Our study represents the first multicenter evaluation of the relationships among circulating cytokines/chemokines, innate immune function, and outcomes in children with critical illness resulting from influenza. We were fortunate to have the infrastructure for this study in place at the time of the 2009 H1N1 pandemic, allowing us to enroll patients with seasonal and 2009 H1N1 influenza infections. The serum cytokine/chemokine profiles of nonsurvivors were characterized by high levels of proinflammatory mediators; however, nonsurvivors were also markedly immunosuppressed with leukopenia and severe impairment of ex vivo TNF-α production capacity. In addition, we report an association between S. aureus bacterial coinfection, often fatal, and a state of severe immune suppression, which was not seen in patients coinfected with other bacteria.
Data from the United States (24
) and abroad (29
) indicate that children undergoing ICU care in the 2008–2009 influenza season tended to be older than previous seasons, had high likelihoods of having an underlying medical condition, and were likely to be treated with invasive mechanical ventilation. Our cohort was similar in these regards although it was enriched, over time, with children with severe disease (mechanical ventilation, shock) and those without comorbidities. Accordingly, ARDS and shock were highly prevalent in our study. In fact, all of the subjects treated with ECMO had circulatory failure in addition to hypoxemic respiratory failure. This may explain why the ECMO survival rate in this study was lower than that reported in other case series of ECMO use for respiratory failure in patients with 2009 H1N1 (32
Although 2009 H1N1 was the most common influenza strain in our cohort and was associated with mortality, it should be noted that our study represents a convenience sample of children admitted to selected pediatric referral centers. In addition, the enrollment strategy was adjusted midstudy to insure representation of subjects across demographic and illness severity types. There were undoubtedly critically ill children with milder 2009 H1N1 disease who were not included in our study, precluding further comment on 2009 H1N1-specific mortality risk relative to other virus strains. The relationships between S. aureus secondary bacterial infection (particularly with MRSA) and mortality were more highly significant.
The relationships between systemic cytokine levels and outcomes from pediatric influenza to date have been largely limited to survivors. For example, elevated systemic levels of IL-6, IL-12, and IFN-γ were found in a ten-patient group of critically ill children with seasonal influenza (all survivors) compared with noninfected control subjects (15
). The multicenter design of our study allowed for the sampling of eight nonsurvivors. Despite a rigorous α level of 0.0016 for our cytokine/chemokine comparisons, we were able to demonstrate statistically significant elevations of serum GM-CSF, IL-6, IL-8, IP-10, MCP-1, and MIP-1α in nonsurvivors. These represent proinflammatory mediators whose primary roles are to mobilize and activate innate immune cells (GM-CSF), activate the acute phase response (IL-6), or serve as chemokines (IL-8, IP-10, MCP-1, and MIP-1α). Mediators that demonstrated a trend toward higher levels in nonsurvivors (p
< 0.01) included counterregulatory, anti-inflammatory mediators such as IL-1 receptor antagonist and IL-10. These data are in agreement with the finding of hypercytokinemia in adult influenza nonsurvivors (2
Adult serum studies and cell culture experiments have suggested that 2009 H1N1 influenza may induce a more robust cytokine response compared with seasonal influenza with elevated levels of mediators such as IL-6, IL-8, IL-10, IP-10, and growth-related oncogene (35
). The serum cytokine profiles of children with 2009 H1N1 influenza have been reported in several small case series in recent years although these studies have focused on influenza survivors. Kim et al (37
) demonstrated higher serum levels of IL-6 and IL-10 in 26 tachypneic children with H1N1 compared with those without tachypnea, a finding reproduced in a mixed adult–pediatric study including nine children with severe 2009 H1N1 (38
). Takano et al (39
) showed higher serum levels of IL-1β, IL-2, IL-4, and MCP-1 in ten children with 2009 H1N1 who were mechanically ventilated compared with 11 who were not, whereas Ohta et al (40
) found higher levels of IL-5 and IL-6 in 15 children with severe 2009 H1N1 compared with children with uncomplicated disease. Our data are in agreement with these findings with elevations seen in the levels of many of the same mediators in children with 2009 H1N1. These findings may not be specific to 2009 H1N1, however, given that S. aureus
secondary infection was so common in our patients with 2009 H1N1 and the cytokine profile in subjects with S. aureus
was so similar.
Despite the fact that only two subjects were immunocompromised at baseline, nonsurvivors demonstrated marked impairment of innate immune function. Whole blood from immunocompetent hosts should produce robust amounts of the proinflammatory cytokine TNF-α on ex vivo stimulation with LPS, a finding we observed in our outpatient control population. Activation of the Toll-like receptor pathway is used, in this context, as a readout of innate immune cell responsiveness given the fact that TNF-α is produced quickly by circulating innate immune cells (predominately monocytes) after a LPS challenge. This allows for a 4-hr incubation time, which is favorable for study logistics. This bioassay for innate immune function has been used in a number of single-center studies of adult and pediatric critical illness with lower TNF-α production capacity being associated with mortality (5
). This study marks the first time that this assay has been used in a multicenter fashion, highlighting the feasibility of prospective, multicenter functional immune-monitoring studies.
Our critically ill children demonstrated a reduced capacity to produce TNF-α compared with outpatient control subjects, a finding that was expected as part of the compensatory anti-inflammatory response syndrome that often follows a proinflammatory insult. However, the degree of reduction in TNF-α production capacity that we observed in nonsurvivors was severe with a level <250 pg/mL being highly predictive of mortality by receiver operating characteristic curve analysis. This threshold compares favorably with the definition of immunoparalysis (200 pg/mL using this assay) that we and others have used to identify subjects at high risk of death in unrelated patient populations (5
). Severe leukopenia was very common in nonsurvivors in our cohort, and this undoubtedly contributed to the development of the immunoparalyzed phenotype that we observed. Indeed, an AMC <288 cells/mm3
was strongly associated with mortality. Twenty percent of survivors, however, demonstrated severely reduced AMC counts, and ex vivo TNF-α production capacity was better able to identify those at risk for death.
The contributors to innate immune suppression associated with critical illness are poorly understood and are thought to include anti-inflammatory cytokine production (9
) as well as genetic and epigenetic factors (41
). Recent evidence suggests that negative regulation of host innate and adaptive immunity may be important for influenza pathogenesis (3
). Treatment of mice with the immunostimulatory cytokine IFN-γ in the early stages of influenza infection has been shown to be protective (18
), and intact airway macrophage function is known to be important for limiting lung injury during influenza infection in mice (46
). Intriguing evidence suggests that influenza infection may specifically increase susceptibility to S. aureus
secondary infection in the lung (16
Subjects in our cohort with S. aureus
infection demonstrated a greater degree of immune suppression than those without secondary bacterial infection or infection with a different organism. S. aureus
is known to produce host immune evasion factors including those that dampen the innate and adaptive immune responses (47
), but an association between S. aureus
infection and the immunoparalyzed phenotype has not been previously reported. Similarly, it is unclear if the increased mortality risk associated with MRSA infection was related to MRSA-specific immunosuppression or to enhanced virulence of the isolates. We view these relationships to be highly deserving of future study.
An important aspect of this study is the demonstration of the feasibility of multicenter, provocative immune function testing. Studies of innate immune function in critical illness to date have relied on single-center reports that largely focus on monocyte HLA-DR expression and/or ex vivo TNF-α production capacity (5
). Monocyte HLA-DR measurement is problematic for multicenter studies in which it requires considerable on-site processing of fresh blood specimens, and the measurements can be difficult to standardize across flow cytometers and antibody lots. Provocative testing such as ex vivo LPS-induced cytokine production profiling poses different challenges. There is currently no single industry standard reagent for ex vivo stimulation or protocol for cytokine determination. Accordingly, it is difficult to determine thresholds of immune responsiveness that are associated with clinical outcomes when stimulation protocols differ between investigators.
We elected to measure ex vivo LPS-induced TNF-α production capacity using a protocol that required only basic laboratory equipment that is likely to be available in most tertiary care medical centers. The stimulation procedure was done on-site and required minimal technician time. The LPS stimulation reagents were produced and quality controlled at a single site and shipped monthly to participating centers, ensuring a ready supply of highly standardized, fresh reagents. Lastly, the resulting supernatants were shipped back to a central laboratory where TNF-α was quantified from all supernatants on the same highly automated, good laboratory practices instrument.
The identification of potential treatment thresholds is important because mounting evidence suggests that innate immune suppression associated with critical illness may be reversible. Several small single-center clinical trials have shown that systemic therapy with GM-CSF can improve monocyte HLA-DR expression and/or ex vivo TNF-α production capacity in critically ill adults and children with immunoparalysis with improvement in clinical outcome (5
). Administration of IFN-γ has shown promise in this population as well (13
). In these studies, augmentation of innate immune function occurred without worsening systemic inflammation. Although our nonsurvivors had higher serum levels of GM-CSF than survivors, these levels (median 52 pg/mL) were far below the peak serum concentrations seen in patients undergoing GM-CSF therapy for reconstitution of bone marrow after chemotherapy, which are typically >1000 pg/mL (54
). Some authors have advocated for immunosuppressive therapies in the setting of severe influenza, including glucocorticoids or peroxisome proliferator-activated receptor-γ agonists (19
). We suggest that patient-specific immune monitoring be incorporated into immunomodulatory trials to aid in treatment assignment. It is possible that immunostimulatory therapies such as GM-CSF targeting both leukopenia and innate immune cell responsiveness may have a role in the treatment of high-risk children with immunoparalysis associated with influenza infection.
Our study has several notable limitations. First, the patient population studied represents a convenience sample, which has been enriched for a higher severity of illness. This has allowed us to investigate the immunobiology of the highest risk subjects in our critically ill population but makes it difficult to compare the epidemiology of our cohort with that of a broader, less ill population. Second, despite the study’s multicenter design, we experienced a small number of deaths in our cohort. A larger study will be needed to more fully explore potentially important confounders including age, race, and socioeconomic status. Still, strong associations between innate immune function and mortality were seen. Also, complete blood count testing was not protocolized for this study, and 25% of survivors did not undergo white blood cell count measurement. Subsequent studies with closely timed complete blood counts and immune function measurements would be informative. It is quite likely that the interplay between innate and adaptive immune function, including relationships among helper, cytotoxic, and regulatory T cells, is important in children with influenza and should be the subject of future study as well. Indeed, mechanisms of immune suppression, including differences in cell populations and expression levels of pattern recognition receptors, were not evaluated in our study and are deserving of investigation. Lastly, this study involved sampling at only one time point. We limited our analyses to samples obtained within the first 72 hrs of ICU admission to identify early predictors of mortality. It is probable that more longitudinal measurements would yield additional information about risk factors for adverse outcomes from critical influenza.