Host Gene Expression Prior to Emergency Paediatric Intensive Care Admission Predicts Severity and Shared Immune Dysfunction in an Inclusive Population of Critically Ill Children.

Abstract

Background: Identification of shared patterns of immune responses in diverse critical illnesses, "endotypes", may enable stratified treatment of immune dysfunction. We aimed to identify and characterise early shared immune responses with prognostic and possible future therapeutic value in children recruited to the Biomarkers of Acute Serious Illness in Children (BASIC) study. Methods: We studied early immune responses in 382 children sequentially recruited children, aged 0-16 years of age with an emergency admission for any cause to paediatric intensive care units in south-east England between 2014 and 2016. We sampled whole blood at retrieval and used Illumina Human-HT-12 version 4 Expression BeadChips (Illumina San Diego, CA, USA) for subsequent host genome-wide gene expression profiling. We used unsupervised k-means clustering of patient gene expression to derive BASIC endotypes and assign endotype membership to patients in the cohort. Following internal cross-validation we assessed the prognostic value of BASIC endotype membership against the primary outcome (ventilator-free days at day 30, VFD-30) and secondary outcomes (mortality; and mortality of ventilation >=30 days) in adjusted Cox proportional hazards models. We converted endotype membership to a linear quantitative score and characterised these early shared immune responses using digital cytometry (Cibersort, Stanford University, CA, USA) and gene set enrichment analysis (Broad Institute, MA, USA). We subsequently investigated the relationship of the BASIC endotypes derived from diverse childhood critical illness to sepsis response syndromes derived from adults with sepsis. Findings: We identified two robust clusters, termed BASIC endotype 1 (122,31.9%, children), and BASIC endotype 2 (260, 68.1%, children), that were present in children with diverse critical illnesses and across age groups. BASIC endotype 1 membership was associated with 4.1 (95% CI 2.0-6.2) days of increased length of mechanical ventilation (hazard ratio for extubation 0.59, 95% CI 0.44-0.80), and a non-significant association with mortality, including after adjustment with syndrome-specific severity scores (such as the Phoenix Sepsis Score), in comparison with BASIC endotype 2 (HR 1.39, 95% CI 0.30-6.41, p=0.7). Following imputation of immune cell subsets, BASIC endotype 1 membership was positively associated naive and resting memory CD4 T cells, while BASIC endotype 2 was associated with a predominant neutrophil response - unlike previously published work in adults with sepsis. Gene set enrichment analyses showed BASIC endotype 2 membership was associated with elevated gene sets associated with tumour necrosis factor (TNF) alpha, interferon gamma, interferon alpha, and interleukin 6/JAK/STAT pathway signaling in comparison to BASIC endotype 1. Interpretation: In summary, we show shared immune features with prognostic and possible future therapeutic significance across different causes of critical illnesses in children. These findings may enable stratified treatment of immune dysfunction in future clinical trials in critical illness.