Eighty-four subjects were selected for untargeted metabolomics analysis, including 38 individuals with MS (15 Yi [YMS] and 23 Han [HMS]) and 46 healthy controls (21 Yi [YCTR] and 25 Han [HCTR]). Inclusion criteria required participants to (1): be aged ≥18 years, (2) reside in Yunnan Province for over five years, (3) provide informed consent, and (4) present with three or more metabolic disorders as defined by the 2004 criteria of the Chinese Diabetes Society (28). These disorders included a body mass index (BMI) ≥25.0 kg/m² (overweight/obesity), systolic blood pressure (SBP) ≥140 mmHg or diastolic blood pressure (DBP) ≥90 mmHg, fasting plasma glucose (FPG) ≥6.1 mmol/L, triglycerides (TG) ≥1.7 mmol/L, or high-density lipoprotein cholesterol (HDL-C) levels <1.0 mmol/L in females or <0.9 mmol/L in males (dyslipidemia). Healthy controls exhibited no metabolic disorders. Exclusion criteria encompassed prior use of medications for hyperglycemia, dyslipidemia, or hypertension, being underweight (BMI<18.5kg/m²), as well as the presence of cancer, advanced liver disease (Child–Pugh classes B/C), significant lung conditions (chronic obstructive pulmonary disease, chronic bronchitis, emphysema, asthma, or pneumonia), severe heart disease (New York Heart Association classes II–IV), renal impairment (eGFR <60 mL/min) (29–31), hyperthyroidism or Hypothyroidism), missing too much data or blood samples were not available. MS patients were matched with healthy controls by age (± 3 years), gender, and geographic location, a detailed case selection process is illustrated in Figure 1 .

Methanol (LC-MS, CNW Technologies), acetonitrile (LC-MS, SIGMA-ALDRICH), ammonium acetate (LC-MS, SIGMA-ALDRICH), ammonium hydroxide (LC-MS, CNW Technologies), double-distilled water (ddH₂O, Watsons), acetic acid (LC-MS, SIGMA-ALDRICH), 2-propanol (LC-MS, CNW Technologies), acetone (LC-MS, CNW Technologies), sodium bicarbonate (NaHCO₃, LC-MS, CNW Technologies), hydrochloric acid (HCl, LC-MS, CNW Technologies), and standard samples were used. The instruments included a Vanquish UHPLC system (Thermo Fisher Scientific, MA, USA), a Thermo Altis TSQ Plus mass spectrometer (Thermo Fisher Scientific, MA, USA), a Heraeus Fresco17 centrifuge (Thermo Fisher Scientific), an analytical balance (BSA124S-CW, Sartorius), an ultrasonic instrument (PS-60AL, Shenzhen Redbon Electronics Co. Ltd.), a homogenizer (JXFSTPRP-24, Shanghai Jingxin Technology Co. Ltd.), a freeze dryer (LGJ-10C, Sihuan Fruike Instrument Technology Development Co. Ltd.), and a freeze centrifugal concentrator (CV600).

Plasma samples were thawed at 4°C, and 50 µL of each sample was mixed with 200 µL of an extraction solution consisting of methanol and acetonitrile (1:1, v/v) containing deuterated internal standards (ISs). The mixture was vortexed for 30 seconds, sonicated for 10 minutes in a 4°C water bath, and incubated for 1 hour at -40°C. Subsequently, the samples were centrifuged at 12,000 rpm (relative centrifugal force: 13,800 × g; rotor radius: 8.6 cm) for 15 minutes at 4°C. The resulting supernatant was transferred to fresh glass vials for analysis. A quality control (QC) sample was prepared by pooling equal volumes of the supernatants. The supernatants were analyzed using ultra-high-performance liquid chromatography-tandem mass spectrometry (UHPLC-MS/MS).

LC-MS/MS analyses were conducted using a Vanquish UHPLC system (Thermo Fisher Scientific) equipped with a Waters ACQUITY UPLC BEH Amide column (2.1 mm × 50 mm, 1.7 µm) and an Orbitrap Exploris 120 mass spectrometer (Thermo Fisher Scientific). The mobile phase comprised 25 mmol/L ammonium acetate and 25 mmol/L ammonium hydroxide in water (pH 9.75) as phase A, and acetonitrile as phase B. The autosampler temperature was maintained at 4°C, and the injection volume was set to 2 µL. The Orbitrap Exploris 120 operated in information-dependent acquisition (IDA) mode, controlled by Xcalibur software (Thermo), which continuously evaluated full-scan MS spectra. Electrospray ionization (ESI) source parameters included: sheath gas flow rate = 50 Arb, auxiliary gas flow rate = 15 Arb, capillary temperature = 320°C, full MS resolution = 60,000, MS/MS resolution = 15,000, collision energy = 20/30/40 stepped normalized collision energy (SNCE), and spray voltage = 3.8 kV (positive mode) or 3.4 kV (negative mode).

The metabolomics raw data were converted to the mzXML format using ProteoWizard and processed with the R package XCMS (v3.5, CA, USA). Before data analysis, peak pretreatment steps included identification, alignment, extraction, and integration. Variability in the MS platform was monitored and adjusted using quality control (QC) spectra to ensure data reproducibility and reliability (QC details were shown in Supplementary 1 ).

In this study, 21,011 peaks were initially detected, and 17,333 peaks remained after relative standard deviation (RSD) de-noising. Missing values were imputed using half of the minimum value, and the internal standard normalization method was applied during data analysis. The final dataset, containing peak numbers, sample names, and normalized peak areas, was imported into the SIMCA 16.0.2 software package (Sartorius Stedim Data Analytics AB, Umeå, Sweden) for multivariate analysis. Data were scaled and log-transformed to reduce noise and the impact of high variable variance. Principal component analysis (PCA), an unsupervised dimensionality reduction method, was performed to visualize sample distribution and grouping. A 95% confidence interval in the PCA score plot was used to identify potential outliers and evaluate metabolic feature separation among groups.

To further investigate group separation and identify significantly altered metabolites, supervised orthogonal projections to latent structures-discriminant analysis (OPLS-DA) was applied. A 7-fold cross-validation was conducted to calculate R² and Q² values, where R² reflects the explained variation and Q² indicates predictive accuracy. Model robustness and predictive ability were assessed through a 200-time permutation test, which evaluated the likelihood of overfitting by examining the cross-validation-derived R² and Q² values (32, 33). A lower Q² intercept value indicated greater model reliability and reduced risk of overfitting.

The variable importance in projection (VIP) score of the first principal component in OPLS-DA summarized each variable’s contribution to the model. Metabolites with VIP > 1 and p < 0.05 (unpaired two-sided Student’s t-test) were identified as significantly altered. Remaining peaks were annotated by comparing retention time and mass-to-charge ratio (m/z) indices with the HMDB (www.hmdb.ca), KEGG (www.kegg.jp), and an in-house Biotree DB (V3.0) library (33, 34).

After thawing the samples in an ice-water bath and vortexing for 30 seconds, 15 μL of each sample was combined with 35 μL of water and 200 μL of extraction solution (methanol: acetonitrile, 1:1 [v/v], containing deuterated internal standards, precooled to -40°C). The mixture was vortexed for 30 seconds, ultrasonicated for 15 minutes in a 4°C water bath, and incubated at −40°C for 1 hour. The samples were then centrifuged at 12,000 rpm (RCF = 13,800 × g, radius = 8.6 cm) for 15 minutes at 4°C. The supernatant was collected and evaporated to dryness. The dried residue was reconstituted in 100 μL of 50% methanol, followed by the addition of 100 μL of derivatizing agent and 50 μL of 1 M NaHCO₃. After mixing, the sample was incubated at 40°C for 1 hour in a water bath. Once cooled to room temperature, 50 μL of 2 M HCl was added, and the sample was evaporated to dryness again. The final residue was re-dissolved in 200 μL of methanol and transferred to a fresh glass vial for analysis.

Stock solutions were prepared by dissolving or diluting each standard to a final concentration of 10 mmol/L. Aliquots of these solutions were combined in a 10-mL flask to create a mixed working standard solution. Absolute quantification was performed using isotope internal standard correction ( Supplementary Table S1 ). Calibration standard solutions were prepared by serial dilution of the mixed working standard, ensuring isotopically labeled internal standards matched the sample concentrations.

Ultra-high-performance liquid chromatography (UHPLC) separation was performed on a Thermo Vanquish UHPLC System (Thermo Fisher) with a Waters ACQUITY UPLC BEH C18 column (100 mm × 2.1 mm, 1.7 μm). The mobile phases consisted of 5 mM ammonium acetate in water (A) and acetonitrile (B). The column temperature was maintained at 45°C, the autosampler at 4°C, and the injection volume was 2 μL.

Mass spectrometric analysis was conducted using a Thermo Altis TSQ Plus Mass Spectrometer (Thermo Fisher, USA) with an electrospray ionization (ESI) interface. Key ion source parameters included a spray voltage of -3300 V, sheath gas at 40 Arb, auxiliary gas at 10 Arb, sweep gas at 1 Arb, ion transfer tube temperature at 325°C, and vaporizer temperature at 350°C.

MRM parameters for targeted analytes were optimized using flow injection analysis. Standard solutions of individual analytes were injected into the mass spectrometer’s atmospheric pressure ionization (API) source. For each analyte, the most sensitive and selective Q1/Q3 transitions were designated as “quantifiers” for quantitative monitoring, while additional transitions were used as “qualifiers” to verify analyte identity.

Supplementary Figure S1 presents the extracted ion chromatograms (EICs) for the targeted analytes from a standard solution ( Supplementary Figure S1A ) and a sample ( Supplementary Figure S1B ) under optimal conditions. The EICs demonstrate (i) symmetrical peak shapes for all analytes, (ii) baseline separations, and (iii) consistent retention times and peak shapes between the standard and the sample.

Supplementary Table S2 details the lower limits of detection (LLODs) and quantitation (LLOQs) for all analytes. The LLODs ranged from 0.33 to 604.36 nmol/L, and the LLOQs ranged from 0.65 to 1208.72 nmol/L. Correlation coefficients (R²) for regression fitting exceeded 0.9933 for all analytes, confirming a robust quantitative relationship between MS responses and analyte concentrations, suitable for targeted metabolomics. Supplementary Table S3 summarizes the analytical recoveries and relative standard deviations (RSDs) for quality control (QC) samples, measured over five technical replicates. Recoveries ranged from 81.0% to 109.0%, with RSDs below 6.6%, indicating that the method provides accurate quantitation of targeted metabolites within the specified concentration range.

Calibration solutions were analyzed using UPLC-MRM-MS/MS methods as described. Supplementary Table S10 outlines calibration curve parameters, where y represents the ratio of the analyte peak area to its corresponding internal standard, and x represents the analyte concentration (nmol/L). Regression fitting was performed using the least-squares method with 1/x weighting, which provided the highest accuracy and correlation coefficients (R²). Calibration levels were excluded if their accuracy fell outside the 80–120% range. Detailed calibration curves for individual analytes are provided in Supplementary Table S10 .

Stepwise dilution of the calibration standard solution, with a dilution factor of 2, was performed for UHPLC-MRM-MS analysis. Signal-to-noise ratios (S/N) were used to define LLODs and LLOQs, corresponding to S/N values of 3 and 10, respectively, in accordance with US FDA guidelines for bioanalytical method validation.

Quantitation precision was assessed by calculating the RSD of replicate QC sample injections. Quantitation accuracy was determined as the analytical recovery of spiked QC samples, calculated as [(mean observed concentration)/(spiked concentration)] × 100%.

In the sample detection process, the final concentration (C _F , nmol/L) equals the calculated concentration (C _C , nmol/L) measured by the instrument, multiplied by the dilution factor. The concentration of the target metabolite (C _M , nmol/L) in the sample is equal to the amount of C _F times the final volume of the sample (V _F , μ L), divided by the sample volume (V _S , μ L), which is expressed as nmol/L. The following calculation formula was used:

MRM data processing was performed using Skyline, while data acquisition utilized Xcalibur (version 4.4.16.14, Thermo Fisher). Metabolites with P<0.05 (unpaired two-sided Student’s t-tests) were identified as significantly altered. Pathway enrichment analysis of these metabolites was conducted using the KEGG database (http://www.genome.jp/kegg/).

A nontargeted LC-MS/MS metabolomics analysis was performed to investigate metabolic differences between the MS and CTR groups in the Yi and Han populations. PCA revealed significant distinctions between the CTR and MS groups within both ethnic groups ( Figures 2A, B ). OPLS-DA and its VIP scores identified key DEMs between the MS and CTR groups in both populations ( Figures 2C, D ). Similar to PCA, the OPLS-DA results demonstrated notable differences in metabolic profiles between the two groups. A permutation test confirmed the robustness of the OPLS-DA model, indicating no overfitting (P < 0.05). For the Han population, the R²Y and Q² intercepts were 0.9 and -0.8, respectively, while for the Yi population, these values were 0.82 and -0.57, respectively ( Figures 2E, F ). These results confirmed the high quality and reliability of the OPLS-DA model.

A total of 21,011 peaks were detected, and 17,333 metabolites were retained after de-noising based on relative standard deviation within the MS and CTR groups of the Han and Yi populations. Using the criteria VIP > 1 and P < 0.05, 2,762 and 1,535 significant DEMs were identified in the Han and Yi populations, respectively. In the Han population, 900 metabolites were upregulated and 1,862 were downregulated in the MS group ( Figure 3A , Supplementary Table S4 ). Of these, 266 DEMs were matched using the KEGG and HDMB databases. The identified DEMs were categorized into 19 groups, with predominant classes including lipids and lipid-like molecules (23.68%), organic acids and derivatives (12.41%), organoheterocyclic compounds (10.53%), benzenoids (8.27%), and fatty acids (8.27%; Figure 3B ). In the Yi population, 587 metabolites were upregulated and 948 were downregulated in the MS group ( Figure 3C , Supplementary Table S4 ). Among these, 280 DEMs were matched using the KEGG and HDMB databases, with major categories comprising organic acids and derivatives (19.63%), organoheterocyclic compounds (15.34%), lipids and lipid-like molecules (12.88%), benzenoids (6.75%), and fatty acids (6.75%; Figure 3D ).

Venn analysis was performed to explore the common and unique DEMs in the MS groups of the two populations. In total, 90 DEMs were changed in both MS groups of the two populations, including 46 upregulated metabolites and 44 downregulated metabolites ( Figures 4A, B , Supplementary Table S5 ). These common DEMs were mainly organic acids and derivatives (24.44%), lipids and lipid-like molecules (18.89%), and organoheterocyclic compounds (13.33%) ( Figure 4C ).

In the Han population, there were 176 DEMs in the MS group, including lipids and lipid-like molecules (30.68%), organic acids and derivatives (12.5%), organoheterocyclic compounds (11.93%), and benzenoids (10.80%) ( Figure 4D , Supplementary Table S5 ). In the Yi population, there were 73 DEMs, including organoheterocyclic compounds (23.29%), organic acids and derivatives (17.81%), benzenoids (9.59%), and lipids and lipid-like molecules (8.22%) ( Figure 4E , Supplementary Table S5 ). Compared with the CTR groups, the serum metabolites in the MS groups in the two populations changed to varying degrees, with the three main metabolites that changed being organic acids and derivatives, lipids and lipid-like molecules, and organoheterocyclic compounds.

KEGG functional enrichment analysis of the metabolites indicated that 41 metabolic pathways may be involved in the pathogenesis of MS in the Han population. Most of these DEMs were predominantly enriched in amino acid synthesis and metabolism pathway (25%), followed by protein digestion and absorption (15.91%), ABC transporters (15.91%), and 2-oxocarboxylic acid (13.64%) ( Figure 5A , Supplementary Table S6 ). Differential abundance (DA) score analysis was performed to determine the overall changes in all DEMs enriched in the same pathway. Galactose metabolism, protein digestion and absorption, 2-monocarboxylic acid metabolism, pantothenic acid metabolism, and CoA biosynthesis metabolism were significantly upregulated, while glycine, serine, and threonine metabolism, arginine synthesis metabolism, alanine, aspartic acid, and glutamic acid metabolism, cancer center carbon metabolism, mineral metabolism, amino acid synthesis metabolism, ABC transporter, thiamine metabolism, β-alanine metabolism, purine and pyrimidine metabolism, and amino acyl-tRNA biosynthesis metabolism were downregulated ( Figure 5B , Supplementary Table S6 ).

For the Yi population, 35 metabolic pathways may be involved in the pathogenesis of MS. Most of these DEMs were predominantly enriched in amino acid synthesis and metabolism pathway (24%), followed by protein digestion and absorption (20%), biosynthesis of cofactors (20%), glycine, serine, and threonine metabolism, galactose metabolism, mineral metabolism, aminoacyl-tRNA biosynthesis, and cancer center carbon metabolism (16%) ( Figure 5C , Supplementary Table S6 ). DA score analysis showed that pathways of phenylalanine and tryptophan synthesis metabolism, galactose metabolism, protein digestion and absorption, glycosylphosphatidylinositol anchor synthesis metabolism, pathogenic Escherichia coli infection, and autophagy were upregulated, while glycine, serine, and threonine metabolism, tryptophan metabolism, and mineral metabolism were downregulated ( Figure 5D , Supplementary Table S6 ).

Bioinformatics reduction and metabolic pathway enrichment analysis showed that seven pathways were significant (P<0.05) among the 39 metabolism pathways in the MS group in the Han population ( Figures 6A, B , Table 2 , Supplementary Table S7 ). Similarly, five pathways were significant (P<0.05) among the 35 metabolism pathways in the MS group in the Yi population ( Figures 6C, D , Table 3 , Supplementary Table S7 ). Metabolic pathways enrichment analysis in the different ethnic groups identified three common metabolic pathways (P<0.05) in both Yi and Han populations, including glycine, serine, and threonine metabolism, nitrogen metabolism, and cyanoamino acid metabolism, representing potential metabolic pathways for further research on MS in different ethnic groups.

PCA analysis further demonstrated clear separation between the CTR and MS groups in both populations ( Figures 7A, B ). Consistent with PCA results, the OPLS-DA model distinguished the MS and CTR groups effectively ( Figures 7C, D ). The permutation test again confirmed the model’s robustness, with no overfitting observed (P < 0.05). For the Han population, R²Y and Q² values were 0.37 and -1.27, respectively, while for the Yi population, these values were 0.9 and -0.55, respectively ( Figures 7E, F ). These results validated the reliability of the OPLS-DA model. Differential amino acid metabolites were identified between the CTR and MS groups in both populations.

A total of 71 amino acid metabolites were detected in plasma samples. We applied rigorous data management processes, retaining only metabolites with no more than 50% missing values in a single group or across all groups. Missing values were imputed using the minimum observed value multiplied by a random factor (0.1–0.5). After preprocessing, 52 metabolites were retained. Based on a screening criterion of P<0.05, 19 significantly altered amino acid metabolites were identified in the MS group of the Han population, including 13 upregulated and 6 downregulated metabolites. In the Yi population’s MS group, 6 significantly altered metabolites were identified, comprising 5 upregulated and 1 downregulated metabolite (P<0.05), as shown in Tables 4 and 5 . The distribution of these differential amino acids is visualized through volcano plots and heatmaps for both populations ( Figures 8A–D ), highlighting metabolites such as D-glutamine, L-citrulline, and symmetric dimethylarginine (SDMA).

To further clarify the mutual regulatory relationship between metabolites and MS, Spearman’s correlation analysis was performed to reveal the synergy of changes between metabolites and several cardiometabolic risk factors. 1-Methyl-L-histidine, D-aspartic acid, D-citrulline, D-methionine, D-valine, and L-2,4-aminobutyric acid were negatively correlated with BMI, BP, TC, TG, and LDL-C but positively correlated with HDL-C. The other differential amino acids were positively correlated with BMI, BP, TC, TG, and LDL-C but negatively correlated with HDL-C ( Figures 9A, B , Supplementary Table S8 ).

KEGG functional enrichment pathway analysis indicated that 25 metabolic pathways may be involved in the pathogenesis of MS in the Han population ( Supplementary Table S9 ). Most of these amino acids were predominantly enriched in D-amino acid metabolism (55.56%), biosynthesis of amino acids (33.33%), 2-oxocarboxylic acid metabolism, biosynthesis of cofactors, glycine, serine, and threonine metabolism, and cysteine and methionine metabolism (22.22%) ( Figure 10A , Supplementary Table S9 ). DA score analysis showed that alanine, aspartate, and glutamate metabolism was significantly downregulated, while the other pathways were upregulated, except for 2-oxocarboxylic acid metabolism and glycine, serine, and threonine metabolism ( Figure 10B , Supplementary Table S9 ).

For the Yi population, KEGG pathway analysis identified the involvement of 16 metabolic pathways in the pathogenesis of MS. Most of these amino acids were predominantly enriched in the biosynthesis of amino acids (80%), 2-oxocarboxylic acid metabolism (40%), glycine, serine, and threonine metabolism (40%), and the other metabolic pathways (20%) ( Figure 10C ). DA score analysis showed that all pathways were significantly upregulated, except for 2-oxocarboxylic acid metabolism, glycine, serine, and threonine metabolism ( Figure 10D , Supplementary Table S9 ).

The significant differential amino acid metabolites were enriched to further analyze the metabolomic pathways involved in MS. There were two significant pathways (P<0.05) among the 25 metabolism pathways in the MS group in Han population, including D-glutamine and D-glutamate metabolism, as well as cysteine and methionine metabolism, as shown in Table 6 and Figures 11A, B . For the Yi population, there were three significant pathways (P<0.05) among the 16 metabolic pathways, namely, namely, D-glutamine and D-glutamate metabolism, sulfur metabolism, and valine, leucine, and isoleucine biosynthesis ( Figures 11C, D , Table 6 ). Notably, D-glutamine and D-glutamate metabolism was significantly enriched in the MS group of both populations, suggesting that these targets should be further explored to investigate the pathological mechanism underlying MS.