Identifying interactions between cytokine levels and clinical outcomes in influenza infection using a novel data-driven method for defining cytokine modules

Abstract Cytokines and chemokines are key signaling molecules of the immune system. Recent technological advances enable measurement of multiplexed cytokine profiles in blood as well as sites of infection such as the nasal and alveolar mucosa. These profiles can then be used to identify potential biomarkers of clinical outcomes following influenza infection, by comparing groups of individuals with different outcomes. Recently, several associations between single cytokines and mortality have been reported. However, single analyte association testing ignores context-dependent covariation in cytokine secretion and decreases statistical power due to multiple testing. We have developed a novel data-driven approach to analyze cytokine and chemokine profiles that uses unsupervised clustering to form functional modules. We applied this method to three cohorts of severely ill children infected with influenza. Innate profiles were assayed using a 42-plex Luminex panel at enrollment. Clinical outcomes were collected including hospitalization, septic shock, ALI/ARDS and death. Though many modules were consistent with the immunological literature, some of the modules included cytokines that have not previously been associated, generating novel hypotheses. Modules generated from blood were different than those generated at the site of infection. Furthermore, we found that the modules were significantly associated with clinical outcomes in all three cohorts, while individual cytokines were not statistically significant. Overall, our approach provides a novel method to characterize cytokine and chemokine profiles that both increases statistical power and provides novel hypotheses about context-specific innate signaling.