Untargeted metabolomic profiling identifies serum metabolites associated with type 2 diabetes in a cross-sectional study of the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study

Abstract

Type 2 diabetes (T2D) is a complex chronic disease with substantial phenotypic heterogeneity affecting millions of individuals. Yet, its relevant metabolites and etiological pathways are not fully understood. The aim of this study is to assess a broad spectrum of metabolites related to T2D in a large population-based cohort. We conducted a metabolomic analysis of 4,281 male participants within the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study. The serum metabolomic analysis was performed using an LC-MS/GC-MS platform. Associations between 1,413 metabolites and T2D were examined using linear regression, controlling for important baseline risk factors. Standardized β-coefficients and standard errors (SEs) were computed to estimate the difference in metabolite concentrations. We identified 74 metabolites that were significantly associated with T2D based on the Bonferroni-corrected threshold (P < 3.5 × 10⁻⁵). The strongest signals associated with T2D were of carbohydrates origin, including glucose, 1,5-anhydroglucitol (1,5-AG), and mannose (β = 0.34, −0.91, and 0.41, respectively; all P < 10⁻⁷⁵). We found several chemical class pathways that were significantly associated with T2D, including carbohydrates (P = 1.3 × 10⁻¹¹), amino acids (P = 2.7 × 10⁻⁶), energy (P = 1.5 × 10⁻⁴), and xenobiotics (P = 1.2 × 10⁻³). The strongest subpathway associations were seen for fructose-mannose-galactose metabolism, glycolysis-gluconeogenesis-pyruvate metabolism, fatty acid metabolism (acyl choline), and leucine-isoleucine-valine metabolism (all P < 10⁻⁸). Our findings identified various metabolites and candidate chemical class pathways that can be characterized by glycolysis and gluconeogenesis metabolism, fructose-mannose-galactose metabolism, branched-chain amino acids, diacylglycerol, acyl cholines, fatty acid oxidation, and mitochondrial dysfunction.

NEW & NOTEWORTHY These metabolomic patterns may provide new additional evidence and potential insights relevant to the molecular basis of insulin resistance and the etiology of T2D.

INTRODUCTION

Type 2 diabetes mellitus (T2D) constitutes over 90% of diabetes cases, results from insulin resistance and insufficient compensatory insulin secretion, and currently poses a substantial public health burden (1). It affects millions of individuals with a global prevalence of 9% and has been projected to increase to 10.4% (642 million individuals) by 2040, with its rates of mortality and disability continuing to rise (2, 3).

Over the past two decades, advanced technologies enabled the characterization and quantification of biochemical metabolites in biospecimens. Such metabolomic profiling offers an overall reflection of the substrates, end products, and metabolic responses from exogenous and endogenous exposures, and facilitates the identification of novel biomarkers, biochemical processes, and impaired pathways involved in disease pathogenesis. Previous metabolomic analyses identified numerous circulating metabolites related to the T2D, including branched-chain amino acids (BCAA), aromatic amino acids, phospholipids, acylcarnitines, glycine, lysophospholipids, and sphingomyelins (4–6). These studies were generally limited by their ability to measure only 50–400 metabolites, whereas additional analyses encompassing a broader array of biochemicals would capture deeper molecular signatures of more complex pathways and afford new windows to identify novel disease associations. Such a comprehensive investigation of novel biomarkers having relevance to metabolic impairment may enhance our understanding of disease etiology and pathophysiology and facilitate targeted prevention and interventions.

To address the disrupted metabolites and biological pathways involved in T2D, we conducted a large cross-sectional serological analysis based on 1,413 measured and identified metabolites in 4,281 individuals nested within a prospective study.

MATERIALS AND METHODS

The Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) study was a randomized controlled primary prevention trial that evaluated whether α-tocopherol or β-carotene supplementation could decrease cancer incidence (7). The study enrolled 29,133 male Finnish smokers aged 50 to 69 who were provided 50-mg α-tocopherol, 20-mg β-carotene, both vitamins, or placebo for a median of 6.1 yr through the end of the trial on April 30, 1993. Information regarding behavioral, lifestyle, and disease risk factors was obtained via self-administrated questionnaires at study entry, and height, weight, blood pressure, and heart rate were measured by skilled research nurses. Baseline overnight fasting blood samples were also obtained and stored at −70°C until assayed, and written informed consent was collected from all participants. The ATBC Study was approved by institutional review boards at the Finnish National Public Health Institute and the US National Cancer Institute and performed in compliance with human subject guidelines and regulations.

The present cross-sectional metabolomic serological analysis is based on participants from several substudies nested in the ATBC Study (subsequently referred to as “metabolomic sets”) (8–15). After removing participants with more than one measurement, a total of 4,281 men with diabetes mellitus data were included in the final analysis, with 135 participants with self-reported, physician-diagnosed T2D.

Metabolite Assessment

Fasting serum metabolites were assayed by Metabolon, Inc. (Durham, NC) using high-resolution, accurate-mass ultrahigh-performance liquid chromatography (LC)-tandem mass spectroscopy (MS/MS) and gas chromatography (GC)-MS/MS, including sample accessioning/preparation, quality control, data extraction, and compound identification (8–10, 16). Raw MS data were processed for peak identification and quality control, with compounds curated from the Metabolon metabolite database of purified standards. A total of 2,196 metabolites remained after excluding those with more than 10 missing values in each metabolomic set. A total of 1,413 identified metabolites were included in the final analysis and were categorized as one of the following mutually exclusive chemical classes: amino acids and amino acid derivatives, carbohydrates, cofactors and vitamins, energy metabolism, lipids, nucleotides, peptides, and xenobiotics. The median and interquartile range of the coefficients of variation and intraclass correlation coefficients across the metabolites was 9% (4%–20%) and 0.85 (0.60–0.96), respectively.

Statistical Methods

Each metabolite was batch-normalized by dividing by batch median value. The levels of each metabolite were then processed through log transformation and normalization for analyses. We conducted multivariable-linear regression to assess the association between each metabolite and T2D, adjusting for age at baseline, body mass index (BMI), cigarettes smoked per day, metabolomic set, and cancer case-control status. Standardized β-coefficients and their corresponding standard errors were calculated from each model, with the standardized β-coefficient representing the difference in T2D status (yes or no) for a given metabolite (Y_{metabolite level} ∼ β₁X_{T2D status} + β₂X_{age at baseline} + β₃X_{body mass index} + β₄X_{cigarettes smoking per day} + β₅X_{metabolomic set} + β₆X_{case-control status}). The Bonferroni correction was applied to control for multiple comparisons (P < 3.5 × 10⁻⁵, across tests for 1,413 metabolites) (2, 17).

We performed metabolic pathway gene set analysis (GSA) to examine whether the predefined chemical class superpathways and subpathways were associated with risk of T2D within each of the metabolic sets (18). For Z values (Z₁, …, Z_k) tested from K metabolites in a predefined chemical class pathway, the GSA test examined the “maxmean” statistic maximum that is the mean of all Z values (both positive and negative) and computed the P values as the proportion of the 10,000 permutations. The P values from each predefined chemical pathway were combined using Fisher’s method, namely, the sum of logs, with all metabolomic sets (Bonferroni-corrected thresholds: P < 6.3 × 10⁻³ and P < 4.7 × 10⁻⁴, across tests for 8 superpathways and 106 subpathways).

Statistical tests and P values were all two-sided. The analyses were conducted using SAS 9.4 (SAS Institute, Inc., Cary, North Carolina) and R, version 4.2.0 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Baseline characteristics of the 4,281 study participants in this analysis are presented in Table 1. Compared with non-T2D subjects, T2D cases had a higher weight and BMI (both P < 0.05), with no other statistically significant differences between the two groups, including age, height, smoking intensity and duration, or dietary components (Table 1).

Table 1.

Selected baseline characteristics of 4,281 Finnish men in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study

	Self-Reported Diabetes: Yes	Self-Reported Diabetes: No
n	135	4,146
Age at blood collection, yr	57.9 (4.8)	57.8 (5.1)
Height, cm	174.2 (6.4)	173.6 (6.2)
Weight, kg	85.3 (15.1)	79.1 (12.4)
BMI, kg/m2	28.0 (4.2)	26.2 (3.6)
Cigarettes per day	21 (9.8)	20 (8.5)
Duration of smoking, yr	35.7 (9.6)	35.9 (8.7)
Dietary intake per day
Total energy, kcal	2613 (765)	2702 (741)
Total protein, g	100.7 (29.1)	94.9 (26.2)
Total fat, g	121.6 (42.0)	123.9 (40.4)
Fruit, g	153.1 (115.0)	130.7 (101.5)
Vegetables, g	124.4 (73.1)	114.5 (70.3)
Red meat, g	74.7 (37.8)	70.9 (32.7)
Alcohol, g	17.0 (20.9)	16.6 (20.1)

Values are means (SD). BMI, body mass index.

In the multivariable-adjusted linear regression models, 74 serum metabolites were significantly associated with T2D at the Bonferroni-corrected threshold (P < 3.5 × 10⁻⁵, Table 2). Of these 74 metabolites, 43 were positively associated with T2D and 31 were inversely associated, including 34 lipids, 20 amino acids, 7 carbohydrates, 7 xenobiotics, 1 energy metabolite, 1 cofactor and vitamin metabolite, and 1 peptide (Table 2, Fig. 1). As expected, the strongest association with T2D was seen for glucose (β = 0.34, SE = 0.013, P = 1.76 × 10⁻¹⁴⁶), followed by the carbohydrates 1,5-AG (β = −0.91, SE = 0.04, P = 1.12 × 10⁻¹¹¹) and mannose (β = 0.41, SE = 0.022, P = 4.1 × 10⁻⁷⁶), the lipids stearoylcholine (β = −0.74, SE = 0.051, P = 5.42 × 10⁻⁴⁷), linoleoylcholine (β = −0.70, SE = 0.051, P = 8.97 × 10⁻⁴⁴), oleoylcholine (β = −0.72, SE = 0.059, P = 7.23 × 10⁻³⁴), and palmitoylcholine (β = −0.65, SE = 0.054, P = 1.90 × 10⁻³⁰), and amino acids β-hydroxypyruvate (β = 0.45, SE = 0.051, P = 4.11 × 10⁻¹⁹), pyroglutamine (β = −0.29, SE = 0.044, P = 2.14 × 10⁻¹¹), 2-hydroxybutyrate, 3-hydroxyisobutyrate, 1-carboxyethylleucine, isoleucine, leucine, and valine (β = 0.066 to 0.36, all P < 7.43 × 10⁻⁸) (Table 2). To investigate whether metabolite-T2D associations were possibly influenced by subclinical cancers, we conducted sensitivity analyses that excluded cancer cases diagnosed after blood collection and found the primary metabolite associations largely unchanged (n = 1,402, Supplemental Table S1 and Supplemental Fig. S1).

Table 2.

Metabolite associations with T2D at 3.5 × 10⁻⁵ level of statistical significance in the ATBC Study*†

Metabolite	Chemical Subclass	No. of Sets	Sample Size	P Value	Estimate	Standard Error	95% CI
Amino Acid
Pyroglutamine	Glutamate metabolism	10	4,274	2.14 × 10−11	−0.294	0.044	(−0.380, −0.208)
2-Aminobutyrate	Glutathione metabolism	10	4,274	2.44 × 10−8	0.135	0.024	(0.087, 0.182)
2-Hydroxyisobutyrate	Glutathione metabolism	3	2,749	9.39 × 10−6	0.204	0.046	(0.114, 0.295)
β-Hydroxypyruvate	Glycine, serine, and threonine metabolism	5	1,044	4.11 × 10−19	0.454	0.051	(0.355, 0.554)
Glycine	Glycine, serine, and threonine metabolism	10	4,274	9.11 × 10−7	−0.083	0.017	(−0.116, −0.050)
2-Hydroxybutyrate (AHB)	Leucine, isoleucine, and valine metabolism	7	1,525	3.40 × 10−10	0.321	0.051	(0.221, 0.421)
3-Hydroxyisobutyrate	Leucine, isoleucine, and valine metabolism	10	4,274	7.83 × 10−10	0.200	0.032	(0.136, 0.263)
1-Carboxyethylleucine	Leucine, isoleucine, and valine metabolism	5	3,230	2.22 × 10−9	0.361	0.060	(0.243, 0.480)
Isoleucine	Leucine, isoleucine, and valine metabolism	10	4,274	3.36 × 10−9	0.078	0.013	(0.052, 0.104)
Leucine	Leucine, isoleucine, and valine metabolism	10	4,274	3.73 × 10−9	0.081	0.014	(0.054, 0.108)
1-Carboxyethylvaline	Leucine, isoleucine, and valine metabolism	7	3,602	1.52 × 10−8	0.239	0.042	(0.157, 0.322)
1-Carboxyethylisoleucine	Leucine, isoleucine, and valine metabolism	5	3,230	4.45 × 10−8	0.363	0.066	(0.233, 0.493)
Valine	Leucine, isoleucine, and valine metabolism	10	4,274	7.43 × 10−8	0.066	0.012	(0.042, 0.090)
3-Methyl-2-oxovalerate	Leucine, isoleucine, and valine metabolism	10	4,274	3.39 × 10−7	0.128	0.025	(0.079, 0.177)
3-Methyl-2-oxobutyrate	Leucine, isoleucine, and valine metabolism	10	4,274	2.83 × 10−6	0.110	0.024	(0.064, 0.156)
3-Dehydrocarnitine	Lysine metabolism	5	3,230	2.36 × 10−6	0.158	0.033	(0.092, 0.224)
2-Aminoadipate	Lysine metabolism	10	4,274	1.44 × 10−5	0.128	0.029	(0.070, 0.185)
α-Ketobutyrate	Methionine, cysteine, SAM, and taurine metabolism	8	4,031	2.10 × 10−5	0.295	0.069	(0.159, 0.430)
1-Carboxyethylphenylalanine	Phenylalanine metabolism	9	4,169	4.52 × 10−9	0.213	0.036	(0.142, 0.284)
1-Carboxyethyltyrosine	Tyrosine metabolism	4	3,105	5.41 × 10−7	0.316	0.063	(0.192, 0.440)
Carbohydrate
Mannose	Fructose, mannose, and galactose metabolism	10	4,274	4.10 × 10−76	0.408	0.022	(0.365, 0.452)
Fructose	Fructose, mannose, and galactose metabolism	10	4,274	1.91 × 10−29	0.322	0.029	(0.266, 0.378)
Sorbitol	Fructose, mannose, and galactose metabolism	3	925	3.03 × 10−8	0.532	0.096	(0.344, 0.720)
Glucose	Glycolysis, gluconeogenesis, and pyruvate metabolism	10	4,274	1.76 × 10−146	0.337	0.013	(0.311, 0.362)
1,5-Anhydroglucitol (1,5-AG)	Glycolysis, gluconeogenesis, and pyruvate metabolism	10	4,274	1.12 × 10−111	−0.909	0.040	(−0.988, −0.830)
Ribonate	Pentose metabolism	4	2,915	1.72 × 10−8	0.219	0.039	(0.143, 0.295)
Ribitol	Pentose metabolism	9	4,169	8.94 × 10−7	0.156	0.032	(0.094, 0.218)
Cofactors and Vitamins
Carotene diol	Vitamin A metabolism	3	2,749	3.14 × 10−6	−0.231	0.049	(−0.328, −0.134)
Energy
α-Ketoglutarate	TCA cycle	10	4,274	1.35 × 10−6	0.199	0.041	(0.118, 0.279)
Lipid
Palmitoyl-sphingosine-phosphoethanolamine (D18:1/16:0)	Ceramide PEs	3	2,749	1.46 × 10−7	−0.132	0.025	(−0.182, −0.083)
Oleoyl-linoleoyl-glycerol (18:1/18:2) [2]	Diacylglycerol	3	2,749	1.28 × 10−5	0.324	0.074	(0.179, 0.470)
Palmitoyl-linoleoyl-glycerol (16:0/18:2)	Diacylglycerol	3	2,749	1.48 × 10−5	0.512	0.118	(0.281, 0.744)
Oleoyl-linoleoyl-glycerol (18:1/18:2) [1]	Diacylglycerol	3	2,749	2.90 × 10−5	0.332	0.079	(0.176, 0.487)
Myristoyl dihydrosphingomyelin (d18:0/14:0)	Dihydrosphingomyelins	3	2,749	2.68 × 10−5	−0.144	0.034	(−0.211, −0.077)
N-stearoylserine	Endocannabinoid	2	2,624	5.25 × 10−13	−0.367	0.051	(−0.467, −0.268)
N-Palmitoylserine	Endocannabinoid	2	2,624	8.50 × 10−9	−0.304	0.053	(−0.408, −0.201)
N-Oleoylserine	Endocannabinoid	2	2,624	1.74 × 10−7	−0.230	0.044	(−0.316, −0.144)
Hydroxybutyrylcarnitine	Fatty acid metabolism (acyl carnitine, hydroxy)	10	4,274	1.63 × 10−6	0.345	0.072	(0.204, 0.487)
Stearoylcholine	Fatty acid metabolism (acyl choline)	10	4,274	5.42 × 10−47	−0.735	0.051	(−0.836, −0.635)
Linoleoylcholine	Fatty acid metabolism (acyl choline)	10	4,274	8.97 × 10−44	−0.703	0.051	(−0.803, −0.604)
Oleoylcholine	Fatty acid metabolism (acyl choline)	3	2,749	7.23 × 10−34	−0.721	0.059	(−0.837, −0.604)
Palmitoylcholine	Fatty acid metabolism (acyl choline)	3	2,749	1.90 × 10−33	−0.652	0.054	(−0.758, −0.546)
Palmitoloelycholine	Fatty acid metabolism (acyl choline)	3	2,749	5.35 × 10−29	−0.842	0.075	(−0.989, −0.694)
Arachidonoylcholine	Fatty acid metabolism (acyl choline)	3	2,749	2.30 × 10−28	−0.632	0.057	(−0.744, −0.520)
Dihomo-linolenoyl-choline	Fatty acid metabolism (acyl choline)	3	2,749	8.22 × 10−25	−0.657	0.064	(−0.782, −0.532)
Docosahexaenoylcholine	Fatty acid metabolism (acyl choline)	3	2,749	1.06 × 10−23	−0.636	0.063	(−0.76, −0.511)
Eicosapentaenoylcholine	Fatty acid metabolism (acyl choline)	3	2,749	1.10 × 10−21	−0.836	0.087	(−1.007, −0.665)
1-Oleoylglycerophosphocholine	Lysophospholipid	10	4,274	1.41 × 10−6	−0.117	0.024	(−0.164, −0.069)
1-(1-Enyl-Palmitoyl)-GPC (P-16:0)	Lysoplasmalogen	3	2,749	1.12 × 10−8	−0.216	0.038	(−0.290, −0.142)
1-Stearoylplasmenylethanolamine	Lysoplasmalogen	5	3,202	1.00 × 10−5	−0.174	0.039	(−0.252, −0.097)
3-Hydroxy-3-methylglutarate	Mevalonate metabolism	5	3,230	2.87 × 10−7	0.166	0.032	(0.102, 0.229)
1-palmitoylglycerol (1-monopalmitin)	Monoacylglycerol	10	4,274	6.22 × 10−7	0.223	0.045	(0.135, 0.310)
1-oleoylglycerol (1-monoolein)	Monoacylglycerol	9	4,232	2.37 × 10−6	0.249	0.053	(0.145, 0.352)
1-Stearoyl-2-linoleoyl-GPE (18:0/18:2)	Phosphatidylethanolamine (PE)	3	2,749	2.05 × 10−6	0.271	0.057	(0.159, 0.384)
1-Stearoyl-2-oleoyl-GPE (18:0/18:1)	Phosphatidylethanolamine (PE)	3	2,749	3.83 × 10−6	0.285	0.062	(0.164, 0.405)
1-Stearoyl-2-arachidonoyl-GPE (18:0/20:4)	Phosphatidylethanolamine (PE)	3	2,749	1.16 × 10−5	0.195	0.044	(0.108, 0.282)
Glycerophosphoinositol	Phospholipid metabolism	3	2,749	4.34 × 10−13	−0.495	0.068	(−0.628, −0.361)
1-(1-enyl-palmitoyl)-2-palmitoleoyl-GPC (P-16:0/16:1)	Plasmalogen	3	2,749	1.08 × 10−8	−0.189	0.033	(−0.253, −0.124)
1-(1-enyl-palmitoyl)-2-oleoyl-GPC (P-16:0/18:1)	Plasmalogen	3	2,749	1.95 × 10−8	−0.161	0.029	(−0.217, −0.105)
Glyco-β-muricholate	Primary bile acid metabolism	5	3,230	1.41 × 10−5	0.581	0.134	(0.319, 0.843)
Glycocholate	Primary bile acid metabolism	10	4,274	3.25 × 10−5	0.385	0.093	(0.204, 0.567)
Taurine	Secondary bile acid metabolism	10	4,274	1.96 × 10−6	−0.134	0.028	(−0.190, −0.079)
Sphingomyelin (D18:2/23:1)	Sphingomyelins	3	2,749	2.21 × 10−6	−0.165	0.035	(−0.234, −0.097)
Nucleotide
Pseudouridine	Pyrimidine metabolism, uracil containing	10	4,274	3.24 × 10−6	−0.073	0.016	(−0.103, −0.042)
5,6-Dihydrouridine	Pyrimidine metabolism, uracil containing	9	4,169	6.65 × 10−6	−0.082	0.018	(−0.117, −0.046)
3-Ureidopropionate	Pyrimidine metabolism, uracil containing	9	4,169	1.10 × 10−5	0.143	0.032	(0.079, 0.206)
Peptide
ADSGEGDFXAEGGGVR	Fibrinogen cleavage peptide	8	3,707	1.70 × 10−12	0.381	0.054	(0.275, 0.487)
Xenobiotics
O-Sulfo-l-tyrosine	Chemical	9	4,169	2.55 × 10−7	−0.132	0.026	(−0.182, −0.082)
2-Ethylhexanoate	Chemical	5	1,044	1.46 × 10−6	−0.203	0.042	(−0.286, −0.120)
Metformin	Drug metabolic	4	2,785	7.53 × 10−13	0.106	0.015	(0.077, 0.135)
Saccharin	Food component/plant	10	4,274	2.53 × 10−39	1.858	0.142	(1.581, 2.136)
Gluconate	Food component/plant	10	4,274	9.47 × 10−19	0.316	0.036	(0.246, 0.386)
2-Keto-3-deoxy-gluconate	Food component/plant	2	2,046	9.33 × 10−15	0.540	0.070	(0.403, 0.676)
Methyl glucopyranoside (alpha+beta)	Food component/plant	5	3,230	9.63 × 10−7	−0.501	0.102	(−0.702, −0.301)

ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study; CI, confidence interval; T2D, type 2 diabetes.

*Each metabolite was analyzed individually, adjusted for age at baseline, body mass index (BMI), cigarettes smoked per day, metabolomic set, and cancer case-control status.

†Estimate is the standardized β-coefficient. Bonferroni-corrected threshold of significance is P < 3.5 × 10⁻⁵.

Figure 1.

Manhattan plot for the associations between serum metabolites and T2D in the ATBC Study. A statistically significant association of T2D was found for 74 serum metabolites: the solid dots indicate positive associations and the hollow dots indicate inverse associations. Dot size reflects the magnitudes of the effect sizes of the standardized β-coefficients (larger circles for stronger associations). Eight classes of metabolites are shown in different colors. The red dashed line indicates the Bonferroni-corrected threshold of significance (P value = 3.5 × 10⁻⁵). Each metabolite was modeled individually, adjusted for age at baseline, body mass index (BMI), cigarettes smoked per day, metabolomic set, and cancer case-control status. ATBC, the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study; T2D, type 2 diabetes.

In the GSA models, the chemical class superpathways of carbohydrates, amino acids, energy metabolites, and xenobiotics were significantly associated with T2D at the Bonferroni-corrected threshold of P < 6.3 × 10⁻³, and lipids were associated at the P < 0.05 threshold (P = 0.048, Table 3). In the chemical subpathway analysis, the fructose-mannose-galactose, glycolysis-gluconeogenesis-pyruvate, acyl choline fatty acid, and leucine-isoleucine-valine classes were most strongly associated with T2D, reaching the Bonferroni-corrected threshold of P < 4.7 × 10⁻⁴ (Table 4).

Table 3.

GSA for associations between chemical class of metabolites and T2D in the ATBC Study

Chemical Class	No. of Set	No. of Contributing Metabolites	No. of Contributing Metabolites after Bonferroni correction	P Value	No. of Positive	No. of Negative
Carbohydrates	7	46	7	1.30 × 10−11	6	1
Amino acids	7	243	20	2.66 × 10−6	18	2
Energy	7	16	1	1.47 × 10−4	1	0
Xenobiotics	7	245	7	1.16 × 10−3	4	3
Lipids	7	599	34	4.79 × 10−2	12	22
Nucleotides	7	45	3	1.00 × 10−1	1	2
Cofactors and vitamins	7	47	1	3.09 × 10−1	0	1
Peptides	7	172	1	4.51 × 10−1	1	0

ATBC, the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study; GSA, gene set analysis; T2D, type 2 diabetes.

Table 4.

GSA for associations between chemical subclasses of metabolites and T2D at 4.7 × 10⁻⁴ level of statistical significance (106 tests) in the ATBC Study

Subpathway	No. of Set	No. of Contributing Metabolites	No. of Contributing Metabolites after Bonferroni correction	P Value	No. of Positive Associations	No. of Negative Associations
Fructose, mannose and galactose metabolism	7	9	3	1.82 × 10−12	3	0
Glycolysis, gluconeogenesis, and pyruvate metabolism	7	7	2	2.60 × 10−12	1	1
Fatty acid metabolism (acyl choline)	7	14	9	1.31 × 10−9	0	9
Leucine, isoleucine, and valine metabolism	7	38	10	1.52 × 10−9	10	0
Drug - neurological	6	3	0	4.53 × 10−9
Drug - metabolic	4	3	1	1.99 × 10−7	1	0
Lysine metabolism	7	17	2	3.45 × 10−6	2	0
Glycine, serine, and threonine metabolism	7	10	2	6.08 × 10−5	1	1
Diacylglycerol	4	14	2	9.30 × 10−5	2	0
Phosphatidylethanolamine (PE)	3	8	3	1.67 × 10−4	3	0
Pentose metabolism	7	13	2	2.70 × 10−4	2	0
TCA cycle	7	10	1	2.88 × 10−4	1	0

ATBC, the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study; GSA, gene set analysis; T2D, type 2 diabetes; TCA, tricarboxylic acid.

DISCUSSION

This present metabolomic analysis of nearly 4,300 men identified 74 metabolites in several chemical pathways that were associated with T2D. Notably, we observed that carbohydrates, lipids, and amino acids were greatly represented among the top signals, with carbohydrates glucose, 1,5-AG, and mannose, lipids stearoylcholine and linoleoylcholine, and amino acids β-hydroxypyruvate and pyroglutamine ranked highest within the three chemical classes. Fructose-mannose-galactose metabolites, glycolysis-gluconeogenesis-pyruvate metabolites, acyl choline fatty acids, and leucine-isoleucine-valine metabolites were also significantly associated with T2D.

A recent systematic review and meta-analysis of 61 studies that included 71,196 participants with 11,771 T2D events identified 412 significantly associated metabolites from 37 metabolic pathways, including in particular amino acids, carbohydrates, and lipids (19). The meta-analysis showed that glucose, mannose and fructose, BCAAs, alanine, glutamate, ceramides, phenylalanine, tyrosine, lysine, and phosphatidylethanolamines and glycerolipids were positively associated with risk of T2D, whereas glycine was inversely associated. These findings were largely consistent with our present results, with the exception of betaine and indolepropionate. Furthermore, we observed that 1,5-AG and acyl choline fatty acid metabolism were also strongly related to T2D (19).

The present study is based on serum obtained after an overnight fast and identified the well-established biomarker for diabetes, glucose (20–22), as a top signal related to an increased T2D. A nested case-control study from the Diabetes Prevention Program with 2,015 participants and T2D cases reported that glucose was associated with an increased risk of T2D (odds ratio per SD = 1.42, 95% confidence interval: 1.25, 1.61) (23), also reinforcing the fact that T2D is defined in part by hyperglycemia (4). The amino acid 1,5-AG is the 1-deoxy form of glucose (24) and has been used for monitoring of short-term glycemic control (25–28). Circulating 1,5-AG reflects its reabsorption in the renal tubules which can be competitively inhibited by glucose via glucose transporters. Rising glucose concentrations lead to urinary loss of 1,5-AG and reduced serum concentrations (26). We also found that other monosaccharides, including fructose and galactose, were positively associated with T2D. Fructose and galactose in the glycogen synthetic pathway have been shown to reduce insulin sensitivity (29, 30) and are significantly associated with an increased risk of T2D (20, 31). For example, galactose can be rapidly converted to glucose and may subsequently lead to a long-term rise in blood glucose levels (30).

Also derived from glucose, mannose is a bioactive monosaccharide and one of the primary carbohydrates in glycoproteins resulting from N-glycosylation reactions (32). An in vivo experiment demonstrated that ingesting glucose increased plasma mannose levels through elevated glucokinase activity (33), with mannose concentrations decreasing following increased glycogen synthesis from intravenous insulin administration (34). Previous studies have shown significant correlations between circulating mannose and BMI, and mannose concentrations were independently associated with insulin resistance and insulin secretion (35). Taken together with the present study, the data suggest that mannose may serve as an early-stage biomarker of insulin resistance and T2D (20, 21, 35, 36).

We identified robust amino acid-T2D associations, including with the BCAAs isoleucine, leucine, and valine and their derivatives branched-chain keto acids 3-methyl-2-oxovalerate and 3-methyl-2-oxobutyrate. These findings corroborated prior evidence from cohorts in the United States, Europe, and Asia showing strong positive BCAAs-T2D associations (19, 37–40), which are supported by data from a Mendelian randomization analysis, which found increased serum BCAAs contributing to insulin resistance and T2D (41). Excess BCAAs may adversely affect insulin signaling by activating mTOR kinase, which stimulates insulin secretion and accelerates pancreatic β-cell exhaustion (42, 43). Expression of several key BCAA oxidation enzymes is suppressed in liver and adipose tissue of insulin-resistant humans and animals through multiple biological processes including hypoxia, endoplasmic reticulum stress, impairment of mitochondrial oxidation, and inflammation (44–49). Our inverse glycine-T2D association is consistent with a recent meta-analysis and previous cohort studies (19, 50), with glycine depletion possibly indicating glutathione consumption as a result of oxidative stress (51). In addition, the positive α-hydroxybutyrate-T2D association was consistent with the findings for α-hydroxybutyrate and dysglycemia in the RISC (Insulin Sensitivity and Cardiovascular Disease) study and the α-hydroxybutyrate-T2D association in the Botnia study (52) and α-hydroxybutyrate is considered a marker of insulin resistance, glucose intolerance, and β-cell dysfunction.

Notably, we identified a strong metabolomic profile of lipid origin in the present study, with increased T2D related to higher levels of the of diacylglycerol (DAG), monoacylglycerol (MAG), and phosphatidylethanolamine (PEs) chemical subgroups which have been linked to insulin sensitivity and impaired glucose tolerance (53–55). These findings are in accord with previous positive associations for DAGs and PEs from meta-analyses (4, 19). In addition to the hypothesis of inflammation-mediated insulin resistance, DAGs-mediated insulin resistance may represent another common form of insulin resistance related to obesity and T2D (54). Elevated acyl choline fatty acids, sphingomyelins, lysophospholipids, plasmalogens, and endocannabinoids were associated with reduced T2D in our study. Sphingomyelins are a major class of phospholipids and key metabolites for lipid and glucose metabolism and decreased sphingomyelin biosynthesis has been associated with increased reactive oxygen species and mitochondrial abnormalities, as well as reduced insulin secretion (56, 57). Moreover, several fatty acids play important roles in the development and progression of diabetes and impaired insulin actions through accumulation of lipid derivatives, increased oxidative stress, inflammation, and mitochondrial dysfunction (56, 58). The metabolic role of acyl cholines in diabetes is not fully understood, but the acyl choline metabolites docosahexaenoylcholine and eicosapentaenoylcholine have been significantly positively associated with serum 25-hydroxyvitamin D (59) and an inverse 25-hydroxyvitamin D-T2D association has been shown in several studies (60–62), with vitamin D deficiency possibly inducing insulin resistance by promoting oxidative stress and inflammation leading to islet β-cell dysfunction (63). Meta-analysis of the Boston Puerto Rican Health Study (BPRHS) and San Juan Overweight Adult Longitudinal Study (SOALS) that measured over 600 fasting plasma metabolites, for example, showed inverse associations of T2D with acyl cholines and sphingolipids, and positive associations with BCAAs and sugar metabolites, generally consistent with our present findings (64). In addition, palmitoylcholine was inversely related to T2D in the Qatari cohort (996 participants with 574 T2D cases) and the Qatar BioBank (2,618 participants with 282 T2D cases) based on over 1,000 metabolites assayed (65). Impaired endocannabinoid metabolism may facilitate the development of T2D by promoting energy storage and impairing glucose and lipid metabolism (66), and endocannabinoids have been related to diabetes-induced oxidative stress, inflammation, fibrosis, and accelerated progression of T2D and its complications (67).

Strengths of this study include the large sample size, and overnight fasting serum samples assayed on a high-quality metabolomics platform with established validity and reliability. We identified >1,000 metabolites that captured a broad range of chemical classes and subpathways not available in smaller, targeted laboratory panels. In addition, these T2D-related metabolic alterations may potentially serve as circulating biomarkers for screening, early detection, prognosis, and therapeutic targets. Study limitations should also be mentioned, including the cross-sectional study design, which limits our ability to draw firm causal conclusions (e.g., temporality) regarding serum metabolites and T2D status. The assessment of diabetes in our study was based on a self-reported physician diagnosis and awareness of disease might otherwise be low, resulting in a true prevalence that may be higher with <100% ascertainment. This limitation would likely underestimate the observed risk estimates and bias them toward null. We also lacked diagnostic information regarding type 1 versus type 2 diabetes, although T2D accounts for >90% of the diabetes in this older age group, and our findings were generally consistent with prior T2D-metabolomic studies. The ATBC comprised a homogeneous population of Finnish men of European ancestry, which may restrict generalizability of our findings to other populations.

In summary, the present study measured >1,000 serum metabolites in >4,000 individuals and identified 74 significant metabolite-T2D associations. Carbohydrate, amino acid, and lipid biochemical pathways were heavily represented among them, including altered molecular signatures characteristic of metabolism related to glycolysis and gluconeogenesis, fructose-mannose-galactose, BCAAs, DAGs, acyl cholines, fatty acid oxidation, and mitochondrial dysfunction. Our results provide new additional evidence and potential insights relevant to the molecular basis of insulin resistance and the etiology of T2D. Future studies, including prospective analyses, are needed to reexamine the specific metabolite findings. If confirmed, underlying mechanisms and roles of these identified metabolites and pathways can be explored with regard to disease etiology and disease-associated risk factors.

DATA AVAILABILITY

Because of previously enacted EU General Data Protection Regulation privacy rules and an existing data use agreement between Finland and the US National Cancer Institute, the ATBC Study data and materials described in the article may not be made publicly available for the purpose of reproducing the findings. The principal investigators of the ATBC Study can be contacted with specific data requests (https://atbcstudy.cancer.gov/). To minimize the possibility of unintentionally sharing information that can be used to identify private information, a subset of the data generated for this study will be made available first on reasonable request.

SUPPLEMENTAL DATA

Supplemental Table S1: https://doi.org/10.6084/m9.figshare.21714602.

Supplemental Fig. S1: https://doi.org/10.6084/m9.figshare.21714629.

GRANTS

The ATBC Study is supported by the Intramural Research Program of the US National Cancer Institute, National Institutes of Health. J.H. was supported by the National Natural Science Foundation of China (82100949) and the Outstanding Young Investigator Award of Hunan Province (2022JJ10094).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

Y.L., L.G., J.H., and D.A. conceived and designed research; Y.L., L.G., J.H., and D.A. analyzed data; Y.L., L.G., J.H., and D.A. interpreted results; Y.L., L.G., J.H., and D.A. prepared figures and tables; Y.L., L.G., J.H., and D.A. drafted manuscript; Y.L., L.G., B.Z., K.Y., Y.W., S.M., S.J.W., J.H., and D.A. edited and revised manuscript; Y.L., L.G., B.Z., K.Y., Y.W., S.M., S.J.W., J.H., and D.A. approved final version of manuscript.