Residual lipid-related proteomic and metabolomic signatures associated with major adverse cardiovascular events after revascularization in peripheral artery disease Authors: Shaobo Cao, Bowen Zhang, Peng Li, Jiayin Luo, Shuping Tan, Chengran Lu, Fang Yu, Chenguang Yang, Zuoguan Chen, Yongjun Li, Yongpeng Diao Published: 2026 ABSTRACT BackgroundPatients with peripheral artery disease (PAD) remain at high risk for major adverse cardiovascular events (MACE) after lower-extremity revascularization. In contemporary practice, widespread lipid-lowering therapy may reduce the discriminatory value of conventional lipid parameters and clinically defined hyperlipidemia (HLD), leaving residual lipid-related molecular risk insufficiently characterized. This study aimed to characterize residual lipid-related proteomic and metabolomic signatures in PAD and examine their association with MACE after revascularization.MethodsWe enrolled a prospective cohort of 165 consecutive patients with PAD who underwent lower-extremity endovascular revascularization. Untargeted plasma metabolomic and proteomic profiling was performed. Lipid and lipid-like metabolite coexpression modules were identified using weighted gene coexpression network analysis, and molecular subtypes were defined using nonnegative matrix factorization clustering. Differential metabolites and proteins between subtypes were identified, and a compact multi-omics molecular score was constructed using LASSO regression. Associations with MACE were evaluated using Kaplan-Meier analysis, time-dependent receiver operating characteristic analysis, and multivariable Cox regression.ResultsWe identified a lipid-related coexpression module associated with HLD but independent of statin use, with enrichment in sphingolipid metabolism, bile acid biosynthesis, and steroid hormone biosynthesis. Molecular clustering stratified patients into two subtypes (C1, n = 64; C2, n = 101) with distinct lipid metabolomic and proteomic profiles. The C1 subtype had significantly poorer MACE-free survival (log-rank P = 0.001). A compact multi-omics score comprising QDPR and four putatively annotated metabolites (deoxycholylphenylalanine, succinylcarnitine, glutarylcarnitine, and acetolein) discriminated between subtypes and was independently associated with MACE (hazard ratio = 1.25, P = 0.032), with a 5-fold cross-validated 15-month AUC of 0.871.ConclusionsIntegrated proteomic and metabolomic profiling revealed residual lipid-related molecular signatures beyond conventional lipid parameters in patients with PAD undergoing lower-extremity revascularization. These signatures were associated with subsequent MACE, supporting further validation of molecular biomarkers for postoperative cardiovascular risk stratification in PAD.Graphical Abstract Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12872-026-06621-y. Introduction Peripheral artery disease (PAD) is a common manifestation of atherosclerotic vascular disease and imposes a growing global burden, driven in part by population aging and metabolic risk [1]. PAD can lead to lower-limb ischemia, impaired walking capacity, reduced quality of life and, in severe cases, chronic limb-threatening ischemia and amputation. Importantly, PAD is not merely a localized lower-extremity arterial disorder, but a peripheral manifestation of systemic atherosclerotic cardiovascular disease. Even after lower-extremity revascularization, patients with PAD often retain a substantial systemic atherosclerotic burden and remain at high risk of major adverse cardiovascular events (MACE), including myocardial infarction, stroke and cardiovascular death [2, 3]. Identifying patients at high risk of MACE is therefore an important issue in cardiovascular risk assessment and long-term PAD management [4, 5]. Dyslipidemia is a key driver of atherosclerosis and an important contributor to MACE risk in patients with PAD [6]. Conventional lipid parameters, including triglycerides, total cholesterol, low-density lipoprotein cholesterol and high-density lipoprotein cholesterol, together with clinically defined hyperlipidemia status, have traditionally been used to assess MACE-related risk [7–10]. However, lipid-lowering therapies, particularly statins, are now widely used in patients with PAD [11, 12]. In this contemporary treatment context, conventional lipid parameters and hyperlipidemia labels may no longer fully reflect the true MACE-related risk of individual patients [13–17]. Whether residual lipid-related risk phenotypes exist beyond standard lipid measurements and are associated with MACE remains to be clarified. High-throughput metabolomics provides a systematic approach to characterize circulating lipid and lipid-like metabolites beyond routine clinical lipid measurements. In parallel, proteomic profiling can capture inflammatory, coagulation-related, and vascular injury networks that interact with metabolic phenotypes. Integrating these molecular layers may help reveal systems-level signatures of lipid-related metabolic heterogeneity in PAD [18–20]. In this prospective cohort study, we integrated plasma proteomic and metabolomic profiling in patients with PAD undergoing lower-extremity endovascular revascularization. We aimed to identify residual lipid-related molecular signatures beyond conventional lipid assessment and examine their association with post-revascularization MACE. Methods Study population and study design This prospective cohort study was approved by the Ethics Committee of Beijing Hospital (Approval No.: 2025BJYYEC-KY171-01), and all participants provided written informed consent. From an ongoing cohort of patients undergoing lower-extremity endovascular revascularization, 313 patients were assessed for eligibility. Eligible patients were aged > 18 years, had PAD diagnosed according to established clinical criteria [15] and confirmed by imaging showing > 70% stenosis in the iliac or infrainguinal arteries, were scheduled for unilateral or bilateral revascularization, and provided a venous blood sample. Patients with active infection, a cardiovascular event within 1 month, major surgery or severe trauma within 3 months, or autoimmune disease were excluded. Of 175 patients who underwent plasma proteomic and metabolomic profiling, 10 with severe renal failure (serum creatinine > 442 µmol/L) were excluded because advanced renal dysfunction can substantially alter circulating omics profiles. The final analytical cohort comprised 165 patients (Fig. S1). Clinical data collection and outcome definition Baseline data included age, sex, height, weight, body mass index (BMI), Rutherford classification, ankle-brachial index (ABI), medication history, hypertension, hyperlipidemia, diabetes, and lipid-related variables, including triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), lipid-lowering therapy, and statin use. Clinical data were independently reviewed and standardized by two experienced physicians. Patients were followed up at 1, 3, and 6 months after revascularization and annually thereafter through outpatient visits, medical record review, or telephone interviews. Follow-up ended at the first MACE, last contact, or June 2025, whichever occurred first. Events were adjudicated by trained vascular surgeons using available clinical and imaging data. The primary outcome was MACE, defined as nonfatal myocardial infarction, stroke, or cardiovascular death. Blood sample collection, omics profiling, and data preprocessing Fasting venous blood was collected within 12 h of admission into EDTA tubes, centrifuged at 1,500 × g for 10 min at 4 °C, and stored as plasma aliquots at − 80 °C until analysis. Metabolomic analysis Plasma samples were extracted with a methanol-acetonitrile mixture containing internal standards, followed by sonication, protein precipitation, centrifugation, nitrogen drying, and reconstitution. Untargeted metabolomic profiling was performed on a UHPLC-Q Exactive HF-X system (Thermo Fisher Scientific) with an ACQUITY UPLC HSS T3 column in positive and negative ion modes. Pooled quality-control samples were injected regularly to monitor analytical stability. Raw data were processed in Progenesis QI software (Waters) for baseline filtering, peak detection, alignment, and normalization. Feature intensities were normalized to internal standards, log-transformed, and median-centered. Of approximately 3,000 detected metabolites, 645 lipid and lipid-like metabolites were retained for lipid-focused analyses (Table S1). Metabolites were annotated using accurate mass-to-charge ratios (m/z), MS/MS spectra, adduct information, and database records. Annotation-related information is reported in Table S1. Metabolites not confirmed using authentic standards were considered putative annotations. Details for the four metabolite features included in the molecular score are provided in Table S4. Proteomic analysis Plasma proteomics was performed using a Vanquish Neo UHPLC system coupled with an Orbitrap Astral mass spectrometer in data-independent acquisition mode. Samples were processed with PTM-Max magnetic nanoparticles (PTM Bio, Hangzhou, China), followed by on-bead digestion, reduction with dithiothreitol, alkylation with iodoacetamide, and C18 cleanup. DIA data were processed in Spectronaut (v18, Biognosys) against the UniProt human reference proteome with a reverse decoy database. Trypsin/P digestion with up to two missed cleavages was specified; carbamidomethylation of cysteine was fixed, and protein N-terminal acetylation and methionine oxidation were variable modifications. The false discovery rate was controlled below 1% at the protein, peptide, and peptide-spectrum match levels. Missing values were imputed using K-nearest neighbors, and protein abundances were center-normalized after filtering. Identification of clinically relevant lipid metabolic co-expression modules Principal component analysis (PCA) was used to assess lipid metabolic profiles in the HLD and No-HLD groups. Lipid and lipid-like metabolites were clustered into co-expression modules using weighted gene co-expression network analysis (WGCNA) [21]. The soft-thresholding power was selected according to the scale-free topology fit index, and modules were identified using dynamic tree cutting. Pearson correlation analysis assessed associations between module eigengenes and clinical variables, including HLD status and statin use. In this study, residual lipid-related molecular features were operationally defined as lipid and lipid-like metabolite patterns associated with HLD status, not explained by statin use, and linked to molecular subtype differences and subsequent MACE risk. To examine the robustness of this definition across current LDL-C levels, complete-case sensitivity analyses were stratified at an LDL-C threshold of 1.8 mmol/L. Multivariable linear regression within each stratum adjusted for statin use, Rutherford classification, age, sex, diabetes, and smoking status. Effect modification was assessed using an HLD-by-LDL-C stratum interaction term, and an additional model included LDL-C as a continuous covariate. Multivariable linear regression assessed associations between candidate modules and clinical variables after adjustment for age, sex, Rutherford classification, smoking status, diabetes, and statin use. Partial correlation analysis examined the association between HLD and candidate modules. Hub metabolites were identified by Pearson correlations between module membership and metabolite significance. KEGG pathway enrichment and lipid class annotation were performed using MetaboAnalyst 6.0. Identification and characterization of lipid metabolic subtypes Non-negative matrix factorization (NMF) [22] was applied to subtype-specific lipid metabolites within candidate modules to identify lipid metabolic subtypes. Data were normalized before clustering, and the optimal number of clusters was selected using the cophenetic correlation coefficient, silhouette score, and reconstruction error. PCA and loading analyses characterized metabolic differences and major contributors to subtype separation, and heatmaps were used to visualize metabolite expression and lipid class composition. Identification of differential metabolites and pathway enrichment analysis between subtypes Orthogonal partial least-squares discriminant analysis was used to identify differential metabolites between lipid metabolic subtypes. Differential metabolites were defined by a variable importance in projection score > 1, an absolute log2 fold change > 1, and a false discovery rate < 0.05. Differential metabolites were visualized using bubble plots, and lipid class composition and KEGG pathway enrichment were subsequently assessed. Differential proteomic analysis and functional annotation Differential protein expression between subtypes was assessed using the limma R package [23], with adjustment for statin use. Differentially expressed proteins were defined by an FDR-adjusted P value < 0.05 and an absolute log2 fold change > 0.58. Protein expression patterns were visualized using volcano plots and heatmaps, and Gene Ontology enrichment analysis was performed using clusterProfiler [24]. Multi-omics integration analysis Spearman correlation analysis assessed protein-metabolite associations, with |r| >0.3 and FDR < 0.05 considered significant. Principal components derived from differential metabolites and proteins were correlated to assess the overall relationship between the two omics layers. Two-way orthogonal partial least-squares discriminant analysis was performed using OmicsPLS [25] to identify core multi-omics features associated with subtype separation. Construction of the multi-omics molecular score and prognostic analysis Feature selection was performed among differential metabolites and proteins using least absolute shrinkage and selection operator regression implemented in glmnet [26]. Hyperparameters were tuned by 10-fold cross-validation. Candidate models were evaluated by receiver operating characteristic curves and area under the curve using pROC [27], and AUCs were compared using the DeLong test. Multivariable logistic regression assessed the independent discriminative value of the molecular score for subtype classification after adjustment for clinical variables. MACE-free survival was compared between molecular score strata using Kaplan-Meier analysis and the log-rank test. Time-dependent ROC analysis was performed using timeROC [28]. Internal validation of 15-month prognostic performance used stratified 5-fold cross-validation. Within each training fold, missing-value imputation, feature standardization, LASSO selection, and logistic model fitting were performed before out-of-fold scores were generated for held-out patients. Pooled out-of-fold scores were used to calculate the cross-validated 15-month AUC. Bootstrap percentile 95% confidence intervals for apparent and cross-validated AUCs were based on 1,000 valid patient-level resamples. Stepwise-adjusted Cox proportional hazards models evaluated exploratory covariate-adjusted associations between the molecular score and MACE. Covariates were selected a priori based on their established clinical relevance to cardiovascular risk in PAD. Model 1 included the molecular score; Model 2 additionally included age, Rutherford classification, and smoking status; and Model 3 further included HLD and statin use. The molecular score and age were modeled continuously, with hazard ratios expressed per one-unit increase in the score and per one-year increase in age; Rutherford classification was modeled as an ordinal variable, and smoking status, HLD, and statin use as binary variables. No covariates in the Cox models had missing values. The proportional hazards assumption was assessed using scaled Schoenfeld residuals and the global test. Given the limited number of events, these Cox analyses were intended to evaluate exploratory covariate-adjusted associations rather than to develop a clinical prediction model. Subgroup analyses were conducted in statin-treated and non-HLD patients, and interaction terms tested modification by HLD and statin use. Restricted cubic spline analysis evaluated the dose-response association between the molecular score and MACE. Statistical analysis All analyses were performed using R version 4.5.0. Continuous variables were summarized as mean ± standard deviation or median (interquartile range) and compared using t tests or Wilcoxon rank-sum tests, as appropriate. Categorical variables were summarized as n (%) and compared using chi-square or Fisher exact tests. All tests were two-sided, with P < 0.05 considered statistically significant. Multiple comparisons were adjusted using the false discovery rate. Figures were generated in R, and pathway and network analyses were performed using MetaboAnalyst 6.0 [29, 30]. Results Baseline demographic and clinical characteristics of participants The final analytical cohort comprised 165 patients with PAD, including 72 with HLD and 93 without HLD (Table 1). Compared with the No-HLD group, patients with HLD were younger (67.75 ± 8.62 vs. 70.88 ± 8.85 years, P = 0.024) and more frequently received lipid-lowering therapy, predominantly statins (P = 0.014). Other baseline characteristics, including sex, BMI, Rutherford classification, smoking status, hypertension, diabetes, TG, HDL-C, and LDL-C, did not differ significantly between groups. Table 1Baseline demographic and clinical characteristics of PAD patients stratified by hyperlipidemia statusVariablesCategoryHLD (n = 72)No-HLD (n = 93)Total (n = 165)P-valueGenderMale57 (79.2%)70 (75.3%)127 (77.0%)0.687Female15 (20.8%)23 (24.7%)38 (23.0%)AgeMean ± SD67.75 ± 8.6270.88 ± 8.8569.52 ± 8.860.024BMIMean ± SD24.05 ± 2.8923.83 ± 3.3023.93 ± 3.120.643RutherfordLow6 (8.3%)8 (8.6%)14 (8.5%)0.946Median41 (56.9%)55 (59.1%)96 (58.2%)High25 (34.8%)30 (32.3%)55 (33.3%)Missing0 (0.0%)0 (0.0%)0 (0%)HypertensionNo23 (31.9%)28 (30.1%)51 (30.9%)0.934Yes49 (68.1%)65 (69.9%)114 (69.1%)DiabetesNo27 (37.5%)42 (45.2%)69 (41.8%)0.406Yes45 (62.5%)51 (54.8%)96 (58.2%)SmokingNo43 (59.7%)54 (58.1%)97 (58.8%)0.956Yes29 (40.3%)39 (41.9%)68 (41.2%)Lipid-lowering drugsNone10 (13.9%)30 (32.3%)40 (24.2%)0.014Atorvastatin43 (59.7%)52 (55.9%)95 (57.6%)Rosuvastatin13 (18.1%)7 (7.5%)20 (12.1%)Others6 (8.3%)4 (4.3%)10 (6.1%)Triglycerides (mmol/L)Median [IQR]1.41 [0.96, 1.73]1.28 [0.88, 1.74]1.33 [0.94, 1.73]0.493Missing15 (20.8%)22 (23.7%)37 (22.4%)HDL-C (mmol/L)Median [IQR]0.97 [0.83, 1.10]0.99 [0.88, 1.18]0.99 [0.88, 1.13]0.599Missing17 (23.6%)21 (22.6%)38 (23.0%)LDL-C (mmol/L)Median [IQR]2.07 [1.71, 3.01]2.22 [1.61, 2.66]2.17 [1.61, 2.73]0.383Missing19 (26.4%)24 (25.8%)43 (26.1%)Differences between the HLD and No-HLD groups were assessed using the chi-square test for categorical variables, the independent-samples t test for normally distributed continuous variables, and the Mann–Whitney U test for non-normally distributed continuous variables. Data are presented as n (%), mean ± standard deviation (SD), or median [interquartile range (IQR)], as appropriate. Missing data are shown where applicable. P < 0.05 was considered statistically significantHLD Hyperlipidemia, BMI Body mass index, HDL-C High-density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, IQR Interquartile range, SD Standard deviation Table 1Baseline demographic and clinical characteristics of PAD patients stratified by hyperlipidemia statusVariablesCategoryHLD (n = 72)No-HLD (n = 93)Total (n = 165)P-valueGenderMale57 (79.2%)70 (75.3%)127 (77.0%)0.687Female15 (20.8%)23 (24.7%)38 (23.0%)AgeMean ± SD67.75 ± 8.6270.88 ± 8.8569.52 ± 8.860.024BMIMean ± SD24.05 ± 2.8923.83 ± 3.3023.93 ± 3.120.643RutherfordLow6 (8.3%)8 (8.6%)14 (8.5%)0.946Median41 (56.9%)55 (59.1%)96 (58.2%)High25 (34.8%)30 (32.3%)55 (33.3%)Missing0 (0.0%)0 (0.0%)0 (0%)HypertensionNo23 (31.9%)28 (30.1%)51 (30.9%)0.934Yes49 (68.1%)65 (69.9%)114 (69.1%)DiabetesNo27 (37.5%)42 (45.2%)69 (41.8%)0.406Yes45 (62.5%)51 (54.8%)96 (58.2%)SmokingNo43 (59.7%)54 (58.1%)97 (58.8%)0.956Yes29 (40.3%)39 (41.9%)68 (41.2%)Lipid-lowering drugsNone10 (13.9%)30 (32.3%)40 (24.2%)0.014Atorvastatin43 (59.7%)52 (55.9%)95 (57.6%)Rosuvastatin13 (18.1%)7 (7.5%)20 (12.1%)Others6 (8.3%)4 (4.3%)10 (6.1%)Triglycerides (mmol/L)Median [IQR]1.41 [0.96, 1.73]1.28 [0.88, 1.74]1.33 [0.94, 1.73]0.493Missing15 (20.8%)22 (23.7%)37 (22.4%)HDL-C (mmol/L)Median [IQR]0.97 [0.83, 1.10]0.99 [0.88, 1.18]0.99 [0.88, 1.13]0.599Missing17 (23.6%)21 (22.6%)38 (23.0%)LDL-C (mmol/L)Median [IQR]2.07 [1.71, 3.01]2.22 [1.61, 2.66]2.17 [1.61, 2.73]0.383Missing19 (26.4%)24 (25.8%)43 (26.1%)Differences between the HLD and No-HLD groups were assessed using the chi-square test for categorical variables, the independent-samples t test for normally distributed continuous variables, and the Mann–Whitney U test for non-normally distributed continuous variables. Data are presented as n (%), mean ± standard deviation (SD), or median [interquartile range (IQR)], as appropriate. Missing data are shown where applicable. P < 0.05 was considered statistically significantHLD Hyperlipidemia, BMI Body mass index, HDL-C High-density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, IQR Interquartile range, SD Standard deviation Baseline demographic and clinical characteristics of PAD patients stratified by hyperlipidemia status Differences between the HLD and No-HLD groups were assessed using the chi-square test for categorical variables, the independent-samples t test for normally distributed continuous variables, and the Mann–Whitney U test for non-normally distributed continuous variables. Data are presented as n (%), mean ± standard deviation (SD), or median [interquartile range (IQR)], as appropriate. Missing data are shown where applicable. P < 0.05 was considered statistically significant HLD Hyperlipidemia, BMI Body mass index, HDL-C High-density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, IQR Interquartile range, SD Standard deviation Anatomical and procedural characteristics are summarized in Table 2. The superficial femoral artery was the most frequently treated territory (67.3%), followed by the iliac (29.1%), infrapopliteal (26.1%), and popliteal (25.5%) arteries. Multiple-site and bilateral treatment were performed in 49.1% and 12.7% of patients, respectively; balloon angioplasty and stent implantation were performed in 58.8% and 41.2%. Anatomical distribution, treatment laterality, multiple-site treatment, and procedural strategy did not differ between C1 and C2 (all P > 0.05). Table 2Anatomical and procedural characteristics of the study populationCharacteristicOverall (n = 165)C1 (n = 64)C2 (n = 101)P valueAnatomical territories treated Iliac artery48 (29.1%)22 (34.4%)26 (25.7%)0.311 Superficial femoral artery111 (67.3%)37 (57.8%)74 (73.3%)0.059 Popliteal artery42 (25.5%)19 (29.7%)23 (22.8%)0.418 Infrapopliteal arteries43 (26.1%)20 (31.3%)23 (22.8%)0.305 Multiple-site treatment81 (49.1%)37 (57.8%)44 (43.6%)0.104Treatment laterality0.052 Left79 (47.9%)30 (46.9%)49 (48.5%) Right65 (39.4%)21 (32.8%)44 (43.6%) Bilateral21 (12.7%)13 (20.3%)8 (7.9%)Primary procedural strategy0.776 Balloon angioplasty97 (58.8%)39 (60.9%)58 (57.4%) Stent implantation68 (41.2%)25 (39.1%)43 (42.6%)Data are presented as n (%). Anatomical territories were not mutually exclusive because some patients underwent treatment at more than one arterial site. Multiple-site treatment was defined as treatment involving two or more side-specific arterial territories, including bilateral treatment at the same anatomical level. P values were calculated using the chi-square test Table 2Anatomical and procedural characteristics of the study populationCharacteristicOverall (n = 165)C1 (n = 64)C2 (n = 101)P valueAnatomical territories treated Iliac artery48 (29.1%)22 (34.4%)26 (25.7%)0.311 Superficial femoral artery111 (67.3%)37 (57.8%)74 (73.3%)0.059 Popliteal artery42 (25.5%)19 (29.7%)23 (22.8%)0.418 Infrapopliteal arteries43 (26.1%)20 (31.3%)23 (22.8%)0.305 Multiple-site treatment81 (49.1%)37 (57.8%)44 (43.6%)0.104Treatment laterality0.052 Left79 (47.9%)30 (46.9%)49 (48.5%) Right65 (39.4%)21 (32.8%)44 (43.6%) Bilateral21 (12.7%)13 (20.3%)8 (7.9%)Primary procedural strategy0.776 Balloon angioplasty97 (58.8%)39 (60.9%)58 (57.4%) Stent implantation68 (41.2%)25 (39.1%)43 (42.6%)Data are presented as n (%). Anatomical territories were not mutually exclusive because some patients underwent treatment at more than one arterial site. Multiple-site treatment was defined as treatment involving two or more side-specific arterial territories, including bilateral treatment at the same anatomical level. P values were calculated using the chi-square test Anatomical and procedural characteristics of the study population Data are presented as n (%). Anatomical territories were not mutually exclusive because some patients underwent treatment at more than one arterial site. Multiple-site treatment was defined as treatment involving two or more side-specific arterial territories, including bilateral treatment at the same anatomical level. P values were calculated using the chi-square test Identification of clinically relevant lipid metabolite co-expression modules Untargeted metabolomics detected approximately 3,000 metabolites, of which 645 lipid and lipid-like metabolites were retained for analysis (Table S1). PCA showed substantial overlap between the HLD and No-HLD groups (Fig. S2A). WGCNA was subsequently used to identify HLD-associated co-expression modules (Fig. S2B). MEbrown was negatively correlated with HLD but was not correlated with statin use (Fig. S2C). In multivariable regression, HLD and smoking status were associated with the MEbrown eigengene, whereas statin use was not (Fig. S2D). The adjusted partial correlation between HLD and MEbrown remained significant (r = − 0.173, P = 0.031; Fig. S2E), and module membership was positively correlated with HLD-related metabolite significance (r = 0.53, P = 2.8 × 10⁻⁶; Fig. 1A). Fig. 1Characterization of the MEbrown lipid module associated with hyperlipidemia (HLD) but independent of statin use. A Scatter plot of module membership versus metabolite significance for HLD showing a significant positive correlation (Cor = 0.53, P = 2.8 × 10⁻⁶). B Bubble chart of enriched KEGG pathways of MEbrown metabolites. C Donut chart of metabolite class distribution in the MEbrown module. The inner ring represents class proportions, and the outer ring represents subclass numbers Characterization of the MEbrown lipid module associated with hyperlipidemia (HLD) but independent of statin use. A Scatter plot of module membership versus metabolite significance for HLD showing a significant positive correlation (Cor = 0.53, P = 2.8 × 10⁻⁶). B Bubble chart of enriched KEGG pathways of MEbrown metabolites. C Donut chart of metabolite class distribution in the MEbrown module. The inner ring represents class proportions, and the outer ring represents subclass numbers In sensitivity analyses stratified by LDL-C level, HLD remained associated with the MEbrown eigengene among patients with LDL-C < 1.8 mmol/L (β = −0.079, 95% CI − 0.137 to − 0.022; P = 0.008), but not among those with LDL-C ≥ 1.8 mmol/L (β = −0.009, 95% CI − 0.047 to 0.029; P = 0.634; P for interaction = 0.023; Table S5). When LDL-C was included as a continuous covariate, the association remained directionally consistent but was attenuated (β = −0.030, 95% CI − 0.060 to 0.001; P = 0.058). MEbrown was enriched in sphingolipid metabolism, alpha-linolenic acid metabolism, biosynthesis of unsaturated fatty acids, primary bile acid biosynthesis, and steroid hormone biosynthesis (Fig. 1B). Its 70 metabolites were predominantly fatty acyls (n = 27, 39.1%) and prenol lipids (n = 20, 29.0%), with additional steroids and steroid derivatives, glycerophospholipids, sphingolipids, and glycerolipids (Fig. 1C). Identification and characterization of lipid metabolic subtypes NMF applied to the 70 metabolites in MEbrown identified rank = 2 as the optimal clustering solution based on the cophenetic correlation coefficient, silhouette score, and reconstruction error (Fig. S3A). Patients were classified into C1 (n = 64) and C2 (n = 101) (Fig. S3B). PCA showed clear metabolic separation between C1 and C2 (Fig. 2A). The leading discriminatory metabolites were mainly fatty acyls, prenol lipids, and steroids (Fig. 2B), and their expression patterns across subtypes are shown in Fig. 2C. Compared with C2, C1 had more smokers (54.7% vs. 32.7%, P = 0.008) and a lower prevalence of HLD (31.2% vs. 51.5%, P = 0.017); other baseline variables did not differ significantly (Table S2). Fig. 2Metabolic and prognostic characteristics of PAD lipid subtypes. A PCA plots of lipidomic profiles in the C1 and C2 subtypes, with different colors indicating different subtypes. B PCA loadings of lipid metabolites, with different colors indicating different metabolite classes. C Heatmap of lipid metabolites grouped by metabolite classes in the C1 and C2 subtypes. Metabolite levels were z-score normalized. D Kaplan–Meier curves for MACE-free survival in the C1 and C2 subtypes Metabolic and prognostic characteristics of PAD lipid subtypes. A PCA plots of lipidomic profiles in the C1 and C2 subtypes, with different colors indicating different subtypes. B PCA loadings of lipid metabolites, with different colors indicating different metabolite classes. C Heatmap of lipid metabolites grouped by metabolite classes in the C1 and C2 subtypes. Metabolite levels were z-score normalized. D Kaplan–Meier curves for MACE-free survival in the C1 and C2 subtypes MACE-free survival was lower in C1 than in C2 (log-rank P = 0.001; Fig. 2D). After adjustment for baseline clinical covariates, C2 remained associated with a lower MACE risk than C1 (P = 0.005; Fig. S3C). Global metabolic differences and functional enrichment of PAD subtypes OPLS-DA showed clear separation between C1 and C2 (Fig. 3A). The top differential metabolites were mainly fatty acyls, prenol lipids, and steroids (Fig. 3B, C). C1 showed higher abundances of fatty acyls, prenol lipids, steroids and steroid derivatives, and glycerophospholipids, whereas glycerolipids and sphingolipids were relatively more abundant in C2 (Fig. 3D). Fig. 3Metabolic differences and functional enrichment of PAD subtypes C1 and C2. A OPLS-DA score plot of individuals in the C1 and C2 subtypes. B Bubble chart of differential metabolites between the C1 and C2 subtypes according to VIP values in the OPLS-DA model. C Bar plot of VIP values for differential metabolites, with different colors indicating different metabolite classes. D Bar plot of metabolite levels across different metabolite classes in the C1 and C2 subtypes. E KEGG pathway enrichment analysis of differential metabolites Metabolic differences and functional enrichment of PAD subtypes C1 and C2. A OPLS-DA score plot of individuals in the C1 and C2 subtypes. B Bubble chart of differential metabolites between the C1 and C2 subtypes according to VIP values in the OPLS-DA model. C Bar plot of VIP values for differential metabolites, with different colors indicating different metabolite classes. D Bar plot of metabolite levels across different metabolite classes in the C1 and C2 subtypes. E KEGG pathway enrichment analysis of differential metabolites Differential metabolites were enriched in steroid hormone biosynthesis, sphingolipid metabolism, primary bile acid biosynthesis, butanoate metabolism, and terpenoid backbone biosynthesis (Fig. 3E). Steroid hormone biosynthesis, sphingolipid metabolism, and primary bile acid biosynthesis were also enriched in MEbrown. Proteomic characteristics and multi-omics cross-talk of PAD subtypes After adjustment for statin use, 28 differentially expressed proteins were identified between C1 and C2, including 3 upregulated and 25 downregulated proteins in C2 (Fig. 4A). PCA showed distinct proteomic profiles between subtypes (Fig. 4B), and the heatmap illustrated opposing expression patterns for selected proteins (Fig. 4C). Gene Ontology enrichment identified coagulation- and lipid signaling-related functions, redox-related functions, and postsynaptic cytosol involvement (Fig. 4D). Fig. 4Integrated proteomic and metabolomic analysis of PAD subtypes C1 and C2. A Volcano plot of differentially expressed proteins between the C1 and C2 subtypes. B PCA plots of the proteomic profiles in the C1 and C2 subtypes, with different colors indicating different subtypes. C Heatmap of differentially expressed proteins in the C1 and C2 subtypes. D GO enrichment network of differentially expressed proteins. E Scatter plots of principal component scores from proteomic and metabolomic data. The correlations were evaluated by Spearman correlation tests. F Correlation network of metabolites and proteins. G O2PLS-DA loading plot of proteins and metabolites Integrated proteomic and metabolomic analysis of PAD subtypes C1 and C2. A Volcano plot of differentially expressed proteins between the C1 and C2 subtypes. B PCA plots of the proteomic profiles in the C1 and C2 subtypes, with different colors indicating different subtypes. C Heatmap of differentially expressed proteins in the C1 and C2 subtypes. D GO enrichment network of differentially expressed proteins. E Scatter plots of principal component scores from proteomic and metabolomic data. The correlations were evaluated by Spearman correlation tests. F Correlation network of metabolites and proteins. G O2PLS-DA loading plot of proteins and metabolites Principal components derived from differential metabolites and proteins were significantly correlated (Fig. 4E). Key lipid classes were correlated with proteins including QDPR, CMTM5, SNAP91, FCN3, and B4GALNT2 (Fig. 4F). O2PLS-DA identified proteins and metabolites jointly associated with subtype separation (Fig. 4G). Construction and prognostic value of the multi-omics molecular score LASSO regression was applied to differential metabolites and proteins to construct candidate molecular scores. The 3 candidate models showed cross-validation AUCs of 0.905, 0.970, and 0.960, respectively. As Models 2 and 3 did not differ significantly by the DeLong test (P > 0.05), Model 3, comprising QDPR and 4 putatively annotated metabolites, was selected for parsimony (Fig. 5A). Molecular scores differed between C1 and C2 (P < 0.001; Fig. 5C) and remained independently associated with subtype classification after adjustment for clinical variables (Fig. S4A). Fig. 5Prognostic value of the molecular score in PAD patients. A Cross-validation ROC curves of prediction models for discriminating the C1 and C2 subtypes. B Bar plot of association effects for features included in the final molecular score model. C Boxplot of molecular scores in the C1 and C2 subtypes. D Kaplan–Meier curves for MACE-free survival according to molecular score groups. E Forest plot of stepwise multivariable Cox regression analyses for the association between molecular score and MACE risk. F Forest plot of subgroup Cox regression analyses for the association between molecular score and MACE risk. G Forest plot of multivariable Cox regression analysis for MACE. H Restricted cubic spline curves for the association between molecular score and MACE risk Prognostic value of the molecular score in PAD patients. A Cross-validation ROC curves of prediction models for discriminating the C1 and C2 subtypes. B Bar plot of association effects for features included in the final molecular score model. C Boxplot of molecular scores in the C1 and C2 subtypes. D Kaplan–Meier curves for MACE-free survival according to molecular score groups. E Forest plot of stepwise multivariable Cox regression analyses for the association between molecular score and MACE risk. F Forest plot of subgroup Cox regression analyses for the association between molecular score and MACE risk. G Forest plot of multivariable Cox regression analysis for MACE. H Restricted cubic spline curves for the association between molecular score and MACE risk During follow-up, 23 patients experienced MACE, including 5 cardiovascular deaths (4 due to myocardial infarction and 1 due to hemorrhagic stroke), 7 nonfatal myocardial infarctions, and 11 nonfatal strokes. Higher molecular scores were associated with worse MACE-free survival (log-rank P = 0.0043; Fig. 5D). The apparent 15-month AUC was 0.872 (bootstrap 95% CI, 0.672–0.987; Fig. S4B), and the 5-fold cross-validated AUC was 0.871 (bootstrap 95% CI, 0.573–1.000; Fig. S4C). The wide cross-validated confidence interval indicated limited precision. After adjustment for age, Rutherford classification, smoking, HLD, and statin use, the molecular score remained associated with MACE (HR = 1.25, P = 0.032; Fig. 5E, G). The proportional hazards assumption was not violated for the molecular score, age, or statin use, whereas some adjustment covariates showed potential time-dependent effects (Fig. S5). The molecular score showed nonsignificant associations with MACE in the statin-treated and non-HLD subgroups, with no significant interactions for statin use or HLD (Fig. 5F; Table S3A). RCS analysis showed a significant overall association with MACE risk (P-overall = 0.013; P-nonlinear = 0.058; Fig. 5H). Discussion In this prospective cohort of patients with peripheral artery disease (PAD) undergoing lower-extremity endovascular revascularization, integrated proteomic and metabolomic profiling identified residual lipid-related molecular subtypes beyond conventional lipid parameters and clinically defined hyperlipidemia (HLD). These subtypes showed distinct molecular features involving lipid metabolism, inflammatory responses, coagulation activation, vascular injury, and tissue remodeling, which are processes previously implicated in atherosclerotic progression and cardiovascular events. A multi-omics score derived from proteomic and metabolomic features was also associated with subsequent major adverse cardiovascular events (MACE). These findings suggest that, in the context of contemporary lipid-lowering therapy, residual lipid-related molecular signatures may provide complementary information for cardiovascular risk characterization in patients with PAD after revascularization. Conventional lipid parameters and clinically defined HLD are commonly used to assess atherosclerosis-related risk. Derived lipid indices, such as the atherogenic index of plasma and the LDL-C/HDL-C ratio, may provide additional information on PAD complexity and outcomes after revascularization [31, 32]. However, in patients with PAD, particularly those at high risk after revascularization, MACE risk may not be fully explained by these conventional indicators. In contemporary practice, widespread use of statins and other lipid-lowering therapies may modify circulating lipid levels and further attenuate their discriminatory value for individual cardiovascular risk. Reliance solely on conventional lipid parameters or HLD classification may therefore fail to identify patients who remain at high risk. Our findings support this concept at the molecular level, suggesting that residual lipid-related risk states in PAD can be captured by proteomic and metabolomic profiling beyond routine lipid testing. Our results are consistent with previous studies on residual cardiovascular risk. Substantial evidence indicates that a considerable proportion of patients remain at non-negligible residual cardiovascular risk even when LDL-C is adequately controlled [33]. In the FOURIER trial, for example, primary endpoint events occurred in approximately 10% of patients whose median LDL-C concentration was reduced to 30 mg/dL [34, 35], suggesting that risk drivers beyond LDL-C persist. Nevertheless, systematic molecular characterization of residual lipid features outside classical lipid parameters remains limited in patients with PAD [36, 37]. Consistent with these observations, principal component analysis in the present study showed substantial overlap between the HLD and non-HLD lipid metabolic profiles. This finding further supports the view that conventional HLD classification and standard lipid parameters do not fully reflect lipid metabolic dysregulation in patients with PAD against the background of widespread statin use. The stronger HLD–MEbrown association observed among patients with LDL-C < 1.8 mmol/L further suggests that this module may capture lipid-related molecular variation not reflected by the contemporaneous LDL-C concentration, although this subgroup finding should be interpreted as exploratory. At the lipid pathway level, the HLD-associated lipid module identified in this study, which was independent of statin use, was mainly enriched in sphingolipid metabolism, bile acid biosynthesis and steroid hormone biosynthesis. Previous studies have associated dysregulation of sphingolipid metabolism with increased cardiovascular risk [38–40], bile acid metabolites with atherosclerosis and coronary artery disease [41], and alterations in steroid hormone biosynthesis with atherosclerosis progression [37, 42, 43]. These findings suggest that the residual lipid-related features identified in our study are not isolated statistical signals, but are consistent with established lipid metabolic pathways implicated in atherosclerosis. At the molecular subtype level, the C1 subtype showed a higher risk of MACE and was characterized by increased fatty acyls, steroids and steroid derivatives, and glycerophospholipids, together with decreased sphingolipids. Previous studies indicate that accumulation of free saturated fatty acids may promote atherosclerosis through endothelial lipotoxicity and pro-inflammatory responses, whereas abnormalities in steroids and their derivatives are also related to atherosclerosis development and progression [37, 42, 43]. In addition, the higher proportion of smokers in C1 and the increase in glycerophospholipids are compatible with oxidative lipid remodeling, although this interpretation requires mechanistic validation. Reduced sphingolipids may further indicate attenuation of protective lipid signaling [44–48]. Together, these results suggest that C1 may represent an exploratory high-risk residual lipid phenotype characterized by pro-inflammatory lipid remodeling and lower sphingolipid abundance. The proteomic findings were directionally consistent with these lipid alterations. Differentially expressed proteins between subtypes were mainly enriched in functions related to oxidoreduction, coagulation and lipid signaling. Enrichment of azurophil granule lumen and NADPH binding suggests that neutrophil-associated oxidative stress may contribute to subtype differences. Previous studies have shown that the myeloperoxidase and NADPH oxidase systems are important sources of vascular reactive oxygen species and participate in atherosclerosis development. This protein-level pattern is consistent with the increased glycerophospholipid profile in C1, suggesting that ROS-mediated phospholipid oxidation may link lipid metabolic abnormalities to vascular inflammatory injury [49, 50]. In addition, enrichment of serine-type endopeptidase inhibitor activity and phosphatidylinositol phosphate binding suggests that coagulation-fibrinolysis imbalance and PI3K-related inflammatory signaling may also contribute to this high-risk phenotype [51–54]. At the representative molecular level, the multi-omics score constructed in this study included one protein, QDPR, and four putatively annotated metabolites: deoxycholylphenylalanine, succinylcarnitine, glutarylcarnitine and acetolein. These molecular features collectively point to oxidative stress, bile acid metabolism, mitochondrial energy metabolism and disruption of lipid homeostasis, processes closely related to atherosclerosis and cardiovascular events. QDPR is involved in tetrahydrobiopterin recycling, endothelial nitric oxide synthase coupling, nitric oxide bioavailability and oxidative stress regulation [55–57]. Changes in circulating QDPR may therefore reflect vascular oxidative injury or inflammatory vascular remodeling [58]. The feature putatively annotated as deoxycholylphenylalanine may reflect host-microbiota-mediated bile acid metabolic alterations [59]. The features putatively annotated as succinylcarnitine and glutarylcarnitine, as acylcarnitine-related metabolites, suggest abnormalities in mitochondrial substrate utilization and energy metabolism [59–62]. The feature putatively annotated as acetolein may represent triglyceride-related disruption of lipid homeostasis. These interpretations should be considered hypothesis-generating until confirmed by targeted assays using authentic standards. This study has several limitations. First, this was a single-center prospective observational cohort with a relatively limited sample size and no independent external validation cohort. The molecular score was developed and evaluated within the same cohort. Although 5-fold cross-validation yielded similar discrimination, internal validation within the same cohort cannot exclude model optimism, and the wide bootstrap confidence interval indicates limited precision. Second, because of the observational design, this study cannot establish causal relationships between lipid-related molecular signatures and MACE. In addition, detailed information on concomitant antithrombotic regimens, other guideline-directed medical therapies, limb events, and non-cardiovascular deaths during follow-up was not available to permit additional adjustment or formal competing-risk analyses; therefore, residual confounding from these factors cannot be excluded. Third, metabolomic and proteomic profiling was based on high-throughput platforms, and the identified features require further validation using targeted assays. Importantly, several metabolite features were annotated from accurate mass and MS/MS spectral matching without authentic chemical standards; these annotations should therefore be regarded as putative rather than definitive structural identifications. In the multivariable Cox analysis, some adjustment covariates showed potential departures from the proportional hazards assumption. Given the limited sample size and number of events, these issues restrict the precision of the analysis, and the corresponding findings should therefore be interpreted as exploratory. Future studies should validate the stability of the score in larger, multicenter external cohorts and simplify it into a measurable and scalable biomarker panel before considering any clinical application. In conclusion, integrated proteomic and metabolomic profiling identified residual lipid-related molecular signatures beyond conventional lipid parameters in patients with PAD undergoing lower-extremity revascularization. These signatures were associated with subsequent MACE and provide a basis for future biomarker validation aimed at improving postoperative cardiovascular risk characterization in PAD. Further validation in larger, multicenter external cohorts is warranted to determine their potential clinical utility. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2. Supplementary Material 3. Supplementary Material 3. Supplementary Material 4. Supplementary Material 4. Supplementary Material 5. Supplementary Material 5. Supplementary Material 6. Supplementary Material 6.