The clinical importance of a cytokine network in the acute phase of sepsis

Sepsis remains a major cause of death. Cytokines interact closely with each other and play a crucial role in the progression of sepsis. We focussed on the associations of a cytokine network with prognosis and disease severities in sepsis. This retrospective study included 31 patients with sepsis and 13 healthy controls. Blood samples were collected from patients on days 1, 2, 4, 6, 8, 11 and 15 and from healthy controls. Levels of PAI-1, IFN-α, IFN-γ, IL-1β, IL-6, IL-8, IL-12/IL-23p40, IL-17A, TNF-α, MCP-1, IL-4 and IL-10 were measured. SOFA, JAAM DIC and ISTH DIC scores were evaluated at the same times blood samples were taken. Network analysis revealed a network formed by PAI-1, IL-6, IL-8, MCP-1 and IL-10 on days 1, 2 and 4 throughout the acute phase of sepsis. There were positive correlations of each cytokine and the combined score (IL-6 + IL-8 + IL-10 + MCP-1) with the SOFA, JAAM DIC and ISTH DIC scores throughout the acute phase. A Cox proportional hazards model focussed on the acute phase showed that the above combined score was significantly related with patient prognosis, suggesting that the cytokine network of IL-6, IL-8, MCP-1 and IL-10 could play a pivotal role in the acute phase of sepsis.

Severities and outcome assessment. The Acute Physiology and Chronic Health Evaluation (APACHE) II score was assessed at the enrolment of the patients with sepsis. The APACHE II score is designed to assess the severity of critically ill patients based on physiologic measurements, age and previous health status and is used for the prediction of outcome in critically ill patients 16 . The Sequential Organ Failure Assessment (SOFA) score was assessed at the same time points that blood samples were taken. The SOFA is a scoring system composed of six organ systems (comprising the respiratory, coagulation, hepatic, cardiovascular, renal and neurologic systems) and can be used for the evaluation of organ failure and prognosis 17 . DIC is a life-threatening complication that occurs based on an imbalance of the coagulation cascade. The severity of DIC was also evaluated at the same time points that blood samples were taken with two definitions of DIC: the Japanese Association for Acute Medicine (JAAM) DIC score 18 and the International Society of Thrombosis and Haemostasis (ISTH) overt DIC score 19 . The JAAM score is sensitive for detecting septic DIC, and the ISTH overt DIC score is specific for diagnosing DIC. Both DIC scores correlate with each other 20 . The outcome measure was death within 28 days of diagnosis. Sepsis was divided into two phases based on a previous report: the acute phase (days 1-4) and the later phase (days [5][6][7][8][9][10][11][12][13][14] 21 . Statistical analysis. The levels of the 11 cytokines and of PAI-1 were transformed to common logarithm values to normalise the distribution of the data before the following analyses. The Dunnet test was used to evaluate the differences in each cytokine between the patients with sepsis and the healthy controls. The patients were respectively divided into two groups based on a previous report 22 on day 1, day 2 and day 4: critical ill patients with sepsis (SOFA score on each day ≥12) and non-critical ill patients (SOFA score on each day <12). The Dunnet test was also used to assess differences between critical ill patients, non-critical ill patients and the healthy controls. Correlations between the 11 cytokines and PAI-1 were assessed by Ward's hierarchical clustering analysis based on Pearson correlation coefficients. Network analysis was performed with Cytoscape ® software (www. cytoscape.org) version 3.5.1 23 . The network was based on the significant Pearson correlation coefficients between the 11 cytokines and PAI-1. The log2 fold changes were calculated by dividing average cytokine levels in the septic patients by the average levels in the healthy controls. The cytokines and PAI-1 with log2 fold change >1.5 were considered differentially increased. The network with major impact was visualised on the basis of the significant Pearson correlation coefficients between the cytokines and PAI-1. The combined score was calculated with the use of IL-1β, IL-6, IL-8, IL-10, MCP-1 and PAI-1, which are included in the networks with major impact, on the basis of a previous report 24 . Patients were divided into two groups based on the 75th percentile of each mediator's levels. The patients with mediator levels greater than or equal to the 75th percentile value were assigned the value "1", and those with mediator levels below the 75th percentile value were assigned the value "0". The combined scores were calculated by adding each of the mediators' values (i.e. the combination of three mediator scores consists of the individual value 0 or 1 or 2 or 3, the combination of four mediator scores consists of the individual value 0 or 1 or 2 or 3 or 4, the combination of five mediator scores consists of the individual value 0 or 1 or 2 or 3 or 4 or 5, and the combination of six mediator scores consists of the individual value 0 or 1 or 2 or 3 or 4 or 5 or 6). The combined scores A, B, C and D were calculated using combined mediator scores of (IL-1β, IL-6, IL-8, IL-10, MCP-1, PAI-1), (IL-6, IL-8, IL-10, MCP-1, PAI-1), (IL-6, IL-8, IL-10, MCP-1) and (IL-6, IL-8, MCP-1), respectively. Correlations of the 11 cytokines and PAI-1 and the combined scores with the SOFA and DIC scores were assessed by Spearman's correlation coefficients. The strengths of the associations were divided into four groups based on the correlation coefficients: strong (>0. Cox proportional hazards analysis with time-dependent covariates was performed based on the 11 cytokines and PAI-1 and the combined scores measured from day 1 to day 4 to evaluate the association of each value in the acute phase with death. It was reported that the maximum cytokine values and rapid increase of PAI-1 with a peak in the acute phase could reflect the severity of sepsis 25,26 . Therefore, the maximum values from three days (day 1, day 2 and day 4) were used for the analysis as time-dependent covariates to reflect the effect of the maximum values in the acute phase (i.e. day 1: the cytokine values measured on day 1; day 2: the maximum values from day 1 or day 2; and day 4: the maximum values from day 1, day 2 or day 4). The hazard ratio is provided as Q1 to Q3 to allow comparison of the strength of the association between the cytokines. We selected SOFA score and APACHE II score as confounders based on their strong relation with outcome in critically ill patients 16,17 . Competing risks were not taken into account because we focussed on overall survival.
To explore new predictive factors of sepsis, receiver operating characteristic (ROC) analysis with penalised maximum likelihood estimations were used to evaluate the associations of outcome with the SOFA score and several cytokines and PAI-1 and the combined scores while controlling for model overfitting. The penalised maximum likelihood estimation is one of the shrinkage approaches, which is really a special case of Bayesian modelling with a Gaussian prior 27 . A P value of <0.05 was considered to indicate statistical significance. Statistical analyses were performed with JMP Pro 13.0 for Windows (SAS Institute Inc., Cary, NC, USA) and R version 3.3.1 (R Foundation for Statistical Computing, Vienna, Austria).

Results
Patient characteristics. Thirty-one patients with sepsis and 13 healthy controls were consecutively enroled in this study (Table 1). In total, 156 blood samples collected from the 31 patients and 13 samples collected from the controls were analysed. At the time of enrolment, patients were in septic shock. Of the patients with sepsis, 23 were men and 8 were women, and the mortality rate of these patients was 22 Table 1. There was no significant difference in age and sex between the patients with sepsis and the controls.

Sepsis 11
Septic shock 20  Cytokine (IL-1β, IL-6, IL-8, IL-10, MCP-1) and PAI-1 levels in critical and non-critical ill patients. We evaluated the levels of cytokines IL-1β, IL-6, IL-8, MCP-1 and IL-10 and of PAI-1, which increased over the acute phase (Figs 1 and 2), in the critical and non-critical ill patients. The levels of IL-10 (days 1, 2 and 4) and IL-6 and PAI-1 (days 2 and 4) and IL-8 (day 4) in the critically ill patients were significantly increased compared to those in the non-critically ill patients. The levels of IL-6, IL-8, IL-10 and PAI-1 in the critically ill patients were significantly increased compared over the acute phase. MCP-1 levels in the critically ill patients showed significant increases compared with those of the controls on days 1 and 2. IL-1β levels showed no significant difference between the critically and non-critically ill patients and the controls in the acute phase (Fig. 2).
Spearman's correlations between 11 cytokines and PAI-1 and combined scores and SOFA and DIC scores. We chronologically investigated Spearman's correlation coefficients between the 11 cytokines and PAI-1 and the combined scores. There were very weak to strong positive correlations of the SOFA score with IL-6, IL-8, MCP-1, IL-10 and PAI-1 and combined scores A, B, C and D throughout the study period (Fig. 4A).  hazards analysis with time-dependent covariates was conducted. The cytokine and PAI-1 values were transformed to common logarithm values to normalise the distribution of the data before the analyses. The maximum value of each cytokine and PAI-1 and the combined scores measured on 3 days in the acute phase was used for the time-dependent covariate to highlight the acute phase. IL-6, IL-8, MCP-1, IL-10, PAI-1, IL-12/23p40 and combined scores A, B, C and D showed significant correlation with patient prognosis. IL-6, IL-8, MCP-1 and combined scores A, B, C and D adjusted by the SOFA score or APACHE II score showed significant associations with prognosis (Fig. 5).

ROC analysis of the cytokines and PAI-1 and combined scores and SOFA score.
To investigate potential prognostic biomarkers, ROC analysis was performed with IL-1β, IL-6, IL-8, MCP-1, IL-10 and PAI-1, which had increased in the acute phase, and the combined scores and the SOFA score on day 1. Assessment with the SOFA score is inevitable for the diagnosis of sepsis 15 and is an important prognostic marker in the clinical setting. Therefore, we assessed whether the AUC analysed by the SOFA score with each cytokine increased compared with the AUC analysed by the SOFA score only.  , combined score C (0.938) and combined score D (0.949) were significantly increased compared with that of the SOFA score only (Fig. 6).

Discussion
In this study, the levels of IL-1β, IL-6, IL-8, MCP-1 and PAI-1, which reflect endothelial injury, significantly increased in the early acute phase (day 1) of septic patients compared with those of the controls (Fig. 1A). The network analysis revealed a network of these mediators with weak to moderate correlations (Fig. 3B). A variety of cytokines are produced from immune cells in sepsis. IL-1β, which is mainly generated by activated monocytes and macrophages 28 , acts on different immune cells including endothelial cells. The activated immune cells then produce cytokines such as IL-6, IL-8 and MCP-1 29,30 . This suggests that IL-1β might promote the cytokine network formed by IL-6, IL-8 and MCP-1.
TNF-α is recognised as an important pro-inflammatory cytokine that produces cytokines such as IL-6, IL-8 and MCP-1 in addition to IL-1β 31,32 . In this study, TNF-α was not related to those cytokines (Fig. 3B), which indicates that the effect of TNF-α as a pro-inflammatory cytokine is limited.
This study showed significant increases of the IL-6, IL-8 and PAI-1 levels of the septic patients compared with those of the controls throughout the acute phase (Fig. 1A). The IL-10 levels of the critically ill patients showed significant increases compared with those of the non-critically ill patients and the controls throughout the acute phase. There were significant increases of MCP-1 levels on days 1 and 2 and on day 4 (t-test, P = 0.045) compared with those of the controls (Fig. 2). Hierarchical clustering analysis revealed a cluster formed by IL-6, IL-8, IL-10, MCP-1 and PAI-1, and network analysis clearly indicated a network of these mediators, which showed weak to strong correlations throughout the acute phase (Fig. 3A,B). Endothelial injury, which is strongly associated with the progression of sepsis 33 , can proliferate to generate cytokines such as IL-6, IL-8 and MCP-1. This indicates that endothelial cell injury might play a role in forming the cytokine network composed of IL-6, IL-8 and MCP-1 throughout the acute phase of sepsis. Also, the augmentation of IL-6, IL-8 and MCP-1 could proliferate the inflammation by endothelial cells as a positive feedback system 32,34,35 . Among these cytokine levels, the level of IL-6 increased most strikingly over the acute phase (Fig. 1A). The high concentration of IL-6 binds to the soluble  The cytokine network in the acute phase was formed by the pro-inflammatory cytokines IL-6, IL-8 and MCP-1 and an anti-inflammatory cytokine, IL-10. Recently, the concurrent presence of both pro-inflammatory and anti-inflammatory mediators from the onset of sepsis has been described 36 . Our results were consistent with this report and suggest that both pro-inflammatory and anti-inflammatory responses correlate with the pathogenesis of sepsis in a mutual relationship 37 .
The pro-inflammatory cytokines are closely related with the progression of the coagulation process in sepsis. Excessive pro-inflammatory cytokines promote the expression of tissue factor, which is predominantly synthesised by activated monocytes 38,39 , causing coagulation disorder and microthrombi formation 2 . The formation of microthrombi could contribute to microcirculatory dysfunction and result in multiple organ failure that leads to death 40 . In our results, cytokines in the cytokine network (IL-6, IL-8, MCP-1 and IL-10) and the SOFA, JAAM and ISTH DIC scores showed similar time-dependent changes (Fig. 1A,B). Very weak to moderate positive correlations of each of the cytokines and combined score C (IL-6 + IL-8 + IL-10 + MCP-1) with the SOFA, JAAM DIC and ISTH DIC scores were seen over the acute phase ( Fig. 4A-C). Also, in the Cox proportional hazards model, which focussed on the acute phase, combined score C (IL-6 + IL-8 + IL-10 + MCP-1) adjusted by SOFA or APACHE II scores showed a significant relation with the prognosis of the patient (Fig. 5). This suggests that the cytokine network composed by the pro-inflammatory cytokines IL-6, IL-8 and MCP-1, when interacting with endothelial cells, could facilitate the progression of sepsis based on the coagulation disorder, leading to a lethal outcome. However, the anti-inflammatory cytokine IL-10 might act as a negative feedback mechanism against the inflammatory response.
To discover clinically useful markers of prognosis on day 1, we compared the AUCs analysed by SOFA score with each cytokine in the cytokine network (IL-1β, IL-6, IL-8, MCP-1 and IL-10) and PAI-1 and the combine scores to the AUC of the SOFA score only. The AUC analysed by the SOFA score with combined score A (IL-1β + IL-6 + IL-8 + IL-10 + MCP-1 + PAI-1) (0.958) was highest and was significantly increased compared with the AUC of the SOFA score only (0.658). The increased AUC of the SOFA score with combined score A was the same as that of combined score A only (0.958) (Fig. 6B). This suggests that the combined scores of IL-1β + IL-6 + IL-8 + IL-10 + MCP-1 + PAI-1 could be a useful prognostic marker without using the SOFA score.
Limitations of this study are the relatively small number of patients included and the use of data from a single institution. Further study is necessary to clarify the role of cytokine networks in the pathogenesis of sepsis. We believe that these findings will have implications for the management of patients with sepsis.

Conclusions
Cytokine profiles were assessed in patients with sepsis. We found that IL-6, IL-8, MCP-1 and IL-10 formed a cytokine network in the acute phase of sepsis and that the combined score of IL-6 + IL-8 + IL-10 + MCP-1 correlated with patient prognosis and disease severity.