Genetic drivers of heterogeneity in type 2 diabetes pathophysiology

Abstract

Type 2 diabetes (T2D) is a heterogeneous disease that develops through diverse pathophysiological processes^1,2 and molecular mechanisms that are often specific to cell type^3,4. Here, to characterize the genetic contribution to these processes across ancestry groups, we aggregate genome-wide association study data from 2,535,601 individuals (39.7% not of European ancestry), including 428,452 cases of T2D. We identify 1,289 independent association signals at genome-wide significance (P < 5 × 10^-8) that map to 611 loci, of which 145 loci are, to our knowledge, previously unreported. We define eight non-overlapping clusters of T2D signals that are characterized by distinct profiles of cardiometabolic trait associations. These clusters are differentially enriched for cell-type-specific regions of open chromatin, including pancreatic islets, adipocytes, endothelial cells and enteroendocrine cells. We build cluster-specific partitioned polygenic scores⁵ in a further 279,552 individuals of diverse ancestry, including 30,288 cases of T2D, and test their association with T2D-related vascular outcomes. Cluster-specific partitioned polygenic scores are associated with coronary artery disease, peripheral artery disease and end-stage diabetic nephropathy across ancestry groups, highlighting the importance of obesity-related processes in the development of vascular outcomes. Our findings show the value of integrating multi-ancestry genome-wide association study data with single-cell epigenomics to disentangle the aetiological heterogeneity that drives the development and progression of T2D. This might offer a route to optimize global access to genetically informed diabetes care.

Fig. 1. Heat map of associations of 37 cardiometabolic phenotypes with 8 mechanistic clusters of index SNVs for T2D association signals.

Each column corresponds to a cluster. Each row corresponds to a cardiometabolic phenotype. The ‘temperature’ of each cell represents the z-score (aligned to the T2D risk allele) of association of the phenotype with index SNVs assigned to the cluster. *Phenotype is adjusted for body mass index.

Fig. 2. Heat map of cluster-specific enrichments of T2D associations for cell-type-specific regions of open chromatin derived from single-cell ATAC-seq peaks in adult and fetal tissue.

a, Cell types (222 types) from 30 human adult tissues and 15 human fetal tissues. b, Cell types (106 types) from the human brain. In each panel, columns represent mechanistic clusters. Each row represents a cell type that was significantly enriched (Bonferroni correction for the number of cell types) for T2D associations in at least one cluster (indicated by an asterisk). The ‘temperature’ of each cell defines the magnitude of the log fold enrichment. The liver and lipid metabolism cluster is not presented because it includes only three T2D association signals and the model parameter estimates were unstable.

Fig. 3. Associations of cluster-specific components of the partitioned PS with five T2D-related vascular outcomes in up to 279,552 individuals from multiple ancestry groups.

Summaries of the associations of each cluster-specific component of the partitioned PS with CAD, ischaemic stroke (IS), peripheral artery disease (PAD), ESDN and proliferative diabetic retinopathy (PDR). The height of each bar corresponds to the log-odds ratio (beta) per standard deviation of the PS, and the grey bar shows the 95% confidence interval. Analyses of T2D-related macrovascular complications (CAD, PAD and IS) were undertaken in all individuals, with adjustment for T2D status. Analyses of microvascular complications were undertaken in individuals with T2D only. *P < 0.05, nominal association; **P < 0.0063, Bonferroni correction for eight clusters. Exact P values are provided in Supplementary Table 21.

Extended Data Fig. 1. Axes of genetic variation separating GWASs of T2D across ancestry groups.

We used SNVs that were reported in all studies to construct a distance matrix of mean effect allele frequency differences between each pair of GWASs. We implemented multi-dimensional scaling of the distance matrix to principal components that represent axes of genetic variation. The first three axes of genetic variation (PC1, PC2 and PC3) from multi-dimensional scaling of the Euclidean distance matrix between populations are sufficient to separate GWASs from six ancestry groups: African American (AFA), East Asian (EAS), European (EUR), Hispanic (HIS), South African (SAF), and South Asian (SAS). Variance explained by each axis: PC1 90.7%; PC2 6.5%; PC3 1.0%.

Extended Data Fig. 2. Cluster-specific associations of index SNVs with defining cardiometabolic phenotypes.

Each bar presents the −log₁₀ P value for association, with effect direction aligned to the T2D risk allele. FG: fasting glucose. FI: fasting insulin. PI: proinsulin. BMI: body mass index. WHR: waist–hip ratio. LDL: low-density lipoprotein cholesterol. HDL: high-density lipoprotein cholesterol. TG: triglycerides. *Trait adjusted for BMI.

Extended Data Fig. 3. Cluster-specific associations of index SNVs with T2D.

The height of each bar corresponds to the log-odds ratio (beta), and the grey bar shows the 95% confidence interval. *P < 0.05, nominal association. **P < 0.0063, Bonferroni correction for eight clusters. Exact P values are presented in Supplementary Table 9.

Extended Data Fig. 4. Cluster-specific associations of T2D risk alleles at index SNVs with insulin-related endophenotypes.

Measures of insulin secretion and insulin sensitivity were derived from hyperinsulinaemic-euglycaemic clamp assessments and oral glucose tolerance tests in up to 1,316 Mexican American participants without diabetes. Homeostatic model assessment measures of beta-cell function (HOMA-B) and insulin resistance (HOMA-IR) were obtained from 36,466 non-diabetic individuals of European ancestry. Each point corresponds to the cluster-specific mean z-score for each trait, and grey bars represent 95% confidence intervals. The liver and lipid metabolism cluster has been removed for ease of presentation.

Extended Data Fig. 5. Cluster-specific associations of T2D risk alleles at index SNVs with insulin-resistance-related disorders.

Association with gestational diabetes mellitus (GDM) was assessed in 5,485 cases and 347,856 female controls of diverse ancestry. Association with polycystic ovary syndrome (PCOS) was assessed in 10,074 cases and 103,164 female controls of European ancestry. The height of each bar corresponds to the mean z-score, and the grey bar shows the 95% confidence interval. The liver and lipid metabolism cluster has been removed for ease of presentation. *P < 0.05, nominal association. **P < 0.0063, Bonferroni correction for eight clusters. Exact P values are presented in Supplementary Table 12.

Extended Data Fig. 6. Ancestry-correlated heterogeneity is driven by differences in allelic effect sizes between AFA, EAS and EUR ancestry groups.

In the scatter plot, index SNVs with significant evidence (P_HET < 3.9 × 10⁻⁵, Bonferroni correction for 1,289 signals) for ancestry-correlated heterogeneity are plotted according to their association (z-score) with the first two axes of genetic variation. The first axis represents differences in allelic effect sizes between AFA/EUR GWASs and EAS GWASs (AFA–EAS axis), whilst the second axis represents differences in effect size between AFA/EAS GWASs and EUR GWASs (AFA–EUR axis). The forest plots present examples of ancestry-correlated heterogeneity at index SNVs. In each forest plot, the allelic log-odds ratio (OR) from each ancestry group-specific fixed-effects meta-analysis is given by the black tick mark, the 95% confidence interval is given by the horizontal line, and the weight (inverse-variance) of each ancestry group by the grey box. AFA: African American ancestry group. EAS: East Asian ancestry group. EUR: European ancestry group. HIS: Hispanic ancestry group. SAF: South African ancestry group. SAS: South Asian ancestry group.

Extended Data Fig. 7. Cluster-specific associations of index SNVs with the first two axes of genetic variation in T2D cases and controls.

a, Unadjusted for BMI. b, Adjusted for study-level mean BMI. Each point corresponds to a cluster, plotted according to the mean z-score for association with the first two axes of genetic variation (PC1 and PC2) on the x axis and y axis, respectively. Grey bars correspond to 95% confidence intervals. The liver and lipid metabolism cluster has been removed for ease of presentation.

Extended Data Fig. 8. Associations of cluster-specific components of the partitioned PS with CAD in up to 279,552 individuals across diverse ancestry groups.

The panel summarizes the associations of each cluster-specific component of the partitioned PS with CAD, with and without adjustment for a previously published multi-ancestry CAD PS. The height of each bar corresponds to the log-OR (beta) per standard deviation of the PS, and the grey bar shows the 95% confidence interval. Analyses were undertaken in all individuals, with adjustment for T2D status. *P < 0.05, nominal association. **P < 0.0063, Bonferroni correction for eight clusters. Exact P values are presented in Supplementary Tables 21 and 22.

Extended Data Fig. 9. Associations of cluster-specific components of the partitioned PS with T2D age of onset in up to 30,288 individuals across diverse ancestry groups.

The panel summarizes the associations of each cluster-specific component of the partitioned PS with age of onset. The height of each bar corresponds to years (beta) per standard deviation of the PS, and the grey bar shows the 95% confidence interval. A negative effect corresponds to earlier age of onset. *P < 0.05, nominal association. **P < 0.0063, Bonferroni correction for eight clusters. Exact P values are presented in Supplementary Table 23.

Extended Data Fig. 10. Associations of the beta cell +PI and obesity cluster-specific components of the partitioned PS with vascular outcomes in up to 29,827 EUR individuals with T2D from six clinical trials from the TIMI Study Group.

Major cardiovascular event is defined as myocardial infarction, ischaemic stroke, or cardiovascular death. Major limb event is defined as acute limb ischaemia or peripheral revascularization. The height of each bar corresponds to the log-hazard ratio per standard deviation of the PS, and the grey bar shows the 95% confidence interval. *P < 0.05, nominal association. **P < 0.0063, Bonferroni correction for eight clusters. Exact P values are presented in Supplementary Table 24.