Confounder-Adjusted Plasma Proteomics Identify Robust Protein Signatures in Sepsis Across Two Independent Perioperative Control Cohorts

Abstract

Background Sepsis is the primary driver of mortality in intensive care. Personalized medicine requires an accurate understanding of the syndrome, and plasma proteomics is increasingly used to characterize its molecular alterations. Comorbidities, demographics and postoperative sterile inflammation complicate this task, and proteins associated with them may be mistaken as sepsis-specific. Methods We compared 343 sepsis patients with two independent perioperative control cohorts (n = 73 and n = 75) sampled before and after surgery. Plasma was analyzed by mass spectrometry, and analyzed using linear models adjusted for age, sex and cohort-specific comorbidities. Propensity score matching was used to characterize renal comorbidities and Elastic Net classifiers were trained per cohort and time point. Results We show that renal comorbidities had the strongest impact on the plasma proteome and affected sepsis-associated proteins including Cystatin-C and Beta-2-microglobulin. Eleven proteins were robustly associated with sepsis, among them established acute phase proteins and the less well characterized Beta-1,4-galactosyltransferase 1. Feature importance analysis confirmed this core set and identified further proteins that contributed in a cohort- and time point-specific manner. Classifiers performed very well except against control patients at day 3 after surgery, when sterile inflammation peaked. Conclusions Sepsis, comorbidities and sterile inflammation may alter the same acute phase and renal proteins, differing in magnitude rather than in the type of response. We disentangled these influences and report a robust set of sepsis-associated proteins together with confounder-associated proteins, providing a basis for future studies and the development of diagnostic and therapeutic strategies.