This study was reviewed and approved by the Institutional Review and Ethics Board at Tongren Hospital, associated with Shanghai Jiao Tong University, with the approval number 2022-049-01, dated 22 June 2023. All participants provided written informed consent to participate in the study and for their data to be published. The research was carried out following the ethical principles of the Declaration of Helsinki and national regulatory standards.

Participants of both sexes, aged 30-55 years, were recruited in a descriptive cross-sectional study design [Figure 1]. Obesity was defined as having a body mass index (BMI) of 28 kg/m² or higher. Participants were stratified into three groups: metabolically healthy obesity (OB, age 38.00 ± 6.26 years), obesity with IGT (age 41.46 ± 13.06 years), and obesity-associated diabetes (OD, age 40.40 ± 6.74 years).

IGT was defined as a 2-h plasma glucose level ranging from 140 to 200 mg/dL (7.8 to 11.0 mmol/L) after consuming 75 g of glucose, while fasting glucose levels remain normal. The diagnostic criteria for T2DM were as follows: (1) diagnosis established within the past month; (2) glucose tolerance results consistent with World Health Organization (WHO) diagnostic criteria; and (3) no previous history of diabetes. Specifically, T2DM diagnosis required fasting blood glucose (FBG) ≥ 126 mg/dL, and/or a 2-h post glucose load oral glucose tolerance test (OGTT) ≥ 200 mg/dL, and/or glycated hemoglobin (HbA1c) ≥ 6.5%. Individuals with any of the following conditions were excluded: familial hypercholesterolemia; serum creatinine ≥ 132.6 μM (males) or ≥ 123.8 μM (females); estimated glomerular filtration rate (eGFR) < 60 mL/min; a history of malignancy or severe systemic disease; recent cardiovascular events.

The study included a normal control (NC) group of 23 lean, metabolically healthy participants, comprising 12 males and 11 females, with a mean age of 41.13 years and a BMI of 22.26 kg/m². Individuals were considered “metabolically healthy” if they had: (1) no prior history of diabetes and FBG < 6.1 mmol/L, 2-h post OGTT < 7.8 mmol/L; (2) no prior history of hypertension with systolic blood pressure (SBP)/diastolic blood pressure (DBP) < 140/90 mmHg; (3) no prior record of elevated cholesterol levels [total cholesterol (TC) < 5.18 mmol/L] and fasting plasma TG < 1.7 mmol/L, with fasting serum high-density lipoprotein cholesterol (HDL-c) at least 0.9 mmol/L for men or at least 1.0 mmol/L for women; and (4) no prior history of heart or endocrine diseases.

Records were taken for BMI, SBP, and DBP. The levels of TC, TG, HDL-c, low-density lipoprotein cholesterol (LDL-c), FBG, HbA1c, alanine aminotransferase (ALT), aspartate aminotransferase (AST), and eGFR were assessed with an automatic biochemical analyzer (AU5800 clinical chemistry analyzer; Beckman Coulter Inc., Brea, CA, USA). Radioimmunoassay was used to measure fasting insulin (FIns) and fasting c-peptide (FCP). The HOMA2 Calculator (https://www.dtu.ox.ac.uk/homacalculator/) was used for homeostasis model assessment (HOMA) of steady-state beta-cell function (HOMA-%β) and insulin resistance (HOMA-IR).

For lipidomic analysis, serum samples were obtained by centrifuging blood at 1,300 × g for 15 min at room temperature and promptly frozen at -80 °C. A Thermofisher Vanquish UHPLC (ultra-high-performance liquid chromatography) system, which includes a binary pump, an autosampler, a vacuum degasser, a column oven, and a Q Exactive Plus mass spectrometer, was used for the lipid analysis. The lipid extracts were introduced into a reversed-phase ACQUITY UPLC BEH C18 column (100 mm × 2.1 mm, 1.7 μm, Waters), which was kept at 55 °C.

The mobile phase consisted of component A, a mixture of acetonitrile and water (60:40, v/v) containing 10 mM ammonium formate and 0.1% formic acid, and component B, a blend of isopropanol and acetonitrile (90:10, v/v) with the same additives. The mobile phase was pumped at a rate of 0.4 mL/min, beginning with a 95/5 ratio and transitioning to 0/100 over 17 min. The column was subsequently returned to its starting conditions for 10 min. A total duration of 20 min was set for the run, and the injection volume was 1 μL. Data acquisition occurred in data-dependent acquisition (DDA) scan mode, spanning from 150 to 2,000 amu in high-resolution mode at 70,000 Hz. The mass spectrometry (MS) analysis was conducted with the following settings: spray voltage at 3.0 kV for negative mode and 3.2 kV for positive mode, and a capillary temperature of 320 ℃. A full scan was conducted, followed by six MS/MS scans, using nitrogen (99.999%) as the gas for collision-induced dissociation. The normalized collision energy applied to fragment the ions was 15, 30, and 45.

Data was processed using Xcalibur 4.3 for acquisition and LipidSearch 5.0 for peak detection, alignment, and normalization. Lipid identification was based on accurate mass and MS/MS fragment analysis, annotated by total acyl carbon number and degree of unsaturation. Adducts included: positive ionization (M+H, M+NH₄, M+NH₄-H₂O, M+Na, M+K, M+2K-H) and negative ionization (M-H, M-H₂O-H, M+C₂H₃O₂, M+HCO₂). In total, 4,001 lipids spanning 45 classes or subclasses were detected. Missing values below detection limits were replaced with half of the minimum observed value. Detailed lipid class information is presented in Supplementary Table 1.

IBM SPSS version 25 was used for statistical analyses, and variables with skewed distributions were either log-transformed or had their square roots taken to achieve normality^[21]. For quantitative variables, either the Student’s t-test or the Mann-Whitney U test was used for analyses between two groups, and either one-way analysis of variance (ANOVA) or the nonparametric Kruskal-Wallis test for analyses across four groups. For qualitative variables, either the chi-square or Fisher’s exact test was used. Quantitative variables are shown as means ± standard deviation (SD) or medians (25th and 75th percentiles), while qualitative variables are displayed as percentages (frequency). A P value of 0.05 or less was considered significant.

Data processing of MS information included initial peak detection, alignment, and normalization using the Progenesis QI software supplied with the instrument (Waters). The lipid profiles of NC, OB, IGT, and OD groups were compared using orthogonal partial least squares discriminant analysis (OPLS-DA). Univariate and multivariate analyses, including variable importance projection (VIP) of peak intensity, fold change (FC) analysis, and false discovery rate (FDR), were used to screen for differential and abundant lipids. A two-tailed Student’s t-test was utilized for comparing FC, with a P value threshold of less than 0.05. The Benjamini-Hochberg method was used to adjust P values to determine the FDR. Lipids with a VIP > 1, an FDR < 0.05, and an FC ≥ 0.5 or ≤ -0.5 were identified as differentially abundant. The relationship between these lipids and clinical categorical outcomes (OB and OD) was evaluated using Spearman’s correlation.

A total of 96 participants were enrolled, 73 obese individuals and 23 NC. Obese individuals were distributed into three diagnostic groups: obesity (OB, n = 24), obesity with IGT (n = 24), and OD (n = 25). No significant differences in age or sex were observed among the four groups [Table 1]. Compared with the NC group, both the OB and IGT groups exhibited significantly higher HOMA-β% and HOMA-IR values (all P < 0.05), but similar levels of HOMA-β%, lipid metabolism indices (TG, TC, LDL-c, HDL-c), and liver and kidney function parameters (ALT, AST, eGFR). In the IGT group, HbA1c, HOMA-β%, and HOMA-IR were markedly elevated. The OD group displayed significantly increased levels of FBG, HbA1c, HOMA-IR, TC, TG, LDL-c, and ALT compared with the other three groups, while HOMA-β% was substantially lower [Table 1].

OPLS-DA score plots revealed distinct clustering of all four groups, indicating clearly differentiated serum lipid profiles [Figure 2A and B]. A total of 4,001 lipid molecules across 21 lipid classes were identified, among which the top six differentially expressed lipid classes were Cer, phosphatidylethanolamine (PE), LPC, DG, phosphatidylcholine (PC), and TG [Figure 2C]. The distribution of these top six lipid classes across all groups is illustrated in Figure 2D.

Fatty acids were categorized by carbon chain length into short-chain fatty acids (SCFA, 2-6 carbons), medium-chain fatty acids (MCFA, 7-10 carbons), long-chain fatty acids (LCFA, 11-20 carbons), very long-chain fatty acids (VLCFA, 21-25 carbons), and ultra-long-chain fatty acids (ULCFA, > 26 carbons). Heatmap analysis demonstrated significant differences among the four groups in the relative abundance of lipid species with varying chain lengths within DG, PC, LPC, and TG [Figure 3A-D]. In particular, the OD group showed markedly higher proportions of VLCFA and ULCFA species in DG, PC, and LPC compared with the OB and IGT groups. As shown in Figure 3D, TG species predominantly contained chain lengths of 35-65 carbons, corresponding mainly to ULCFA.

Analysis of double bond numbers across lipid chains further revealed significant group-specific patterns. Lipids were classified as saturated (no double bonds), monounsaturated (one double bond), or polyunsaturated (two or more double bonds). In the DG, PC, LPC, and TG, the IGT and OD groups displayed increased levels of saturated fatty acids compared with the OB group, whereas polyunsaturated fatty acids were more abundant in the IGT group [Figure 3E].

Twenty-three lipids were differentially expressed between the OB and the NC groups, including 22 downregulated and one upregulated species [Figure 4A and B]. The OB group primarily exhibited reduced levels of PC and LPC species [Figure 4C]. Spearman correlation analysis demonstrated strong correlations between different lipids and metabolic indices [Figure 4D]. Eight significantly downregulated PC species showed positive correlations with LDL-c and negative correlations with BMI and HDL-c. Similarly, six LPC species were positively correlated with LDL-c and FBG, but negatively correlated with BMI, HOMA-IR, HDL-c, and FCP, indicating that decreased PC and LPC levels may reflect lipid metabolic disturbances associated with obesity [Figure 4D].

In obese individuals with IGT, 208 lipids were differentially expressed compared with control individuals, including 165 significantly upregulated species [Figure 5A]. The heatmap plots highlighted distinct lipidomic shifts [Figure 5B], with TG showing the most pronounced upregulation [Figure 5C].

To further elucidate lipidomic changes associated with diabetes progression, we compared the OD and the NC groups and identified 305 differentially expressed lipids, of which 220 were significantly upregulated [Figure 6A]. The heatmap analysis [Figure 6B] showed that TG constituted the dominant lipid class in the OD group, accounting for 49.5% of total serum lipids [Figure 6C]. Notably, TG species were the most strongly increased lipids during the transition from IGT to diabetes, suggesting that enhanced TG accumulation and remodeling play a central role in lipotoxicity and glucose metabolism dysfunction.

A Venn diagram illustrates 12 overlapping differential lipids shared by the OB, IGT, and OD groups [Figure 7A]. Receiver operating characteristic (ROC) curve analysis identified LPC (16:0), Cer (t33:7), and lysophosphatidylethanolamine (LPE) (O-19:0) as strong diagnostic markers for T2DM, with area under the curve (AUC) values of 0.975, 0.986, and 0.948, respectively [Figure 7B].

Correlation analysis revealed strong positive correlations between the lipids Cer (t33:7), LPC (18:0), LPC (O-16:0), and LPE (O-19:0) and the clinical parameters TC, BMI, FBG, and HbA1c, while the correlation with HOMA-β% was inverse [Figure 7C]. Conversely, PC (36:4) exhibited a positive correlation with HOMA-β% but a negative correlation with TC, BMI, FBG, HbA1c, and liver function indicators, suggesting its potential protective role in maintaining glucose homeostasis [Figure 7C].