Abstract Background Metabolites are pivotal in the biological process underlying type 2 diabetes (T2D) and its cardiovascular complications. Nevertheless, their contributions to these diseases have not been comprehensively evaluated, particularly in East Asian ancestry. This study aims to elucidate the metabolic underpinnings of T2D and its cardiovascular complications and leverage multi-omics integration to uncover the molecular pathways involved. Method This study included 1180 Chinese participants from the Zhejiang Metabolic Syndrome Cohort (ZMSC). A total of 1912 metabolites were profiled using high-coverage widely targeted and non-targeted metabolic techniques. Multivariable logistic regression models and orthogonal partial least squares discriminant analysis were used to identify T2D-related metabolites. A metabolome-wide genome-wide association study (GWAS) in ZMSC, followed by two-sample Mendelian randomization (MR) analyses, was conducted to explore potential causal metabolite-T2D associations. To enhance cross-ancestry generalizability, MR analyses were conducted in European ancestry to explore the potential causal effects of serum metabolites on T2D and its cardiovascular complications. Furthermore, multi-omics evidence was integrated to explore the underlying molecular mechanisms. Results We identified six metabolites associated with T2D in Chinese, supported by metabolome analysis and genetic-informed causal inference. These included two potential protective factors (PC [O-16:0/0:0] and its derivative LPC [O-16:0]) and four potential risk factors ([R]-2-hydroxybutyric acid, 2-methyllactic acid, eplerenone, and rauwolscine). Cross-ancestry metabolome-wide analysis further revealed four shared potential causal metabolites, highlighting the potential protective role of creatine for T2D. Through multi-omics integration, we revealed a potential regulatory path initialized by a genetic variant near CPS1 (coding for a urea cycle-related mitochondrial enzyme) influencing serum creatine levels and subsequently modulating the risk of T2D. MR analyses further demonstrated that nine urea cycle-related metabolites significantly influence cardiovascular complications of T2D. Conclusion Our study provides novel insights into the metabolic underpinnings of T2D and its cardiovascular complications, emphasizing the role of urea cycle-related metabolites in disease risk and progression. These findings advance our understanding of circulating metabolites in the etiology of T2D, offering potential biomarkers and therapeutic targets for future research. Research insights What is currently known about this topic? Metabolites are crucial for understanding diabetes biology.Multi-omics integration aids in revealing complex mechanisms. What is the key research question? How do serum metabolites affect diabetes and its cardiovascular outcomes? What is new? Novel diabetes-related metabolites identified in Chinese populations.Consistent metabolites associated with diabetes and glycemic traits in East Asians and Europeans.Emphasizing the role of urea cycle pathway in cardiometabolic disease. How might this study influence clinical practice? Findings could guide diabetes prevention and personalized management strategies. Supplementary Information The online version contains supplementary material available at 10.1186/s12933-025-02718-4. Keywords: Metabolome, Type 2 diabetes, Cardiometabolic diseases, East Asians, Mendelian randomization, Genetic association studies, Zhejiang Metabolic Syndrome Cohort Introduction Cardiometabolic diseases (CMD), including cardiovascular disease (CVD) and type 2 diabetes (T2D), are leading contributors to global morbidity and mortality [[40]1, [41]2]. T2D, characterized by hyperglycemia due to the diminished or inappropriate secretion of insulin by pancreatic β cells, significantly increases the risk of CVD [[42]3]. Understanding T2D pathophysiology is essential for developing targeted prevention and treatment strategies to reduce the risk of CMD. T2D arises from intricate interactions between genetic predisposition and environmental factors [[43]3]. However, translating findings from genome-wide association studies (GWAS) into practical intervention targets remains a major challenge, primarily due to the limited understanding of how GWAS-identified risk variants influence T2D pathogenesis [[44]4]. While hyperglycemia is a key diagnostic marker, metabolic changes often occur before its onset [[45]5]. Therefore, discriminating metabolic alterations is essential for improving early detection, guiding effective management, and uncovering the molecular processes driving T2D. Metabolites, which connect genotype, environment, and phenotype, serve as indicators of physiological or pathological states [[46]6]. Metabolomics, a high-throughput biological technology, refers to the qualitative and quantitative analysis of small molecule metabolites (< 1 kDa) [[47]6]. Recent observational metabolomics studies have provided clues into the metabolic changes and etiology contributing to T2D [[48]7–[49]9]. However, these studies have predominantly concentrated on a limited range of metabolites and may be vulnerable to bias due to confounding factors and reverse causality under a classic case–control or cohort study design. Besides, most studies focus on European (EUR) ancestries, which may constrain the generalizability of causal insights across diverse ancestries. Recent advancements in both non-targeted and targeted metabolomics technologies have promoted the dissection of the genetic architecture underlying metabolites [[50]10–[51]13]. Shared genetic determinants between metabolites and T2D form a robust basis for investigating their relationship using genetic variants as instrumental variables through genetic-informed approaches such as Mendelian randomization (MR). MR is less prone to confounding and reverse causality bias compared to traditional observational approaches [[52]14]. Moreover, integrating metabolomics with other omics layers, such as genomics, transcriptomics, and proteomics, enables a more comprehensive understanding of the T2D metabolic network, thereby strengthening causal inferences and offering thorough insights into the complex biological mechanisms driving the disease. In this study, we aim to investigate the metabolic underpinnings of T2D and its cardiovascular complications based on the Zhejiang Metabolic Syndrome Cohort (ZMSC) toward the Chinese population. Non-targeted and widely targeted approaches were adopted to provide a high-coverage serum metabolomic profile. Then, metabolomic and genomic data of ZMSC were integrated to reveal the potential causality of metabolite-T2D associations through metabolome-wide GWAS and MR analysis. To assess cross-ancestry consistency, we conducted metabolome-wide MR analyses for T2D and glycemic traits in both East Asian (EAS) and EUR ancestries. Furthermore, a multi-omics integration analysis was employed to uncover the putative molecular mechanisms underlying these metabolite-T2D associations. Additionally, we extended our MR analysis to explore the potential role of T2D-related metabolites in cardiovascular complications of T2D, providing further insights into the metabolic pathways involved in T2D-related cardiovascular risks. Method Study design Figure [53]1 depicts the study design. We conducted a cross-sectional survey, genotyping, and serum metabolome profiling (both widely targeted and non-targeted) of participants from the Zhejiang Metabolic Syndrome Cohort (ZMSC) in Zhoushan, Zhejiang, China. A case–control study was conducted to identify T2D-associated metabolites. Then, we performed GWAS for 1912 metabolites and subsequent two-sample MR analysis to investigate the putative metabolite-T2D/glycemic traits associations. To enhance cross-ancestry generality, MR analyses were further carried out in European ancestry. For the identified metabolite-T2D associations, a multi-omics integration strategy was used to explore the underlying molecular mechanisms. Fig.1. [54]Fig.1 [55]Open in a new tab Flowchart of the study design. ZMSC Zhejiang Metabolic Syndrome Cohort, SNP Single nucleotide polymorphism, LC–MS/MS Liquid chromatography-tandem mass spectrometry, GWAS Genome-wide association analysis, OPLS-DA Orthogonal partial least squares discriminant analysis, SNP Single nucleotide polymorphism, T2D Type 2 diabetes, CVD Cardiovascular disease, METSIM The Metabolic Syndrome in Men study, CLSA The Canadian Longitudinal Study on Aging Furthermore, for metabolites supported by multi-omics evidence, we extended our MR analysis to explore their potential impact on cardiovascular complications of T2D. Study participants and data collection The present study was based on the ZMSC, a community-based prospective cohort established in Zhejiang Province, Southeastern China. The cohort was initiated between 2010 and 2014, enrolling 22,649 participants aged 19 to 80 years through a multi-stage clustered sampling approach. During the baseline survey, each participant underwent a comprehensive assessment, including a questionnaire interview, clinical health examination, and routine biochemical tests. Details of the study design have been described elsewhere [[56]15, [57]16]. Participants enrolled in 2015 from a sub-population of the ZMSC, namely the Liuheng sub-cohort, were included in our analysis. Baseline characteristics, including age, sex, body mass index (BMI), blood pressure, smoking status and drinking status, were collected through questionnaires and physical examination. Blood cell and serum samples were obtained after a minimum of 8 h of overnight fasting, then separated and stored at − 80 °C for subsequent genomic and metabolomic analysis. Our study enrolled participants aged 18 years or older. The exclusion criteria were as follows: diagnosis with cancer, coronary heart disease, or stroke; missing metabolomic data or covariate data. The participant selection process was outlined in Supplementary Material [58]2: Fig. S1. Finally, a total of 1180 participants were included in the metabolomic analyses. Among them, 1124 samples that passed both genome and metabolome quality control were included for metabolite-wide GWAS. The study was approved by the Ethics Committee of the School of Public Health, Zhejiang University, China (ZGL202312-7), and conducted in accordance with the principles of the Helsinki Declaration. Written informed consent was obtained from all participants. Laboratory measurements Metabolic profiling Using non-targeted metabolomics and widely targeted profiling, a total of 1912 metabolites were quantified. Non-targeted metabolomic profiling was conducted using an ultra-performance liquid chromatography (UPLC, ExionLC AD) system coupled with quadrupole-time of flight (TripleTOF 6600, AB SCIEX). Widely targeted metabolomic profiling was conducted on UPLC (ExionLC AD) and tandem mass spectrometry (MS/MS, QTRAP®). All metabolomic profiling was conducted on serum samples at Wuhan Metware Biotechnology in August 2023. Metabolites were identified based on the Metware database (MWDB), which includes both standard reference compounds and annotated non-targeted compounds. All samples were processed within the same experimental condition. Details are provided in Supplementary Material [59]1. Genotyping, quality control, and imputation Genotyping was carried out using the Illumina ASA-750K array. Genome-wide imputation was performed via the University of Michigan imputation server ([60]http://imputationserver.sph.umich.edu/index.html). The genotype phasing was conducted using Beagle with the 1000 Genomes Project (1 KG) Phase 3 V5 of EAS as the reference panel [[61]17]. Standard quality controls were performed using PLINK (v1.90b7.2). Details of the quality control procedure are explained in Supplementary Material [62]1. After filtering, 4192,584 SNPs remained for analysis. Ascertainment of T2D Blood glucose status was assessed according to the diagnostic criteria by the Chinese Diabetes Society guideline [[63]18]. T2D was identified based on physician diagnosis (either from self-report or from medical records), the use of hypoglycemic medication or insulin, or fasting glucose (FG) ≥ 7.0 mmol/L. Prediabetes was defined as 6.1 ≤ FG < 7 mmol/L. MR data source Exposure data Metabolite-wide GWAS in ZMSC We used the GWAS summary statistic data of 1912 metabolites from ZMSC for MR analysis, as detailed in the Methods Sect. “[64]Statistical analysis” included below. METSIM metabolomics study The Metabolic Syndrome in Men (METSIM) study includes approximately 10,197 Finnish men aged 45–74 years at baseline [[65]19]. Non-targeted plasma metabolomics profiling via the Metabolon DiscoveryHD4 mass spectrometry platform was performed in 6136 randomly selected individuals with normoglycemia [[66]11]. GWAS was conducted for 1391 metabolites, with the summary-level GWAS statistics publicly available at [67]https://pheweb.org/metsim-metab. Canadian longitudinal study on aging study The Canadian Longitudinal Study on Aging (CLSA) comprises over 50,000 Canadians aged 45–85 years at baseline [[68]20]. Plasma metabolomics profiling was conducted using the Metabolon DiscoveryHD4 platform. GWAS of 1091 metabolites were performed in 8192 individuals of European ancestry [[69]21]. Summary-level statistics for these metabolites were obtained from the GWAS Catalog, corresponding to accession numbers GCST90199621-GCST90201020. Outcome data T2D Summary-level statistic data were obtained from the DIAbetes genetics replication and meta-analysis (DIAGRAM) consortium. The DIAGRAM consortium performed a multi-ancestry genome-wide meta-analysis among 180,834 participants with T2D and 1,159,055 participants with normoglycemia [[70]4]. The present MR analysis was based on data from Europeans (80,154 cases; 853,816 controls) and East Asians (56,268 cases; 227,155 controls). Glycemic traits Summary-level data for fasting glucose (FG), fasting insulin (FI), 2-h glucose after an oral glucose challenge (2hGlu), and glycated hemoglobin (HbA1c) were obtained from the Meta-Analyses of Glucose and Insulin-related Traits Consortium (MAGIC) [[71]22]. The present MR analysis was based on data from 196,991 Europeans and 36,584 East Asians without diabetes. Cardiovascular complications of T2D Summary-level data for cardiovascular complications related to T2D were obtained from the FinnGen [[72]23], including atrial flutter/fibrillation (AF; 50,743 cases, 210,652 controls), chronic heart failure (CHF; 29,672 cases, 357,772 controls), ischemic stroke (IS; 27,497 cases, 371,723 controls), myocardial infarction (MI; 26,060 cases, 343,079 controls), and cardiomyopathy (6338 cases, 312,154 controls). Statistical analysis Baseline characteristics The normally distributed variable was summarized as mean (standard deviation, SD) and was compared using the ANOVA among the groups of prediabetes, diabetes, and normoglycemia. The non-normally distributed variable was presented as median (interquartile range, IQR) and compared using the Kruskal–Walli test among the groups. The categorical variable was presented as a number (percentage) and was compared using the chi‐square test among the groups. Metabolome analysis The quality of metabolomic data was assessed by excluding raw signals with a coefficient of variation greater than 50% in quality control samples (details provided in Supplementary Material [73]1). Missing values exceeding 50% within a group were set to zero; others were imputed using k-nearest neighbors (KNN). To address batch effects and improve data integration, the support vector regression (SVR) algorithm was applied. Metabolite levels were log-transformed with outliers exceeding three standard deviations removed. The data were normalized using Z-score transformation to achieve a standard normal distribution (mean: 0, SD:1). For each metabolite, a multinomial logistic regression was employed to estimate the odds ratio (OR) and the corresponding 95% confidence interval (CI) for participants with diabetes, prediabetes and normoglycemia (reference category). The analysis was adjusted for covariates including age, sex, BMI, hypertension status (yes or no), smoking status (never, ever, or now), and drinking status (never, ever, sometimes, or often). Hypertension was defined as systolic blood pressure > 140 mmHg, diastolic blood pressure > 90 mmHg, or antihypertensive medication [[74]24]. Drinking status was categorized into four groups based on the frequency of alcoholic beverage intake: never, ever, sometimes (1–3 times per month), and often. To explore potential sex-specific influences, a subgroup analysis by sex was also conducted using the same model. The false discovery rate (FDR) adjustment was conducted using the Benjamini–Hochberg method. Orthogonal partial least squares discriminant analysis (OPLS-DA) was utilized to identify a linear combination of features for the distinction of cases and controls. To mitigate potential confounding factors and achieve more robust findings, propensity score matching (PSM) was further utilized to equilibrate covariates between participants with and without T2D. Specifically, case–control matching was performed using the “MatchIt” package in R, balancing for age and sex with the nearest-neighbor matching method at a 1:1 ratio. For the OPLS-DA analysis, we used the “ropls” package in R (version 1.26.4). To evaluate the robustness of the detected OPLS-DA model, we calculated the R^2Y(cum) and Q^2(cum) by tenfold cross-validation, which indicates the fit and prediction ability of the model. The variable importance in projection (VIP) scores were calculated from the OPLS-DA model to select individual variables that substantially enhance the explanatory capability of the model. The metabolites with VIP values > 1 and P[FDR] < 0.05 were classified as significant differential metabolites. Pathway analysis and enrichment of significant differential metabolites were conducted by MetaboAnalyst v6.0 ([75]https://www.metaboanalyst.ca/) using the human Kyoto Encyclopedia of Genes and Genomes (KEGG) [[76]25], the human metabolome database (HMDB) [[77]26], and the Small Molecule Pathway Database (SMPDB) [[78]27]. Metabolite-wide GWAS analysis The metabolite-wide GWAS was conducted for 1912 metabolites by the mixed linear model-based association analysis (MLMA) implemented in the GCTA software (version 1.94.0 beta) [[79]28]. Covariates age, sex, and the first ten ancestral principal components were included as covariates. Manhattan plots were drawn via the R package “CMplot”. Two-sample Mendelian randomization MR analysis, leveraging genetic variants as instrumental variables (IVs), enhances causal inference by mitigating confounding and reverse causality. This method relies on three vital assumptions: (1) the genetic instruments must be strongly associated with metabolites, (2) the genetic instruments should influence the risk of T2D through metabolites, without direct effects or alternative pathways, and (3) the genetic instruments must be independent of confounders. For the EAS ancestry, metabolites from ZMSC were used as the exposure, with T2D and glycemic traits from DIAGRAM (EAS) serving as the outcome. For the EUR ancestry, metabolites from CLSA/METSIM were used as the exposure, with T2D, glycemic traits, and CVD from DIAGRAM (EUR) or FinnGen serving as the outcomes. For the selection of IVs for metabolites in EUR and EAS ancestries, we applied the same criteria. A relatively loose IV threshold with P < 1 × 10^–5, commonly used in metabolome-wide MR analysis, was applied [[80]29, [81]30]. We performed linkage disequilibrium (LD) clumping using the 1 KG Phase3 V5 EAS/EUR as the reference panel to obtain independent IVs. The parameters for LD clumping were as follows: LD r^2 < 0.01 within 1 Mb distance. The random-effect inverse variance weighting (IVW) method was implemented as the primary method to investigate putative causal effects of metabolites on traits. To ensure the robustness of estimates, we also employed MR Robust Adjusted Profile Scoring (MR-RAPS), a method designed to account for weak instrument bias, pleiotropy and extreme outliers. Sensitivity analyses were conducted to enhance the reliability of our results, including tests for pleiotropy and heterogeneity. The MR-Egger intercept test was applied to detect potential horizontal pleiotropy, while Cochran’s Q statistic was used to assess heterogeneity among the IVs. Only potential causal effects that passed both pleiotropy and heterogeneity tests (P > 0.05) were considered for further investigation. We sought to provide genetic evidence supporting the case–control study through MR analysis; however, the evidence from the IVW method alone was not sufficiently robust. To strengthen the findings, we further applied the MR-RAPS method. Thus, metabolite-trait associations were deemed significant with P < 0.05 estimated by the IVW and MR-RAPS method. To exclude potential reverse causation, we performed reverse MR analysis for the metabolite-T2D association in ZMSC. For T2D, the threshold of IV selection was set to P < 5 × 10^−8 for T2D, with other parameters remaining the same as those used for metabolites as mentioned above. To further mitigate the potential effects of T2D on the genetic-metabolite associations, we conducted sensitivity analyses by excluding T2D participants from the ZMSC and subsequently re-performing both metabolome-wide GWAS and MR analyses. The “TwoSampleMR” package (version 0.5.8) and the “mr.raps” package were adopted to perform MR analysis. Potential nonlinear metabolite-T2D association We further investigated the association between metabolites and T2D in ZMSC using restricted cubic spline analysis (RCS) to access potential nonlinear trends. We focus on metabolite-T2D associations with consistent observational and genetic evidence. In the analysis, we adjusted for several confounders, including age, sex, BMI, smoking, and alcohol consumption, which might influence metabolites. We used the minimum Akaike Information Criterion (AIC) criterion to select optimal nodes for the RCS model. A two-tailed test was used for statistical analyses. All statistical analyses were conducted using R software 4.3.2. Results Characteristics of participants The characteristics of the Chinese participants stratified by their blood glucose status (normoglycemia, prediabetes and diabetes) are summarized in Supplementary Material [82]3: Table S1. Among the 1180 participants, the prevalence of prediabetes was 11.62% (138), and that of diabetes was 10.94% (130). The median age was 61.92 (IQR: 55.50, 67.52), 59.44 (IQR: 52.67, 65.32), and 58.49 (IQR: 51.61, 64.99) years for the three groups, respectively. Participants with diabetes tended to have higher age, BMI and prevalence of hypertension compared to the normoglycemia and prediabetes groups. Other characteristics of participants between the three groups were comparable. A detailed description of the European cohort characteristics is provided in Supplementary Material [83]3: Table S2. Metabolome measurement in ZMSC In the ZMSC, a total of 1912 serum metabolites were identified by the UPLC-MS-based metabolomics profiling approach (Supplementary Material [84]3: Table S2-S3). Widely-targeted profiling and non-targeted metabolome identified and quantified 1601 (83.73%) and 311 (16.27%) metabolites, respectively. These metabolites spanned 23 major classes, with amino acids and their derivatives (27.67%) comprising the most abundant class. Associations of metabolites with T2D in ZMSC Multinomial logistic regression identified 225 metabolites significantly associated with T2D (FDR-adjusted P-value < 0.05), after adjusting for potential confounding factors including age, sex, BMI, hypertension status, smoking, and drinking consumption (Fig. [85]2A, Supplementary Material [86]3: Table S4). Among these significant metabolites, 80 metabolites were potential protective factors, while 145 metabolites were potential risk factors. Subgroup analysis by sex revealed 22 metabolites with FDR-adjusted P-value < 0.05 in 369 male participants, and 168 metabolites with FDR-adjusted P-value < 0.05 in 811 female participants (Supplementary Material [87]3: Table S5). Besides, to enhance the ascertainment reliability of T2D, we incorporated drug purchase records from the medical insurance system, defining T2D based on the purchase of T2D medications on at least two occasions in 2015, or FG ≥ 7.0 mmol/L for the sensitivity analysis. The sensitivity analysis based on T2D drug purchase records from the medical insurance system (171 diabetes, 127 prediabetes, and 885 normoglycemia) identified 94.44% (165) T2D-associated metabolites of the base analysis, indicating the reliability of our analysis (Supplementary Material [88]3: Table S6). Fig. 2. [89]Fig. 2 [90]Open in a new tab Metabolomic data analyses in the ZMSC. A. The score plot of the orthogonal partial least squares discriminant analysis (OPLS-DA) model. Propensity scores (by age and sex) were used to match participants with diabetes and normoglycemia. B. Pathway enrichment analysis of T2D-related metabolites. The bubble plot displays significant metabolic pathways identified by the overlap of multinomial logistic regression and OPLS-DA results Depending on the OPLS-DA scores plot (Fig. [91]2B), T2D cases and controls based on PSM could be successfully distinguished from each other. The R^2 was 0.89 and the Q^2 was 0.67 calculated by tenfold cross-validation, suggesting the robustness of the model. Furthermore, the OPLS-DA model identified 428 metabolites with VIP scores > 1 (Fig. [92]2A, Supplementary Material [93]3: Table S7). Metabolic pathway analysis Functional enrichment analysis and pathway analysis were conducted to identify potential metabolic pathways potentially associated with T2D. Only metabolites that met the statistical significance threshold of FDR-adjusted P < 0.05 from the multivariable logistic regression and had VIP scores > 1 were included in the pathway analysis (164 metabolites). We found that the T2D-associated metabolites were significantly enriched in pathways involving the “Warburg effect”, “Amino Sugar Metabolism”, “Aspartate Metabolism”, and “Fatty Acid Biosynthesis”, with P-values of 0.0152, 0.0299, 0.0363, and 0.0363, respectively (Fig. [94]2B). Genetic-informed inference supports metabolite-T2D associations in East Asian GWAS was performed to identify metabolite-associated genetic variants as IVs. The Manhattan plot shows GWAS results aggregating all the 1912 metabolites (Supplementary Material [95]2: Fig. S2). The genetic inflation factors for metabolites ranged from 0.946 to 1.004, indicating little evidence of population stratification. To provide evidence from genetic regulations, we performed metabolome-wide two-sample MR analyses to evaluate the potential causal relationships between serum metabolites and T2D. Based on the GWAS summary statistics of the ZMSC cohort and the publicly available data of the DIAGRAM consortium (EAS), metabolome-wide MR analysis identified 89 metabolites significantly associated with T2D (both P[IVW] < 0.05 and P[MR-RAPS] < 0.05, Supplementary Material [96]2: Fig. S3). The MR estimated effects of 89 genetically predicted serum metabolites on T2D risk were detailed in Supplementary Material [97]3: Table S8. To estimate potential reverse causation, we performed reverse MR analyses for the metabolite-T2D associations. Our results revealed that 75 metabolites could be influenced by T2D (Supplementary Material [98]3: Table S9), and only three of the 89 metabolite-T2D associations were perturbed. These discoveries indicate little evidence of reverse causality. In the sensitivity analyses, excluding 130 T2D participants from the ZMSC, metabolome-wide GWAS followed by MR analyses identified 98 T2D-associated metabolites (Supplementary Material [99]3: Table S10), with 93.26% (83 metabolites) of the original findings being replicated. These results further support the robustness and reliability of our findings. The Venn diagram (Fig. [100]3A) summarizes the results from multinomial logistic regression, OPLS-DA and two-sample MR analysis. Using a criterion that combined observational analysis (P[FDR] < 0.05, VIP > 1) and metabolome-wide MR analysis (P[IVW] < 0.05, P[MR-RAPS] < 0.05, P[Pleiotropy] > 0.05 and P[Heterogeneity] > 0.05), we identified 13 potential causal metabolites-T2D associations (Supplementary Material [101]3: Table S11). Among these, six metabolites exhibited consistent effects in both observational studies and genetic analyses (Fig. [102]3B, C). Notably, we found that PC (O-16:0/0:0) and its derivative LPC (O-16:0) were potential protective factors, and (R)-2-hydroxybutyric acid, 2-methyllactic acid, eplerenone and rauwolscine were potential risk factors for T2D. Fig. 3. [103]Fig. 3 [104]Open in a new tab The potential causal relationships between serum metabolites and T2D. A. The Venn diagram shows the number of identified metabolites across three methods based on metabolomic data analysis, including multinomial logistic regression, OPLS-DA, and MR. B–C. The forest plots show significant metabolite-T2D associations with the same effect direction for both multinomial logistic regression (Figure B) and Mendelian randomization analyses (Figure C). ZMSC Zhejiang metabolic syndrome cohort, OPLS-DA Orthogonal partial least squares discriminant analysis, MR Mendelian randomization. Nonlinear association between serum metabolites and T2D To further explore the potential nonlinear relationship between serum metabolites and T2D, we employed RCS analysis in ZMSC. Our analysis focused on six metabolite-T2D associations that were consistently supported by both observational and genetic evidence. We adjusted for several confounders, including age, sex, BMI, smoking, and alcohol consumption, which might influence metabolites. The minimum AIC was used to select the optimal number of RCS nodes, with the final models for each metabolite incorporating three or four nodes. Our results indicated that the associations of PC (O-16:0/0:0), LPC (O-16:0), 2-methyllactic acid, 2-hydroxybutyric acid, and rauwolscine with T2D primarily followed a linear trend, with no significant evidence of nonlinear effects or inflection points (Fig. [105]4). In contrast, the association between eplerenone and T2D displayed a “J” shape (P[nonlinear] = 0.02). Fig. 4. [106]Fig. 4 [107]Open in a new tab Testing for nonlinear association between serum metabolites and T2D in ZMSC Restricted cubic spline analysis based on logistic regression was adjusted for age, sex, smoking, drinking and body mass index. ZMSC, Zhejiang metabolic syndrome cohort. The solid line is the estimated odds ratio, and the shaded area is the 95% confidence interval. Blue represents protective factors and red represents risk factors. Metabolites and glycemic traits in East Asians We conducted comprehensive MR investigations to explore the potential causal role of metabolites in T2D and glycemic traits in EAS ancestry. Associations of genetically predicted metabolite levels with 2hGlu, FG, FI, and HbA1c are shown in Supplementary Material [108]3: Tables S12–S15. Figure [109]5 displays 12 metabolites that were significantly associated with T2D and at least one of four glycemic traits among EAS ancestry. Of 12 verifiable metabolites-T2D associations, 9 pairs were further validated in glycemic traits. Notably, genetic predisposition to higher serum creatine levels was associated with lower HbA1c levels (beta = − 0.02, P = 0.03) and T2D risk (beta = − 0.05, P = 0.02). Pelargonin was found to be a potential risk factor for T2D, existing a positive genetic correlation with FG and 2hGlu. Isofenphos-methyl (IFP), an organophosphorus pesticide, was also found to be a risk factor for T2D, positively associated with FG and HbA1c. L-lysine and FAHFA (6:0/18:3), as an essential amino acid and fat acid, were found to be a protective factor for T2D, existing a protective effect for HbA1c. Deoxycholic acid, a metabolite produced by intestinal microbiota, was positively associated with increased T2D risk and higher HbA1c levels. Additionally, we observed that elevated levels of three amino acid derivatives (Tyr-Val-Ser-Arg, Ile-Glu-Leu-Lys and Asp-Arg-Gln-Arg) were linked to an increased risk of T2D, with Tyr-Val-Ser-Arg and Ile-Glu-Leu-Lys further corroborating their association with glycemic traits. Fig. 5. [110]Fig. 5 [111]Open in a new tab Mendelian randomization associations of serum metabolites with T2D and four glycemic traits. Beta was derived from the random-effect inverse-variance weighted analysis. T2D Type 2 diabetes, 2hGlu 2-h glucose after an oral glucose challenge, FG Fasting glucose, FI Fasting insulin, HbA1c Glycated hemoglobin T2D/glycemic traits-associated metabolites in Europeans By leveraging GWAS summary data from CLSA/METSIM and DIAGRAM, metabolome-wide MR analyses identified 54 and 83 metabolites for T2D with nominal significance (P[IVW] < 0.05, Supplementary Material [112]2: Fig. S4–S5; Supplementary Material [113]3: Tables S16–S17), respectively. After FDR correction, five and one serum metabolites were significantly associated with T2D using CLSA and METSIM, respectively. Supplementary Material [114]2: Fig. S6 displays metabolites that were significantly associated with T2D and at least one of four glycemic traits among EUR ancestry. MR analysis based on CLSA found that 4 of 12 verifiable pairs of metabolite-T2D association yielded further evidence in other glycemic traits (Supplementary Material [115]2: Fig. S6A). MR analysis based on METSIM found that 9 verifiable pairs of metabolite-T2D association yielded further validation in other glycemic traits, and 4 pairs of metabolite-T2D had supporting evidence with the same effect direction (Supplementary Material [116]2: Fig. S6B). In addition, 8 pairs of metabolite-T2D/glycemic traits were able to verify each other in CLSA and METSIM (Supplementary Material [117]2: Fig. S7). Cross-ancestral analysis probes metabolites associated with T2D/glycemic traits Notably, we found 4 metabolite-trait pairs of association showed cross-ancestry consistency (Fig. [118]6). Spermidine and creatine were potential protective factors for T2D. Specifically, spermidine was negatively associated with the level of 2hGlu in EAS ancestry and the risk of T2D in EUR ancestry. Higher level of creatine was associated with lower HbA1c levels and risk of T2D in EAS ancestry, and lower levels of FI in EUR ancestry. Meanwhile, N-acetyl-L-alanine and estrone 3-sulfate were potential risk factors for glycemic traits. N-acetyl-L-alanine was positively associated with the level of FI in EAS ancestry and HbA1c in EUR ancestry. A higher level of estrone 3-sulfate was associated with higher levels of HbA1c in EUR ancestry and FG in EAS ancestry. Fig. 6. [119]Fig. 6 [120]Open in a new tab Mendelian randomization associations of metabolites with T2D and four glycemic traits across two ancestries. The beta value was derived from the fixed-effect IVW analysis of Mendelian randomization. IVW Inverse-variance weighted, EUR Europeans, EAS East Asians. T2D Type 2 diabetes, 2hGlu 2-h glucose after an oral glucose challenge, FG Fasting glucose, FI Fasting insulin, HbA1c Glycated hemoglobin Multi-omics evidence for the metabolite-T2D association By integrating multi-omics data along the gene-transcript-protein-metabolite-phenotype axis, we systematically dissected the molecular mechanisms underlying T2D across multiple biological layers. For the creatine-T2D association supported by cross-ancestry evidence, we integrated additional omics data, including genomics, transcriptomics, and proteomics, and metabolomics, to explore potential mechanistic links (Fig. [121]7A). At the genomic level, we found that the SNP rs1047891 on CPS1 was a metaboQTL for creatine in the ZMSC (P = 3.25 × 10^−9), with supporting from the CLSA (P = 2.32 × 10^−41, Supplementary Material [122]2: Fig. S8). At the transcriptomic and proteomic levels, the CPS1 gene is highly expressed in the liver and encodes a mitochondrial enzyme (CPS1), which plays a key role in catalyzing the first step of the urea cycle [[123]31, [124]32]. At the metabolomic level, arginine produced by the urea cycle is involved in creatine synthesis [[125]33]. At the phenomics level, we found that genetically predicted creatine was negatively associated with FI (beta = − 0.009, P = 0.038), and HbA1c (beta = − 0.019, P = 0.025), in EUR ancestry and T2D in EAS ancestry (beta = − 0.05, P = 0.016). Mechanistically, creatine could exert beneficial effects on glucose metabolism through two key pathways: (1) enhancing beta-cell insulin secretion and (2) improving glucose uptake via type 4 glucose transporter (GLUT-4) [[126]34]. Taken together, our results provide a putative genetics-protein-metabolite-phenotypic evidence chain for the molecular regulation of creatine. Fig. 7. [127]Fig. 7 [128]Open in a new tab The putative molecular mechanism chain for metabolite-T2D association. A. The rs1047891-CPS1-creatine-T2D pathway. At the genomic level, we found that the SNP rs1047891 on CPS1 was a metaboQTL for serum creatine. At the transcriptomic and proteomic levels, the CPS1 gene is highly expressed in the liver and encodes a mitochondrial enzyme (CPS1), which plays a key role in catalyzing the first step of the urea cycle. At the metabolomic level, arginine produced by the urea cycle is involved in creatine synthesis. At the phenomics level, we found that genetically predicted creatine was negatively associated with T2D in both European and East Asian ancestries. Mechanistically, creatine may exert beneficial effects on glucose metabolism through two key pathways: (1) enhancing beta-cell insulin secretion and (2) improving glucose uptake via type 4 glucose transporter (GLUT-4). B. The rs28946889-UGT1A1-isofenphos-methyl-T2D pathway. Isofenphos-methyl (IFP) is a commonly used organophosphorus pesticide. At the genomic level, we found that the SNP rs28946889 on UGT1A1 was a metaboQTL for serum isofenphos-methyl. At the transcriptomic and proteomic levels, the UGT1A1 gene is highly expressed in the liver and encodes a drug-metabolizing enzyme (UGT1A1), which plays a key role in metabolism and detoxification. IFP could be degraded into IFPO, ICP, and ICPO in liver microsomes. At the phenomics level, we found that genetically predicted IFP was a risk factor for T2D in Chinese. The potential mechanisms underlying this association may involve IFP-induced insulin resistance. The red arrows represent the results of our research, and the blue represents the evidence reported in the literature. T2D, type 2 diabetes; metaboQTL, metabolite quantitative trait loci; MR, Mendelian randomization. GLUT-4, type 4 glucose transporter; CPS1, carbamoyl phosphate synthetase I; UGT1A1, UDP glucuronosyltransferase family 1 member A1. IFPO, isofenphos-methyl oxon; ICP, isocarbophos; ICPO, isocarbophos oxon We also integrated multi-dimensional evidence to explore the potential molecular pathways by which environmental exposure, IFP, influences T2D (Fig. [129]7B). IFP could be degraded into IFPO, ICP, and ICPO in liver microsomes [[130]35, [131]36]. At the genomic level, we found that the SNP rs28946889 on UGT1A1 gene was a metaboQTL for serum isofenphos-methyl in the ZMSC (effect allele: T, beta = − 0.24, P = 2.13 × 10^−8). At the transcriptomic and proteomic levels, the UGT1A1 gene is highly expressed in the liver and encodes a drug-metabolizing enzyme (UDP glucuronosyltransferase family 1 member A1, UGT1A1), which plays a key role in metabolism and detoxification (GTEx v10, effect allele: T, normal effect size = 0.23, P = 8.30 × 10^−6) [[132]37]. At the phenomics level, we found that genetically predicted IFP was a risk factor for T2D in Chinese [[133]38]. Mechanistically, the IFP-T2D association involves IFP-induced insulin resistance [[134]38]. Collectively, these findings support a putative genetics-expression-protein-metabolite-phenotypic evidence chain for the molecular regulation of IFP. Urea cycle-related metabolites and cardiovascular complications of T2D By leveraging GWAS summary data from METSIM and FinnGen, metabolome-wide MR analyses identified 15 significant associations between 9 urea cycle-related metabolites and 6 cardiovascular complications linked to T2D (Supplementary Material [135]2: Fig. S9, Supplementary Material [136]3: Table S18). Among these, N-acetylarginine, N-delta-acetylornithine, and N2, N5-diacetylornithine were each associated with at least one cardiovascular trait. Notably, N-acetylarginine and N2, N5-diacetylornithine were identified as risk factors for AF, IHD and HF. In contrast, N-delta-acetylornithine demonstrated a protective role against AF, IHD, and cardiomyopathy, highlighting the nuanced metabolic contributions of urea cycle-related metabolites to cardiovascular health. Discussion In this study, we performed a comprehensive metabolome-wide analysis to uncover the associations between metabolites and CMD with the support of genetic regulations, providing novel insights into the metabolic alterations underlying CMD. Notably, we identified six metabolites with multiple lines of evidence for T2D in EAS ancestry, including PC(O-16:0/0:0), LPC(O-16:0), 2-methyllactic acid, 2-hydroxybutyric acid, eplerenone, and rauwolscine. Furthermore, our study revealed four metabolites implicated in glucose metabolism shared across EAS and EUR ancestries, including creatine, spermidine, N-acetyl-L-alanine and estrone 3-sulfate. By integrating multi-omics data, we elucidated the putative mechanisms underlying the creatine-T2D association and identified the urea cycle pathway as a potential therapeutic target for T2D and its cardiovascular complications. MS facilitates comprehensive metabolome profiling, offering valuable opportunities to uncover novel metabolites associated with T2D. For instance, our findings suggested PC(O-16:0/0:0) and its derivative LPC(O-16:0) as potential protective factors in T2D development among Han Chinese. Besides, case–control analyses in ZMSC revealed several novel LPC subclasses, including LPC(O-18:1), LPC(O-16:1), and LPC(O-18:3), which were inversely associated with T2D risk. Prior research has shown that PC and LPC classes exert beneficial effects on glucose metabolism [[137]9], and our results further expand the potential causal evidence for specific PC and LPC subclasses. In line with our findings, the Hong Kong Cardiovascular Risk Factor Prevalence Study (CRISPS) revealed LPC(O-16:0) was independently associated with decreased T2D risk [[138]39]. Further functional investigations demonstrated that LPC (O-16:0) significantly potentiated glucose-induced insulin secretion in a dose-dependent manner [[139]39]. Corroborating and extending the previous studies, the consistent protective effects of LPC(O-16:0) and its upstream metabolite PC (O-16:0/0:0) observed in our study strengthen the genetic evidence supporting their roles in T2D. However, evidence on other LPC subclasses remains limited, underscoring the need for further research to validate their effects on T2D and elucidate the underlying mechanisms. We identified several key metabolic pathways that may play a role in the etiology of T2D. Specifically, four metabolic pathways emerged as significantly involved in T2D, including the Warburg effect, amino sugar metabolism, aspartate metabolism, and fatty acid biosynthesis. These findings align with and extend the existing evidence. Firstly, T2D is associated with metabolic alterations resembling the Warburg effect, characterized by upregulated anaerobic glycolysis and diminished mitochondrial function [[140]40, [141]41]. Notably, the antidiabetic drug metformin has been shown to disrupt the Warburg effect, further supporting the link between this pathway and T2D pathogenesis [[142]42]. Secondly, both amino sugar metabolism and fatty acid biosynthesis have been implicated in the development of diabetes, with mounting evidence pointing to their roles in glucose homeostasis and lipid dysregulation. Thirdly, aspartate metabolism has also been linked to T2D, with previous studies revealing that elevated asparagine-to-aspartate ratios are associated with an increased risk of the disease [[143]43, [144]44]. Collectively, these results deepen our understanding of the metabolic underpinnings of T2D and offer a foundation for further research into therapeutic interventions targeting these pathways. MR provides an effective strategy for identifying potential causal associations between metabolites and T2D. Our study demonstrated strong consistency in MR findings, with sensitivity analyses excluding T2D participants in the ZMSC showing consistent results for 83 out of 89 (93.26%) of the initially identified metabolite associations. However, the six non-replicated metabolites, including tetramisole, barbituric acid, 5-methoxytryptamine, hydroxyquinoline, octadecapentaenoic acid, and ketoleucine, should be interpreted cautiously. We note that there is a delicate balance in deciding whether to include the T2D samples. Including them may increase the potential complexity of the model, which may result in a biased estimate, while excluding them may slightly increase the standard error of effect size estimation. Therefore, we emphasize the importance of prioritizing metabolites that demonstrate consistent results in sensitivity analysis. Furthermore, metabolites with concordant evidence from multiple analytical approaches, such as MR analysis, case–control studies, and cross-ancestry multi-omics studies, are more likely to represent causal factors in T2D pathogenesis. Certain metabolites exhibit conserved genetic regulation in relation to T2D across diverse ancestries, providing compelling evidence for their robustness and generalizability as causal molecular signatures of T2D. Among the identified associations through our MR analysis, four metabolites exhibited cross-ancestry consistent effects on T2D and glycemic traits. Creatine and spermidine were found to exert beneficial effects on glucose regulation, while N-acetyl-L-alanine and estrone 3-sulfate were associated with detrimental impacts. Interestingly, previous observational studies have reported ethnic differences in creatine-T2D association. For instance, the observational analysis in the METSIM study indicated that creatine was associated with a higher risk of T2D in EUR ancestry [[145]45]. However, the residual confounding factors in the observational study may obscure the actual association. Our MR analysis indicated potential beneficial effects of creatine on T2D in EAS ancestry and FI levels in EUR ancestry, providing consistent evidence supporting the protective role of creatine in reducing T2D risk. These findings align with a recent randomized controlled trial (RCT) demonstrating that creatine supplementation, in combination with exercise, improves glycemic control in individuals with diabetes [[146]34]. Besides, spermidine, a natural polyamine derived from dietary intake, gut microbiota, and cellular biosynthesis, is known for its health-promoting effects [[147]46]. Aligned with our MR results, a cross-sectional study among Chinese rural adults and the National Health and Nutrition Examination Survey (NHANES) 2009–2010 also reported an inverse association of spermidine with T2D and glucose levels [[148]47, [149]48]. Animal experiments also demonstrated that spermidine could modulate glucose homeostasis and mitigate insulin resistance [[150]48]. Additionally, supporting our findings, previous studies among African Americans and mixed populations have demonstrated that higher levels of N-acetyl-l-alanine and estrone sulfate were associated with the risk of T2D [[151]49, [152]50]. Exploring the influence of metabolites on blood glucose traits is conducive to revealing the potential mechanism through which metabolites influence T2D. For example, our study identified IFP, a commonly used organophosphorus pesticide, as a potential risk factor for T2D, with its effects on FG and HbA1c aligning in a consistent, adverse direction. Supporting our findings, a case–control study conducted in rural Chinese populations demonstrated a positive association between isofenphos exposure and both impaired FG and T2D [[153]38]. The potential mechanisms underlying this association may involve IFP-induced lipid toxicity, inflammatory stimulation, and oxidative stress, which collectively impair the action of proinsulin, promote insulin resistance, and ultimately lead to the onset of T2D [[154]38]. Notably, the integration of multi-omics evidence further provides a potential framework linking genes, transcription, metabolism, and phenotype. The rs28946889, a genetic IV for IFP in our MR analysis, is located in the UGT1A1 gene, which plays a pivotal role in metabolizing endogenous and exogenous compounds, including pesticides [[155]51, [156]52]. Additionally, GTEx v10 data (liver tissue) revealed a positive correlation between the rs28946889-T allele and UGT1A1 expression [[157]37], while our analysis demonstrated a negative correlation between rs28946889-T and IFP levels. These results suggest that UGT1A1 polymorphisms influence UGT1A1 expression, thereby modulating IFP metabolism and potentially contributing to T2D pathogenesis. Our investigation further strengthens the evidence linking IFP exposure to disrupted glucose metabolism and increased T2D risk. These findings underscore the need to evaluate the metabolic and public health impacts of widespread pesticide use and highlight the broader implications of environmental exposures on metabolic health. In addition to identifying novel metabolite-trait associations, we sought to explore the regulatory mechanisms underlying these metabolites through a multi-omics integration strategy. By integrating genomics to reveal genetic predisposition, transcriptomics to link variants with gene expression, proteomics to connect expression with function, metabolomics to map metabolic changes, and phenotypic analysis to relate molecular alterations to T2D, we provided deeper mechanistic insights into disease processes. For instance, we elucidated the rs1047891-CPS1-creatine-T2D regulatory pathway. We found that genetically predicted creatine was negatively associated with T2D in both EUR and EAS ancestries. Multi-omics integration revealed that rs1047891 in CPS1 is a significant metaboQTL for creatine, demonstrating consistent genetic regulation across ancestries. The CPS1 gene, highly expressed in the liver, encodes a mitochondrial enzyme central to the urea cycle, which produces arginine, a precursor for creatine synthesis. Mechanistically, creatine may improve glucose metabolism by enhancing beta-cell insulin secretion and promoting glucose uptake via GLUT-4 [[158]34]. These findings highlight the critical role of integrating multi-omics data to uncover regulatory pathways in T2D. Multi-omics approaches enable a comprehensive understanding of biological processes by linking genetic variants to functional molecules and their downstream effects. This integrative strategy is especially important for identifying ancestry-consistent mechanisms, bridging knowledge gaps, and informing targeted therapeutic development in precision medicine. Our findings also provide novel insights into the role of urea cycle-related metabolites in the pathophysiology of T2D-associated cardiovascular complications. By integrating cross-ancestry multi-omics evidence, we identified urea cycle-related metabolites as key biomarkers of T2D, with MR analysis further supporting their involvement in cardiovascular pathology. Among the identified metabolites, N-acetylarginine, N-delta-acetylornithine and N2, N5-diacetylornithine belong to the arginine-ornithine metabolic pathway, where ornithine and arginine serve as core components of the urea cycle. Notably, N-acetylarginine and N2, N5-diacetylornithine were identified as potential risk factors, suggesting that dysregulated arginine metabolism may contribute to cardiovascular pathology in individuals with T2D. Mechanistically, arginine and its derivatives play a crucial role in nitric oxide (NO) synthesis, endothelial function, and vascular homeostasis [[159]53]. Conversely, N-delta-acetylornithine exhibited a potential cardioprotective effect, likely mediated through its involvement in ornithine metabolism, which is closely linked to polyamine synthesis and oxidative stress regulation that may mitigate cardiovascular damage [[160]54, [161]55]. These findings suggest that targeting specific metabolic pathways within the urea cycle might offer novel therapeutic strategies for mitigating cardiovascular complications in high-risk T2D populations. Furthermore, our multi-omics approach highlights the power of integrating genetics, metabolomics, and phenotypic data to uncover potential therapeutic targets for cardiometabolic diseases. Future studies should explore whether metabolic interventions, such as dietary modulation, pharmacological targeting, or gut microbiome-based therapies, could beneficially modulate these metabolites to improve cardiovascular outcomes. There were several limitations in the current study. Firstly, reliable causal inference in MR relies on three key assumptions: relevance, independence, and exclusion, which can be challenging to fully validate, especially in the presence of potential pleiotropy. We have used multiple methods and sensitivity analyses to minimize potential biases arising from weak instruments and unbalanced horizontal pleiotropy. Besides, we prioritized metabolite-trait associations supported by multi-dimensional evidence, including epidemiological, genetic, or functional studies, to strengthen our findings. Secondly, although this study covered a broad spectrum of metabolites, the roles and underlying mechanisms of certain metabolites in disease processes remain unclear, limiting the depth of interpretation of some findings. Thirdly, due to data limitations, FG was the main diagnostic criterion accessible for T2D and prediabetes. However, exclusive reliance on FG may fail to capture cases of isolated postprandial hyperglycemia, which is common in the early stages of T2D. Future studies should incorporate postprandial glucose measurements to provide a more comprehensive assessment of T2D and prediabetes. Fourthly, in the cross-sectional study of ZMSC, the lack of HbA1c information may lead to an underestimation of T2D cases. However, the robustness of our findings was supported by a sensitivity analysis using Medicare data, which mitigates potential biases and strengthens the validity of our results. Fifthly, our metabolite GWAS in ZMSC included a small proportion of T2D participants. While genetic regulation of metabolites is generally robust, T2D may perturb these relationships, complicating the interpretation of genetic effects. Sixthly, the proposed molecular pathways linking metabolites to T2D were primarily inferred from multi-omics integration. While these findings provide valuable mechanistic insights, further experimental studies are necessary to confirm the proposed pathway and elucidate the underlying biological mechanisms. Seventhly, while we adjusted for key lifestyle factors in the ZMSC case–control analyses, residual confounding from unmeasured environmental or lifestyle variables may still exist. To enhance reliability, we also conducted MR analyses, which leverage the random allocation of alleles to minimize environmental confounding, providing more robust causal inference. Lastly, although our study identified promising molecular signatures for T2D and its cardiovascular complications, their clinical applicability remains preliminary. Validation in larger cohorts and clinical trials is necessary, as the current sample size and follow-up period limit the ability to assess long-term outcomes. Further research is needed to explore how these biomarkers can complement existing clinical tools, such as HbA1c and FG, and inform personalized treatment strategies. Conclusion This study provides a comprehensive metabolome-wide investigation into the potential etiology of T2D, offering new insights into its metabolic basis and associated cardiovascular complications. Through an integrated multi-omics approach, we identified critical metabolites and molecular pathways associated with CMD. These findings not only enhance our understanding of the complex metabolic networks underlying CMD pathophysiology but also provide valuable biomarkers for early disease detection and potential therapeutic targets. Supplementary Information [162]Supplementary Material 1.^ (20.1KB, docx) [163]Supplementary Material 2.^ (1.3MB, docx) [164]Supplementary Material 3.^ (3.8MB, xlsx) Acknowledgements