Background
Sepsis is characterized by a detrimental unbalanced host response to an infection, resulting in damage to tissues and dysfunction of organs [
1]. The outcome of sepsis is influenced by elements associated with the causative pathogen, the primary source of the infection and the host. With respect to the latter, genetic composition, age and comorbidities are considered to contribute [
1].
Diabetes mellitus is one of the most common comorbidities in patients with sepsis [
2]. Diabetes mellitus is associated with increased susceptibility to infection and sepsis [
3], likely due, at least in part, to compromised immune responses, such as adhesion, chemotaxis, phagocytosis and bacterial killing by immune cells [
3,
4]. Diabetes mellitus can be accompanied by chronic organ dysfunction together with a state of low-grade chronic inflammation and activation of the vascular endothelium and the coagulation system [
4‐
6], pathological alterations that in a more acute way also feature in patients with sepsis [
1]. Nonetheless, several studies have reported that diabetes does not change the mortality of sepsis [
7‐
10]. Knowledge of the possible influence of diabetes mellitus on the host response to severe sepsis is limited. One investigation reported similar cytokine and procoagulant responses in critically ill patients with sepsis with and without pre-existing diabetes mellitus [
9], while another study reported elevated levels of endothelial cell activation markers in patients with diabetes mellitus and septic shock, relative to patients without diabetes mellitus [
11].
The primary objective of the present study was to provide more insight into the association between pre-existing diabetes mellitus and the host response to sepsis. To this end, we studied the host response in a large prospective cohort of well-documented critically ill patients with sepsis stratified according to the presence or absence of diabetes mellitus, using both a targeted approach (by measuring 15 plasma biomarkers reflective of pathways implicated in sepsis pathogenesis) and an unbiased approach (by analyzing the whole genome expression profiles in blood leukocytes). Additionally, considering that the common anti-diabetic drugs insulin [
12‐
15] and metformin [
16‐
18] can exert immune modulatory effects, the secondary objective of this study was to determine whether patients with diabetes mellitus treated with either of these medications presented with an altered host response to sepsis.
Methods
Study design, setting and patient identification
This study was conducted as part of the Molecular Diagnosis and Risk Stratification of Sepsis (MARS) project, a prospective observational study in the mixed intensive care units (ICUs) of two tertiary teaching hospitals (Academic Medical Center in Amsterdam and University Medical Center in Utrecht) (ClinicalTrials.gov identifier NCT01905033) [
19‐
21]. Consecutive patients above 18 years of age admitted between January 2011 and July 2013 with an expected length of stay longer than 24 hours were included via an opt-out method approved by the medical ethical committees of the participating hospitals. All patient data were encrypted for privacy reasons and no separate ethical approval was required for sub-studies such as the one described here. Analyses were limited to the first 2.5 years of the MARS project, because host response measurements were restricted to this period for financial reasons. For every admitted patient the plausibility of an infection was assessed using a 4-point scale (ascending from none to possible, probable or definite) using the Center for Disease Control and Prevention and International Sepsis Forum consensus definitions, making use of all clinical, radiological and microbiological data, as previously described in detail [
19]. Sepsis was defined as the presence of infection diagnosed within 24 hours after ICU admission with a probable or definite likelihood, accompanied by at least one additional parameter as described in the 2001 International Sepsis Definitions Conference [
22]. Dedicated research physicians prospectively collected demographic data, details of comorbidities (including the Charlson comorbidity index [
23]), and daily clinical data and severity scores, including Acute Physiology and Chronic Health Evaluation (APACHE) IV [
24] and Sequential Organ Failure Assessment (SOFA) scores (the central nervous system was excluded) [
25].
Diabetes mellitus was defined as known history of type I or type II diabetes mellitus prior to ICU admission or by the use of oral anti-diabetic medication or insulin as chronic medication. The timing of the last dose of insulin or metformin prior to ICU admission was not registered; metformin was stopped after ICU admission. Cardiovascular insufficiency was defined as a medical history of congestive heart failure, chronic cardiovascular disease, peripheral vascular disease or cerebrovascular disease. Malignancy was defined by medical history of either non-metastatic solid tumor, metastatic malignancy or hematologic malignancy. Patients with a history of chronic renal insufficiency or with chronic intermittent hemodialysis or continuous ambulatory peritoneal dialysis were marked as renal insufficient [
21].
Specific organ failure was defined by a SOFA score ≥3, except for cardiovascular failure for which a score ≥1 was used (the central nervous system was excluded) [
26]. Shock was defined by the use of vasopressors (noradrenaline) for hypotension in a dose >0.1 μg/kg/min during at least 50 % of the ICU day. Acute kidney injury and acute lung injury were defined using strict pre-set criteria [
27,
28]. ICU-acquired complications were defined as events that started 2 days or more after ICU admission. The Municipal Personal Records Database was consulted to calculate survival after ICU admission. Patients readmitted to the same ICU or transferred from other ICUs were excluded, except when patients were referred to one of the study centers on the day of admission.
Plasma protein assays
Daily (on admission and at 6 a.m. thereafter) EDTA anti-coagulated leftover plasma harvested from blood obtained for regular patient care was stored within 4 hours after blood draw at -80 °C. Measurements were done in EDTA anti-coagulated plasma. Tumor necrosis factor (TNF)-α, interleukin (IL)-6, IL-8, IL-1β, IL-10, IL-13, interferon-γ, soluble E-selectin, soluble intercellular adhesion molecule (ICAM)-1 and fractalkine were measured by FlexSet cytometric bead array (BD Biosciences, San Jose, CA, USA) using FACS Calibur (Becton Dickenson, Franklin Lakes, NJ, USA). Matrix metalloproteinase (MMP)-8, angiopoietin-1, angiopoietin-2, protein C, antithrombin (all R&D systems, Abingdon, UK) and D-dimer (Procartaplex, eBioscience, San Diego, CA, USA) were measured by Luminex multiplex assay using BioPlex 200 (BioRad, Hercules, CA, USA). C-reactive protein (CRP) was determined by immunoturbidimetric assay (Roche diagnostics), prothrombin time (PT) and activated partial thromboplastin time (aPTT) by using a photometric method with Dade Innovin Reagent or by Dade Actin FS Activated PTT Reagent, respectively (both Siemens Healthcare Diagnostics). Normal values were obtained in plasma from 27 age-matched and gender-matched healthy volunteers, included after providing written informed consent, with the exception of CRP, PT and aPTT (routine laboratory reference values).
Blood gene expression microarrays
Whole blood was collected in PAXgene™ tubes (Becton-Dickinson, Breda, the Netherlands) within 24 hours after ICU admission. Total RNA was isolated using the PAXgene blood mRNA kit (Qiagen, Venlo, the Netherlands) in combination with QIAcube automated system (Qiagen, Venlo, the Netherlands), according to the manufacturer’s instructions. RNA (RNA integrity number >6.0) was processed and hybridized to the Affymetrix Human Genome U219 96 array and scanned using the GeneTitan instrument at the Cologne Center for Genomics (CCG), Cologne, Germany, as described by the manufacturer (Affymetrix). Raw data scans (.CEL files) were read into the R language and environment for statistical computing (version 2.15.1; R Foundation for Statistical Computing, Vienna, Austria;
http://www.R-project.org/).
Pre-processing and quality control was performed using the Affy package version 1.36.1. Array data were background-corrected by Robust Multi-array Average, quantile-normalized and summarized by medianpolish using the expresso function (Affy package). The resultant 49,386 log-transformed probe intensities were filtered by means of a 0.5 variance cutoff using the genefilter method [
29] to recover 24,646 expressed probes in at least one sample. The occurrence of non-experimental chip effects was evaluated by means of the Surrogate Variable Analysis (R package version 3.4.0) and corrected by the empirical Bayes method ComBat [
30]. The non-normalized and normalized MARS gene expression data sets are available at the Gene Expression Omnibus public repository of the NCBI [GEO:GSE65682].
The 24,646 probes were assessed for differential abundance across healthy subject and patient samples by means of the limma method (version 3.14.4) [
31]. Supervised analysis (comparison between predefined groups) was performed by moderated
t statistics. Throughout, significance was defined using the Benjamini-Hochberg (BH) multiple comparison adjusted probabilities, correcting for the 24,646 probes (false discovery rate <5 %). Ingenuity Pathway Analysis (Ingenuity Systems IPA,
www.ingenuity.com) was used to identify the associated canonical signaling pathways stratifying genes by over-expressed and under-expressed patterns. The Ingenuity gene knowledgebase was selected as reference and human species specified. All other parameters were left at default. The significance of association was assessed using Fisher’s exact test.
Statistical analysis
All data distributions were tested for normality using the Shapiro-Wilk test and histogram plots. The Mann-Whitney U test or Kruskal-Wallis test was used to analyze continuous nonparametric data, presented as median and interquartile range (IQR, 25th and 75th percentiles). Continuous parametric data, presented as numbers (percentages) or as means ± standard deviation (SD), were analyzed using Student’s t test or analysis of variance when appropriate. All categorical data were analyzed using the chi square test. As the biomarker data were not normally distributed, the Kruskal-Wallis test was used to analyze non-parametric data. A multivariable cox proportional hazard model was used to determine the association between diabetes mellitus and mortality. The covariables included in the model were BMI, patient age, gender, cardiovascular insufficiency, renal insufficiency and hypertension. A sensitivity analysis was conducted, correcting for the APACHE IV score.
All data were analyzed using R studio built under R version 3.0.2 (R Core Team 2013, Vienna, Austria) [
32]. The R package
survival was used for the survival analysis. Multiple-comparison-adjusted (BH)
p values <0.05 were used to define the significance of plasma biomarkers.
Discussion
The main finding of this investigation is that sepsis patients with a medical history of diabetes mellitus have a similar host response upon ICU admission when compared with sepsis patients without known diabetes mellitus. In addition, pre-existing diabetes mellitus did not alter sepsis-associated mortality up to 90 days after ICU admission. We also demonstrated that amongst patients with diabetes mellitus two common diabetes mellitus therapies, insulin and metformin, are not associated with a modified host response or sepsis outcome.
Our data on similar case fatality rates of sepsis in critically ill patients with or without a known history of diabetes mellitus are in accordance with previous studies [
7,
9,
10]. Notably, however, other investigations have reported increased [
33,
34] or decreased [
35] sepsis mortality in patients with diabetes mellitus. These latter studies differ from ours in that they did not merely include patients with sepsis in need of intensive care, and as a consequence thereof reported much lower mortality rates than observed in our sepsis ICU population. Together these data suggest that the association between diabetes mellitus and sepsis outcome depends, at least in part, on the (sub)acute severity of disease.
Diabetes mellitus was not associated with altered host responses during sepsis, suggesting that diabetes-mellitus-related low-grade chronic inflammation, activation of the vascular endothelium and the coagulation system [
4‐
6], does not influence acute infection-induced alterations in these pathways. In accordance, a previous study likewise reported comparable markers of coagulation and inflammation in patients with diabetes mellitus and sepsis [
9]. Similarly, diabetes mellitus was not associated with altered cytokine or procoagulant responses in patients presenting at the emergency room with community-acquired pneumonia [
36].
We measured several plasma biomarkers of endothelial cell activation (soluble E-selectin, soluble ICAM-1 and fractalkine) [
37] and barrier function (angiopoietin-1 and angiopoietin-2) [
37], neither of which was modified in patients with diabetes mellitus. The vascular endothelium is of great interest in the context of diabetes mellitus and sepsis, because chronic diabetes mellitus can cause a dysfunctional endothelium leading to diabetes-mellitus-related atherosclerosis [
38], and upregulation of ICAM-1, vascular cell adhesion molecule-1 (VCAM-1) and E-selectin, resulting in increased adherence of neutrophils [
39]. An earlier investigation reported endothelial cell activation plasma markers in patients presented to the emergency department with clinically suspected infection [
11]. In this cohort soluble ICAM-1 and VCAM-1 did not differ between patients with and without diabetes mellitus; however, plasma-soluble E-selectin and soluble fms-like tyrosine kinase-1 were higher in patients with diabetes mellitus; notably, unlike our investigation, diabetes mellitus patients presented with more severe disease when compared with patients who did not have diabetes mellitus, possibly confounding the interpretation of plasma biomarker measurements [
11].
Similarities in patients with sepsis who did and did not have diabetes mellitus were not exclusive to plasma biomarkers, as we also observed no differences in their blood leukocyte transcriptomes. Almost identical alterations were evident in both groups of patients as compared to healthy controls; alterations that were strongly attuned to a “common host/sickness response” [
20]. Over-expression of typical pro-inflammatory, anti-inflammatory and metabolic pathways in parallel with under-expression of a plethora of T cell related pathways were overarching features of the common response. Notably, key cellular metabolic pathways (mitochondrial dysfunction, EIF2 and MTOR signaling) were also similarly altered in patients with and without diabetes mellitus.
Our study is the first to analyze sepsis patients with insulin-treated compared to patients with non-insulin-treated diabetes mellitus, revealing no differences between groups in the presentation or outcome of sepsis, or the host responses. An earlier investigation reported increased disease severity in critically ill patients with insulin-treated diabetes mellitus relative to all other patients on ICU admission [
10]. However, this study is difficult to compare with ours, because it was not restricted to patients with sepsis and the comparator group consisted of both patients with diabetes mellitus not treated with insulin and patients without diabetes mellitus [
10]. As insulin can modify inflammatory and procoagulant responses independently of glucose levels [
12‐
15] we sought to compare the host response in insulin-treated versus non-insulin-treated patients with diabetes mellitus. Pre-existing insulin treatment did not affect any of the plasma host response biomarkers or the blood genomic response in our cohort, possibly due to the more profound alterations caused by the sepsis response in both groups.
Considering the established anti-inflammatory effects of metformin [
16,
17], we compared the presentation and outcome of sepsis and the host responses in metformin-treated and non-metformin-treated patients with diabetes mellitus. Metformin had no effect on the outcome or host response, despite the fact that patients with non-metformin-treated diabetes mellitus were admitted with significantly higher rates of shock. Metformin may decrease mortality in other patient populations. In patients with diabetes mellitus who underwent cardiac surgery, preadmission metformin use was associated with a decreased postoperative morbidity rate, including from infections, and with a substantial decrease in inpatient mortality [
40]. In a general population of medical and surgical ICU patients with type II diabetes mellitus, preadmission metformin use was associated with reduced 30-day mortality; however, this was not statistically significant in the subgroup of patients with sepsis [
41]. Blood transcriptome data obtained from patients with diabetes mellitus and sepsis, who were or were not treated with metformin suggests no influence of gluconeogenesis antagonism on systemic genomic responses.
Our study has strengths and limitations. It is the first to report on the potential influence of diabetes mellitus on the blood genomic response to sepsis and the first of its size to extensively study the association between treatments for diabetes mellitus and the host response to sepsis. Due to our prospective data collection we present a well-defined sepsis population with detailed clinical and laboratory information. As diabetes mellitus was registered based on medical history or the use of anti-diabetic medication, the possibility of previously unrecognized or new-onset diabetes mellitus cannot be excluded and could lead to misclassification of diabetic status. The type of diabetes mellitus was not registered and we cannot provide information on diabetes mellitus control prior to ICU admission, because HbA1c was not routinely measured and diabetes mellitus care is usually provided by general practitioners. In addition, as plasma protein biomarkers were measured at intervals, relative rapid changes in biomarker distribution could not be analyzed.
Abbreviations
APACHE-IV, Acute Physiology and Chronic health Evaluation IV; APS, Acute Physiology Score; aPTT, activated partial thromboplastin time; BH, Benjamini-Hochberg; BMI, body mass index; CCG, Cologne Center for Genomics; CRP, C-reactive protein; EDTA, ethylenediaminetetraacetic acid; HR, hazard ratio; ICU, intensive care unit; IL, interleukin; IPA, Ingenuity Pathway Analysis; IQR, interquartile ranges; MARS, Molecular Diagnosis and Risk Stratification of Sepsis; MMP-8, matrix metalloproteinase-8; PT, prothrombin time; RNA, ribonucleic acid; SD, standard deviation; sICAM-1, soluble intercellular adhesion molecule-1; SOFA, Sequential Organ Failure Assessment; TNF-α, tumor necrosis factor-α
Acknowledgements
The authors acknowledge all members of the MARS consortium of both study centers of this multicenter study for the participation in data collection. A list of participating investigators can be found in the
Appendix. In addition, we would like thank Marek Franitza and Mohammad R. Toliat (Cologne Center for Genomics (CCG), University of Cologne, Cologne, Germany) for performing the micro-arrays.