Integrated epigenome, whole genome sequence and metabolome analyses identify novel multi-omics pathways in type 2 diabetes: a Middle Eastern study

The worldwide prevalence of Type 2 diabetes (T2D) in 2021 was estimated to be 9.8% [1]. However, the prevalence in countries from the Middle East and North Africa was 18.1%. The prevalence of T2D in Qataris has been estimated to be 14–17% [2], similar to other Middle Eastern populations. Such high prevalence of T2D in Qatar compared to western populations increases the risk of diabetes complications, including cardiovascular disease, retinopathy, nephropathy and early mortality. More than 400 independent genetic loci have been found to be associated with T2D across different ethnicities and yet explain only a modest portion of the prevalence [3‐5]. T2D is caused by complex interactions of multiple factors including genetic, epigenetic and environmental influences. Current evidence indicates that lifestyle and environmental factors can influence gene expression and clinical phenotype through epigenetic mechanisms such as changes in DNA methylation. Identifying the epigenetic factors driving T2D are thus important to our understanding of T2D etiology.

DNA methylation in blood or pancreatic beta cells has been associated with T2D in multiple populations [6‐13]. Most used a previous version of the Infinium EPIC array utilizing 450K CpGs (rather than the newer 850K array) and involved western populations. A previous study done on Qataris (n = 123) demonstrated heterogeneity in methylation between Qatari and UK populations, indicating the importance of studying multiple world populations with varying prevalence of T2D [14].

Genetic variants have also been associated with methylation sites, referred to as methylation quantitative trait loci (meQTLs), and have further elucidated factors affecting methylation. For example, 4.7 million cis- and 630,000 trans-meQTL variants were identified in one study using the 450K array [15]. Using the 850K EPIC arrays, another study identified a large number of significant SNP-CpG pairs related to various features of diabetic kidney disease [16]. Those, and other previous studies, used imputed genotypes from available SNP arrays.

Circulating metabolites have also been associated with T2D and its complications [17, 18]. Since metabolite levels are influenced by both genetics [19] and epigenetics, and methylation affects gene expression which in turn affects metabolic pathways, T2D methylation would be expected to be associated with metabolites and metabolic pathways related to T2D.

This study focuses on the epigenetic associations with T2D in a large number of subjects from the Qatari population to highlight the similarities and differences in T2D methylation compared to other populations. The integration of genomics and metabolomics with epigenetics was used to better understand the biological mechanisms underlying methylation changes associated with diabetes. First, the larger Infinium 850K EPIC array was used to identify novel methylation associations with T2D using Qatari samples. Second, whole genome-wide sequencing (> 5.5M SNPs) was used to identify methylation quantitative trait loci (meQTLs) for the T2D-associated CpGs to highlight genetically driven CpGs. Third, an untargeted panel of metabolites (Metabolon) was used to identify the associations of the T2D CpGs with the metabolome. Finally, combining these results with metabolic correlations, multi-omics networks were constructed to find functional metabolic pathways connecting the methylated genes and to understand their biological relation to T2D in this high-risk Middle Eastern population.

Figure 1 summarizes the overall study design. Three cohorts of samples were obtained from the Qatar Biobank over three years with 43.5% being T2D patients (Table 1 shows the cohort characteristics). Cohorts 1 and 2 were used for the T2D Epigenome Wide Association Study (EWAS) (n = 835) and cohort 3 was added to improve the statistical power of SNP-CpG association analysis (n = 1026). Samples from the 3 cohorts that had both metabolomics and methylation were used for the CpG-metabolite association analyses (n = 708).

This study used a large number of subjects from the Qatari population that has a high prevalence of diabetes and used the largest methylation EPIC array. Whole genome sequencing and metabolomics data were incorporated to enrich the identification of biological pathways associated with methylation and diabetes.

The most significantly-associated T2D methylated gene, thioredoxin-interacting protein (TXNIP), plays a role in insulin sensitivity, is a tumor suppressor that is upregulated in diabetes, is induced when glucose levels are elevated, and its deficiency improves glucose tolerance and increases insulin sensitivity in high fat diet-induced obesity [34, 39]. TXNIP suppresses glucose uptake by directly binding and suppressing the glucose transporter, GLUT1 (SLC2A1), by facilitating its clathrin-mediated endocytosis [34, 40]. When elevated, TXNIP induces β-cell apoptosis, while its deficiency protects against type I and type II diabetes by promoting β-cell survival [41, 42]. It is also known to be involved in kidney injury and with IL-1 \(\upbeta\) in the pathogenesis of T2D [43, 44]. Thus, findings from the meQTL analysis, linking SLC2A1 to TXNIP and its several metabolic associations support and add to those known functions of the gene. Besides TXNIP, pathway analysis and the known biology of the novel methylated genes support their relevance to T2D. For example, PPP1R3E, PRKAR2A and HK1 are involved in insulin signaling and resistance (Additional file 2: Table S5). Together with the several T2D-related pathways reported for HK1, further evidence of its relevance to T2D has been provided by GWAS associations of HK1 genetic variants with HbA1c [45], with the strongest signals reported in [46]. HK1 is in the family of hexokinases, which are known to phosphorylate glucose to produce glucose-6-phosphate, the first step in most glucose metabolism pathways [47]. Its association with the insulin signaling pathway and hyperinsulinemia [48], along with being part of the carbohydrate absorption, glycolysis, fructose and mannose metabolism pathways further emphasizes the statistically inferred causality relation. Other methylated genes have been associated with T2D incidence [49] and cardiovascular disease [50]. For example, CACNA2D2 is involved in pathways related to T2D-associated coronary heart disease, cardiomyopathy, cardiac muscle contraction [51]. Variants of DOCK6 have been associated with GWAS studies on coronary heart disease and total and HDL cholesterol levels [52, 53]. THBS4 has been associated with more advanced peripheral artery disease and T2D [54]. It has also been reported to mediate breast cancer inflammation and growth in mouse models in response to hyperglycemia and TGF-beta [55]. IDO2, one of the isoforms of IDO (Indoleamine 2, 3-dioxygenase) which forms part of the tryptophan metabolism and catalyzes the conversion of tryptophan to kynurenine, is upregulated in T2D patients with baseline tryptophan being associated with higher risk of incident T2D [56].Variants of PBRM1 have been reported to be associated with BMI-related traits and diabetes pathways [57, 58]. While the pathways and biological function of the identified genes could partially be explained from previous findings, our study reveals other biological functions through linking methylation with genomic variants and metabolic pathways.

The meQTL analysis indicated the effect of genomic variation on methylation levels of 22 T2D CpGs, whereas the remaining 44 T2D CpGs are more likely affected by environment/lifestyle. The association of the missense SNP rs1136287 in the SERPINF1/SMYD4 locus with cg11692409 aligns with the previously reported GWAS association of SMYD4 with T2D that is harbored in the SRR locus [4]. The DOCK10 association with diabetic atherosclerosis [33] and its gene expression association with glucose in pancreatic islets [59] suggest that the nonsynonymous meQTL SNPs in DOCK10 may link the gene’s methylation to T2D genetic factors, diabetes complications or mechanisms in islet cells. The presence of a 5’UTR variant in SLC2A1 that codes the GLUT1 protein and that controls methylation in TXNIP is another novel finding that confirms previous biological links found between TXNIP and SLC2A1 [34, 35, 40] as discussed above.

GWAS associations of meQTL SNPs with T2D-related traits such as HbA1c (HK1 and PFKFB2), BMI (EFEMP2, CFL1), eGFR and creatinine (PFKFB2), diabetic nephropathy, urate levels (EFEMP2), bilirubin (HK1), and white blood cell counts (DOCK10, EFEMP2, SLC2A1, CFL1, PFKFB2) [51, 60, 61], further emphasize the involvement of the meQTL genes in the pathogenesis of T2D. In other words, SNP variation may alter methylation levels in T2D CpGs inducing various biological perturbations and complications in T2D patients. Furthermore, the mendelian randomization analysis suggested a causal relationship of DNA methylation in HK1 with HbA1c levels, supporting previous findings of HK1’s GWAS associations with HbA1c and HK1 pathways involving glucose [47], insulin signaling and hyperinsulinemia [48].

This study reports novel T2D methylation-metabolite associations. Compared to a few previously reported metabolic associations with TXNIP CpG cg19693031, this study reveals a much larger role of TXNIP in T2D metabolic pathways that span several pathways of interest, including branched chain amino acids, alanine, lipid metabolism, and sugars among others. DQX1, previously identified to be methylated with T2D in an Arab population, has no meQTL associated with it, suggesting that the metabolic associations of DQX1 with lipid metabolism and ribitol could be largely affected by lifestyle factors. Metabolic associations with genes known to associate with bladder cancer (BLCAP), ovarian cancer (OCIAD1), and infertility caused by azoospermia (DAZAP1) could highlight the importance of integrating both methylation and metabolomics for assessing T2D complications.

Integrating methylation with both genomics and metabolomics has an important role in understanding the cause-effect direction of methylation-metabolism in T2D, i.e., whether methylation drives metabolic changes through gene expression or changes in metabolites cause changes in methylation. For example, methylation of TXNIP is controlled by SNPs in SLC2A1, known to be involved in glucose metabolism, and thus TXNIP association with the glycolysis pathway could be through this meQTL. Similarly, we showed that the meQTL of PFKFB2 had SNPs associated with HbA1c, eGFR, and creatinine, and at the same time the gene had associations with carbohydrates, lipids, phenylalanine and alanine that are involved in both T2D [17, 18] and kidney function [62, 63], suggesting that the metabolic perturbations could be an end effect of the meQTL. Linking methylated genes to metabolism may help future investigation of the population-specificity of some methylated genes to Qataris. Perturbation of the methylation-metabolic pathway by changing lifestyle factors or nutrition in Qataris may result in different methylation patterns in Qataris compared to the other populations and thus may help explain T2D mechanisms in Qataris.

Multi-omics networks have facilitated the identification of metabolic pathways linked to the methylated genes, thus revealing their functionality through metabolites that were associated with T2D risks/complications. One example is the alanine subnetwork that links PFKFB2 and TXNIP to the urea cycle and to amino acids, where amino acids were associated with triglycerides (alanine association with triglycerides: discovery p-value of p = 6.4 × 10^–12, replication p-value of p = 9.3 × 10^–13) and where both urea cycle and alanine have been associated with kidney failure [64]. Moreover, 1-carboxyethylphenylalanine in the alanine subnetwork was associated with triglycerides (p-value discovery: 1.42 × 10^–34, p-value replication: 4.6 × 10^–21) and is known to be associated with hypertension [65]. Both links extend the GWAS findings for PFKFB2 described above and reveal the involvement of triglycerides in T2D methylation. The association of 1-palmitoyl-2-oleoyl-GPE(16:0/18:1) with triglycerides (discovery p-value p = 6.47 × 10^–100, replication p-value p = 1.39 × 10^–80) and its link to 11 methylated genes suggests a possible cause of methylation linked to a diabetes risk/complication that is common between those genes. Mannonate, a xenobiotic (food component/plant) is another example that shares a network with six genes, namely TXNIP, BLCAP, THBS4, PEF1, DAZAP1, and KIAA1211L and may suggest a possible xenobiotic effect on methylation. The networks also highlighted a STYXL1–POR–sphingosines/ceramides–LDL,HDL–steroids connection, where the unique association of POR with a lipid subnetwork of sphingosines, GPCs, and ceramides is supported by the fact that gene’s production of the enzyme cytochrome P450 oxidoreductase is required for the synthesis of cholesterol and steroid hormones (https://​medlineplus.​gov/​genetics/​gene/​por/​, and references therein). Moreover, since reproductive and fertility complications may arise from diabetes [66], it is interesting to note that STYXL1, the meQTL gene of the cg01676795 (POR) is known to be associated with seminal vesicle tumor and male reproductive organ benign neoplasm [67], that may help in understanding the link of POR to a network of sphingosines and ceramides which play a role in forming steroids [68].

Limitations of the study include measuring methylation from whole blood rather than pancreatic islets as that was the only available tissue from the Qatar Bio Bank. Another limitation was the inability to correct for medication as that information was not yet annotated by the provider at the time of the study.

A total of 66 CpG sites were identified to be associated with T2D, of which 63 were novel and linked to biological pathways of T2D. Genomic drivers of the CpG sites were identified in 688 significant CpG-SNP pairs, among which 181 CpG-SNP pairs were novel associations or in novel loci. Several GWAS associations in meQTLs’ genes revealed various underlying factors in pathways linked to T2D. The study identified novel nonsynonymous and 5’UTR variants associated with T2D methylation in SERPINF1, DOCK10, and TXNIP, as well as the statistically-inferred causal relation between HbA1c and each of the HK1 and PFKFB2 methylation sites. A total of 61 novel methylation-metabolite associations revealed association of several methylated genes with T2D metabolic pathways including population specific genes such as DQX1, among others. The multi-omics network suggested a number of T2D biological mechanisms associated with methylation, including the association of triglyceride metabolism and xenobiotics to 11 and 6 methylated genes respectively, indicating a common factor among those methylated genes and may have a role in their methylation. The pathways identified linked to steroid metabolism, and to metabolites associated with BMI, lipoprotein cholesterols and kidney function, namely STYXL1-POR, SLC2A1-TXNIP, and PFKFB2, may reveal the association of T2D risk factors or complications with methylation mechanisms. Finally, this study revealed several novel methylated genes related to T2D, with their associated genomic variants and metabolic pathways.