Sepsis is a leading cause of mortality in intensive care, and improved molecular phenotyping is needed to support personalized approaches. Plasma proteomics is increasingly applied to characterize sepsis-associated alterations, but interpretation is complicated by patient comorbidities, demographic variation and noninfectious inflammatory states such as postoperative sterile inflammation. These confounders can cause proteins to be incorrectly attributed to sepsis.
The report compared plasma proteomes from a large sepsis cohort against two independent perioperative control cohorts sampled before and after surgery. The primary aim was to identify robust sepsis-associated proteins while explicitly adjusting for key confounders and to separate sepsis-specific signatures from proteomic changes driven by comorbidity or sterile inflammation.
The study analyzed plasma from 343 patients diagnosed with sepsis and two independent perioperative control cohorts with sample sizes of 73 and 75. The perioperative cohorts provided pre- and post-surgery time points to represent sterile inflammatory responses in a noninfectious setting. Plasma was profiled using mass spectrometry–based proteomics.
Proteomic data processing and downstream analyses were used to compare sepsis patients with controls across cohorts and time points. The authors note that the work is a preprint and has not been certified by peer review.
To reduce bias from demographics and chronic disease, the analyses used linear models adjusted for age, sex and cohort-specific comorbidities. Propensity score matching was applied specifically to characterize the influence of renal comorbidities on the plasma proteome. Elastic Net classifiers were trained separately for each cohort and sampling time point to evaluate discriminatory performance of proteomic signatures.
This multi-layered approach aimed to disentangle overlapping proteomic effects from sepsis, comorbidity and sterile postoperative inflammation rather than attributing all observed protein changes to sepsis alone.
Among examined comorbidities, renal disease had the strongest impact on plasma protein levels. Renal comorbidity altered proteins that are frequently reported as sepsis-associated in the literature, indicating potential confounding when renal impairment is present in sepsis cohorts.
Notably, renal comorbidities affected levels of Cystatin-C and Beta-2-microglobulin, both proteins commonly measured in clinical and research settings. The authors emphasize that such overlap can lead to misclassification of proteins as sepsis-specific if renal status is not considered in analyses.
After confounder adjustment, the study identified eleven proteins that were robustly associated with sepsis across cohorts and time points. This core set included established acute phase proteins as well as the less well characterized Beta-1,4-galactosyltransferase 1. Feature importance analyses from predictive models supported the relevance of this core set.
The authors additionally report other proteins that contributed to classification in a cohort- or time point–specific manner, reflecting context-dependent proteomic signals that may not generalize across settings.
Elastic Net classifiers trained per cohort and time point generally performed well in distinguishing sepsis patients from controls. An important exception occurred when classifiers were evaluated against control patients at postoperative day 3: performance deteriorated at that time point, coinciding with a peak in sterile inflammation following surgery.
This finding suggests that temporal dynamics of postoperative inflammation can confound sepsis detection based on plasma proteomics and highlights the need to account for timing relative to inflammatory insults when developing diagnostic models.
The results indicate substantial overlap in the types of proteins altered by sepsis, chronic comorbidities (particularly renal disease) and sterile postoperative inflammation. According to the authors, differences among these conditions are more quantitative—differences in magnitude of change—than qualitative differences in which proteins are affected.
By adjusting for demographics and comorbidities and by comparing to time-matched perioperative controls, the study disentangled these influences and produced a list of sepsis-associated proteins alongside confounder-associated proteins. The authors present these findings as a resource to inform future diagnostic and therapeutic research rather than as immediate clinical recommendations.
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE repository with the dataset identifier PXD082539. The authors conclude that their confounder-adjusted proteomic atlas offers a foundation for follow-up studies aiming to develop diagnostics or therapeutic strategies that account for comorbidity and sterile inflammation.
Competing interests and ethical approvals were declared: the authors report no competing interest, and ethical approval was granted by the Ethics committee of the Medical Faculty of the Ruhr-University Bochum. The report is a preprint and has not undergone peer review.