSTN1 and AMD from PheWAS analysis*.
Full size table

Candidate drug prediction

We utilized the DSigDB to predict potential therapeutic agents for six proteins encoded by druggable genes. Several candidate drugs were identified, with the top ten ranked based on FDR correction (Table 3). Notably, DLD, and RELA possessed multiple potentially binding agents.

Table 3 Candidate agents predicted using DSigDB.
Full size table

Molecular docking

To understand the druggability of the six drug targets and evaluate their binding affinity with candidate agents, molecular docking was performed. We presented the binding sites of the top five candidate agents for target protein. These agents were docked to the binding sites of proteins encoded by the druggable genes DLD, RELA, and KCNJ11 (ebselen CTD 00001949 3D structure not found) (Fig. 4). These docked structures can provide insights into the mechanism of action of the agents and guide further optimization of their structure.

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Docking results between druggable target proteins and agents. (A) DLD docking with RUTIN CTD 00006712; (B) RELA docking with RUTIN CTD 00006712; (C) DLD docking with 1,3,5[10]-Estratriene- 3,4,17 beta-triol CTD 00000733; (D) RELA docking with 1,3,5[10]-Estratriene- 3,4,17 beta-triol CTD 00000733; (E) DLD docking with ellagic acid CTD 00005891; (F) RELA docking with ellagic acid CTD 00005891; (G) KCNJ11 docking with D-glucose BOSS; (H) DLD docking with D-glucose BOSS.

Validation of the potential drug targets

In the matched case-control study, the mean age was 60.27 ± 12.29 years for T2DM and 60.64 ± 12.72 years for control (t = − 1.37, P = 0.179). Women comprised 56.82% of the participants in both the T2DM and control (χ2 = 0.00, P = 1.000). With the exception of DLD, the findings of the remaining 5 potential drug targets aligned with the predictions of MR (Fig. 5). The expressions of CLSTN1, KCNJ11, MLX were shown to have a 0.71-, 0.65-, and 0.57-fold reduce in T2DM (P < 0.05). In contrast, RELA and ULK1 exhibited a 3.44- and 1.68-fold increases in the T2DM group (P < 0.05). No difference in DLD was observed between T2DM and control (P = 0.114).

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Log-transformed serum levels of 6 target proteins in T2DM and control. (A) Serum levels of CLSTN1 (5.41 ± 0.68 vs. 5.75 ± 0.45, P = 0.008); (B) Serum levels of DLD (1.09 ± 0.89 vs. 0.82 ± 0.87, P = 0.144); (C) Serum levels of KCNJ11(3.44 ± 0.56 vs. 3.86 ± 0.29, P = 9.61e- 05); (D) Serum levels of MLX(6.25 ± 0.47 vs. 6.80 ± 0.37, P = 4.75e- 08); (E) Serum levels of RELA(6.37 ± 0.72 vs. 5.14 ± 0.74, P = 5.11e- 10); (F) Serum levels of ULK1(1.18 ± 0.20 vs. 0.66 ± 0.40, P = 7.83e- 09).

Discussion

Leveraging genetic insights in drug development can accelerate the discovery of novel therapeutic agents. In this study, we combined two-sample MR analysis with colocalization analysis to identify potential therapeutic targets for T2DM. Among the 4,302 protein-coding genes, CLSTN1, DLD, KCNJ11, MLX, RELA, and ULK1 were identified as candidate drug targets. External validation confirmed the MR-predicted associations for all genes except DLD. Furthermore, potential therapeutic compounds targeting these proteins were identified through DSigDB-based drug prediction and molecular docking analysis.

CLSTN1, a member of the cadherin superfamily, is a transmembrane protein predominantly expressed in the central nervous system. It plays a critical role in synaptic function by converting extracellular proteolysis in the synaptic cleft into postsynaptic Ca²⁺ signaling22,23. CLSTN1 has been demonstrated to regulate trafficking and processing of amyloid precursor protein via interacting directly with kinesin- 124. It was proposed that CLSTN1 involved in the onset of Alzheimer’s disease. CLSTN1 was down-regulated in Alzheimer’s disease brains, with the extent of reduction inversely correlated with amyloid β-protein accumulation25. The role of CLSTN1 in T2DM remains unexplored. Our study, to the best of our knowledge, was the first to report a downregulation of CLSTN1 in T2DM. CLSTN3, another member of the cadherin superfamily with a structural resemblance to CLSTN1, had been implicated in metabolic regulation. Animal study showed that CLSTN3 deficiency induced insulin resistance and hepatic steatosis, whereas its overexpression ameliorated lipid and glucose metabolism as well as insulin sensitivity26. However, whether CLSTN1 is involved in insulin resistance remains unclear. Further research is needed to elucidate the role of CLSTN1 in the pathogenesis of T2DM.

KCNJ11, termed as Kir6.2 protein, constitutes ATP-sensitive potassium (KATP) channel by binding with sulfonylurea receptor 1 (SUR1). KATP channel is responsible for regulating insulin secretion in pancreatic β-cells27. Animal studies have confirmed that KATP channel modulates insulin secretion and glucose metabolism. In Kir6.2-deficient mice, a reduction in insulin secretion and impaired glucose tolerance were observed28. Furthermore, a variety of genetic variations were found in KCNJ11, some of which have been shown to reduce expression of KATP channel29and increase susceptibility of neonatal diabetes mellitus30and T2DM31. Our findings revealed that downregulation of KCNJ11 was associated with an increased risk of developing T2DM. Based on existing evidence, dysregulation of KCNJ11 appears to induce T2DM via reducing the activity of the KATP channel. Therefore, there is an underlying mechanism for the development of T2DM drug targeting KCNJ11.

MLX, a glucose-sensing transcription factor, belongs to the family of basic helix-loop-helix leucine zipper (bHLH-Zip). In response to high intracellular glucose concentrations, MLX migrate from the cytoplasm to the nucleus, where it binds to carbohydrate response element binding protein (ChREBP) to form a heterodimer. The MLX/ChREBP heterodimer regulates the expression of glycolytic, gluconeogenic, and lipogenic genes to mediate glucose and lipid metabolism32,33. Recently, MLX was confirmed to mediate lipid and glucose metabolism in vitro study. MLX knockdown lowered lipid accumulation via decreasing de novolipogenesis and increasing fatty acid oxidation in hepatic cell. Furthermore, MLX knockdown promoted gluconeogenesis34. Few epidemiological studies showed genetic variants in the MLX were associated with insulin resistance and T2DM34,35. Our study illustrated the relationship between MLX expression and T2DM at the population level, revealing that MLX downregulation was associated with an increased risk of developing T2DM.

RELA, also known as p65, is part of NF-κB transcription factors family. NF-κB is involved in various cellular processes, including inflammation, metabolism, and immunity36. Insulin resistance and T2DM are considered as chronic inflammatory diseases. NF-κB-inducing kinase (NIK), a component of the non-canonical pathway of NF-κB, activates NF-κB pathway37. Animal study showed that NIK was involved in glucose metabolism. Deletion of NIK led to hypoglycemia, improved glucose tolerance, and reduced hepatic gluconeogenesis38. However, human data on investigating the role of RELA in T2DM remains limited. Our findings revealed that upregulation of RELA was associated with an increased risk of developing T2DM. An animal study showed that deletion of RELA in hepatocytes reduced inflammation and improved systemic insulin sensitivity39. However, another animal study reported that RELA deletion in pancreatic β-cells resulted in loss of insulin secretion, resulting glucose intolerance40. Deletion of RELA in different tissues yielded conflicting results. Further research is needed to determine whether RELA exhibits tissue-specific functions.

ULK1, a serine/threonine kinase, is recognized as an initiator in the autophagic cascade41. Autophagy is an essential degradation process in maintaining cellular homeostasis by eliminating damaged organelles and misfolded proteins, recycling amino acids or other degradation products42. Dysfunctional autophagy has been shown to be involved in several diseases including T2DM, and cancer42. Autophagy exhibited a dual role in pancreatic β-cells. Under normal physiological conditions, autophagy was crucial for regulating pancreatic β-cells, supporting their differentiation and functional maintenance. Additionally, autophagy contributed to insulin homeostasis by controlling the process of proinsulin conversion to insulin. On the other hand, inhibiting autophagy has been observed to protect β-cells from glycolipid-induced toxicity, while excessive autophagy impaired insulin secretion and led to β-cell loss. The deletion of autophagic enhanced insulin secretion43. ULK1 also exhibited a dual role in glucose metabolism. Hepatocyte-specific knockout of ULK1 in mice resulted in impaired glucose tolerance and insulin resistance44. In contrast, endothelial-specific knockout of ULK1 in mice improved glucose tolerance45. Our findings showed that upregulation of ULK1 was linked to an increased risk of developing T2DM. Given these conflicting findings, further research is needed to clarify the role of ULK1 in glucose metabolism and its potential implications for T2DM.

In summary, we identified 42 genes associated with T2DM, five of which (CLSTN1, KCNJ11, MLX, RELA, and ULK1) were validated through external datasets and functional studies. While further research is required to establish their therapeutic potential, this study underscores the importance of integrating genetic data into drug development, with the potential to enhance the success of novel T2DM therapies.