Elucidating the Epigenetic Landscape of Type 2 Diabetes Mellitus: A Multi-Omics Analysis Revealing Novel CpG Sites and Their Association with Cardiometabolic Traits
Article information
Abstract
Background
Type 2 diabetes mellitus (T2DM) is a complex, multifactorial disease with a significant global burden. Although genome-wide association studies (GWAS) have identified many T2DM-associated variants, most lie in non-coding regions, making it difficult to interpret their functional roles.
Methods
We aimed to identify genetically regulated Cytosine–phosphate–Guanine (CpG) sites associated with T2DM by conducting a methylome-wide association study (MWAS), followed by Mendelian randomization (MR) and functional validation using human pancreatic cells and mouse models. MWAS was performed using summary statistics from large-scale GWAS and a DNA methylation (DNAm) prediction model to test associations between genetically predicted DNAm and T2DM.
Results
We identified 111 CpG sites significantly associated with T2DM in Europeans, including 8 novel sites near genes not previously linked to T2DM. These findings were replicated in independent datasets. Many CpGs also showed associations with cardiometabolic traits, highlighting shared epigenetic mechanisms. Trans-ethnic MR analysis confirmed consistent effects for six CpGs in East Asians. Functional analysis revealed that several CpGs regulate gene expression in human pancreatic α- and β-cells. Among them, 2´-5´-oligoadenylate synthetase like (OASL) expression, regulated by a significant CpG, was differentially expressed in α-cells of T2DM cases compared to controls. Supporting evidence from mouse models suggests a role for OASL in glucose regulation.
Conclusion
Our study identifies novel genetically regulated CpG sites associated with T2DM risk and highlights OASL as a potential epigenetic regulator of glucose metabolism in α-cells. These findings provide mechanistic insights into the epigenetic architecture of T2DM and suggest potential targets for cross-ethnic biomarker development and therapeutic intervention.
Highlights
• We identified 111 CpGs linked to T2DM in Europeans, including 8 novel CpG sites.
• Six CpGs showed trans-ethnic effects, replicated in East Asian populations.
• OASL gene expression was altered in T2DM alpha cells, linked to CpG methylation.
• Many CpGs overlapped with cardiometabolic traits like TG and blood pressure.
• Mouse data supported OASL’s role in regulating glucose metabolism in T2DM.
INTRODUCTION
Type 2 Diabetes (T2DM) is a complex multifactorial disease with significant global health impact. Genome-wide association studies (GWAS) have identified numerous genetic variants associated with T2DM [1], but understanding the functional mechanisms by which these variants contribute to the disease remains challenging. This difficulty arises from the polygenic nature of T2DM, the complexity of distinguishing causal variants from those in linkage disequilibrium (LD), and the fact that most GWAS hits are located in non-coding regions of the genome [2].
Integrated approaches combining GWAS data with functional genomics have been employed to address these challenges [3]. One such approach is transcriptome-wide association studies (TWAS), which leverage expression quantitative trait loci (eQTL) to impute gene expression levels and link them to phenotypic traits [4,5]. TWAS has successfully uncovered novel gene-disease associations in complex diseases, including T2DM, providing insights into disease etiology and highlighting potential therapeutic targets [6,7].
The principles underlying TWAS are effectively applicable to the study of DNA methylation (DNAm), a pivotal mechanism for regulation of gene expression. DNAm involves the addition of a methyl group to cytosine residues within Cytosine–phosphate–Guanine (CpG) sites, influencing gene activity by altering the accessibility of transcriptional machinery [8]. Abnormal DNAm levels are associated with various complex diseases, including T2DM, as identified through epigenome-wide association studies (EWAS) [9]. Importantly, DNAm is not just an epigenetic phenomenon but also a heritable one. Studies have demonstrated that CpG methylation possesses a significant heritable component, with heritability estimates ranging from 16% to 20% [10]. Consequently, many methylation quantitative trait loci (meQTL), which are genetic variants affecting the DNAm levels, have been identified [11]. The combination of DNAm’s heritability and the availability of meQTL data paves the way for methylome-wide association studies (MWAS). MWAS tests the association between imputed DNAm levels—based on meQTL from GWAS—and diseases, providing a powerful approach to uncover genetically regulated DNAm associations with diseases and offering insights into disease mechanisms that may not be detectable through traditional GWAS or TWAS methods.
Despite its potential, the application of MWAS in T2DM remains unexplored, and studies focusing on non-European populations are notably absent. Fryett et al. [10] pioneered the construction of a methylation prediction model and identified over 700 significant CpGs in primary biliary cholangitis. However, similar studies in T2DM and diverse populations are lacking. This gap highlights the need for MWAS to uncover genetically regulated DNAm associations with T2DM, particularly in underrepresented populations.
To address this gap, we conducted a MWAS to identify CpGs where DNAm levels regulated by meQTL are associated with T2DM risk in Europeans. We replicated significant findings using Mendelian randomization (MR) in independent datasets and validated them with measured methylation levels. Additionally, we evaluated the associations of these CpGs with other cardiometabolic traits and assessed their trans-ethnic effects in East Asians by leveraging large-scale genomic data from the Taiwan Biobank (TWB) [12] and the Asian Genetic Epidemiology Network (AGEN) consortium [13]. The inclusion of these datasets strengthens the generalizability of our findings by extending the analysis beyond European populations, addressing the scarcity of MWAS in East Asians. Recognizing that both pancreatic β-cells, which produce insulin, and α-cells, which secrete glucagon, play significant roles in T2DM pathophysiology [14], we integrated methylation and gene expression data from human pancreatic α- and β-cells to assess the associations of gene expression regulated by CpGs identified in our study with T2DM. Dysregulated glucagon secretion from α-cells contributes to hyperglycemia by promoting hepatic glucose production [15], and DNAm patterns in α-cells may influence gene expression related to glucagon secretion and glucose metabolism [16]. Therefore, investigating epigenetic modifications in both α- and β-cells can provide a more comprehensive understanding of T2DM mechanisms. By evaluating functional implications using databases of mouse models, we aimed to elucidate the biological mechanisms underlying these associations. Our study provides novel insights into the epigenetic regulation of T2DM and highlights potential therapeutic targets.
METHODS
Overview of datasets and models
A comprehensive summary of the datasets and models used in this study is provided in Supplementary Table 1, including details on sample size, ancestry, data type, and references. Our analyses primarily relied on large-scale GWAS summary statistics for T2DM, including data from the Diabetes Meta-analysis of Trans-ethnic Association Studies (DIAMANTE) consortium (80,154 T2DM cases and 853,816 controls of European ancestry) [17] and the FinnGen study (38,657 cases and 310,131 controls) [18]. For the MWAS, we employed a DNAm prediction model trained using genotype and methylation data from the Understanding Society cohort (n=1,120) [19]. meQTL information was obtained from the MeQTL EPIC database (n=2,358) [11], which provided single-nucleotide polymorphism (SNP)-CpG associations profiled using the Illumina EPIC array (San Diego, CA, USA). We performed MWAS analyses across ten cardiometabolic traits using trait-specific GWAS summary statistics from European populations obtained through the GWAS catalog. To assess trans-ethnic generalizability, we incorporated East Asian datasets, including meQTL summary statistics from the TWB (n=2,150) [12] and T2DM GWAS summary statistics from the AGEN consortium (77,418 cases and 356,122 controls) [13]. Functional validation was conducted using methylation and gene expression data from human pancreatic α- and β-cells in Data portal of The Human Pancreas Analysis Program (PANC-DB) (https://hpap.pmacs.upenn.edu) [20], and further biological insights were derived from the Mouse Genome Informatics (MGI) database.
MWAS for type 2 diabetes mellitus
We performed a MWAS to identify CpG sites associated with T2DM using S-PrediXcan from the MetaXcan software [21], a tool designed for TWAS that leverages GWAS summary statistics to infer associations between imputed molecular traits (e.g., methylation levels) and phenotypes. This method does not require individual-level genotype or methylation data, making it particularly useful for large-scale studies.
S-PrediXcan estimates the association between genetically predicted methylation levels and T2DM by integrating: (1) a methylation prediction model trained on an independent reference dataset, which imputes methylation levels using nearby meQTLs, and (2) GWAS summary statistics, which provide association signals between these meQTLs and T2DM. This approach effectively links methylation regulation to disease risk while accounting for LD among genetic variants.
The methylation prediction model used in this study was developed by Fryett et al. [10], based on data from the Understanding Society cohort [19], which includes genotype and methylation data from 1,120 individuals. The model applies penalized regression techniques, including least absolute shrinkage and selection operator (LASSO), ridge regression, and elastic net, to predict methylation levels (i.e., β values, which represent the proportion of methylated alleles at a particular CpG site). These methods use genotypic data at nearby SNPs within a specific window as predictors for methylation levels at CpG sites. The models were optimized using cross-validation to identify the optimal window sizes and penalization parameters to maximize prediction accuracy. Prediction accuracy, measured by the correlation between observed and predicted methylation levels, was validated using an independent dataset from the Accessible Resource for Integrated Epigenomics Studies (ARIES) [22], which consists of 841 samples. In total, the model predicts methylation levels at over 179,000 CpG sites, achieving a prediction accuracy greater than 0.1 in both the training and validation datasets. Most of the CpGs selected by this threshold achieved the upper bound of heritability [10], ensuring the reliability of the predicted methylation levels for downstream association testing.
We used GWAS summary statistics from the DIAMANTE consortium, comprising 80,154 T2DM cases and 853,816 controls of European ancestry [17]. The dataset was obtained from the DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) website (https://diagram-consortium.org/downloads.html) under the file name ‘Ancestry-specific GWAS meta-analysis summary statistics: European.’ Using S-PrediXcan, we tested the association of methylation levels at 179,523 CpG sites with T2DM. CpG sites with association P values below 2.78×10−7 were considered significant after Bonferroni correction for multiple testing. Further details on the S-PrediXcan algorithm and its application in MWAS are provided in the Supplementary Methods.
Replication of MWAS results using MR analysis
To strengthen the findings from our MWAS and establish a causal relationship between DNAm and T2DM, we used MR for replication. MR is particularly well-suited for addressing limitations in MWAS by helping to infer causality rather than mere association [23]. In the context of our study, MR allows us to test whether the CpGs identified as significant in MWAS have a causal effect on T2DM, independent of confounding factors such as pleiotropy (where a genetic variant affects multiple traits via different biological pathways).
In this study, we conducted a two-sample MR analysis using the inverse-variance weighted (IVW) [24] algorithm, the most widely used method in MR, alongside the MR-Egger [25] intercept test to evaluate the potential effect of pleiotropy. Both IVW and MR-Egger tests were conducted using the R package MendelianRandomization v0.10.1 (R Foundation for Statistical Computing, Vienna, Austria) [26]. The F-statistic [27] and Cochran’s Q statistic, derived from the IVW estimate, and its associated P value were also calculated using MendelianRandomization v0.10.1 to assess weak instrument bias and heterogeneity.
For the MR analysis, we selected CpGs that were significant in the MWAS and used meQTL as instrumental variables. The association signals for these meQTL with CpGs were obtained from the MeQTL EPIC database [11], which contains data on 724,499 CpGs profiled on the Illumina Infinium Methylation-EPIC array in 2,358 blood samples from three UK cohorts. We chose clumped meQTL with low LD for independence and applied a false discovery rate <0.05 to exclude weak instrumental variables, ensuring robust MR analysis.
Association statistics of the instrumental variables with T2DM were obtained from the FinnGen study [18], another large-scale GWAS for T2DM in Europeans, consisting of 38,657 T2DM cases and 310,131 controls. Importantly, the meQTL and T2DM GWAS datasets used in the MR analysis were independent from those used in the initial MWAS, ensuring unbiased replication of the signals. In adhering to the key assumptions of two-sample MR, we ensured that the summary statistics for the Exposure-SNP (i.e., meQTL information) and Outcome-SNP (i.e., T2DM associations) were derived from independent but related populations. This approach aligns with the principle of using non-overlapping datasets for the exposure and outcome in two-sample MR to avoid bias.
Validation of the CpGs using measured DNAm levels
To validate the signals identified from the MWAS and MR analyses, we compared our findings against external studies based on measured DNAm levels. For this purpose, we used the EWAS Catalog [9], which compiles association statistics for CpGs and traits from published EWAS. We downloaded the full summary statistics from the Generation Scotland (GS) cohort [28], calculated based on 757 incident T2DM cases and 16,992 controls—the largest T2DM EWAS dataset available in the catalog.
MWAS for cardiometabolic traits
To further explore the broader implications of the CpGs identified in our study, we extended our analysis to investigate their effects on cardiometabolic traits that are closely related to T2DM. These traits included body mass index (BMI), fasting glucose, glycosylated hemoglobin, systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse pressure (PP), total cholesterol, triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C). We accessed GWAS summary statistics for these traits in European populations from the GWAS Catalog [29-33]. Using S-PrediXcan, we then conducted MWAS analyses for each trait, employing Fryett’s DNAm prediction model alongside the respective trait’s GWAS summary statistics to assess whether the CpGs identified for T2DM also play a role in these related cardiometabolic traits.
Trans-ethnic analysis of the significant CpGs
To determine the broader applicability and generalizability of our findings beyond the European population, we conducted a trans-ethnic analysis to assess whether the CpGs identified in our study exhibit consistent effects on T2DM among East Asians. This analysis aimed to explore whether the relationships between these CpGs and T2DM hold across different ethnicities, which is crucial for validating the global relevance of these CpGs. We also used a two-sample MR analysis to investigate the causal influence of these CpGs on T2DM. The meQTL summary statistics were derived from 2,150 samples profiled based on the Illumina Infinium MethylationEPIC BeadChip, with GWAS data from the TWB [12]. Supplementary Methods provide in-depth details regarding the QC procedures for the TWB samples and the methodology for computing the meQTL summary statistics. GWAS summary statistics for T2DM were obtained from the AGEN consortium, based on a cohort of 77,418 T2DM cases and 356,122 controls [13].
Expression quantitative trait methylation and differential gene expression analyses
To investigate whether the significant CpGs function as expression quantitative trait methylation (eQTM) loci in human pancreatic α- and β-cells, we utilized whole-genome bisulfite sequencing (WGBS) and RNA sequencing (RNA-Seq) data from PANC-DB [20]. Given the essential role of α-cells in glucagon secretion and β-cells in insulin secretion, examining DNAm patterns in both cell types is crucial, as their dysfunction contributes to T2DM pathophysiology [34]. The processing details for the WGBS and RNA-Seq data are elaborated in the Supplementary Methods. Our analysis was confined to European-descent samples without a diabetes history, comprising 17 α-cell samples and 10 β-cell samples with both WGBS and RNA-Seq data.
For eQTM analysis, we modeled the transcript per million (TPM) for each gene using a linear regression model, accounting for age and sex, with R version 4.3.2. The residuals from this model were then correlated with methylation levels using Kendall’s τ coefficient, with significance determined by a nonparametric τ test. Given that CpGs often regulate gene expression from positions upstream, our focus was on CpGs located within 100 kb upstream of genes. To maximize the identification of potential genes regulated by CpGs, we did not apply multiple testing correction at this stage. Instead, genes with correlation P<0.05 were selected for further analysis, and multiple testing correction was performed in the subsequent differential gene expression analysis to ensure robust findings.
To explore the expression dynamics of genes potentially regulated by the CpGs identified in the eQTM analysis, we conducted a differential gene expression analysis between T2DM cases and controls without a diabetes history. This analysis utilized RNA-Seq data from PANC-DB, focusing on samples from pancreatic α- and β-cells. Specifically, the dataset included six T2DM patients and 26 controls for α-cells, and six T2DM patients and 22 controls for β-cells. We first adjusted the TPM for each gene by age and sex using a linear regression model, then derived residuals for further analysis. These residuals were assessed for their association with T2DM status employing a permutation test. Detailed methodology of the permutation test is elaborated in the Supplementary Methods. Genes that exhibited significant associations with T2DM status, as determined by permutation P values and further adjusted for multiple testing using Bonferroni correction, were earmarked for subsequent functional analysis, aiming to uncover the biological implications of these expression differences in the context of T2DM.
Functional analysis using existing mouse model data
To delve deeper into the functional significance of the genes identified in the differential gene expression analysis, we searched the MGI database. This comprehensive resource integrates data from the Mouse Genome Database (MGD) [35], Mouse Gene Expression Database (GXD) [36], and Mouse Tumor Biology Database (MMHCdb) [37]. Our investigation focused on examining the phenotypic outcomes of mutational changes in these genes, aiming to infer potential implications for diabetes pathogenesis and progression. This approach allows us to extrapolate the biological impact of these genes based on phenotypic alterations observed in mouse models, providing valuable insights into their roles in disease mechanisms.
Ethnics approval and consent to participate
Written informed consent for participants in the TWB was obtained from all participants. Research in this study was approved by the Institutional Review Board of the National Health Research Institutes in Taiwan (reference number: EC1091202-E).
RESULTS
Our analysis workflow is illustrated in Fig. 1, and the resources utilized throughout our study are detailed in Supplementary Table 1. The MWAS and MR analyses primarily leveraged GWAS summary statistics from datasets comprising hundreds of thousands of samples.
Study overview and key findings. Methylome-wide association study (MWAS): Using S-PrediXcan and Diabetes Meta-analysis of Trans-ethnic Association Studies (DIAMANTE) genome-wide association study (GWAS) summary statistics (80,154 cases and 853,915 controls), 1,120 cytosine-guanine dinucleotides (CpGs) associated with type 2 diabetes mellitus (T2DM) were identified in Europeans. Mendelian randomization (MR): Among them, 111 CpGs were supported by MR analysis using independent FinnGen GWAS data (38,657 cases and 310,131 controls). Validation with measured methylation data: Of the 111 CpGs, eight were validated using measured DNA methylation data from the Generation Scotland (GS) cohort from the epigenome-wide association study (EWAS) catalog (757 cases and 16,992 controls). Cross-trait associations: MWAS analysis of the 111 CpGs with 10 cardiometabolic traits identified 15 CpGs showing significant associations with T2DM, triglycerides (TG), body mass index (BMI), diastolic blood pressure (DBP), systolic blood pressure (SBP), pulse pressure (PP), and low-density lipoprotein cholesterol (LDL-C). Trans-Ethnic Replication: Using Asian Genetic Epidemiology Network (AGEN) East Asian GWAS data (77,418 cases and 356,122 controls) and methylation data from the Taiwan Biobank Illumina Infinium MethylationEPIC BeadChip (Taiwan Biobank [TWB] EPIC) (n=2,150), six CpGs demonstrated trans-ethnic effects. Functional analyses: Expression quantitative trait methylation (eQTM) analysis using Data portal of The Human Pancreas Analysis Program (PANC-DB) identified six CpGs in α-cells and seven in β-cells. Differential gene expression analysis revealed that 2´-5´-oligoadenylate synthetase like (OASL) was significantly regulated in α-cells. Functional relevance of OASL in glucose homeostasis was supported by data from the Mouse Genome Informatics (MGI) database. MeQTL, methylation quantitative trait loci; FG, fasting glucose; HbA1c, glycosylated hemoglobin; TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol.
Identification and replication of CpGs
The MWAS analysis identified 1,120 significant CpGs. For MR analysis, we required a minimum of three instrumental meQTL per CpG to enable pleiotropy tests, resulting in 495 out of the 1,120 CpGs being eligible for further analysis. Of these, 111 CpGs were successfully replicated, showing significant MR P<1.01×10−4—considering the 495 tests conducted—without evidence of directional pleiotropy (MR-Egger P>0.05) or heterogeneity among instrumental variables (Cochran’s Q statistic P>0.05). Moreover, all 111 CpGs had F-statistics >10, indicating that the instrumental variables were sufficiently strong and unlikely to introduce weak instrument bias.
Novelty and genomic context of replicated CpGs
Among the 111 replicated CpGs, eight CpGs are located within or proximal to genes not previously highlighted as T2DM candidate genes in the GWAS catalog or the Type 2 Diabetes Knowledge Portal [38], where candidate genes are defined as those with a human genetic evidence (HuGE) score ≥10 [39]. Detailed MWAS and MR statistics for these eight novel CpGs are presented in Table 1, whereas comprehensive details regarding all 111 CpGs can be found in Supplementary Table 2. Across MWAS and MR analyses, the direction of effect generally concurs, where positive effects indicated that higher genetically predicted DNAm levels increased T2DM risk.
Significant CpGs in novel genes for type 2 diabetes mellitus
Genetic variants associated with replicated CpGs
Supplementary Table 3 outlines the lead SNPs—those with the smallest GWAS P values—associated with each of the 111 CpGs, identified either in the MWAS as prediction variables or in the MR analysis as instrumental variables. These lead SNPs generally exhibited significant GWAS P<5×10−8. The median proximity of these SNPs to the corresponding CpGs was approximately 27 kb in the MWAS context and about 51 kb in the MR analysis.
Validation of CpGs using measured DNAm data
Out of the 111 identified CpG sites, eight were successfully validated through measured DNAm levels from the GS study, with P<
CpGs that were validated using samples with measured DNA methylation levels
Cross-trait associations of significant CpGs
Fig. 2 presents an upset plot of the MWAS results for the 111 CpGs across the 10 cardiometabolic traits, illustrating both the number of significant CpGs (with P<4.50×10−4 considering 111 tests) associated with each trait and the intersections among these significant CpGs across different traits. Notably, TG showed the most substantial overlap with T2DM, featuring 66 significant CpGs, followed by DBP with 52, and HDL-C with 48 significant CpGs within the set of 111. Additionally, TG, BMI, DBP, SBP, PP, and LDL-C collectively exhibited the most considerable overlap with T2DM, evidenced by 15 CpGs being significant across these traits. FG was unique in having four distinct significant CpGs that overlapped with T2DM but not with the other traits. Supplementary Table 4 contains comprehensive details of the MWAS findings.
Intersection analysis of methylome-wide association study (MWAS) findings for 111 cytosine-guanine dinucleotides (CpGs) across cardiometabolic traits. This figure presents an upset plot visualizing the distribution of significant CpG sites identified through MWAS among 10 cardiometabolic traits associated with type 2 diabetes mellitus (T2DM). The ‘set size’ section quantifies the significant CpGs associated with each individual trait. The ‘intersection size’ part details the count of CpGs that are significant across multiple traits. Traits associated with each significant CpG group are denoted by black dots under the ‘group’ category. Key traits include triglycerides (TG), body mass index (BMI), high-density lipoprotein cholesterol (HDL-C), diastolic blood pressure (DBP), systolic blood pressure (SBP), pulse pressure (PP), low-density lipoprotein cholesterol (LDL-C), fasting glucose (FG), total cholesterol (TC), and glycosylated hemoglobin (HbA1c).
Trans-ethnic replication in East Asians
Of the 111 significant CpGs, 56 possessed three or more instrumental meQTL, identified through the TWB. Among them, six CpGs exhibited significant MR effects in the East Asian population, achieving MR P values less than
CpGs with trans-ethnic effects in East Asians
eQTM and differential gene expression analyses
We assessed whether the 111 CpGs could be potential eQTM in pancreatic α- and β-cells. In α-cells, six CpGs were found to be associated with the expression of six genes (P<0.05). In β-cells, seven CpGs were associated with the expression of seven genes. Table 4 shows the results of differential gene expression analysis between T2DM patients and controls for the 13 genes identified by the eQTM analyses in pancreatic α- and β-cells. Among them, 2´-5´-oligoadenylate synthetase like (OASL) was identified as significant with a P value of 2.20E-03, surpassing the Bonferroni correction threshold (3.84E-03 for 13 tests) in α-cells, but no significant associations were found in β-cells. Supplementary Fig. 1 displays the residual plot of OASL gene expression levels in α-cells, adjusted for age and sex.
Differential gene expression analysis results for genes regulated by eQTM
Functional insights from mouse models
The findings from our database search for functional studies on the OASL gene in mouse models are detailed in Supplementary Table 5. We discovered that in previously published studies, mutations in the mouse OASL2 gene were associated with reduced circulating glucose levels, indicating a potential role in glucose metabolism.
DISCUSSION
The complexity of T2DM etiology necessitates a comprehensive approach that examines the disease from multiple biological dimensions. By integrating data from different omics layers—genomics, epigenomics, and transcriptomics—our comprehensive multi-omics analysis provides significant insights into the regulatory mechanisms contributing to T2DM. This approach allows us to validate findings across multiple data sources, cross-referencing results to enhance the robustness of our conclusions.
Our results demonstrated that GWAS-identified SNPs influenced DNAm in blood samples at CpGs that could be located tens or hundreds of base pairs away. Notably, certain CpGs we identified are likely to regulate gene expression in human pancreatic α- and β-cells, and DNAm levels at the CpG cg25150715, regulated by T2DM SNPs, may lead to differential gene expression of OASL in α-cells, when comparing normal controls to T2DM patients. Evidence from previous studies using mouse models further corroborates the critical roles of OASL in glucose homeostasis, reinforcing the biological relevance of our findings.
The identification of eight novel CpGs located in or near genes not previously associated with T2DM through GWAS highlights the untapped potential of epigenetic modifications in revealing new genetic susceptibilities to metabolic diseases. For example, the presence of CpGs in genes such as BLK proto-oncogene, Src family tyrosine kinase (BLK) suggests that genetically regulated methylation levels may influence β-cell function and insulin secretion, as prior research has implicated BLK in glucose-stimulated insulin synthesis through upregulation of key pancreatic transcription factors [40]. Although Borowiec et al. [40] proposed a strong link between rare BLK mutations and monogenic forms of diabetes (MODY), subsequent analyses have questioned the penetrance and direct causality of these mutations [41]. Our findings indicate that methylation at CpG sites near BLK could serve as an additional regulatory mechanism affecting insulin secretion pathways, potentially contributing to the risk of T2DM. These epigenetic insights complement genetic studies and reinforce the complexity of metabolic disease etiology, underscoring the value of integrating multi-omics approaches to better understand disease mechanisms.
Furthermore, the upset plot showing the intersections of significant CpGs among 10 cardiometabolic traits provides valuable insights into the shared epigenetic basis of these conditions, highlighting the substantial overlap of significant CpGs between T2DM and traits such as TG, BMI, blood pressure, and LDL-C. This overlap reinforces the idea that T2DM is part of a broader metabolic syndrome, with these traits being well-established risk factors. Epigenetic modifications, such as DNAm, may serve as a regulatory mechanism linking these metabolic abnormalities. For instance, alterations in DNAm may drive changes in lipid metabolism or insulin sensitivity, leading to the dysregulated glucose levels observed in T2DM. Hypomethylation of the monocyte chemoattractant protein-1 (MCP-1) promoter, for example, is linked to increased MCP-1 levels, which promote inflammation, exacerbating insulin resistance and lipid metabolism dysregulation [42]. This inflammatory response may further contribute to metabolic abnormalities like hyperglycemia, dyslipidemia, and hypertension in individuals with T2DM.
The identification of six CpGs with trans-ethnic effects in East Asians suggests that these epigenetic markers may be relevant across different populations. The consistent direction of effects observed in both Europeans and East Asians enhances the generalizability of our findings between these groups and indicates that these CpGs could serve as potential biomarkers for T2DM risk in these populations. However, further studies are needed to examine these associations in additional ethnic groups to determine their broader applicability.
The unique expression of OASL in α-cells and its mutation-induced glucose level reduction highlight a novel link between immune mechanisms and metabolic regulation. OASL, through its modulation of the type I interferon (IFN) response, plays a significant role in the innate immune system’s response to various pathogens, including its dual role in enhancing antiviral defenses and inhibiting autophagic mechanisms [43]. While type I IFN signaling, to which OASL contributes, is involved more directly in type 1 diabetes mellitus in terms of immune system regulation [44], its role in modulating the immune response suggests potential implications for T2DM as well, given the chronic inflammation and innate immune system dysfunction associated with T2DM. These discoveries underscore the potential of targeting OASL for diabetes management, emphasizing the need to understand gene functions across cell types for effective interventions.
While our study has uncovered significant associations and novel CpGs, it is not without limitations. The sample size for some analyses, such as the GS study and PANC-DB, was relatively small, which may affect the generalizability of those findings. Future studies with larger cohorts are necessary to validate our results and to explore the functional consequences of methylation changes in greater detail. Moreover, while the MWAS and MR associations were observed in whole blood samples, we have used human pancreatic α- and β-cells to delve into their specific functional roles in relation to the pathology of T2DM. However, investigating the effects in other tissue types, including adipose tissue, liver, and skeletal muscles, could provide further insights into the systemic nature of T2DM and its epigenetic regulation.
In conclusion, our study underscores the importance of epigenetic mechanisms in the etiology of T2DM and related cardiometabolic diseases. The identification of novel CpGs and their association with T2DM risk opens new avenues for research into the genetic and epigenetic underpinnings of metabolic diseases. This could potentially lead to the development of novel diagnostic and therapeutic strategies that target the epigenetic landscape of T2DM.
SUPPLEMENTARY MATERIALS
Supplementary materials related to this article can be found online at https://doi.org/10.4093/dmj.2025.0041.
Datasets/models used in our analysis
Summary statistics for significant and independent CpGs identified by MWAS and MR for type 2 diabetes mellitus
Lead SNP information from the MWAS and MR analyses
MWAS results for the 10 cardiometabolic traits
Functional analysis of the differentially expressed genes in mouse models
Box plot illustrating the residuals of 2´-5´-oligoadenylate synthetase like (OASL) gene expression, adjusted for age and sex, in pancreatic alpha cells across the normal and type 2 diabetes mellitus (T2DM) groups. The displayed P value signifies the difference in means between the two groups, determined through a permutation test.
Notes
CONFLICTS OF INTEREST
Wayne Huey-Herng Sheu has been an International editorial board members of the Diabetes & Metabolism Journal since 2024. He was not involved in the review process of this article. Otherwise, there was no conflict of interest.
AUTHOR CONTRIBUTIONS
Conception or design: R.H.C., C.C.W., D.D.O., W.H.H.S., H.Y.C.
Acquisition, analysis, or interpretation of data: all authors.
Drafting the work or revising: all authors.
Final approval of the manuscript: all authors.
FUNDING
This study was supported by grants PH-112-GP-04 and PH-112-PP-10 from the National Health Research Institutes and MOST 114-2221-E-400-003-MY3 from the National Science and Technology Council in Taiwan.
ACKNOWLEDGMENTS
We thank the participants from the Taiwan Biobank. This manuscript used data acquired from the Human Pancreas Analysis Program (HPAP-RRID:SCR_016202) Database (https://hpap.pmacs.upenn.edu), a Human Islet Research Network (RRID: SCR_014393) consortium (UC4-DK-112217, U01-DK-123594, UC4-DK-112232, and U01-DK-123716).
The Taiwan Biobank data can be applied through the Taiwan Biobank at https://www.twbiobank.org.tw/. The sources for publicly available datasets used in this study are provided in Supplementary Table 1.