An Endotoxin Tolerance Signature Predicts Sepsis and Organ Dysfunction at Initial Clinical Presentation

Abstract

Background: Sepsis involves aberrant immune responses to infection, but the exact nature of this immune dysfunction remains poorly defined. Bacterial endotoxins like lipopolysaccharide (LPS) are potent inducers of inflammation, which has been associated with the pathophysiology of sepsis, but repeated exposure can also induce a suppressive effect known as endotoxin tolerance or cellular reprogramming. It has been proposed that endotoxin tolerance might be associated with the immunosuppressive state that was primarily observed during late-stage sepsis. However, this relationship remains poorly characterised. Here we clarify the underlying mechanisms and timing of immune dysfunction in sepsis.

Methods: We defined a gene expression signature characteristic of endotoxin tolerance. Gene-set test approaches were used to correlate this signature with early sepsis, both newly and retrospectively analysing microarrays from 593 patients in 11 cohorts. Then we recruited a unique cohort of possible sepsis patients at first clinical presentation in an independent blinded controlled observational study to determine whether this signature was associated with the development of confirmed sepsis and organ dysfunction.

Findings: All sepsis patients presented an expression profile strongly associated with the endotoxin tolerance signature (p < 0.01; AUC 96.1%). Importantly, this signature further differentiated between suspected sepsis patients who did, or did not, go on to develop confirmed sepsis, and predicted the development of organ dysfunction.

Interpretation: Our data support an updated model of sepsis pathogenesis in which endotoxin tolerance-mediated immune dysfunction (cellular reprogramming) is present throughout the clinical course of disease and related to disease severity. Thus endotoxin tolerance might offer new insights guiding the development of new therapies and diagnostics for early sepsis.

Keywords: Cellular reprogramming; Diagnosis; Endotoxin tolerance; Immune dysfunction; Sepsis; Severe sepsis; Signature.

Fig. 1

Definition of the ‘endotoxin tolerance signature’. Schematic representation of the method used to define the endotoxin tolerance signature and the inflammatory signature. The endotoxin tolerance signature was obtained from our previously published dataset (Pena et al., 2011) and defined as 99 genes uniquely differentially expressed in endotoxin-tolerant human PBMCs (treated twice with LPS), but not inflammatory human PBMCs (treated once with LPS), as compared to controls (fold change > 2, p-value < 0.05) (please see ref. Pena et al., 2011 for more details regarding this dataset). The inflammatory signature was reduced from the 178 genes uniquely differentially expressed in inflammatory human PBMCs to a common 93 gene signature, by selecting genes that were consistently differentially expressed in an in vivo human volunteer endotoxin challenge dataset (Calvano et al., 2005).

Fig. 2

Sepsis patients from published datasets showed a strong association with the ‘endotoxin tolerance signature’. A gene-set test approach, ROAST (Wu et al., 2010), testing the statistically significant presence of a signature (collection) of genes, was used to characterise the enrichment of ‘Endotoxin Tolerance’ in sepsis patients versus controls from a small study performed by us and 10 previously published datasets. All datasets contained sepsis patients recruited at day 1 or 3 post-ICU admission and were compared to ‘healthy’ controls. The ROAST gene-set test was run with 99,999 rotations so the most significant p-value resulting from this test is 0.00001. p-Values from the ROAST gene-set test were graphed as log (1/p-value), and untransformed p-values are shown for ease of visualization.

Fig. 3

The ‘endotoxin tolerance signature’ was strongly associated with sepsis patients at first clinical presentation. A gene-set test approach (Wu et al., 2010) was used to characterise the enrichment, cf. controls as well as non-sepsis critically ill patients, of the ‘Endotoxin Tolerance’ and ‘Inflammatory’ signatures in prospective sepsis patients from a unique in-house cohort recruited on first clinical suspicion of sepsis (i.e. generally in the emergency ward and before ICU admission cf. the studies described in Fig. 2 that were post-ICU admission). Patient groups were subsequently defined based on retrospective clinical characteristics as ‘Sepsis’ or ‘No Sepsis’ consistent with the current sepsis criteria^3,15,16,36 (Supplemental Table 4). Analyses were performed comparing (a) the ‘Sepsis’ and ‘No Sepsis’ groups vs. controls and (b) the ‘Sepsis’ and the ‘No Sepsis’ groups to each other. Additionally, enrichment of the signature was also analysed based on microbial culture results within (c) the ‘Sepsis’ group and (d) the ‘No Sepsis’ group.

Fig. 4

The ‘endotoxin tolerance signature’ was strongly associated with sepsis patients at first clinical presentation and was associated with the severity of the disease. A gene-set test approach (Wu et al., 2010) was used to characterise the enrichment, cf. surgical controls, of the ‘Endotoxin Tolerance’ and ‘Inflammatory’ signatures in prospective sepsis patients from a unique in-house cohort recruited on first clinical suspicion of sepsis (i.e. generally in the emergency ward prior to ICU admission). (a) Patients were grouped into individual-, combined- (3 +), individual type of organ failure and no-organ failure groups. NB. No patients were observed with sepsis-associated encephalopathy. (b) Patients were also grouped into those requiring and those not-requiring transfer to the ICU.

Fig. 5

A core-set of endotoxin tolerance genes characteristic of sepsis patients. A core-set of 31 of the 99 genes from the ‘endotoxin tolerance signature’ was determined based on the most frequently differentially expressed genes observed literature sepsis datasets. For better visual comparison across different studies, each individual dataset was further transformed by dividing gene expression values into six equal bins. Data is presented as a heatmap with blue and red representing relatively low and high expression, respectively.