Understanding the Inflammatory Cytokine Response in Pneumonia and Sepsis:  Results of the Genetic and Inflammatory Markers of Sepsis (GenIMS) Study FREE

Severe sepsis, a systemic response to infection complicated by acute organ dysfunction, develops in 900 000 people each year in the United States, a third of whom die.¹ Early models of sepsis using animals² and human volunteers injected with bacterial products³ demonstrated a short, intense elevation of proinflammatory cytokine levels. Initially intended to control and eliminate infection, the cytokine activation became widespread and was presumed to precipitate organ failure characteristic of the clinical syndrome.⁴ This theory is supported by the cytotoxic effects of various cytokines⁵ and has underpinned many large-scale studies of interventions intended to blunt the proinflammatory response in sepsis.^6- 21 Unfortunately, these trials generally failed.^22- 23 One explanation is that death from sepsis may follow an excessive counterregulatory anti-inflammatory response.²⁴ However, both theories may be oversimplified, arising from an incomplete understanding of the complex pathophysiologic nature of human sepsis.

Typically in medicine, large-scale interventional trials are initiated only after the natural history of the target condition is well understood. The sepsis field is unusual in that there are surprisingly limited data on inflammatory cytokine responses in humans with infection and severe sepsis. The few clinical studies conducted thus far have often yielded results inconsistent with each other or with those of laboratory experiments and human volunteer studies. These clinical studies were often small, single-center studies, which limited power and generalizability, and they often focused on subjects with established severe sepsis due to a myriad of infections. Also, comparators were often normal adults rather than infected but less ill subjects. These design limitations compromised the ability to study the timing of the response and to distinguish protective from harmful aspects.

We conducted a large, multicenter inception cohort study of subjects presenting to the emergency department (ED) with community-acquired pneumonia (CAP), the leading cause of severe sepsis.¹ Our goals were to describe the systemic cytokine response to infection and to determine if there were specific patterns associated with severe sepsis and death.

The community-acquired Genetic and Inflammatory Markers of Sepsis (GenIMS) study enrolled subjects at 28 academic and community hospitals in southwestern Pennsylvania, Connecticut, southern Michigan, and western Tennessee. We included patients 18 years or older with a clinical and radiologic diagnosis of pneumonia, per the criteria of Fine et al,²⁵ namely, one or more symptoms suggestive of pneumonia and radiographic evidence of pneumonia within 24 hours of presentation. Exclusion criteria included transfer from another hospital; discharge from a hospital within the prior 10 days; an episode of pneumonia within the prior 30 days; chronic mechanical ventilation, cystic fibrosis, or active pulmonary tuberculosis; admission for palliative care; previous enrollment in the study; incarceration; and/or pregnancy. Participants or their proxies provided written consent. We obtained approval from the institutional review boards of the University of Pittsburgh and all participating sites.

All subjects were enrolled while still in the ED, if possible, or as soon after admission as could be accomplished. After enrollment, we gathered detailed baseline and sequential clinical information using structured subject or proxy interviews, bedside assessments, and medical record abstraction. We obtained blood for cytokine assays immediately following enrollment, daily for the first week, and weekly thereafter while subjects remained in hospital. We determined survival after discharge by telephone and a National Death Index search.

We tracked clinical data and blood samples using unique anonymized identification numbers, merging data only after assays had been completed. We observed strict data confidentiality and audited clinical data and assays for accuracy, including random chart audits, additional blood assays, and computer flags for inconsistencies.

We measured circulating levels of tumor necrosis factor (TNF) and IL-6 (interleukin 6) as markers of the proinflammatory and IL-10 as a marker of the anti-inflammatory cytokine response. Generally, day 1 blood samples were drawn at enrollment (day of ED presentation), and subsequent samples were drawn at 8 AM. We did not obtain day 1 samples from subjects presenting after 11 PM or on weekends or holidays for logistic reasons. Each blood sample was drawn into pyrogen-free vials containing heparin and within 1 hour the plasma was separated by centrifugation. The sample was then divided into four 1.5-mL tubes, frozen at −80°C, and batched and shipped on dry ice. We measured cytokine concentrations by chemiluminescent immunoassay using an automated analyzer (IMMULITE; Diagnostic Products Corp, Los Angeles, California), thawing samples only once before assay. We measured IL-6 and IL-10 levels in all samples. We measured TNF levels on day 1 samples for all subjects and on all samples for a subset (the first 1190 subjects).

Subjects were assumed to have pneumonia if they met entry criteria and the primary diagnosis was not revised by the treating physicians in the subsequent 3 days. We defined severe sepsis as pneumonia plus acute organ dysfunction following the 2001 International Consensus Criteria.²⁶ We defined acute organ dysfunction as a new sepsis-related Organ Failure Assessment (SOFA)²⁷ score of 3 or higher in any of 6 organ systems, based on the recent international Sepsis Occurrence in the Acutely ill Patient (SOAP) study.²⁸ We used 90-day mortality as our primary measure of survival, based on end-point recommendations for sepsis trials from 2 recent international expert panels.^29- 30 We defined systemic inflammatory response syndrome (SIRS) criteria following the 1992 American College of Chest Physicians/Society of Critical Care Medicine Sepsis Definitions Consensus criteria.³¹

Statistical analyses were performed using SAS software, version 9.1 (SAS Institute, Cary, NC), with statistical significance set at P<.05. We assumed a log-normal distribution for cytokine concentrations and analyzed data in natural log scale. We constructed Tobit models to account for data that were truncated because they fell below detection thresholds (ranging from 27% to 71% of data).^32- 33 For data from single time points (eg, TNF concentrations on day 1), we used Tobit models to generate daily means and to compare the cytokine concentrations between groups.

For sequential data (eg, serial IL-6 measurements), we conducted regression analysis with mixed models that accounted for correlation of repeated measures over time,³⁴ incorporating Tobit models as necessary.³³ Models included linear, quadratic, and cubic terms to allow evaluation of trends. We determined differences across outcome groups by testing the significance of the regression coefficient in the models. We adjusted for potential confounders by expanding the models to include age, sex, race, and Charlson Comorbidity Index³⁵ as covariates. We also assessed the frequencies with which cytokine concentrations were elevated using the following upper limits of normal: IL-6, 5.9 pg/mL; IL-10, 9.1 pg/mL; and TNF, 8.1 pg/mL, per the manufacturer's specifications. We compared differences in the proportion of subjects with elevated concentrations using logistic regression based on generalized estimating equations.³⁶

To explore whether discrete cytokine patterns existed, we used a group-based longitudinal analysis of IL-6 and IL-10 data to determine the likely number of informative subgroupings, the pattern of profiles in each subgroup, and the probability of subgroup membership for each individual.³⁷ This approach handles censored data directly.³⁸ For each cytokine, we generated 1, 2, 3, and 4 subgroup models, allowing different relationships (linear, quadratic, or cubic) for each subgroup, and selecting the best-fitting model based on the Bayesian Information Criterion.³⁷ We determined the frequency of different combinations of patterns for the 2 cytokines and, with Cox proportional hazards models, the potential association of these combinations with survival.

We enrolled 2320 subjects, of whom 291 (13%) were discharged from the ED. Once admitted, 134 (6%) were excluded because their treating physicians subsequently revised their primary diagnosis, and 9 (0.4%) had inadequate blood specimens. The remaining 1886 subjects comprised the inpatient CAP cohort on whom we conducted our analyses (Figure 1 and Table 1). In this cohort, 583 subjects (31%) developed severe sepsis. In 47% of those developing severe sepsis (n = 274), criteria were met on day 1; in 85% (n = 496), criteria were met by day 4. The mortality rate rose steeply and was significantly higher for subjects who developed severe sepsis than for those who did not (P<.001) (Figure 2).

We obtained blood for assay of cytokine concentrations on ED presentation from 1429 of the 1886 subjects (76%) (Table 2). Not surprisingly, mean cytokine concentrations were elevated, although 17%, 64%, and 63% of subjects had normal concentrations of IL-6, TNF, and IL-10, respectively. For all 3 cytokines, day 1 concentrations were higher in those with severe sepsis than in those without severe sepsis (P<.001) and in nonsurvivors than in survivors (P = .01). In addition, among the 1221 subjects who presented without severe sepsis (85%), IL-6 concentrations were higher for those who subsequently developed severe sepsis than for those who did not (P = .03). Similarly, among subjects who developed severe sepsis at any point, nonsurvivors had higher concentrations of IL-6 and IL-10 than survivors (P = .01). Day 1 cytokine concentrations were also available on 285 of the 291 ED discharges (98%) and 101 of the 134 subjects (75%) who were excluded for having a change in primary diagnosis. The differences in day 1 concentrations between survivors and nonsurvivors remained significant when these additional subjects were included in the analysis (P<.001).

Cytokine concentrations over time are presented in Figure 3. The mean IL-6 concentration fell rapidly from day 1 over the subsequent 2 days but remained elevated throughout the first week. Mean IL-6 concentrations (Figure 3A, top) and the proportion of subjects with elevated concentrations (Figure 3B, top) were higher for those who developed severe sepsis (survivors and nonsurvivors) compared with those who did not (P<.001 and P<.001) and for those who died following severe sepsis compared with those who survived (P<.003). Of note, although mean levels were elevated, a number of subjects had normal concentrations on any given day and 12% of survivors (n = 198) and 4% of nonsurvivors (n = 8) had normal levels throughout.

The mean TNF concentrations were generally lower than those observed for IL-6, with a large number of subjects displaying normal levels (53% of nonsurvivors and 61% of survivors). Otherwise, the pattern was similar to that seen with IL-6, where levels fell over the first 2 days and remained mildly but persistently elevated thereafter, with higher levels seen among those with severe sepsis (survivors and nonsurvivors) vs those without severe sepsis (P = .01 and P<.001).

Concentrations of IL-10 were also lower than those observed for IL-6, with more noticeable decay over the first 2 days. A very high proportion of subjects had normal concentrations (64% of survivors and 42% of nonsurvivors had normal levels throughout), but higher levels were associated with survivors with severe sepsis compared with survivors without severe sepsis (P<.001) and nonsurvivors (P<.001) and for nonsurvivors with sepsis vs survivors with sepsis (P = .002).

For all 3 cytokines, differences between groups remained significant after adjusting for differences in baseline patient characteristics. Restricting analysis to the subset of subjects who developed severe sepsis after day 1 (n = 309), no rise in concentrations or spike occurred for any cytokine immediately prior to or on the day that severe sepsis developed. Of note, although cytokine measurements were often normal, only 4 of the 149 fatal cases of severe sepsis (2.7%) exhibited no cytokine concentration elevation. On the other hand, some amount of systemic cytokine concentration elevation was also present in those who fared well. For example, in the 364 of the 1886 patients who survived (19%) with virtually no organ dysfunction (sum of worst SOFA score for all organs <1), 79% had elevated cytokine levels on day 1, and 74% after day 1 (days 2-30).

Cytokine levels remained elevated much longer than clinical signs. For example SIRS criteria resolved quickly, with fewer than 50% of subjects having more than 2 SIRS criteria after day 2, whereas more than 50% of subjects continued to have elevated IL-6 levels by day 7.

Trajectory analyses suggested very distinct cytokine patterns. For both IL-6 and IL-10, 3-group models (high, medium, and low concentrations) with a quadratic trend for each group exhibited the best fit compared with all other combinations of 1-, 2-, or 4-group models with quadratic, cubic, or linear trends (Figure 4A). For IL-6, the mean day 1 concentration of the high group was 50-fold higher than that of the low group. Model selection and group membership were similar when the analyses were repeated with inclusion of baseline characteristics as covariates.

Figure 4B shows the distribution and outcomes of the cohort across the combined IL-6 and IL-10 concentration patterns. Of the 9 possible combinations of concentration, 3 accounted for more than 75% of subjects: medium IL-6/low IL-10, 36% (n = 672); low IL-6/low IL-10, n = 492 (26%); and medium IL-6/medium IL-10, n = 259 (14%). Both severe sepsis and mortality rates varied by combination, with the highest rates observed for combinations of higher cytokine expression (Table 3 and Figure 4B). These findings persisted in Cox proportional hazards models (Table 4 and Figure 5) that also controlled for age, sex, race, and comorbidity. Patterns of relative overexpression or underexpression of IL-6 with respect to IL-10 were uncommon and not obviously associated with very high or very low adverse outcomes.

We found that several of the past interpretations of the inflammatory response to sepsis—at least in patients with CAP—may be incorrect or misleading, with important implications for the development of future therapies. First, although systemic cytokine activation was common, it was not universal. Second, although we initiated our study at the time of presentation in the ED, cytokine concentrations had already peaked, and very few subjects had markedly elevated levels later, suggesting the classic cytokine cascade was already fully activated by the time patients sought hospital care. Third, although cytokines may be released secondary to organ damage or dysfunction,^40- 41 we did not see an increase in cytokine concentrations with the onset of organ dysfunction. Fourth, although concentrations were higher in those who fared worse, differences between groups with different outcomes were modest. Furthermore, systemic activation was a prominent feature in many subjects who mounted a successful response to infection, suggesting such systemic activation may be neither overly exuberant nor necessarily deleterious.

Fifth, for many subjects, cytokine activation persisted throughout the hospital course, far longer than traditional models would suggest, and far longer than the typical courses of anticytokine strategies in prior negative trials. For example, Halm et al⁴² showed that clinical signs of sepsis (tachycardia, tachypnea, and fever) resolve within 2 to 3 days after admission for CAP, and in both failed anti-TNF antibody studies,^6,9 a drug was administered for only 3 days. The duration of cytokine activation was also longer than clinical signs such as SIRS criteria. Finally, although different subjects had different patterns of cytokine response, these patterns could best be described as high, medium, and low generalized activation. Immunologic dissonance, defined as a preponderance of either proinflammatory or anti-inflammatory cytokine activation,⁴³ was neither common nor obviously associated with poor outcome.

Although our findings are somewhat discordant with conventional thinking, they reinforce the findings of 3 small single-center studies of serial cytokine concentrations in severe sepsis.^44- 46 These studies reported modest cytokine elevations with an absence of a spike and continued activation over several days. More recently, Kinasewitz et al⁴⁷ reported the cytokine patterns in subjects enrolled in the PROWESS trial (Recombinant Human Activated Protein C Worldwide Evaluation in Severe Sepsis).⁴⁷ These data were also similar, with low levels of IL-10 and persistent, slowly decaying IL-6 concentrations.

The current effort to delineate the natural history of sepsis provides insight into the potential reasons for failure in prior anticytokine strategies and guidance for future therapies. First, treatments designed to manipulate early cytokine activation are at best impractical, especially for community-acquired sepsis, since the classic cascade seen in human volunteer and animal studies either does not occur or occurs before presentation. Second, a “1 size fits all” anticytokine approach is likely ineffective and potentially dangerous because not all patients mount a cytokine response, and many patients who mount a cytokine response fare well. Third, the highest risk of death was seen in patients who expressed high levels of both proinflammatory and anti-inflammatory cytokines, suggesting that therapy targeted at individual cytokines might not be useful. Finally, typical dosing over 2 or 3 days may be too short, since cytokine activation is often persistent. It is notable that successful trials of corticosteroids for CAP⁴⁸ and septic shock⁴⁹ used a 1-week course.

There are important caveats and limitations to our work. First, limiting our study to CAP reduced unwanted heterogeneity but may affect the generalizability of our findings. Extrapolating our findings to severe sepsis arising from other types of infection is speculative. Second, recruiting a large number of patients from multiple centers provided power to explore different patterns but placed logistic constraints on our blood sampling frame. It was impractical to draw blood samples more frequently than daily, which limited our ability to explore whether more time-sensitive patterns existed. It was also impractical to draw blood samples after hospital discharge. However, we did not find evidence of discharge-related informative censoring affecting our estimates of the serial patterns. Third, we relied on statistical inference to determine the presence of discrete cytokine patterns, which necessarily is limited by power to find rare patterns. Fourth, we relied on the circulating concentrations of just 3 cytokines as biomarkers of the innate immune response. There may well be important patterns at the local or tissue level, within the cellular component of the immune response, or within other acute response cascades, which are not reflected by changes in these biomarkers. Furthermore, these cytokines themselves have multiple, context-specific effects and are not easily classified as proinflammatory or anti-inflammatory.

In conclusion, we have, for the first time to our knowledge, described the systemic cytokine response to infection and severe sepsis in a large, multicenter inception cohort of subjects presenting to the ED with CAP. Our results show that this response is much longer in duration and much more heterogeneous than suggested by previous studies. Nevertheless, individuals with high circulating levels of both proinflammatory and anti-inflammatory cytokines had a markedly increased risk of severe sepsis and death.

Correspondence: Derek C. Angus, MD, MPH, The CRISMA Laboratory, Department of Critical Care Medicine, University of Pittsburgh School of Medicine, 604 Scaife Hall, 3550 Terrace St, Pittsburgh, PA15261 (angusdc@upmc.edu).

Accepted for Publication: March 18, 2007.

Author Contributions:Study concept and design: Kellum, Fink, Weissfeld, Yealy, Pinsky, Fine, Krichevsky, Delude, and Angus. Acquisition of data: Yealy, Fine, and Krichevsky. Analysis and interpretation of data: Kellum, Kong, Weissfeld, Pinsky, Fine, and Delude. Drafting of the manuscript: Kellum, Kong, Weissfeld, Yealy, Krichevsky, and Angus. Critical revision of the manuscript for important intellectual content: Kellum, Kong, Fink, Yealy, Pinsky, Fine, Krichevsky, Delude, and Angus. Statistical analysis: Kong and Weissfeld. Obtained funding: Kellum, Fink, Pinsky, Krichevsky, Delude, and Angus. Administrative, technical, and material support: Yealy, Fine, Krichevsky, Delude, Angus. Study supervision: Yealy, Fine, Krichevsky, Delude, Angus.

Financial Disclosure: None reported.

Funding/Support: The GenIMS study was funded by National Institute of General Medical Sciences, National Institutes of Health grant R01 GM61992 with additional support from GlaxoSmithKline for enrollment and clinical data collection and Diagnostic Products Corporation for the cytokine assays.

Group Information: A list of the GenIMS Investigators was published in Crit Care Med. 2007;35(4):1061-1067.

Role of the Sponsor: None reported.

Additional Contributions: We are indebted to the nurses, respiratory therapists, phlebotomists, physicians, and other health care professionals who participated in GenIMS as well as the patients and their families who supported this trial.