Type 2 diabetes is a complex disorder characterized by multiple metabolic abnormalities and preventable by lifestyle changes. We aimed to identify metabolites associated with glucose metabolism in individuals at risk of type 2 diabetes and those affected by a lifestyle intervention. LC-MS metabolomics was performed on baseline and 1-year samples from 631 individuals at increased risk of type 2 diabetes, categorized into four groups by baseline glucose metabolism. The 1-year samples were from 456 non-diabetic individuals randomized to the intervention. Significant differences in the metabolite signature were observed between baseline glucose metabolism groups, particularly in amino acids, acylcarnitines, and phospholipids. Fatty acid amides, phospholipids, amino acids, DMGV, and 5-AVAB responded most to the lifestyle intervention. Lysophosphatidylcholines containing odd-chain fatty acids showed associations with improved glucose metabolism. Twenty-five metabolites differed between the baseline groups, responded to the intervention, and were associated with changes in glucose metabolism. The findings suggest a metabolite panel could be used in distinguishing individuals with varying degrees of glucose metabolism for early prediction of type 2 diabetes onset. A substantial proportion of these metabolites responded to the lifestyle intervention. These results suggest that metabolites associated with abnormal glucose tolerance potentially reflect responses to personalized interventions.
LC-MS analysis
The plasma samples were prepared for the metabolomics analysis according to Klåvus et al. (2020). Briefly, cold acetonitrile was added in a ratio of 400 µL per 100 µL of plasma into filter plates (Captiva ND filter plate 0.2 µm) and mixed by pipetting. The samples were then centrifuged for 5 min at 700 × g at 4 °C and kept at 10 °C until analysis. The quality control (QC) samples were prepared by pooling aliquots from the experimental samples.
An ultra-high performance liquid chromatography (LC) system (Vanquish Flex UHPLC, Thermo Scientific, Bremen, Germany) was used for the analysis, which was coupled online to a high-resolution mass spectrometry (MS, Q Exactive Classic, Thermo Scientific). The samples were measured using two distinct chromatographic techniques: reversed phase (RP) and hydrophilic interaction chromatography (HILIC). The analyses in RP include the utilization of both electrospray ionization (ESI) polarities, which were ESI positive (ESI+) and ESI negative (ESI−). Only ESI+ data was collected in HILIC. The centroid mode was used to obtain a full scan data. The data was collected over a mass to charge ratio (m/z) range of 120 to 1200 in the RP technique and 75 to 750 in HILIC technique. To identify metabolites, we acquired data-dependent product ion spectra (MS/MS) from pooled quality control (QC) samples at the beginning and end of the analysis for each mode. In addition, QC samples were incorporated in the analysis at the beginning and subsequently after every 12 samples. The configuration and specifications for the LC-MS instrument have been previously published (Lapatto et al. 2023).
Data analysis
The signal detection and alignment were performed in MS-DIAL version 4.48 (Tsugawa et al. 2015) according to Klåvus et al. (2020). Briefly, the minimum peak height for the mass spectrometry data was set at 200 000 counts, and for the alignment, retention time tolerance was set at 0.6 min for HILIC data and 0.5 min for RP data. A sample-to-blank filter was used to remove the solvent background by requiring that the ratio of the maximum peak height in the samples and the average peak height in the solvent blank injections was more than 5. The gap filling by compulsion option was used to reduce missing values in the data. The data matrix containing the aligned peak areas across all samples was then pre-processed in notame R package (Klåvus et al. 2020) to correct the drift in signal intensity across the QC samples and analytical batches and to filter out poor-quality molecular features. The k-means clustering analysis was performed with Multiple Experiment Viewer 4.9.0.
Statistical analyses
The differential molecular features between the NGT, isolated IGT, IGT+IFG, and T2D groups at baseline were determined with the Kruskal–Wallis test. Multivariable linear mixed-effects models (LME) were applied to determine the contribution of each molecular feature on the changes in clinical outcome variables (Supplemental Table S4) during the intervention in DIGI and DIGI+F2F. The covariate p-value from the model signifies whether there was a statistically significant association between the molecular feature and the clinical outcome variable, whereas the time * group interaction p-value indicates how the intervention modified the association between the molecular features and the clinical outcomes. Both models were adjusted for age and sex and repeated-measures structure was taken into account with the mixed model using the participant as subject for random effects. Benjamini–Hochberg false discovery rate correction was performed on the p¬-values (reported as q-values) to account for the potential false positive results caused by multiple measurements. The PERMANOVA analysis was performed on the good-quality molecular features using vegan package in R.
Metabolite identification
The differential metabolites between the baseline groups and altered by the intervention were identified with MS-DIAL by comparing the observed MS/MS spectra with our in-house spectral database (level 1), publicly available spectral databases, such as MassBank, ReSpect, GNPS, RIKEN, and HMDB (level 2), and MS/MS spectra generated in silico in MS-FINDER 3.50 15 (level 3). The level of the annotation reliability is based on the Metabolomics Standards Initiative (MSI) recommendations (Sumner et al. 2007).
Columns A–E: LC-MS characteristics
Columns F–AQ: Metabolite annotation & classification
Columns AR–BC: Results from the data preprocessing by notame R script
Columns BO–FL: Statistical results
Columns FN–AUZ: Preprocessed signal abundances (as peak areas) of the molecular features in all experimental samples
