Discriminatory plasma biomarkers predict specific clinical phenotypes of necrotizing soft-tissue infections

Abstract

BACKGROUNDNecrotizing soft-tissue infections (NSTIs) are rapidly progressing infections frequently complicated by septic shock and associated with high mortality. Early diagnosis is critical for patient outcome, but challenging due to vague initial symptoms. Here, we identified predictive biomarkers for NSTI clinical phenotypes and outcomes using a prospective multicenter NSTI patient cohort.METHODSLuminex multiplex assays were used to assess 36 soluble factors in plasma from NSTI patients with positive microbiological cultures (n = 251 and n = 60 in the discovery and validation cohorts, respectively). Control groups for comparative analyses included surgical controls (n = 20), non-NSTI controls (i.e., suspected NSTI with no necrosis detected upon exploratory surgery, n = 20), and sepsis patients (n = 24).RESULTSThrombomodulin was identified as a unique biomarker for detection of NSTI (AUC, 0.95). A distinct profile discriminating mono- (type II) versus polymicrobial (type I) NSTI types was identified based on differential expression of IL-2, IL-10, IL-22, CXCL10, Fas-ligand, and MMP9 (AUC >0.7). While each NSTI type displayed a distinct array of biomarkers predicting septic shock, granulocyte CSF (G-CSF), S100A8, and IL-6 were shared by both types (AUC >0.78). Finally, differential connectivity analysis revealed distinctive networks associated with specific clinical phenotypes.CONCLUSIONSThis study identifies predictive biomarkers for NSTI clinical phenotypes of potential value for diagnostic, prognostic, and therapeutic approaches in NSTIs.TRIAL REGISTRATIONClinicalTrials.gov NCT01790698.FUNDINGCenter for Innovative Medicine (CIMED); Region Stockholm; Swedish Research Council; European Union; Vinnova; Innovation Fund Denmark; Research Council of Norway; Netherlands Organisation for Health Research and Development; DLR Federal Ministry of Education and Research; and Swedish Children's Cancer Foundation.

Keywords: Bacterial infections; Diagnostics; Immunology; Infectious disease.

Conflict of interest statement

Conflict of interest: MBM is a steering committee member on SKIN-ICU, an observational study on NSTI. KS has received financial support for a study on sepsis biomarkers from Gentian. ANT and OH report research project funds from CIMED, Region Stockholm, the Swedish Research Council, Vinnova, and Innovation Fund Denmark.

Figures

Figure 1. Flow chart of the study pipeline.

The samples included in each test are displayed inside solid line boxes, light gray boxes show the reasons for exclusion at different stages of the study, and dark gray boxes indicate the specific test applied to the different set of samples. *Plasma samples from the INFECT cohort were excluded from the study if there was no positive microbiological culture in blood or tissue. Samples from patients with NSTI in nonamputable sites (i.e., neck, abdomen, and thorax) (**) or who had undergone amputation before admission (***) were not included for the prediction model for amputation.

Figure 2. Thrombomodulin is a plasma protein with biomarker potential for discrimination of NSTI and non-NSTI.

Concentrations of the soluble factors in plasma were compared among NSTI patients (n = 251), surgical controls (S. control; n = 20), and non-NSTI controls (n = 20). (A) The median protein levels in each cohort are depicted in the heatmap. All individual values are shown in Supplemental Figure 1. The measured proteins are divided by categories: I, chemokines; II, interleukins; III, soluble adhesion molecules; IV, matrix metalloproteases; and V, others. Significant differences between the measured concentrations were tested using Kruskal-Wallis (KW) test followed by Dunn’s post hoc test or Mann-Whitney U test (MW). Asterisks indicate the q cutoff obtained in at least 95% of the iterations. *q = 0.05; **q = 0.01; ***q = 0.005. The AUCs from the ROC analyses are given as the mean values of the iterations. The confidence intervals, specificities, and sensitivities of this test are included in Supplemental Table 1. (B) The RF result for discriminating NSTI versus non-NSTI is presented as the mean decrease Gini for each variable. The displayed P values are the result of the model including clinical data (Supplemental Table 1). SS, septic shock; type, microbiological classification of NSTI.

Figure 3. Biomarker panel for discrimination of type I and type II.

Levels of the soluble factors in plasma were compared between type I (n = 117) and type II (n = 134) patients within the NSTI discovery cohort (Table 1). (A) Heatmap depicting the median protein levels in each NSTI type. The measured proteins are divided by categories: I, chemokines; II, interleukins; III, soluble adhesion molecules; IV, matrix metalloproteases; and V, others. Significant differences between the measured concentrations were tested using Mann-Whitney U test. Asterisks indicate the q cutoff obtained in at least 95% of the results. *q = 0.05; **q = 0.01; ***q = 0.005. AUCs from the ROC analyses are shown as the mean values of the iterations. The confidence intervals, specificities, and sensitivities of this test are shown in Supplemental Table 2. (B) The RF result is shown as the mean decrease Gini for each variable. The displayed P values are the result of the model including clinical data (Supplemental Table 2).

Figure 4. Differential production of selected proteins by PBMCs after in vitro stimulation with GAS compared with B.

fragilisplusE. coli(mix). Stimulations were conducted in 6 repeat experiments using PBMCs from different donors stimulated with bacterial supernatant (S) or HK bacteria. (A–E) Scatter plots of each measured analyte in the supernatant after 24 hours of stimulation. The graphs display the individual values and the median with interquartile range. *P < 0.05, Wilcoxon’s matched pairs signed rank test.

Figure 5. Biomarker signatures associated with septic shock differ depending on etiology of NSTI.

Levels of the soluble factors in plasma were compared between patients with (n = 134) and without septic shock (n = 117) at admission within the NSTI discovery cohort (Table 1). (A) Heatmaps of the median protein concentrations in each phenotype. The measured proteins are divided by categories: I, chemokines; II, interleukins; III, soluble adhesion molecules; IV, matrix metalloproteases; and V, others. Significant differences between the measured concentrations were tested using Mann-Whitney U test. Asterisks indicate the q value cutoff obtained in at least 95% of the results. *q = 0.05; **q = 0.01; ***q = 0.005. The results from the ROC analysis are shown as the mean AUC values from the iterations. The confidence intervals, specificities, and sensitivities of this test are shown in Supplemental Table 3. (B) RF results are shown as the mean decrease Gini for each variable. Displayed P values are the results of the models including clinical data (Supplemental Table 4).

Figure 6. Predictive power of plasma biomarkers assessed in the validation cohort.

Selected biomarkers were tested for their potential to detect (A) NSTI (necrosis), (B) NSTI type, and (C and D) septic shock. Scatter plots display the individual values and the median with interquartile range. The discovery cohort consist of 60 NSTI patients, of which 39 were type I and 29 developed septic shock (Supplemental Table 5). In panel B, empty squares indicate type II NSTI caused by GAS (n = 7). The control group of 24 sepsis patients included 11 patients with septic shock. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; Mann-Whitney U test. ROC plots display results of the indicated biomarkers or clinical markers.

Figure 7. Type I and type II NSTIs display contrasting association networks.

The colors of the circles indicate the categories of the analytes. The strength of the partial correlation between analytes is indicated by the color and the weight of the connection.