ACADM-mediated fatty acid β-oxidation pathway in atherosclerosis and abdominal aortic aneurysm

Animal studies

All animal experiments were conducted in accordance with the National Institutes of Health Guidelines on the Care and Use of Laboratory Animals (NIH Publication, 8th Edition, 2011) and were approved by the Animal Ethics Committee of Renji Hospital (RJ2022-1211). Both male and female mice aged 8–12 weeks were used in this study. ApoE^−/− mice on a C57BL/6J background were kept in a specific pathogen-free facility under controlled setting (temperature: 22 ± 2°C; relative humidity: 55 ± 15%; light/dark cycle: 12/12 h) and provided ad libitum access to food and water. Two AS animal models, induced by wire injury- or high-fat diet (HFD, 40% kcal fat and 1.25% cholesterol), and two AAA animal models, induced by angiotensin II (Ang II) or calcium phosphate (CaPO₄), were established. For all experiments, animals were randomly assigned to experimental groups using simple randomization, ensuring age and sex matching.

Ang II- or CaPO₄-induced AAA models were established as described in our previous studies.^15,16 To generate an Ang II-induced AAA model, ApoE^−/− mice were infused with Ang II (1000 ng/kg/min; A9525; Sigma, USA) via subcutaneous osmotic minipumps while being maintained on an HFD for 4 weeks. To generate a CaPO₄-induced AAA model, the infrarenal region of the abdominal aorta of the ApoE^−/− mice was treated with 0.5 mol/L calcium chloride for 10 min, followed by phosphate-buffered saline for another 5 min to form CaPO₄. An aneurysm was defined as a diameter increase of at least 50%. To compare with AAA, an abdominal aortic AS mouse model was generated using a wire injury plus 4-week HFD approach to induce atherosclerotic plaque formation in the abdominal aorta.¹⁷ Briefly, a 0.36-mm guide wire was advanced into the abdominal aorta under ultrasound guidance and then retracted and advanced in a sawing motion 20–30 times to injure the endothelium of the abdominal aorta, followed by an HFD for 4 weeks. To confirm findings from wire injury-induced AS model, a diet-induced AS mouse model was established by maintaining ApoE^−/− mice on an HFD for 12 weeks.¹⁸

Human studies

All human samples used in this study were processed under Institutional Review Board approved protocols at Shanghai Jiao Tong University. All study participants were recruited after providing informed consent with approval by the Ethics Committee of Renji Hospital, School of Medicine, Shanghai Jiao Tong University (KY2021-129-B, KY2019-101), and the study was conducted according to the criteria set by the Declaration of Helsinki (2013).

Statistical analysis

Orthogonal projections to latent structures-discriminant analysis (OPLS-DA) were conducted using SIMCA software (version14.0, Sartorius Stedim Data Analytics AB, Sweden). Model validity was evaluated using seven-fold cross-validation based on R²Y and Q² values, and permutation testing (200 iterations) was performed to assess robustness and exclude overfitting. Significantly regulated metabolites and lipids were identified by variable importance in projection (VIP) scores > 1 and P-value < .05 from two-group comparisons using Student’s t-test, Welch’s t-test, or Mann–Whitney U test according to data distribution and variance assumptions, as described below. Descriptive statistics were employed to summarize clinical and demographic characteristics. Continuous variables were expressed as mean ± standard deviation (SD) for approximately symmetric distributions and as median (interquartile range, IQR) for skewed distributions. Categorical variables were presented as frequency (percentage). Effect estimates are reported with corresponding 95% confidence intervals (CIs). The normality of data distribution was assessed using Q-Q plots and the Shapiro–Wilk test. For two-group comparisons, homogeneity of variances was evaluated using Levene’s test. Normally distributed data were analysed using Student’s t-test (for equal variances) or Welch’s t-test (for unequal variances), whereas non-normally distributed data were analysed using Mann–Whitney U test. For comparisons among three or more groups, homogeneity of variances was assessed using the Brown–Forsythe test. Normally distributed data were analysed using one-way ANOVA with Tukey’s post hoc test (for equal variances) or Welch’s ANOVA with Tamhane’s T2 post hoc test (for unequal variances). Two-way ANOVA with Tukey’s post hoc analysis was used when more than two groups and variables were compared. For non-normally distributed data, the Kruskal–Wallis test was applied. Correlations were assessed using both Spearman’s rank correlation (ρ) and Pearson correlation (r). Linear regression lines were fitted using ordinary least squares (OLS) for visualization purposes only. Bivariate comparisons of categorical variables were performed with the χ² test (all expected cell counts were ≥5) or with the Fisher’s exact test (any expected cell count was <5). Associations between treatment (vehicle vs LD·ATTEC) and systolic blood pressure over time in AS and AAA mice were assessed using linear generalized estimating equations. Representative images were chosen to represent the average results achieved within the experiment. All statistical analyses were performed using GraphPad Prism (version 8), Python (3.10.14), or R (version 4.3.2), with a P-value < .05 considered statistically significant.

For UK Biobank data, hazard ratios and 95% CIs were calculated using Cox proportional hazards models with the date of recruitment as the origin time, adjusted for gender, age at enrolment, body mass index (BMI), smoking status, medication use, and assessment centre. Covariate selection was guided by a directed acyclic graph (DAG) constructed based on established cardiovascular risk determinants summarized in the PREVENT study.²¹ For the proteomics analysis, we also accounted for the first 10 genetic principal components and genotype arrays. The proportional hazards assumption was evaluated using Schoenfeld’s tests, and linearity of quantitative predictors was assessed using restricted cubic splines. To ensure consistency in baseline characteristics across CAD, AAA, and control groups, propensity scores were estimated using a multivariable logistic regression model and used for matching to generate matched cohorts. Age, sex, BMI, current smoking status, and medication use were included as variables in the propensity score model to estimate scores for matching. Matching was then performed using 1:3 nearest-neighbour matching with a caliper width equal to 0.2 of the SD of the logit of the propensity score.^22,23 For the matched cohorts, associations were estimated using Cox proportional hazards models incorporating a shared frailty term at the matched-set level. Since AAA group had the least number of participants, we randomly selected three matched CAD participants and three matched controls for each AAA participant. Absolute standardized mean differences were calculated to compare differences between matched groups, with >10% suggesting imbalance. Analysis was further adjusted for competing risks using the Fine and Gray proportional sub-distribution hazards regression model, with incident CAD and all-cause death specified as competing events in the AAA analysis, and incident AAA and all-cause death in the CAD analysis. The proportions of competing events were 11.47% and 7.23%, respectively. Differences in association magnitude for incident CAD and AAA were estimated using Z_diff = (β_CAD-β_AAA)/$\sqrt{(\mathrm{se}2\mathrm{CAD}+\mathrm{se}2\mathrm{AAA})}$.²⁴ A control group sample-splitting sensitivity analysis was additionally performed to avoid covariance arising from the shared controls. Sample size sufficiency was evaluated using the FDRsamplesize2 method for both lipidomic analysis and Cox proportional hazards analysis.²⁵

Lipid metabolism differs between atherosclerosis and abdominal aortic aneurysm mouse models

The divergent role of lipid metabolic disturbances in the pathogenesis of AS vs AAA remains largely unknown. To elucidate this, we first established two mouse models of AS: one induced by wire injury surgery with a 4-week HFD¹⁷ and the other by continuous HFD for 12 weeks.²⁶ Atherosclerosis severity was assessed by measuring the percentage of aortic lesion area. We analysed correlations between plasma lipid levels (including total cholesterol, triglycerides, and LDL cholesterol) and AS severity and found a strong positive correlation between elevated these plasma lipids and AS progression in both models (see Supplementary data online, Figure S1A and B). We also developed two AAA models: the Ang II-infused model^15,16 and the CaPO₄-induced model.^15,16 Abdominal aortic aneurysm severity was measured by the maximal diameter of the suprarenal abdominal aorta. In contrast to AS, we observed weak correlations between lipid levels and AAA severity in both models (see Supplementary data online, Figure S1C and D). These results suggest that lipid levels have a stronger correlation with AS than with AAA in mouse models. Given the differential involvement of lipid metabolism in AS and AAA, we performed lipidomic analyses on mouse aortic tissues to define disease-specific lipid metabolic features. Lipidomic is a high-throughput analytical technique widely used to study lipid metabolism by quantifying changes in specific lipid classes and their concentrations.²⁷ For comparison, we selected the wire injury-induced abdominal aortic AS model and the Ang II-induced AAA model (Figure 1A and B), as both target the same lesion site in the abdominal aorta and follow the same dietary intervention. The lipidomic analysis identified 328 significantly altered lipid metabolites in AS and 115 in AAA (Figure 1C). Analysis of the number of altered lipid metabolites revealed that the top lipid class in AS were glycerolipids (with 186 altered lipid metabolites), accounted for 56.7% of the total altered lipid metabolites. In contrast, in AAA, phosphoglycerides were the top lipid class (with 63 altered lipid metabolites), constituting 54.7% of the total altered lipid metabolites (Figure 1D). Of the 23 lipid subclasses detected, 15 displayed significant changes in AS and 16 in AAA. Triglycerides were the most significantly elevated lipid subclass in AS, followed by cholesterol esters, while phosphatidylethanolamines were the most prominently increased lipid subclass in AAA, followed by phosphatidylcholines (Figure 1E–H). Direct comparison between AS and AAA identified 40 significantly differentially expressed lipid metabolites (Figure 1I; Supplementary data online, Figure S2), belonging to three lipid classes (glycerolipids, fatty acyls, and sterol lipids) (Figure 1J). Further lipid subclass analysis revealed that triglycerides, diglycerides, free fatty acids, and cholesterol esters exhibited significant increase in AS (vs AAA) (Figure 1K and L). Together, our lipidomic data demonstrated a pronounced increase of neutral lipids (e.g. triglycerides and cholesterol esters) in AS, while phosphoglycerides (e.g. phosphatidylethanolamines and phosphatidylcholines) were more prominently elevated in AAA.

Lipid metabolism differs between patients with atherosclerosis and abdominal aortic aneurysm

To verify our findings from animal experiments, we conducted lipidomic analyses on plasma samples from 363 individuals from Renji hospital (including 121 patients with documented CAD, 121 patients with documented AAA, and 121 age- and sex-matched healthy controls). The analysis identified 705 lipid metabolites, of which 313 were significantly elevated in CAD patients and 174 in AAA patients compared to controls (Figure 2A). In CAD patients, glycerolipids accounted for the largest proportion of the increased lipid metabolites (60.4%). In contrast, phosphoglycerides were the most abundant lipid class among the differential metabolites in AAA patients, accounting for 68.4% (Figure 2B). Consistent with the results from animal studies, further lipid subclass analysis revealed that triglycerides were the most significantly elevated lipid subclass in CAD, followed by diglycerides, while phosphatidylcholines were the most significantly increased lipid subclass in AAA, followed by phosphatidylethanolamines (Figure 2C–F). Direct comparison between CAD and AAA patients identified 135 differentially expressed lipid metabolites (100 upregulated and 35 downregulated in CAD relative to AAA) (Figure 2G), belonging to three lipid classes (glycerolipids, phosphoglycerides, and sterol lipids). Subclass analysis showed that triglycerides, diglycerides, and cholesterol ester exhibited significant increase in CAD. Conversely, phosphatidylcholines displayed the greatest increase in AAA, followed by lyso-phosphatidylethanolamines and monogalactosyldiacylglycerols (Figure 2H–J). The top five significantly upregulated and downregulated lipid metabolites in CAD compared to AAA are shown in Figure 2K. Importantly, when comparing lipidomic data from human cohorts with those from animal models, we observed a similar pattern of altered lipid metabolite subclasses in human vs mouse samples (Figure 2L). Overall, these human findings align with lipidomic analyses from animal models, underscoring the distinct lipidomic features characterizing AS and AAA.

To further explore longitudinal associations between lipid traits and the risks of AS and AAA, we leveraged data from the UK Biobank study. Our analysis included 479 256 adults without CAD or AAA at baseline, with a median age of 57 years, 55.7% of whom were female. During a median follow-up of 14.8 years (IQR: 14.1–15.5 years), 22 675 incident CAD cases and 2159 incident AAA cases were identified. Consistently, the metabolic and proteomic data from the UK Biobank highlight the distinct roles of lipid metabolic disturbances in the further risk of AS and AAA (details in Supplementary data online, Results). Using both Cox proportional hazards model and propensity score matching adjusted analysis, the results showed strong positive associations of triglycerides, cholesterol, and fatty acid with the risk of CAD, while phosphoglycerides were negatively associated with the risk of AAA (details in Supplementary data online, Results, Figures S3 and S4, and Tables S3–S5).

Fatty acid metabolism is identified as the top altered lipid-related pathway contributing to the lipid metabolic differences between atherosclerosis and abdominal aortic aneurysm

To explore the mechanisms underlying the differences in lipid metabolism between AS and AAA, we performed high-throughput RNA sequencing (RNA-seq) on mouse aortas from wire injury-induced AS and the Ang II-induced AAA models. Principal component analysis of the transcriptomes showed a clear separation between the AS and AAA groups (Figure 3A). Differential gene expression analyses (fold change ≥ 2, P < .05) showed significant alterations in both AS and AAA compared to controls (Figure 3B). Gene Ontology (GO) enrichment analysis of differentially expressed genes (DEGs) indicated that both immune-related and inflammatory-related responses were upregulated in both AS and AAA (Figure 3C and D); however, pathways enriched among downregulated genes differed substantially between these two diseases: FA metabolic processes among lipid metabolism were predominantly downregulated in AS, while pathways related to cell adhesion and multicellular organism development were more significantly downregulated in AAA (Figure 3C and D). To gain deeper insights, we directly compared the transcriptomes of AS vs AAA (Figure 3E). Gene Ontology enrichment analysis revealed that lipid metabolic pathways relative to FA metabolism were the most significantly downregulated pathway in AS (Figure 3F), while pathways related to cell adhesion, extracellular matrix organization, and cell migration were most significantly suppressed in AAA (Figure 3G). Furthermore, metabolomic analyses of plasma samples demonstrated that FA biosynthesis was the top enriched pathway in AS, while suppression of FA metabolism was not seen in AAA (see Supplementary data online, Figure S5A–D). In contrast, phosphatidylethanolamines biosynthesis was the top enriched pathway in AAA (see Supplementary data online, Figure S5E). These findings suggest that FA metabolism is suppressed in AS, but not in AAA.

To further validate the aforementioned findings, we conducted transcriptomic and metabolomic analyses using another HFD-induced AS model and another CaPO₄-induced AAA model (see Supplementary data online, Figure S6A and B). Gene Ontology and Reactome enrichment analyses of DEGs confirmed that FA metabolism was significantly downregulated in HFD-induced AS (see Supplementary data online, Figure S6C–E) but not in CaPO₄-induced AAA models (see Supplementary data online, Figure S6F–H). Metabolomic pathway enrichment analyses of plasma samples showed that FA biosynthesis was the top enriched pathway in HFD-induced AS, while ammonia recycling was the top enriched pathway in CaPO₄-induced AAA (see Supplementary data online, Figure S6I and J). Overall, these multi-omics analyses consistently demonstrate the crucial role of downregulated FA metabolism in atherogenesis in a model-independent manner.

We further screened for genes involved in FA metabolism that were altered during AS and AAA progression (Figure 3H). Reactome pathway enrichment analysis highlighted FA β-oxidation (FAO) as the top downregulated pathway in AS mice (Figure 3I). The heatmap further illustrated differentially regulated genes associated with FA metabolism in AS and AAA (Figure 3J). Furthermore, protein–protein interaction network analysis identified Acadm, which encodes medium-chain acyl-CoA dehydrogenase (a key mitochondrial enzyme responsible for FAO),²⁸ as a critical node in FA metabolism (Figure 3K). RT-qPCR and Western blot confirmed that both Acadm mRNA and ACADM protein levels were downregulated in AS mice but not in AAA (Figure 3L). To confirm the above findings, we analysed the human AS microarray dataset (GSE57691) from the Gene Expression Omnibus database. Differentially expressed genes visualized by Venn diagram revealed that four FA metabolism-related DEGs were co-expressed in both human and mouse AS: Acadm, Cpt2, Acadl, and Ndufab1 (see Supplementary data online, Figure S7A). Among these four candidate genes, Acadm was the most significantly downregulated gene in the abdominal aortas of AS mice (see Supplementary data online, Figure S7B). The downregulation of ACADM mRNA and ACADM protein was further confirmed in human AS tissues by RT-qPCR and Western blot, whereas no significant change was observed in AAA tissues (see Supplementary data online, Figure S7C and D). Immunofluorescence confirmed that ACADM is predominantly expressed in aortic macrophages and that its expression is markedly reduced in AS but not in AAA (see Supplementary data online, Figure S7E and F). To further validate the downregulation of Acadm in macrophages, bone marrow–derived macrophages (BMDMs) were isolated from mice and treated with oxidized LDL (ox-LDL). Both Acadm mRNA and ACADM protein levels were significantly reduced in ox-LDL–treated BMDMs (see Supplementary data online, Figure S7G and H). Acadm deficiency results in the accumulation of carnitine derivatives (metabolic markers of impaired FAO), particularly octanoylcarnitine.^29,30 We therefore measured acylcarnitine concentrations in both serum and vascular tissues from both mice and humans. Compared with controls, the levels of medium-chain acylcarnitines (i.e. octanoylcarnitine, hexanoylcarnitine, and decanoylcarnitine) were significantly elevated in AS, whereas no significant changes were observed in AAA (see Supplementary data online, Figure S8A–F). These findings provide compelling evidence that FA metabolism, particularly the FAO pathway, is the most significantly altered lipid-related pathway, contributing to the lipid metabolic differences between these two diseases.

Lipid droplet clearance by lipid droplet autophagy-tethering compound treatment alleviates atherosclerosis but not abdominal aortic aneurysm in mice

FAO is the primary pathway for the degradation of FAs.³¹ Its impairment leads to the accumulation of excess FAs in the form of neutral lipids, predominantly triglycerides and cholesteryl ester, within lipid droplets (LDs).³¹ Lipid droplets regulate cellular FA levels through dynamic cycles of formation and breakdown.³² Therefore, understanding the role of LDs is important to gain deeper insights into FA metabolism in both AS and AAA. Based on the concept of ATTEC, we recently developed a novel LD-degrading strategy, LD·ATTEC, which tethers LDs to autophagosomes for autophagic degradation in several cells including adipocytes and embryonic fibroblasts.^33,34 However, the effects of LD-degrading strategies in vascular diseases have never been investigated. Given the excessive accumulation of LDs in macrophages can lead to foam cell formation, we further investigated whether LD·ATTEC could reduce LDs in macrophages. We found that in macrophages, LD·ATTEC reduced LD accumulation (Figure 4A–D) by enhancing their degradation via autophagy (details in Supplementary data online, Results and Figure S9).

To clarify the different pathological roles of LDs in AS and AAA, we administered LD·ATTEC to mouse models of these two diseases. We first examined LD accumulation during the progression of AS and AAA in mice and observed a more pronounced accumulation of LDs in the lesion area of AS compared to AAA (see Supplementary data online, Figure S10A). Next, we evaluated the effects of LD·ATTEC on blood pressure and plasma lipid levels. Lipid droplet autophagy-tethering compound treatment significantly lowered plasma lipid levels of triglycerides, total cholesterol, and LDL cholesterol in both AS and AAA, without affecting blood pressure (see Supplementary data online, Figure S10B–E). We then evaluated the therapeutic efficacy of LD·ATTEC in AS mice. Lipid droplet autophagy-tethering compound treatment significantly reduced atherosclerotic lesions, as evidenced by Oil Red O staining of abdominal aortas and in vivo micro-ultrasound imaging (MUI) (Figure 4E and F). Histological analysis revealed that LD·ATTEC reduced plaque area and intraplaque lipid contents (Figure 4G–I).

Given that vulnerable plaques are associated with a high risk of acute cardiovascular events in humans, we further generated a rupture-prone vulnerable plaque model, as we previously described.^35–37 Treatment with LD·ATTEC markedly reduced carotid plaque burden (Figure 4J and K). Moreover, LD·ATTEC treatment significantly decreased the incidence of plaque rupture, and reduced atherosclerotic area and lesional lipid deposition (Figure 4L–O). These results suggest that LD·ATTEC reduced lipid accumulation and enhanced the stability of vulnerable atherosclerotic plaques.

To determine whether LD·ATTEC had similar therapeutic effects in AAA, we administered LD·ATTEC to AAA mice. However, LD·ATTEC treatment failed to ameliorate AAA formation, with no significant differences observed in AAA incidence, maximal suprarenal aortic diameter, or total aortic weight/body weight ratio (Figure 4P–S). Similar results were observed by visualization of the MUI (see Supplementary data online, Figure S11A and B). Histological analysis showed that LD·ATTEC treatment significantly decreased lipid deposition without affecting collagen contents (see Supplementary data online, Figure S11C and D). Collectively, although LD·ATTEC treatment successfully eliminated LDs in both AS and AAA mouse models, it only alleviated the pathological progression of AS (but not AAA), suggesting distinct roles of LDs in the pathogenesis of AS vs AAA.

Lipid droplet clearance by lipid droplet autophagy-tethering compound treatment restores FA metabolism in atherosclerosis but not in abdominal aortic aneurysm mice

To explore the mechanisms underlying the differential pathogenic role of LDs in AS and AAA, we performed transcriptomic and metabolomic analyses to investigate the impact of LD·ATTEC on these two vascular diseases (Figure 5A). Differentially expressed genes are presented in Figure 5B and C. Key genes regulated by LD·ATTEC were identified by determining the intersection of the DEGs (Figure 5D). Gene Ontology analysis revealed that LD·ATTEC treatment upregulated pathways primarily associated with lipid metabolism related to FA metabolism in AS-LD·ATTEC group, while downregulating pathways related to inflammatory and immune responses (Figure 5E). In contrast, the AAA-LD·ATTEC group did not exhibit any significant changes in lipid metabolism-related pathways (Figure 5F). Reactome analysis of FA metabolism-related genes identified FAO as the most significantly enriched pathway in AS, whereas no such enrichment was observed in AAA after LD·ATTEC treatment (Figure 5G). Specifically, LD·ATTEC significantly increased the expression of FA metabolism-related genes in AS but did not significantly alter these genes in AAA (Figure 5H). Further RT-qPCR and Western blot analyses validated the significant upregulation of Acadm mRNA and ACADM protein following LD·ATTEC treatment in AS mice but not in AAA mice (Figure 5I and J).

To further investigate the effects of LD·ATTEC on metabolism, we performed metabolomic analysis (see Supplementary data online, Figure S12A). Lipid droplet autophagy-tethering compound administration predominantly impacted lipid metabolism in both AS and AAA (see Supplementary data online, Figure S12B), with different pattern. The differences in metabolite expression are shown in the volcano plot (see Supplementary data online, Figure S12C), with the top 20 differentially expressed metabolites displayed in the heatmap (see Supplementary data online, Figure S12D). Notably, acylcarnitines (the metabolite produced during FAO and accumulating upon the disrupted FAO pathway³⁰) were significantly reduced in plasma following LD·ATTEC treatment (see Supplementary data online, Figure S12E). In summary, transcriptomic and metabolomic analyses revealed that LD·ATTEC restored FA metabolism and FAO pathway in AS, but did not show significant changes in AAA.

To achieve a better understanding of lipid metabolic alterations, we further performed lipidomic analysis on abdominal aortic samples from LD·ATTEC-treated AS mice. Principal component analysis revealed distinct lipidomic signatures between the LD·ATTEC-treated and vehicle-treated AS mice (see Supplementary data online, Figure S13A). Lipid droplet autophagy-tethering compound treatment reduced the total lipid contents in the aortas by 67.6 ± 14.3% (calculated based on the sum of detected lipids) in AS mice. A total of 215 lipid metabolites were significantly downregulated in the LD·ATTEC-treated AS mice compared to the vehicle-treated controls (see Supplementary data online, Figure S13B). Glycerolipids were the most affected lipid category (see Supplementary data online, Figure S13C). Of the 23 lipid subclasses detected, 11 displayed significant changes with LD·ATTEC treatment; the top 4 changed lipid subclasses were triglycerides, diglycerides, cholesteryl esters, and free fatty acids (see Supplementary data online, Figure S13D and E). Specifically, LD·ATTEC reduced the levels of neutral lipids (e.g. triglycerides, diglycerides, cholesterol esters) and FFA to near-control levels (see Supplementary data online, Figure S13F–I).

Acadm overexpression represses lipid droplet formation and mitigates atherosclerosis but not abdominal aortic aneurysm in mice

Acadm serves as a key enzyme in mitochondrial FAO and plays an essential role in regulating FA metabolism.³⁸ However, the functional role of Acadm in the pathogenesis of vascular diseases including AS and AAA has never been investigated. Our aforementioned findings identified Acadm as a hub DEG among the FA metabolism-related genes between AS and AAA. Therefore, we further explored its potential role in AS and AAA formation. First, we overexpressed Acadm in BMDM via lentivirus infection (Figure 6A). Acadm overexpression significantly reduced LD accumulation in ox-LDL-treated BMDM (Figure 6B). To further investigate the role of Acadm in the development of AS and AAA in vivo, we generated myeloid-specific Acadm overexpressing mice, which were subjected to AS or AAA model induction (see Supplementary data online, Figure S14; Figure 6C). Acadm overexpression significantly reduced atherosclerotic lesions compared to the controls in the AS models (Figure 6D–H), but failed to attenuate AAA progression in the AAA models (Figure 6I–P). The protective role of Acadm against AS found in animal models was further supported by the human whole-exome sequencing data from the UK Biobank, which demonstrated that loss-of-function mutations in ACADM were associated with increased AS burden (details in Supplementary data online, Results, Figure S15 and Table S6). Overall, the key FAO metabolic enzyme Acadm exerts differential effects in AS vs AAA, further supporting the distinct roles of FA metabolism in the pathogenesis of these two diseases.

Acadm deficiency has been reported to induce the accumulation of intracellular lipotoxic intermediates (e.g. FFA) generated during impaired FAO, thereby provoking mitochondrial dysfunction.³⁹ To further elucidate the underlying mechanism, we performed both Acadm loss-of-function and gain-of-function experiments in BMDMs. Acadm deficiency by Acadm-knockdown lentivirus transduction increased intracellular total FFAs, accompanied by decreased acetyl-CoA and free-CoA pools (Figure 7A). Oxygen consumption analysis revealed that Acadm knockdown significantly decreased basal, maximal, and ATP-coupled mitochondrial oxygen consumption rates (OCRs) (Figure 7B and C). Moreover, Acadm knockdown disrupted mitochondrial membrane potential and increased mitochondrial reactive oxygen species (mtROS) (Figure 7D). To determine whether the mitochondrial dysfunction resulted from lipotoxic metabolite accumulation, BMDMs were treated with 2.5 μmol/L etomoxir, a pharmacological inhibitor that limits mitochondrial fatty acid uptake. Etomoxir effectively rescued the mitochondrial defects caused by Acadm depletion, restoring mitochondrial respiratory capacity and membrane potential, and reducing mtROS levels (Figure 7E–J). To further validate the protective effects of Acadm, BMDMs were transduced with Acadm-overexpression or control lentivirus and subsequently exposed to ox-LDL. Acadm overexpression restored lipid metabolic balance, as indicated by reduced intracellular lipotoxic metabolites FFAs, accompanied by increased acetyl-CoA and free-CoA levels (see Supplementary data online, Figure S16A). Acadm overexpression markedly improved mitochondrial respiration, membrane potential, and reduced oxidative stress (see Supplementary data online, Figure S16B–I). Collectively, these findings indicate that Acadm preserves mitochondrial homeostasis primarily by preventing the intracellular accumulation of lipotoxic metabolites.

Construction of machine learning predictive models based on lipid metabolic features

The above multi-omics and validation studies in human and preclinical animal models revealed different lipid metabolic features in AS vs AAA. Thus, we sought to further establish disease prediction models based on their distinct lipid metabolic features. Both AS and AAA could lead to life-threatening conditions. Commonly used biomarkers such as troponin, myoglobin, and creatine kinase are specific to acute myocardial infarction, while most AAA patients remain asymptomatic until lethal rupture.^19,40 Given the distinct roles of lipid metabolism in AS and AAA, predictive models based on lipid features may provide unique advantages.

We first developed predictive models for CAD vs healthy controls (HC) as well as those for AAA vs HC based on our clinical lipidomic data using machine learning algorithms. Participants were randomized in a ratio of 7:3 into training and validation cohorts.⁴¹ We screened 13 significantly differential lipid metabolites in CAD patients and 25 metabolites in AAA patients in their respective training sets using a combination of t-test, least absolute shrinkage and selection operator (LASSO) regression, and partial least squares discriminant analysis (PLS-DA). Then, we developed logistic regression models using selected lipidomic features, adjusted for age, sex, and BMI. The area under receiver operating characteristic curve (AUC) for CAD vs HC model in the training cohort was 0.974 (95% CI 0.954–0.994), while the AUC for the AAA vs HC model in the training cohort was 0.913 (95% CI 0.870–0.956) (Supplementary data online, Table S7). In CAD, CE(20:4) showed the strongest association, followed by TAG(52:5), FA(20:3), and CE(20:3) (see Supplementary data online, Figure S17A). In contrast, for AAA, PC(16:0/20:4) exhibited the highest association, followed by PC(16:0/18:0), PC(18:2/18:3), PC(18:1/16:1), LPC(20:1), and LPE(18:0) (see Supplementary data online, Figure S18A). To enhance model interpretability, nomograms were constructed to visualize the predictive influence of lipid metabolites (see Supplementary data online, Figures S17B and S18B). Calibration curves demonstrated excellent concordance between predicted and observed risks (see Supplementary data online, Figures S17C and S18C). In the validation cohort, the AUC value for CAD vs HC was 0.832 (95% CI 0.741–0.923) and the AUC value for AAA vs HC was 0.789 (95% CI 0.734–0.929). To validate the stability of screened lipid metabolites for disease identification, we further constructed four additional machine learning models, including random forest (RF), support vector machines (SVM), extreme gradient boosting (XGBoost), and the Super Learner ensemble model. All models exhibited AUCs greater than 0.75 in the validation cohort (see Supplementary data online, Figures S17D and S18D). Supplementary data online, Figure S17E summarized the evaluation metrics—including accuracy, sensitivity, specificity, precision, F1 scores, and AUROCs—of the four models in the CAD vs HC validation cohort, while Supplementary data online, Figure S18E displayed those metrics of the four models in AAA vs HC validation cohort. Additionally, we measured the abundance of key lipids with significant coefficients in the logistic regression analysis for the CAD vs HC model (see Supplementary data online, Figure S17F) and the AAA vs HC model (see Supplementary data online, Figure S18F).

Next, we developed predictive models to differentiate between AS and AAA based on lipidomic features using machine learning algorithms. We identified 20 differential lipid metabolites between patients with CAD and AAA for model construction. The AUC of the logistic regression model using selected lipidomic features was 0.968 (95% CI 0.946–0.990) in the training set (n = 169) and 0.841 (95% CI 0.747–0.935) in the validation set (n = 73). The forest plot revealed that patients with CAD had higher levels of CE(20:2) and CE(20:4), whereas patients with AAA showed higher levels of PC(16:0/18:3) and PC(18:0/18:2) (Figure 8A). The nomogram constructed with selected lipid metabolites effectively distinguished CAD from AAA (Figure 8B). Furthermore, calibration curves demonstrated excellent agreement between predicted and observed risks (Figure 8C). Area under receiver operating characteristic curve values for all four models (logistic regression, RF, SVM, and XGBoost) exceeded 0.8 in the validation cohort (Figure 8D), confirming the robustness of the classifiers. Other evaluation metrics, including accuracy, sensitivity, specificity, precision, and F1 scores, are shown in Figure 8E. Finally, we examined the concentrations of these key lipids with significant coefficients in the logistic regression analysis and found that patients with CAD had significantly higher concentrations of cholesterol esters, whereas the phosphatidylcholine concentration was increased in patients with AAA (Figure 8F).

Study limitations

Several limitations should be considered. First, residual and unmeasured confounding cannot be entirely excluded despite adjustment for a broad range of demographic, lifestyle, clinical, and biological covariates, owing to imperfect measurement of confounders as well as the presence of factors that were unknown, unavailable, or not included in the models. Some confounders (e.g. smoking status) were self-reported and may be subject to measurement error or misclassification, which could bias effect estimates. In addition, hazard ratios in this study should be interpreted cautiously, as they are conditional on individuals who remain event-free and may vary with changes in the risk set over time, rather than reflecting a constant causal effect.⁶⁶ Finally, although current findings support differential roles of ACADM in AS and AAA, the animal models used in this study are limited in their ability to fully recapitulate the complex pathophysiology of AS and AAA development in humans.

Disclosure of Interest

B.L. is one of the inventors of the LD·ATTEC-related patents. The authors declare no other competing interests.

Data Availability

All raw data supporting the findings of the study are available upon the requests from the corresponding authors. Detailed description of materials and methods can be found in the Supplementary data. We also provided a web portal for easy data exploration (https://pu-lab.sjtu.edu.cn/shiny/lipidomics/).

Funding

This work was supported by grants from the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0503200 to J.P., 2026ZD0554401 to Y.Z.), Innovative Research Group Project of the National Natural Science Foundation of China (82421001 to J.P.), National Natural Science Foundation of China (82230014 to J.P., 81930007 to J.P., 92468301 to Y.J., 81800048 to L.S., 82570549 to L.S.), National Science Fund for Distinguished Young Scholars (81625002 to J.P.), Shanghai Municipality Science and Technology Commission (19DZ2230300 to J.P., 22DZ2292400 to J.P., 23ZR1439100 to L.S.), Shanghai Municipal Health Commission (2022JC013, 2023ZZ02021, 2022ZZ01008 and GWVI-11.1-26 to J.P., 20254Y0115 to F.S.), Shanghai Magnolia Talent Plan Pujiang Project (24PJD056 to F.S.), Shanghai Municipal Education Commission (SHSMU-ZDCX20210700 to J.P.), China Postdoctoral Science Foundation (2024M762036 to H.Z.), and Medical Science and Technology Project of Zhejiang Province (2025KY307, 2025JK078 to X.H.). This work was also supported by the Shanghai Municipal Science and Technology Major Project.

Ethical Approval

For animal, all animal experiments were conducted in accordance with the National Institutes of Health Guidelines on the Care and Use of Laboratory Animals (NIH Publication, 8th Edition, 2011) and were approved by the Animal Ethics Committee of Renji Hospital (RJ2022-1211). For human, all study participants were recruited after providing informed consent with approval by the Ethics Committee of Renji Hospital, School of Medicine, Shanghai Jiao Tong University (KY2021-129-B, KY2019-101), and the study was conducted according to the criteria set by the Declaration of Helsinki (2013).

Pre-registered Clinical Trial Number

Not applicable.