A systems view of type 2 diabetes-associated metabolic perturbations in saliva, blood and urine at different timescales of glycaemic control

Abstract

Aims/hypothesis: Metabolomics has opened new avenues for studying metabolic alterations in type 2 diabetes. While many urine and blood metabolites have been associated individually with diabetes, a complete systems view analysis of metabolic dysregulations across multiple biofluids and over varying timescales of glycaemic control is still lacking.

Methods: Here we report a broad metabolomics study in a clinical setting, covering 2,178 metabolite measures in saliva, blood plasma and urine from 188 individuals with diabetes and 181 controls of Arab and Asian descent. Using multivariate linear regression we identified metabolites associated with diabetes and markers of acute, short-term and long-term glycaemic control.

Results: Ninety-four metabolite associations with diabetes were identified at a Bonferroni level of significance (p < 2.3 × 10(-5)), 16 of which have never been reported. Sixty-five of these diabetes-associated metabolites were associated with at least one marker of glycaemic control in the diabetes group. Using Gaussian graphical modelling, we constructed a metabolic network that links diabetes-associated metabolites from three biofluids across three different timescales of glycaemic control.

Conclusions/interpretation: Our study reveals a complex network of biochemical dysregulation involving metabolites from different pathways of diabetes pathology, and provides a reference framework for future diabetes studies with metabolic endpoints.

Fig. 1

Workflow for the generation of the GGM. Starting with 2,178 metabolites and 2.3 million partial correlations, two steps were conducted. (a) Step 1: filtering on significant partial correlations (3,742) by removing metabolites with no significant correlation to any other metabolite, leaving 1,907 metabolites in the GGM network. (b) Step 2: filtering on metabolites nominally associated with type 2 diabetes (p < 0.05), i.e. 546 metabolites, resulted in 33 subnetworks containing at least three metabolites and covering 243 metabolites

Fig. 2

Venn diagram of metabolites specific to and overlapping with the three glycaemic control timescales

Fig. 3

Selected GGM subnetworks. (a) 1,5-AG subnet, (b) glycolysis–BCAA subnet, (c) urinary ketone body subnet, (d) carbohydrates subnet. Included are metabolites nominally associated with diabetes (p < 0.05); edges indicate significant partial correlations (2.1 × 10^–8) between two metabolites. Node size is proportional to the absolute β value in the regression analysis with diabetes. Node colour and shape denote the biofluid: white triangle, saliva; red circle, plasma; yellow diamond, urine; arrows indicate the direction of the association (upward, higher in diabetes; downward, lower in diabetes); star indicates an association with all three glycaemic timescales; number indicates an association with glucosuria (1), 1,5-AG (2) or HbA_1c (3). For metabolites that are only nominally associated with diabetes, no association with glycaemic control was tested.