In Journal of personalized medicine
BACKGROUND : Sepsis in patients after cardiovascular surgery with cardiopulmonary bypass (CPB) has a high rate of mortality. We sought to determine whether changes in lipidomics can predict sepsis after cardiac surgery.
METHODS : We used high-performance liquid chromatography coupled to tandem mass spectrometry to explore global lipidome changes in samples from a prospective case-control cohort (30 sepsis vs. 30 nonsepsis) hospitalized with cardiovascular surgery. All patients were sampled before and within 48-72 h after surgery. A bioinformatic pipeline was applied to acquire reliable features and MS/MS-driven identifications. Furthermore, a multiple-step machine learning framework was performed for signature discovery and performance evaluation.
RESULTS : Compared with preoperative samples, 94 features were upregulated and 282 features were downregulated in the postoperative samples of the sepsis group, and 73 features were upregulated and 265 features were downregulated in the postoperative samples of the nonsepsis group. "Autophagy", "pathogenic Escherichia coli infection" and "glycosylphosphatidylinositol-anchor biosynthesis" pathways were significantly enriched in the pathway enrichment analysis. A multistep machine learning framework further confirmed that two cholesterol esters, CE (18:0) and CE (16:0), were significantly decreased in the sepsis group (p < 0.05). In addition, oleamide and stearamide were increased significantly in the postoperative sepsis group (p < 0.001).
CONCLUSIONS : This study revealed characteristic lipidomic changes in the plasma of septic patients before and after cardiac surgery with CPB. We discovered two cholesterol esters and two amides from peripheral blood that could be promising signatures for sepsis within a dynamic detection between the preoperative and postoperative groups.
Ding Wenyan, Xu Shaohang, Zhou Baojin, Zhou Ruo, Liu Peng, Hui Xiangyi, Long Yun, Su Longxiang
2022-Nov-03
cardiac surgery, lipidomics, machine learning, sepsis, signature
