Novel plasma protein biomarkers from critically ill sepsis patients

Abstract

Background: Despite the high morbidity and mortality associated with sepsis, the relationship between the plasma proteome and clinical outcome is poorly understood. In this study, we used targeted plasma proteomics to identify novel biomarkers of sepsis in critically ill patients.

Methods: Blood was obtained from 15 critically ill patients with suspected/confirmed sepsis (Sepsis-3.0 criteria) on intensive care unit (ICU) Day-1 and Day-3, as well as age- and sex-matched 15 healthy control subjects. A total of 1161 plasma proteins were measured with proximal extension assays. Promising sepsis biomarkers were narrowed with machine learning and then correlated with relevant clinical and laboratory variables.

Results: The median age for critically ill sepsis patients was 56 (IQR 51-61) years. The median MODS and SOFA values were 7 (IQR 5.0-8.0) and 7 (IQR 5.0-9.0) on ICU Day-1, and 4 (IQR 3.5-7.0) and 6 (IQR 3.5-7.0) on ICU Day-3, respectively. Targeted proteomics, together with feature selection, identified the leading proteins that distinguished sepsis patients from healthy control subjects with ≥ 90% classification accuracy; 25 proteins on ICU Day-1 and 26 proteins on ICU Day-3 (6 proteins overlapped both ICU days; PRTN3, UPAR, GDF8, NTRK3, WFDC2 and CXCL13). Only 7 of the leading proteins changed significantly between ICU Day-1 and Day-3 (IL10, CCL23, TGFα1, ST2, VSIG4, CNTN5, and ITGAV; P < 0.01). Significant correlations were observed between a variety of patient clinical/laboratory variables and the expression of 15 proteins on ICU Day-1 and 14 proteins on ICU Day-3 (P < 0.05).

Conclusions: Targeted proteomics with feature selection identified proteins altered in critically ill sepsis patients relative to healthy control subjects. Correlations between protein expression and clinical/laboratory variables were identified, each providing pathophysiological insight. Our exploratory data provide a rationale for further hypothesis-driven sepsis research.

Keywords: Biomarkers; Critical care; ICU; Proteomics; Sepsis.

Fig. 1

Targeted proteomics accurately differentiates critically ill sepsis patients from healthy controls. In the upper section, t-SNE plots depict the separation between sepsis patients and healthy controls. Age- and sex-matched subjects plotted in 2D following dimensionality reduction of their respective proteomes by t-distributed stochastic neighbour embedding. Axes are dimensionless. The dimensionality reduction shows that based on plasma proteome, the two cohorts are distinct and easily separable. A Black dots represent ICU Day-1 patients and green dots represent healthy controls. B Orange dots represent ICU Day-3 patients and green dots represent healthy controls. C Proteins that feature selection identified as classifying proteins on both ICU Day-1 and Day-3 are shown in the Venn-diagram with overlap. Proteins in Red text were upregulated when compared to healthy controls, whereas proteins in Blue text were downregulated when compared to healthy controls

Fig. 2

Significant temporal differences in plasma protein expression from sepsis patients. Box plots illustrating statistically significant differences between sepsis patients on ICU Day-1 versus ICU Day-3. The green shading illustrates healthy control protein expression (5–95% percentiles). Three ICU Day-1 classifying proteins significantly decreased in concentration by Day-3 (IL-10, CCL23, and TGFα1), and three ICU Day-3 classifying proteins significantly decreased in concentration by Day-3 (ST2, CNTN5, and ITGAV). Only one ICU Day-3 classifying protein was upregulated by ICU Day-3 (VSIG4). Median differences in protein concentrations were analyzed using Wilcoxon signed-rank tests, with P < 0.01 deemed statistically significant

Fig. 3

Correlations between normalized plasma protein expression and sepsis patient parameters on ICU Day-1 and Day-3. Heat maps of rank-based classifying proteins reported in Fig. 1 on ICU Day-1 (A) and ICU Day-3 (B) are illustrated (y -axis) along with patient parameters (x-axis). Only proteins that showed a significant correlation (P < 0.05) with at least one biochemical or clinical parameter are illustrated. Significant correlations had a Pearson R-value of ≥ 0.5 or ≤ -0.5 and P < 0.05, denoted by *. Positive correlations are depicted in red and negative correlations in blue. MODS Multiple Organ Dysfunction Score, SOFA Sequential Organ Failure Assessment, P/F arterial partial pressure of oxygen divided by the fraction of inspired oxygen concentration, MV mechanical ventilation, Vaso vasopressors, HgB hemoglobin concentration, WBC white blood cell count, PMN neutrophil count; Lymph lymphocyte count, PLT platelet count, PTT partial thromboplastin time, INR international normalized ratio, Creat creatinine concentration