The 'analysis of gene expression and biomarkers for point-of-care decision support in Sepsis' study; temporal clinical parameter analysis and validation of early diagnostic biomarker signatures for severe inflammation andsepsis-SIRS discrimination

Abstract

Introduction: Early diagnosis of sepsis and discrimination from SIRS is crucial for clinicians to provide appropriate care, management and treatment to critically ill patients. We describe identification of mRNA biomarkers from peripheral blood leukocytes, able to identify severe, systemic inflammation (irrespective of origin) and differentiate Sepsis from SIRS, in adult patients within a multi-center clinical study.

Methods: Participants were recruited in Intensive Care Units (ICUs) from multiple UK hospitals, including fifty-nine patients with abdominal sepsis, eighty-four patients with pulmonary sepsis, forty-two SIRS patients with Out-of-Hospital Cardiac Arrest (OOHCA), sampled at four time points, in addition to thirty healthy control donors. Multiple clinical parameters were measured, including SOFA score, with many differences observed between SIRS and sepsis groups. Differential gene expression analyses were performed using microarray hybridization and data analyzed using a combination of parametric and non-parametric statistical tools.

Results: Nineteen high-performance, differentially expressed mRNA biomarkers were identified between control and combined SIRS/Sepsis groups (FC>20.0, p<0.05), termed 'indicators of inflammation' (I°I), including CD177, FAM20A and OLAH. Best-performing minimal signatures e.g. FAM20A/OLAH showed good accuracy for determination of severe, systemic inflammation (AUC>0.99). Twenty entities, termed 'SIRS or Sepsis' (S°S) biomarkers, were differentially expressed between sepsis and SIRS (FC>2·0, p-value<0.05).

Discussion: The best performing signature for discriminating sepsis from SIRS was CMTM5/CETP/PLA2G7/MIA/MPP3 (AUC=0.9758). The I°I and S°S signatures performed variably in other independent gene expression datasets, this may be due to technical variation in the study/assay platform.

Keywords: SIRS; biomarker; diagnostic; mRNA signature; sepsis; severe inflammation.

Figure 1

Schematic overview of clinical study, recruitment, sample collection and processing, microarray hybridization and data analysis.

Figure 2

Radar plot of white blood cell, neutrophil, lymphocyte, basophil, free platelet and CRP counts at Days1, 2, 5 and discharge. PLMN DNS , PLMN S , SIRS DNS , SIRS S , ABDM DNS , ABDM S .

Figure 3

(A) PCA analysis of CNTRL versus combined SIRS&Sepsis biomarker groups (each symbol depicting an individual within each group) (B) volcano plot of log₁₀ p-value vs log fold-change of all gene entities, using a 2-fold change cutoff and with select I°I genes highlighted (C) PCA analysis of CNTRL versus combined SIRS&Sepsis biomarker groups (each symbol depicting an individual within each group) using select I°I genes only (from Table 2 ) (D) heat map of select I°I biomarkers from Table 2 across all control, SIRS, ABDM and PLMN sepsis groups stratified by day and prognosis (died/survived).

Figure 4

(A) Volcano plot of T-Test results from analysis of SIRS versus Sepsis biomarker groups (B) heat map of select S°S biomarkers from Table 3 across all control, SIRS, ABDM and PLMN sepsis groups stratified by day and prognosis (died/survived).

Figure 5

Selection of I°I signatures (A) Dot plot depiction of individual inflammatory biomarker candidates, including data from multiple biomarker probes where present. CNTRLs , SIRS , Sepsis (B) Random forest classification of validation data into controls and inflammation groups with ‘mtry’ of 31, ‘ntree’ of 2001 (C) Visualization of random forest models features of importance ranked by mean decrease accuracy and mean decrease Gini score (D) I°I candidate panel: ADM+CD177+FAM20A+ITGA7+MMP9+OLAH (E) I°I candidate panel: ADM+FAM20A+ OLAH+ITGA7+MPP9 (F) I°I candidate panel: ADM+OLAH+ FAM20A (G) I°I candidate panel: OLAH+FAM20A (H) I°I candidate panel: ADM+FAM20A+ OLAH+ITGA7+MMP9 across all time-points (I) ROC curves of ADM+FAM20A+OLAH ITGA7+MMP9 and OLAH+FAM20A comparing CNTRL vs SIRS/sepsis across day 1, day 2, day 5 and discharge time points (J) I°I candidate panel: OLAH+FAM20A across all timepoints.

Figure 6

Selection of S°S signatures (A) Dot pot depiction of individual SIRS/sepsis discriminatory biomarker candidates, with multiple versions of probes for some biomarkers. SIRS , ABDM , PLMN . (B) Random forest classification of validation data into SIRS and Sepsis ‘mtry’ of 11, ‘ntree’ of 2001 (C) Visualization of random forest models features of importance ranked by mean decrease accuracy and mean decrease Gini score (D) S°S candidate panel: CETP+CMTM5+MIA-MPP3-PLA2G7 (E) ROC curves of CETP+CMTM5+MIA-MPP3-PLA2G7 for SIRS vs Sepsis, SIRS vs Abdominal sepsis and SIRS vs Pulmonary sepsis comparisons.

Figure 7

Summary of best performing signatures and cut-off values to maximize discriminatory performance of (A) the I°I signature; CNTRL , SIRS , SEPSIS (B) the S°S Signature; SIRS , PLMN SEPSIS , ABDM SEPSIS .

Figure 8

Correlation plot of diagnostic performance of SIRS and sepsis-specific biomarkers to each other and to CRP.

Figure 9

Evaluation of I^OI Signature (CMTM5+ITGB3-PLA2G7-GPR124-ARHGEF10L) performance on published datasets (A) GSE131761 comparing healthy controls and septic shock , healthy controls and non-septic shock , non septic shock and septic shock (B) GSE9960 comparing healthy controls and sepsis (mixed infection) , healthy controls and sepsis (gram positive) , healthy controls and sepsis (gram negative) healthy controls and sepsis (C) GSE154918 comparing healthy controls and sepsis , healthy controls and follow up of sepsis , healthy controls and septic shock healthy controls and follow up of septic shock (D) GSE154918 comparing uncomplicated infection and sepsis , uncomplicated infection and follow up of sepsis , uncomplicated infection and septic shock , uncomplicated infection and follow up of septic shock .