Circulating metabolomic markers linking diabetic kidney disease and incident cardiovascular disease in type 2 diabetes: analyses from the Hong Kong Diabetes Biobank

Chronic kidney disease (CKD) develops in approximately 40% of people with type 2 diabetes [1] and is associated with increased risk of CVD and mortality [2]. Diabetes is associated with two- to fourfold increased risk of CVD [3], whereas higher CVD incidence was reported in people with CKD than in diabetes [4], suggesting particularly high CVD risk in diabetic kidney disease (DKD). Despite multifactorial management and agents with pleiotropic cardiorenal benefits, DKD prognosis remains poor.

Type 2 diabetes is characterised by atherogenic dyslipidaemia: elevated triglyceride-rich lipoproteins (TRLs) and reduced HDL, contributing to substantial residual risk despite optimal LDL-cholesterol (LDL-C) levels [5]. In CKD, TRLs are increased owing to impaired lipoprotein lipase activities and diminished clearance caused by altered apolipoprotein C-3 (ApoC-3) metabolism [6]. The major structural protein of TRLs, apolipoprotein B (ApoB), can flux across endothelium and be trapped in the artery wall, initiating atherosclerosis by releasing cholesterol to macrophages [7]. Beyond lipids, the kidney also can regulate circulating metabolites via filtration, reabsorption, secretion, catabolism and anabolism [8]. With advances in technologies, metabolites can be quantified simultaneously in a high-throughput manner and multiple metabolites have been associated with DKD [8‐10].

Higher TRLs, ApoB, phenylalanine, inflammation markers and lower HDL and apolipoprotein A-1 (ApoA-1) have been associated with decreased eGFR in people with type 2 diabetes [9], and replicated in a larger study [10], indicating that altered lipoprotein and metabolic profiles may reflect impaired kidney function in diabetes. Furthermore, TRLs, ApoB and phenylalanine have been associated with CVD in people with CKD or type 2 diabetes [11‐13], suggesting that the altered metabolome in DKD may partly explain the increased CVD risk. Although the causal relation between the metabolites and CVD in people with DKD is not yet fully understood, Mendelian randomisation studies have suggested TRLs and ApoB are causally associated with CVD [14, 15]; phenylalanine has been associated with type 2 diabetes [16], impaired kidney function [9], heart failure [17] and CVD [18] in large cohort studies. Better understanding of the potential metabolic links between DKD and CVD is therefore warranted.

Herein, we investigated the metabolomic signature of DKD and examined its association with incident CVD in a well-characterised prospective cohort of individuals with type 2 diabetes. Metabolomic biomarkers were selected among metabolites associated with CVD and were evaluated for their prognostic value towards CVD prediction. External validation of the identified biomarkers for incident CVD was performed in Chinese [19] and Dutch cohorts [20].

Applying NMR metabolomics in a well-characterised type 2 diabetes cohort, we comprehensively examined the cross-sectional associations of lipoproteins, lipids and LMWMs with DKD and prospective associations of DKD-related metabolites with incident CVD, identified and assessed metabolomic biomarkers for incident CVD prediction. We found that: (1) TRLs associated with both DKD and incident CVD; (2) HDL inversely associated with DKD and the inverse association with incident CVD appeared mainly driven by smaller (medium and small) HDL; (3) triglycerides across all lipoproteins associated with CVD; and (4) replicated in both Chinese and Europeans, metabolomic biomarkers performed comparably to conventional risk factors and improved CVD risk stratification beyond established prediction models. The results demonstrate profound metabolomic alterations in DKD and close relation with development of CVD, highlighting potential molecular links between DKD and CVD and potential application of metabolomics for diabetes complication prediction.

Some metabolic alterations associated with decreased eGFR are common across different populations and we further identified metabolites associated with severely increased albuminuria in Chinese (ESM Tables 12–13). Consistently, TRLs were associated with decreased eGFR [9, 10] and also with severely increased albuminuria in our study. TRLs have been associated with CVD [31]; larger differences in TRLs in decreased eGFR were found in people with vs without diabetes [10], suggesting a potential role of TRLs for residual CVD risk in people with DKD. In our prospective analysis, TRLs were associated with incident CVD, with VLDL exhibiting the strongest association, although VLDL, IDL and ApoB were all associated with CVD. Hepatic VLDL production and secretion is increased by insulin resistance [32] and altered metabolism of ApoC-3 in CKD further elevates TRLs by overproduction and impaired clearance [6]. All ApoB-containing lipoproteins, including TRLs, can enter the arterial intima leading to cholesterol deposition [7]. In contrast to LDL for which oxidative modification is usually required before phagocytosis, larger TRLs can be trapped more easily and can be directly phagocytised by macrophages to form foam cells [33]. Moreover, hydrolysis of triglycerides in TRLs by lipoprotein lipase can liberate NEFA, inducing inflammation, promoting atherosclerosis [34].

CKD modifies HDL structure and composition, which may partly explain the increased CVD risk in CKD [35]. Consistent with previous findings [9], we found that HDL was negatively associated with CKD and severely increased albuminuria; the association with CKD was stronger. HDL was inversely associated with CVD in our prospective analysis and the association appeared limited to medium and small HDL. However, in previous population-based studies the inverse association between HDL and CVD was limited to large and medium HDL [18, 25]. HDL’s potential modification by diabetes [36] and CKD [35] may partly explain the contrasting results. Furthermore, a recent MR analysis found that medium and small HDL were CVD-protective [37]. Our observed association of small HDL appeared independent of DKD, which is consistent with findings that small HDL has greater atheroprotective capacities via reverse cholesterol transport, anti-inflammatory, antioxidant and endothelial protection [8, 24]. Further studies are warranted to investigate whether detailed HDL composition (proteins, lipids or enzymes) or HDL function may be potential modulators [35]. We replicated previous findings that LDL was associated with DKD [9, 10] and that small LDL was associated with higher CVD risk.

Triglycerides across all lipoproteins were associated with DKD and incident CVD, including TRLs, LDL and HDL. Despite the fact that 67% of participants were on statins, and cholesterol in LDL was not associated with incident CVD, triglycerides in LDL were associated with CVD in our analysis. In people with prediabetes (impaired glucose tolerance and/or impaired fasting glucose) or diabetes and stable coronary artery disease (73.9% on statins), LDL triglycerides were associated with CVD and improved CVD risk prediction, indicating the prognostic value of LDL triglycerides for residual risk [38].

Lower albumin has been associated with DKD [10] and frailty in older people with type 2 diabetes [39] and albumin levels are inversely associated with CVD or mortality in people with CKD [40], suggesting that as a marker linked with malnutrition, liver and kidney dysfunction and inflammation, albumin may partly capture the integrated altered metabolic signature in diabetes and thus associates with adverse outcomes. As a validated marker for systemic inflammation, GlycA was associated with DKD [9, 10] and incident CVD [18, 25], although further adjustment for DKD attenuated the association with CVD. Taken together, our findings suggest that low-grade inflammation in diabetes may be one of the pathogenetic pathways for diabetes complications.

Other lipids, including sphingomyelins, were also associated with DKD [9, 10], however, none were associated with incident CVD, consistent with previous findings that sphingomyelins were associated with DKD but not CVD [41]. In line with previous studies, MUFAs were positively associated with DKD [10], however, PUFAs were negatively linked with DKD in our analysis and the inverse association was mainly driven by DHA. DHA was inversely associated with macrovascular events in the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) study [42], which supports our result that DHA as a marker negatively linked with DKD was associated with lower risk of CVD. In CKD, dysfunctional activity of phenylalanine hydroxylase impairs the conversion of phenylalanine to tyrosine [43]. Accordingly, we found that higher phenylalanine and lower tyrosine were associated with DKD [9, 10] and were both associated with risk of CVD. Similar to findings from ADVANCE [11], further adjustment for kidney function attenuated associations with CVD, suggesting that the link between dysregulated phenylalanine or tyrosine and CVD may be mediated by kidney dysfunction. Leucine and isoleucine have been associated with decreased eGFR [9], and branched-chain amino acids (BCAAs) have been negatively associated with CKD in a larger study [10]. We found that leucine and valine were negatively associated with CKD and isoleucine was positively associated with severely increased albuminuria. Leucine and valine were also inversely associated with CVD, in line with the inverse association of leucine and valine with all-cause mortality in ADVANCE [11]. The different associations between BCAAs and DKD across studies might be attributed to participant characteristics, dietary intake, medications or analytical strategies.

Integrating information from gene expression and environmental factors and interacting with the microbiome, metabolites may carry molecular information that is not captured by traditional risk factors [8]. Among metabolites associated with CVD independent of conventional risk factors, three metabolites, albumin, triglycerides in large HDL and phospholipids in small LDL, were identified to be most informative for CVD prediction by machine learning method. The metabolite score comprising these three metabolites was strongly associated with CVD, which was validated independently in both Chinese and European cohorts. The selected metabolites performed comparably to conventional risk factors for CVD prediction and improved risk stratification beyond well-established prediction models, highlighting the prognostic value of metabolomic biomarkers for diabetes complications.

Extending the cross-sectional associations between metabolites and DKD, we found some DKD-related metabolites were associated with incident CVD. We further replicated the association between the identified metabolites and incident CVD in HKDR and the Dutch DCS cohort. Other strengths include the extensively phenotyped data and complete follow-up, well-established metabolomics platform with stringent quality control and consistent results across sensitivity analyses. Nevertheless, there are limitations. Only Chinese individuals were included in the discovery analysis, which might limit generalisability of our findings, however, most metabolites associated with DKD in previous studies were replicated in our study and the selected metabolomic biomarkers were validated in two independent cohorts. Around 70% of participants were on lipid-lowering drugs and we could not account for their potential influence on lipoprotein metabolism, although medication use was accounted for and our findings were consistent with a study in people not on lipid therapies [10]. UACR was based on single measurement and to account for intra-individual variability, we used severely increased albuminuria to define albuminuria. Among metabolites ranked by 1000-times bootstrapping priority-Lasso, an arbitrary cut-off (>70%) was applied to select prognostic metabolomic biomarkers. Although fasting samples were profiled, dietary intake and physical activity that may modulate the metabolome [8] were not captured in our cohort. Given the observational design, residual confounding cannot be ruled out and causal inference is not feasible. Although the study population included slightly more men than women, analyses have been adjusted for the sex of the study participants, and the findings should be applicable to both men and women with diabetes.

In conclusion, DKD is linked with alterations in multiple metabolites, including TRLs, HDL, fatty acids, amino acids, albumin and inflammation. Some DKD-related metabolites (TRLs, smaller HDL, leucine and albumin) are also associated with incident CVD. Metabolomic biomarkers provided comparable predictive utility to traditional risk factors and improved CVD risk stratification over established prediction models. Further investigations on pathophysiology and disease prediction of metabolites are warranted.