Background
Sepsis is a leading cause of death among critically ill patients [
1]. The current paradigm of sepsis pathophysiology suggests that some patients will die early in the disease course as a result of a dysregulated proinflammatory phase (systemic inflammatory response syndrome [SIRS]). However, many more patients will die later of subsequent insults during a predominating immunosuppressive recovery phase (compensatory anti-inflammatory response syndrome [CARS]) [
2]. The SIRS response is characterized by high levels of interleukin-1β (IL-1β), tumor necrosis factor-α (TNF-α), and expression of human leukocyte antigen (HLA)-DR on monocytes. In contrast, anti-inflammatory IL-10 and reduced expression of HLA-DR on monocytes are characteristics of the CARS response.
Recently, investigators have observed that certain septic patients never mount a clinically evident SIRS response, suggesting that the presence of SIRS should not be considered an essential component of the sepsis clinical syndrome [
3]. The reason that some patients with infections and associated organ failure do not exhibit a SIRS response may be that preexisting chronic conditions influence the nature of the SIRS and CARS responses during the course of infection [
4]. Studies of the septic immune response often exclude patients with disorders of the immune system or patients undergoing treatments that reduce immunity [
5‐
8]. As a result, the septic immune response is not well characterized for at least 25% of patients with sepsis [
9]. Moreover, any immune response identified as beneficial or harmful in these studies may not be generalizable, owing to the exclusion of patients with immunocompromising comorbidities.
One of the main pathogens responsible for sepsis is
Staphylococcus aureus. Prior studies of immune dysregulation in sepsis due to
S. aureus infection were focused on the innate immune response, particularly TNF-α and IL-10 [
10,
11]. These cytokines, along with IL-6, have long been known to be prognostic in different populations with sepsis. However, numerous attempts to block these cytokines have not led to improved outcomes [
12]. Notably, monocytes isolated from septic patients that express higher levels of HLA-DR also produce more TNF-α after lipopolysaccharide stimulation [
13]. In addition, polymorphisms in HLA-DR increase risk for
S. aureus susceptibility, indicating that antigen presentation by monocytes to T cells may be a critical driver of disease outcome [
14]. However, little is known about the resulting T-cell responses in patients with
S. aureus bacteremia. One recent study indicated that T-helper type 1 (Th1) cells are expanded in immunocompetent patients and are protective in a murine model of infection [
15]. In murine models, T-cell polarization during
S. aureus infection is dependent on the route of administration, the dose of bacteria delivered, and postinfection day of analysis, making these studies difficult to translate to patients with naturally acquired bacteremia [
16‐
21]. Interestingly, whereas Th17 cells have been implicated in protection in murine models of
S. aureus infection, the role of Th17 cells in humans has not been established. Finally, the presence of
S. aureus-specific antibodies does not correlate with resistance to infection [
22]. Therefore, a protective memory T-cell response may be critical in patients with sepsis by maintaining a proper SIRS/CARS balance.
In the present study, we focused on the associations between differentially polarized immune responses after
S. aureus bloodstream infection and 90-day mortality in a prospectively enrolled patient cohort. This approach allowed for evaluation of the host immune response independent of pathogen or pathogen-associated molecular pattern variability. We previously found that over 30% of patients with
S. aureus bacteremia had either immunocompromising hematopoietic disorders (hematologic malignancies or human immunodeficiency virus [HIV] infection) or received medications that interfered with immune function for the management of solid malignancies, solid organ transplants, or rheumatologic conditions [
23]. Thus, our secondary goal was to explore whether the immune responses associated with mortality after
S. aureus bacteremia occurred more or less frequently among these patient groups that are often excluded from studies of the septic immune response.
Methods
Sample collection
This study was conducted at the University of Chicago Medical Center (Chicago, IL, USA), a 547-bed, university-affiliated urban teaching hospital, between July 1, 2013, and October 24, 2014. The University of Chicago Institutional Review Board approved this study. All adult inpatients with at least one positive blood culture for S. aureus within the previous 4 days were approached for participation. Informed consent was obtained by the patients or their surrogates.
Blood samples were drawn into ethylenediaminetetraacetic acid tubes at three distinct time points (2–4 days, 6–9 days, and 12–18 days after the first day of positive
S. aureus blood cultures). It was necessary to include day ranges within time points, because there was variability in the time it took blood cultures to turn positive and we were unable to process samples every day of the week. The first two time points were chosen on the basis of previous sepsis biomarker studies [
24,
25], and we added a third time point to strengthen our analysis. Within 2 hours of collection, plasma and peripheral blood mononuclear cells (PBMCs) were isolated via differential centrifugation over Histopaque-1077 (Sigma-Aldrich, St. Louis, MO, USA). The samples were cryopreserved until analysis. Other investigators have reported consistent measurements when performing flow cytometry on fresh vs. thawed cells [
26,
27]. Clinical and laboratory data (including complete blood count with differential) were abstracted from the patients’ medical records.
Cytokine analysis
A multiplex assay (EMD Millipore, Darmstadt, Germany) was used to determine plasma cytokine concentrations at the first two time points (days 2–4 and 6–9). Cytokine levels that were below the limit of detection were assigned the lowest extrapolated value for each cytokine. Cytokine concentrations were log-transformed and were treated as continuous variables.
Flow cytometry
Thawed PBMCs were washed twice in fluorescence-activated cell sorting buffer (PBS containing 0.1% sodium azide and 0.2% bovine serum albumin) and incubated with a viability dye for 15 minutes (Zombie Aqua, BioLegend, San Diego, CA, USA). Cells were incubated for 10 minutes with pooled human immunoglobulin G to block nonspecific antibody binding (FcX solution; BioLegend), and surface staining was performed using fluorescently conjugated antibodies CD3-fluorescein isothiocyanate, CD4-allophycocyanin (APC), CCR4-phycoerythrin (PE), CCR6-Brilliant Violet 605, CD45RO-Brilliant Violet 711, CD25-APC/cyanine 7 (Cy7), and CD127-PE/Cy7 (BioLegend). Flow cytometry data were acquired on an LSRFortessa (BD Biosciences, San Jose, CA, USA) and analyzed with FlowJo software (FlowJo, Ashland, OR, USA).
Statistical analysis
The primary outcome was death during the first 90 days after bacteremia, which was determined by reviewing the medical record or contacting the patient’s surrogate. Our planned enrollment was 90 patients, of which we expected at least 20 to die during follow-up. On the basis of this enrollment target, our study was powered to include at least two immune markers in a multivariable model. In our secondary analysis, we explored whether the immune responses associated with mortality occurred more or less frequently among the patients with immunocompromising conditions or taking immunosuppressive medications.
The
t test, Mann-Whitney
U test, χ
2 test, or Fisher’s exact test was used in bivariate testing, as appropriate, to determine if differences between groups were significant. In our primary analysis, we determined the associations between immune marker levels and 90-day mortality using Cox proportional hazards modeling. In multivariable models, we removed variables that were not statistically significant using backward selection (
p > 0.05). In our secondary analysis, where outcomes were trends in cellular immune markers over time, we used linear mixed models to account for correlated measurements from the same patient [
28]. We determined that a linear model approximated these relationships because an additional “time after infection squared” term did not reach statistical significance (
p > 0.05). All tests were two-sided. All analyses were performed with STATA 13.1 software (StataCorp, College Station, TX, USA).
Discussion
In this study of immune responses to S. aureus bloodstream infections, we found that two immune marker ratios were associated with increased risk of death. Interestingly, high neutrophil counts and Th17 cytokine responses were not associated with survival in isolation. Rather, neutrophil counts and Th17 cytokine responses were associated with survival only when they were high relative to lymphocyte counts and Th1 cytokine responses, respectively. The average values for these immune marker ratios were similar for patients with and without immunocompromising conditions, owing to the fact that the presence of an immunocompromising condition was associated with proportionately lower neutrophil counts, lymphocyte counts, Th17 cytokine scores, and Th1 cytokine scores.
Patients who are immunocompromised because they have hematopoietic conditions or because they require treatment with immunosuppressive medications are typically viewed as having increased risk of death from sepsis. However, patients with these conditions in our cohort did not have increased 90-day mortality, confirming the results of a prior chart review [
23] . The reason for this finding may be that immunocompromising conditions may have both beneficial and harmful effects on the host’s immune response to
S. aureus. For instance, such a condition may attenuate a harmful proinflammatory neutrophil or Th17 response. However, being immunocompromised was associated with a lower lymphocyte and Th1 response, perhaps compromising the ability of the patient to clear the infection or increasing the risk of a subsequent infection.
A strength of our study is that we avoided introducing heterogeneity with respect to causative pathogens. We included only patients with a definite infection due to a single pathogen that typically causes systemic inflammation and severe acute illness that can result in the development of sepsis. Because sepsis is defined by a clinical response to a presumed infection, it is difficult to identify prognostic immune biomarkers when the pathogens in a study are heterogeneous and some patients may not even be infected [
4]. In fact, prior investigators have shown that the type of immune dysregulation in sepsis depends on the type of pathogen [
36,
37].
Th17 responses have been considered protective during staphylococcal infections, based largely on preclinical mouse studies, as well as the observation that patients with inborn errors in Th17 cells are prone to invasive staphylococcal infections [
19]. Although our findings that Th17 responses were associated with increased mortality may seem counterintuitive, we note that patients who received the Merck V710 staphylococcal vaccine had markedly
increased mortality due to
S. aureus infection compared with placebo recipients, suggesting that the vaccination strategy generated a lethal immune response [
38]. In light of our data, we hypothesize that the optimal Th17 response must be adequate to clear pathogen but not so overwhelming as to precipitate inflammatory damage and death to the host. In addition, the Th17 response must be considered relative to other elements of the adaptive immune response, including Th1, Th2, and Treg responses, all of which may be protective.
The observation that Th2 responses were more likely to be elevated among survivors than non-survivors is interesting, especially in light of recent work indicating an association between type 2/Th2 diseases (such as asthma and allergy) and development of sepsis or sepsis outcomes [
39,
40]. These epidemiologic studies suggest that the presence of allergic diseases or asthma reduces the risk for developing sepsis as well as sepsis-associated mortality. We do not have information about allergic comorbidities among these patients. However, given that Th2 responses evolved to facilitate tissue repair and can attenuate proinflammatory Th17 or Th1 responses, the presence of a preexisting Th2 disease may facilitate the resolution of the inflammatory insult.
Tregs provide a crucial check on the early proinflammatory response, but their presence is thought to contribute to the vulnerability of patients to subsequent infections in the later phase of sepsis. When Treg percentage was analyzed using multivariable analysis, adjusting for Th2 and Th17 responses, a high early Treg percentage was associated with increased mortality. During the course of the infection, an increasing Treg response relative to Th17 response was associated with decreased mortality. These results indicate that consideration of all elements of the adaptive immune response for a given infection provides a more accurate indication of the immune dysregulation of sepsis associated with death.
Our study has several limitations: First, our study was limited by its small size. The cytokine aspect of our study was powered to include two variables in a Cox proportional hazards model. We did not perform a sample size calculation prior to our flow cytometry experiment, because we explored infrequently studied immune markers in a group of patients that is typically excluded from sepsis biomarker research. Second, there is no universally accepted definition of an immunocompromised patient [
9,
41‐
43]. Because there is a wide range of types and degrees of severity of immune impairment among patients who are commonly classified as immunocompromised, we categorized patients who had hematopoietic disorders separately from those who were taking immune-altering medications. Third, our evaluation of how immunocompromising conditions affect the immune response to
S. aureus requires confirmation in future adequately powered studies. There were fewer than 10 patients with hematologic malignancies or HIV infections, and none of them died within 90 days of infection. As a result, we did not include any of these patients in the flow cytometry portion of our study. Finally, although we studied infections of the same pathogen type, there were likely genetic differences among
S. aureus organisms that we did not account for and that could have affected the immune response and risk of death. The majority of the
S. aureus infections in our study were community-acquired and methicillin-sensitive; neither of these characteristics was associated with mortality.
Acknowledgements
Kelly Blaine and Ryan Duggan provided technical assistance by processing blood samples and performing cytokine/flow cytometric analysis. JAG was supported by T32 HL007605, and UL1 TR000430, PAV was supported by K08 HL132109, and Clinical and Translational Science Pilot Award, University of Chicago.