Urinary metabolomics provide insights into coronary artery disease in individuals with type 1 diabetes

Abstract

Background: Type 1 diabetes increases the risk of coronary artery disease (CAD). High-throughput metabolomics may be utilized to identify metabolites associated with disease, thus, providing insight into disease pathophysiology, and serving as predictive markers in clinical practice. Urine is less tightly regulated than blood, and therefore, may enable earlier discovery of disease-associated markers. We studied urine metabolomics in relation to incident CAD in individuals with type 1 diabetes.

Methods: We prospectively studied CAD in 2501 adults with type 1 diabetes from the Finnish Diabetic Nephropathy Study. 209 participants experienced incident CAD within the 10-year follow-up. We analyzed the baseline urine samples with a high-throughput targeted urine metabolomics platform, which yielded 54 metabolites. With the data, we performed metabolome-wide survival analyses, correlation network analyses, and metabolomic state profiling for prediction of incident CAD.

Results: Urinary 3-hydroxyisobutyrate was associated with decreased 10-year incident CAD, which according to the network analysis, likely reflects younger age and improved kidney function. Urinary xanthosine was associated with 10-year incident CAD. In the network analysis, xanthosine correlated with baseline urinary allantoin, which is a marker of oxidative stress. In addition, urinary trans-aconitate and 4-deoxythreonate were associated with decreased 5-year incident CAD. Metabolomic state profiling supported the usage of CAD-associated urinary metabolites to improve prediction accuracy, especially during shorter follow-up. Furthermore, urinary trans-aconitate and 4-deoxythreonate were associated with decreased 5-year incident CAD. The network analysis further suggested glomerular filtration rate to influence the urinary metabolome differently between individuals with and without future CAD.

Conclusions: We have performed the first high-throughput urinary metabolomics analysis on CAD in individuals with type 1 diabetes and found xanthosine, 3-hydroxyisobutyrate, trans-aconitate, and 4-deoxythreonate to be associated with incident CAD. In addition, metabolomic state profiling improved prediction of incident CAD.

Keywords: Cardiac complication; Coronary artery disease; Machine learning; Metabolomics; Network analysis; Oxidative stress; Survival modeling; Type 1 diabetes; Urine.

Fig. 1

Study design with the 10-year incident coronary artery disease (CAD) cohort

Fig. 2

Metabolome-wide survival analysis for 10-year coronary artery disease (CAD) risk in individuals with type 1 diabetes (N = 2501). Survival analyses with adjustment settings: 1. non-modifiable risk factors (age, sex, diabetes onset calendar year), 2. non-modifiable risk factors + kidney function (eGFR, albuminuria), 3. non-modifiable risk factors + kidney function + CAD risk factors (systolic blood pressure, LDL cholesterol, waist-to-height ratio, HbA_1c, smoking). Forest plot displays metabolite-CAD associations (p < 0.05). Darker colors indicate metabolome-wide significant associations (p < 0.00102)

Fig. 3

3-hydroxyisobutyrate (a) and xanthosine (b) quartile’s survival probability within the 10-year CAD risk Cox proportional hazard survival models adjusted for age, sex, diabetes onset calendar year, eGFR, albuminuria, HbA_1c, systolic blood pressure, LDL cholesterol, waist-to-height ratio, and smoking. Inserted are the hazard ratios for 10-year incident CAD, correspondingly adjusted, between individuals belonging to the higher metabolite quartiles (25–50%, 50–75%, 75–100%) and individuals belonging to the lowest quartile (0–25%); *statistically significant (p < 0.05)

Fig. 4

Metabolome-wide survival analysis for 10-year coronary artery disease (CAD) risk in individuals with type 1 diabetes and stratified into participants with and without albuminuria (i.e., albuminuria vs. normoalbuminuria). Adjusted for age, sex, diabetes onset calendar year, eGFR, systolic blood pressure, LDL cholesterol, waist-to-height ratio, HbA_1c, smoking. Forest plot displays metabolite-CAD associations, whenever significant in individuals with or without albuminuria (p < 0.05). Darker color indicates metabolome-wide significance (p < 0.00102). Metabolite-albuminuria interaction effects on 10-year CAD were non-significant (p > 0.05)

Fig. 5

Local centrality measures for the complete baseline metabolic correlation network (p < 2 × 10^–5, N_patient = 2501) (a), and the cases’ baseline metabolic correlation network (p < 2 × 10^–5, N_patient = 209) (b). Here, degree is the number of links leaving a node. Betweenness centrality describes the amount of non-weighted network shortest paths between other nodes passing through the node. 10-year coronary artery disease risk associated metabolites are highlighted with pink (p < 0.05). HPHPA: 3-(3-hydroxyphenyl)-3-hydroxypropionic acid

Fig. 6

Correlation coefficient difference network between individuals with and without 10-year incident coronary artery disease (CAD) (r_{case-to-control}, p_difference < 0.05, p_{case or control} < 2 × 10^–5) (a). Red link represents a more positive correlation in individuals with than in individuals without incident CAD, blue link contrarily, and the link weight represent the magnitude. In addition, links included in the difference network are represented separately for individuals with (b) and without incident CAD (c); such that a red link represents a positive correlation coefficient, and a blue link negative. Clinical variable nodes are colored as beige and metabolite nodes as pink, darker pink for metabolites associated with 10-year incident CAD (p < 0.05). 4-Hydroxyhippurate (4-HH), dimethylamine (DMA), trimethylamine N-oxide (TMAO), 2-hydroxyisobutyrate (2-HIB), 3-hydroxyhippurate (3-HH), 3-(3-hydroxyphenyl)-3-hydroxypropionic acid (HPHPA), 3-aminoisobutyrate (3-AIB), 1-methylnicotinamide (1-MNA), 4-deoxyerythronic acid (4-DEA), 3-methylhistidine (3-MH), 4-deoxythreonate (4-DTA), 3-hydroxyisovalerate (3-HIV), 3-hydroxyisobutyrate (3-HIB), waist-to-height ratio (WHtR), triglyceride (TG), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean arterial pressure (MAP)

Fig. 7

Coronary artery disease (CAD) prediction with support vector machine for 10-year incident CAD (a), and for 5-year incident CAD (b). Mean (± SD) area under receiver operating characteristic curve (AUC) across 10 bootstraps (i.e., replicate). CAD-associated NMR: Metabolites with missingness rate < 15% and associated with incident CAD after full adjustment (p < 0.05) (a/b: 3-hydroxyisobutyrate, xanthosine, 4-deoxythreonate, trans-aconitate, 4-hydroxyhippurate, arabinose, cis-aconitate, glucose, tyrosine, b: leucine). NMR: Metabolites with missingness rate < 15%. Clinical: sex, diabetes onset calendar year, diabetes onset age, age, diabetes duration, body-mass-index, systolic blood pressure, diastolic blood pressure, mean arterial pressure, triglyceride, total cholesterol, HDL cholesterol, LDL cholesterol, HbA_1c, eGFR, albuminuria