Identification of plasma proteomic markers underlying polygenic risk of type 2 diabetes and related comorbidities

Abstract

Genomics can provide insight into the etiology of type 2 diabetes and its comorbidities, but assigning functionality to non-coding variants remains challenging. Polygenic scores, which aggregate variant effects, can uncover mechanisms when paired with molecular data. Here, we test polygenic scores for type 2 diabetes and cardiometabolic comorbidities for associations with 2,922 circulating proteins in the UK Biobank. The genome-wide type 2 diabetes polygenic score associates with 617 proteins, of which 75% also associate with another cardiometabolic score. Partitioned type 2 diabetes scores, which capture distinct disease biology, associate with 342 proteins (20% unique). In this work, we identify key pathways (e.g., complement cascade), potential therapeutic targets (e.g., FAM3D in type 2 diabetes), and biomarkers of diabetic comorbidities (e.g., EFEMP1 and IGFBP2) through causal inference, pathway enrichment, and Cox regression of clinical trial outcomes. Our results are available via an interactive portal ( https://public.cgr.astrazeneca.com/t2d-pgs/v1/ ).

Fig. 1. Study workflow.

The centre of the flowchart corresponds to the primary objectives or aims of the study (in Blue). Branching off each objective are short summaries of the analyses corresponding to that objective (in Red).

Fig. 2. Polygenic score (PGS) associations with circulating proteins.

A Volcano plot of PGS_{T2D_gw}-protein beta coefficients (obtained from linear regression) and the unadjusted -log₁₀ p-values (two-sided), with the colour indicating the magnitude of the -log10 p-values. Labelled proteins are among the top 1% in terms of variance (R²) explained by the PGS_{T2D_gw}. B Beta-beta plot of PGS_{T2D_gw} beta coefficients on circulating proteins with (y-axis) and without (x-axis) BMI adjustment. The diagonal is dashed grey, while the regression line is solid grey. Each point represents a protein; light blue points indicate replicated proteins that remained significant with the adjustment, red points indicate replicated proteins that were no longer significant after the adjustment, and dark blue points indicate proteins that did not significantly replicate prior to adjusting for BMI or pQTLs. C Pearson’s correlations of PGS beta coefficients from the regression on circulating protein levels. Red indicates pairs of PGS with positively correlated effect sizes; blue indicates negatively correlated effect sizes. “*” indicates correlations with a p-value < 0.05 and “**” indicates correlations with a p-value < 0.001 (a Bonferroni correction for 45 comparisons). P-values are unadjusted and two-sided t test as the test statistic follows a t distribution. D Bar plot indicating the overlap between proteins significantly associated with the T2D PGS and the other cardiometabolic PGS. The x-axis is the PGS label, and the y-axis is the percentage of PGS_{T2D_gw}-associated proteins that are also associated with another PGS (e.g., over 60% of proteins were also associated with the PGS_BMI). E Beta-beta plot of PGS_{T2D_gw} effect sizes on circulating proteins with (y-axis) and without (x-axis) pQTL adjustment, with the same definitions as panel (B) albeit for a pQTL adjustment.

Fig. 3. Two-sample Mendelian randomisation analysis in the UK Biobank.

A Type 2 diabetes (T2D) Mendelian randomisation (MR) with cis instruments for each protein as the exposure. B T2D MR with both cis and trans instruments for each protein as the exposure. C Cis colocalization using T2D and protein quantitative loci (pQTL) genome-wide association study (GWAS) information. D Coronary artery disease (CAD) MR with cis instruments for each protein as the exposure. E CAD MR with both cis and trans instruments for each protein as the exposure. F CAD colocalization using CAD and pQTL GWAS information. In the MR plots, four conventional MR methods are displayed (simple median, weighted median, IVW, MR-Egger), plus an additional MR method called MR-Link-2. All proteins displayed in this figure had a median p-value across the four conventional MR methods < 0.05 (FDR-adjusted) and no pleiotropy as detected by MR-Egger (MR-Egger intercept p-value > 0.05). For panels (A, B, D, and E), the points represent the causal estimate obtained by each MR method (in the log-odds scale) and error bars represent the 95% confidence interval calculated using the standard error for each MR estimate. Note that MR-Link-2 estimates are on a different scale than the other MR methods but show consistency in the direction of effect. Finally, “*” signifies proteins with colocalization evidence. For panels A through F, the exposure (protein) GWAS had a sample size of 34,557 European-ancestry UK Biobank participants, and the outcome GWAS (T2D, CAD) had a sample size of 409,048 non-overlapping European-ancestry UK Biobank participants.

Fig. 4. Association of proteins with clinical trial outcomes and survival analyses.

A Summary of results in EXSCEL, with the x-axis corresponding to the three different models used (see Methods) and the y-axis corresponding to the number of proteins significant for each of the three outcomes (Bonferroni p-value < 0.05). B Summary of the replication results in DECLARE, with the x-axis corresponding to the three different models used and the y-axis corresponding to the number of proteins significant for each of the three outcomes (FDR p-value < 0.05). C Results from the survival analysis of the study-specific renal outcome in the placebo arms in EXSCEL and DECLARE. All displayed proteins replicated in DECLARE for the base model (with age, sex, age², age*sex, and genetic PCs 1–10 as covariates). D Results from the survival analysis of the major adverse cardiovascular event (MACE) outcome in the placebo arms in EXSCEL and DECLARE. All displayed proteins replicated in DECLARE for the base model. E Results from the survival analysis of the hospitalisation for heart failure (HHF) outcome in the placebo arms in EXSCEL and DECLARE. For panels (C–E), the points represent the hazard ratios and error bars represent the 95% confidence interval obtained using the standard error for each hazard ratio. Note that all displayed proteins replicated in DECLARE for the base model. In panels (C–E), EXSCEL had a sample size of 1407 study participants from the placebo arm with available proteomics information, while DECLARE had a sample size of 497 study participants from the placebo arm with available proteomics information.

Fig. 5. IGF regulation by IGFBPs pathway.

A PGS associations with IGF binding proteins. A single asterisk (*) indicates the association was nominally significant, while two (**) indicates significance using FDR and three (***) indicates significance using a Bonferroni correction from linear regression of circulating protein levels in the UK Biobank. B Associations of proteins in this pathway with clinical trial outcomes in EXSCEL using Cox proportional hazards regression, adjusting for demographic covariates and clinical risk factors (see Methods). C Associations of proteins in this pathway with clinical trial outcomes in DECLARE using Cox proportional hazards regression, adjusting for demographic covariates and clinical risk factors (see Methods). For panels B and C, the dashed line corresponds the p-value threshold where FDR < 5% (when applied to the proteins in this pathway). Note that panels (B and C) display unadjusted, two-sided p-values obtained from Cox proportional hazards regression.