中文摘要

第一部分 基于心外膜脂肪组织的多组学分析揭示冠心病的潜在分子机制

目的

       内脏脂肪与异位脂肪的增加在冠心病（Coronary artery disease, CAD）的发生中起着重要作用，其中心外膜脂肪组织（Epicardial adipose tissue, EAT）通过释放多种脂肪因子和炎症介质，显著影响冠状动脉粥样硬化的进展。目前关于EAT在冠状动脉粥样硬化病变中的具体病理生理机制尚未完全阐明。本研究对EAT及皮下脂肪组织（Subcutaneous adipose tissue, SAT）进行了非靶向代谢组学、转录组学以及多组学联合分析，旨在探索EAT在CAD发病中的潜在机制。

方法

       本研究纳入阜外医院诊断为CAD并接受CABG治疗的患者32例，术中留取EAT及SAT样本分别进行非靶向代谢组学和转录组学检测，同时收集患者的临床数据。代谢组学数据采用正交偏最小二乘法-判别分析（Orthogonal projections to latent structures-discriminant analysis, OPLS-DA）鉴定差异代谢物，利用MetaboAnalyst 6.0对差异代谢物进行富集分析。转录组数据采用DESeq2和加权基因共表达网络分析（Weighted correlation network analysis, WGCNA）鉴定差异基因并取交集，对交集基因进行基因本体（Gene Ontology, GO）富集分析和京都基因与基因组百科全书（Kyoto Encyclopedia of Genes and Genomes, KEGG）通路分析。随后，进行代谢组和转录组的联合分析。将鉴定的差异代谢物和交集基因进行共富集分析。采用皮尔森相关性分析探究代谢物与基因之间的相关性，并利用Cytoscape软件构建代谢物与基因的相关性网络。根据KEGG数据库的代谢通路绘制关键代谢途径的代谢物-基因协同调控网络。最后，结皮尔森相关分析及基因集富集分析（Gene Set Enrichment Analysis, GSEA）探索与中心代谢物密切相关的信号通路和生物学过程。

结果

       本研究纳入32例患者，平均年龄为61.3 ± 9.7岁，其中男性26人，占总人数的81.3%。通过OPLS-DA共筛选出205种差异代谢物，其中正离子模式88种，负离子模式117种。富集分析显示差异代谢物在组氨酸代谢、甘油磷脂代谢、丙氨酸、天冬氨酸和谷氨酸代谢、精氨酸生物合成、类固醇激素生物合成、氨基糖和核苷酸糖代谢，以及牛磺酸和次牛磺酸代谢中显著富集。转录组学数据共鉴定出966个交集基因，其富集的KEGG通路涉及脂质代谢、糖胺聚糖生物合成与代谢、氨基酸代谢、辅因子与维生素代谢、外源性异物的生物降解与代谢、糖代谢及其他代谢途径，表明EAT处于代谢活跃状态。

       差异代谢物和交集基因的联合分析发现氨基酸代谢（组氨酸代谢、丙氨酸、天冬氨酸和谷氨酸代谢以及牛磺酸和次牛磺酸代谢）、脂质代谢（甘油磷脂代谢）是EAT参与冠状动脉粥样硬化的重要代谢通路。相关性网络分析中，L-天冬氨酸、牛磺酸以及组胺是该网络中的关键代谢物，PEMT, CARNS1以及ETNK2是该网络的关键基因。代谢调控网络图显示，L-谷氨酸、L-天冬氨酸可串联多个代谢途径，是氨基酸代谢网络的中心代谢物；磷脂酰胆碱36:5与多个代谢物和基因具有相互作用，是甘油磷脂代谢的中心代谢物。GSEA分析发现L-谷氨酸、L-天冬氨酸以及磷脂酰胆碱36:5均显著富集于补体系统；L-谷氨酸和L-天冬氨酸与NK细胞毒性和IL-12通路相关；L-谷氨酸和磷脂酰胆碱36:5与CD22介导的B细胞受体调节相关。L-谷氨酸相关基因还显著富集于JAK-STAT信号通路；L-天冬氨酸相关基因富集于FcεRI介导的MAPK激活通路、T/B细胞受体信号通路以及IL-7通路；磷脂酰胆碱36:5则与IL-2家族信号通路和NLRP3炎症小体通路等生物学功能显著相关。

结论

       本研究结合非靶向代谢组学与转录组学技术，探究了EAT促进CAD进展的潜在分子机制。多组学分析表明，氨基酸代谢与脂质代谢是EAT参与冠状动脉粥样硬化的重要通路。L-谷氨酸、L-天冬氨酸和磷脂酰胆碱36:5在氨基酸代谢和甘油磷脂代谢中起关键作用。此外，GSEA分析表明，L-谷氨酸、L-天冬氨酸及磷脂酰胆碱36:5显著关联于免疫、炎症及细胞信号通路。本研究初步阐明了EAT在CAD中的可能作用机制，为后续深入探索其调控网络提供了重要线索。

第二部分 基于血浆蛋白质组学的急性心肌梗死生物标志物筛选及预测模型构建

目的

       急性心肌梗死（Acute myocardial infarction, AMI）是心血管疾病（Cardiovascular disease, CVD）患者猝死的主要原因之一，也是冠心病（Coronary artery disease, CAD）的常见类型。目前基于一级和二级预防的传统风险评分系统对AMI事件的预测能力有限，这一局限性限制了临床实践中对高危个体的精准识别与早期干预。本研究通过整合血浆蛋白质组学和遗传数据，旨在鉴定与AMI具有潜在因果关联的生物标志物，并构建一个适用于基线有或无动脉粥样硬化性心血管疾病（Atherosclerotic cardiovascular disease, ASCVD）个体的AMI通用风险预测模型。

方法

       本研究纳入了英国生物样本库（UK Biobank）中的52002名参与者，其中包括3668名有ASCVD病史的个体和48334名无ASCVD病史的个体。采用OLINK Explore技术对所有参与者血浆中的1461种蛋白质水平进行检测。采用多变量Cox比例风险模型初步筛选与AMI结局相关的血浆蛋白质，并进行功能富集分析。利用最小绝对收缩与选择算子（Least Absolute Shrinkage and Selection Operator, LASSO）算法结合10折交叉验证进一步优化特征选择过程，确定最具预测能力的蛋白质。随后，利用逻辑回归模型建立AMI结局的通用预测模型，并采用受试者工作特征曲线（Receiver operating characteristic curve, ROC）、校准曲线（Calibration curve）、以及决策曲线分析（Decision curve analysis, DCA）等方法对模型性能进行评估。此外，本研究采用巢式病例对照研究设计，结合局部加权回归散点平滑曲线（Locally weighted scatterplot smoothing, LOWESS）方法，系统分析了关键血浆蛋白的时间轨迹特征。通过孟德尔随机化（Mendelian randomization, MR）分析方法，评估了关键血浆蛋白与AMI风险的潜在因果关系。

结果

       在中位随访13.4年期间，研究人群中共观察到2100例AMI事件，占总人数的4.0%。通过多变量Cox比例风险模型筛选出206种与5年、10年和所有AMI发生相关的蛋白质。功能注释分析揭示这些蛋白质主要参与细胞间配体-受体相互作用和炎症反应调控等生物学过程。LASSO回归分析进一步鉴定出29种蛋白质（如MMP12和NTproBNP）与AMI风险显著相关。通过建立的风险预测模型评估发现，基于蛋白质标志物的模型预测性能显著优于临床特征变量，蛋白质模型对5年AMI的AUC为0.80，10年AMI的AUC为0.76，以及所有AMI事件的AUC为0.75，优于SCORE2模型的性能（分别为0.76、0.73和0.72）。整合蛋白质标志物与临床特征构建的通用预测模型显示出更优的诊断效能（AUC = 0.81，95% CI: 0.79-0.83）。并且，该预测模型在具有和不具有ASCVD病史的人群中均表现出良好的区分度。特别是在5年AMI事件的预测中，C统计量分别为0.73和0.78。在无ASCVD个体中，风险最高五分位组的AMI发生率显著高于有ASCVD个体的风险第二五分位组（所有AMI事件：8.58% vs 7.08%）。此外，对血浆蛋白的时间轨迹分析表明，在AMI发生前的15年期间，大多数蛋白质（如MMP12、GDF15和HAVCR1）在AMI人群中处于持续高水平，而RAB6A的表达水平较低。最后MR分析发现，遗传预测的血浆PTGDS升高与AMI风险增加显著相关（P = 0.006）。

结论

       本研究基于大规模蛋白质组学数据，筛选并鉴定出与AMI显著相关的29种蛋白质，并为PTGDS与AMI之间的因果关联提供了强有力的遗传学证据。整合蛋白质标志物和临床因素所构建的通用预测模型在基线有无ASCVD个体的AMI风险预测中均表现出较高的预测效能。并且，该模型能够识别出无ASCVD病史但AMI风险与ASCVD患者相当的个体，为进一步在临床实践中识别无症状高危人群并实施针对性干预提供了理论依据。

ABSTRACT

Part 1 Multi-omics Analysis of Epicardial AdiposeTissue Reveals Potential Molecular Mechanisms of Coronary Artery Disease

Objective

The increase of visceral fat and ectopic fat plays an important role in the development of coronary artery disease (CAD). Among them, epicardial adipose tissue (EAT) significantly affects the progression of atherosclerosis by releasing various adipokines and inflammatory mediators. However, the specific pathophysiological mechanism of EAT in coronary atherosclerotic lesions has not been fully elucidated yet. Therefore, in this study, we performed untargeted metabolomics, transcriptomics and multi-omics combined analysis of EAT and subcutaneous adipose tissue (SAT) to explore the potential mechanism of EAT in the pathogenesis of CAD.

Methods

This study included 32 CAD patients who received coronary artery bypass graft (CABG) in Fuwai Hospital. EAT and SAT were collected during the operation for untargeted metabolomics and transcriptomics analysis, respectively. Meanwhile, the clinical data of the patients were also collected. The metabolomics data were analyzed using orthogonal partial least squares-discriminant analysis (OPLS-DA) to identify differential metabolites, and MetaboAnalyst 6.0 was used for enrichment analysis of the differential metabolites. The transcriptomics data were analyzed using DESeq2 and weighted correlation network analysis (WGCNA) to identify differential genes and take the intersection. Gene Ontology (GO) enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis were performed on the intersection genes. Subsequently, a combined analysis of metabolomics and transcriptomics was conducted. Co-enrichment analysis was performed on the identified differential metabolites and intersection genes. Pearson correlation analysis was used to explore the correlation between metabolites and genes, and a correlation network between metabolites and genes was constructed using Cytoscape software. Then, based on the metabolic pathways in the KEGG database, a metabolite-gene co-regulation network of key metabolic pathways was constructed. Finally, specific pathways and biological processes closely related to central metabolites were explored by combining Pearson correlation analysis with Gene Set Enrichment Analysis (GSEA).

Results

This study included 32 patients with an average age of 61.3 ± 9.7 years, among whom 26 were male, accounting for 81.3% of the total. A total of 205 differential metabolites were screened out by OPLS-DA, including 88 in positive ion mode and 117 in negative ion mode. Enrichment analysis showed that the differential metabolites were significantly enriched in histidine metabolism, glycerophospholipid metabolism, alanine, aspartate and glutamate metabolism, arginine biosynthesis, steroid hormone biosynthesis, amino sugar and nucleotide sugar metabolism, and taurine and hypotaurine metabolism. A total of 966 intersection genes were identified from the transcriptomic data, and the enriched KEGG pathways involved lipid metabolism, glycosaminoglycan biosynthesis and metabolism, amino acid metabolism, cofactor and vitamin metabolism, xenobiotic biodegradation and metabolism, carbohydrate metabolism and other metabolic pathways, indicating that EAT was in a metabolically active state.

The combined analysis of differential metabolites and intersecting genes revealed that amino acid metabolism (histidine metabolism, alanine, aspartate and glutamate metabolism, and taurine and hypotaurine metabolism) and lipid metabolism (glycerophospholipid metabolism) are important metabolic pathways of EAT in atherosclerosis. In the correlation network analysis, L-aspartic acid, taurine, and histamine were the key metabolites, while PEMT, CARNS1, and ETNK2 were the key genes. The metabolic regulatory network diagram showed that L-glutamic acid and L-aspartic acid could connect multiple metabolic pathways and were central metabolites in the amino acid metabolism network; phosphatidylcholine 36:5 interacted with multiple metabolites and genes and was the central metabolite in glycerophospholipid metabolism. GSEA analysis found that L-glutamic acid, L-aspartic acid, and phosphatidylcholine 36:5 were all significantly enriched in the complement system; L-glutamic acid, and L-aspartic acid were related to natural killer cell cytotoxicity and IL-12 pathway; L-glutamic acid and phosphatidylcholine 36:5 were related to CD22-mediated B cell receptor regulation. L-glutamic acid -related genes were also significantly enriched in the JAK-STAT signaling pathway; L-aspartic acid-related genes were enriched in the FcεRI-mediated MAPK activation pathway, T/B cell receptor signaling pathway, and IL-7 pathway; phosphatidylcholine 36:5 was significantly related to biological functions such as the IL-2 family signaling pathway and the NLRP3 inflammasome pathway.

Conclusions

This study combined non-targeted metabolomics and transcriptomics techniques to explore the potential molecular mechanisms by which EAT promotes the progression of CAD. Multi-omics analysis indicated that amino acid metabolism and lipid metabolism are important pathways of EAT in atherosclerosis. L-glutamic acid, L-aspartic acid, and phosphatidylcholine 36:5 play key roles in amino acid metabolism and glycerophospholipid metabolism. Additionally, GSEA analysis revealed that L-glutamic acid, L-aspartic acid, and phosphatidylcholine 36:5 are significantly associated with immune, inflammatory, and cell signaling pathways. This study initially clarified the possible mechanism of EAT in CAD and provided important clues for further exploration of its regulatory network.

Part 2 Plasma Proteomics Analysis Indentifies Biomarkers and Risk Prediction Models of Acute Myocardial Infarction

Objective

Acute myocardial infarction (AMI) is one of the main causes of sudden death in patients with cardiovascular disease (CVD) and a common type of coronary artery disease (CAD). Currently, traditional risk scoring systems based on primary and secondary prevention have limited predictive ability for AMI events, and this limitation restricts the precise identification and early intervention of high-risk individuals in clinical practice. This study aims to identify new biomarkers and potential causal associations with AMI by integrating plasma proteomics and genetic data, and to construct a universal risk prediction model for AMI applicable to individuals with or without atherosclerotic cardiovascular disease (ASCVD).

Methods

This study included 52002 participants from the UK Biobank, among whom 3668 had a history of ASCVD and 48334 did not. The levels of 1461 proteins in the plasma of all participants were detected using the OLINK Explore technology. A multivariate Cox proportional hazards model was used to initially screen plasma proteins associated with AMI outcomes, and functional enrichment analysis was performed. The Least Absolute Shrinkage and Selection Operator (LASSO) algorithm combined with 10-fold cross-validation was utilized to further optimize the feature selection process and determine the most predictive proteins. Subsequently, a universal predictive model for AMI outcomes was established using a logistic regression model, and the model was evaluated using methods including the receiver operating characteristic curve (ROC), calibration curve, and decision curve analysis (DCA). Additionally, this study adopted a nested case-control study design and combined the locally weighted scatterplot smoothing (LOWESS) method to systematically analyze the temporal trajectory characteristics of key plasma proteins. Meanwhile, the potential causal relationship between key plasma proteins and AMI risk was evaluated using the mendelian randomization (MR) analysis.

Results

During a median follow-up of 13.4 years, a total of 2100 AMI events were observed in the study population, accounting for 4.0% of the total. Through a multivariate Cox proportional hazards model, 206 proteins were identified as being associated with 5-year, 10-year, and all incident AMI. Functional enrichment analysis revealed that these proteins were mainly involved in biological processes such as intercellular ligand-receptor interactions and regulation of inflammatory responses. LASSO regression analysis further identified 29 proteins (such as MMP12 and NTproBNP) significantly associated with AMI risk. The risk prediction model was evaluated and it was found that the prediction performance of the model based on protein markers was significantly better than that of clinical characteristic variables. Specifically, the AUC of the protein model for 5-year AMI was 0.80, for 10-year AMI was 0.76, and for all AMI events was 0.75, outperforming the SCORE2 model (0.76, 0.73, and 0.72, respectively). The universal prediction model constructed by integrating protein markers and clinical features demonstrated superior diagnostic performance (AUC = 0.81, 95% CI: 0.79 - 0.83). The prediction model showed good discrimination in both populations with and without a history of ASCVD, particularly in the prediction of 5-year AMI events, the C-statistics were 0.73 and 0.78, respectively. Importantly, in individuals without ASCVD, the incidence of AMI in the highest quintile of risk was significantly higher than that in the second quintile of risk in individuals with ASCVD (all AMI events: 8.58% vs 7.08%). The time trajectory analysis of plasma proteins indicated that most proteins (such as MMP12, GDF15, and HAVCR1) remained at high levels in the AMI population for 15 years before AMI occurrence, while the expression level of RAB6A was relatively low. Finally, MR analysis found that a genetically predicted increase in plasma PTGDS was significantly associated with an increased risk of AMI (P = 0.006).

Conclusions

Based on large-scale proteomics data, this study identified 29 proteins significantly associated with AMI, and provided strong genetic evidence for the causal relationship between PTGDS and AMI. The universal predictive model constructed by integrating protein markers and clinical factors demonstrated high predictive efficacy in AMI risk prediction for both individuals with and without ASCVD. Moreover, this model can identify individuals without a history of ASCVD but with an AMI risk comparable to that of ASCVD patients, providing a theoretical basis for further identification of asymptomatic high-risk individuals in clinical practice and implementation of targeted interventions.