Endotype Discovery in Acute Respiratory Distress Syndrome Dr Romit Samanta Acute respiratory distress syndrome (ARDS) affects 10% of critical care patients and is characterised by acute refractory hypoxaemia and bilateral pulmonary infiltrates on thoracic imaging. Mortality from severe ARDS is approximately 40%, and has not changed in 50 years despite decades of study. Randomised controlled trials of therapies for ARDS have been unsuccessful due to the heterogeneity of the patient population. This has led repeatedly to potentially promising therapies being discarded. The primary reason for the failure is that the underlying biological processes occurring in ARDS are poorly understood. This thesis attempts to address this heterogeneity, and explores the underlying biology by using an integrated, unsupervised bioinformatics approach to describe different mechanistic subtypes (endotypes) of ARDS. The endotypes described here are derived from analysis of data collected by three UK-based studies: an observational study of sepsis (GAinS), an observational study of severe influenza (MOSAIC), and a randomised controlled trial of simvastatin in ARDS (HARP-2). A combination of automated clustering methods and network analysis tools have been used to integrate blood biomarkers and gene expression (transcriptomic) data to define distinct endotypes of ARDS. Three endotypes of ARDS were identified in each of the studies. Integration of protein biomarker and transcriptomic data from patients recruited to the GAinS study identified three endotypes, one of which was characterised by severely dysregulated cytokine release, which we termed hyper-inflammatory. Two gene modules discriminated these patients from a hypo-inflammatory endotype, consisting of patients with globally depressed cytokine concentrations. Enrichment of the genes in these two modules identified genes that were important in vesicle fusion and cytokine release. Mutations in these genes cause the familial type of haemophagocytic lymphohistiocytosis (HLH). The implication here is that these genes played a role in the severely dysregulated cytokine concentrations observed within patients with hyper-inflammatory, sepsis-associated ARDS. Analysis of the cytokines and transcriptomic data collected during the MOSAIC study identified three endotypes we named: adaptive, endothelial leak and neutrophil driven. The endothelial leak endotype was characterised by enrichment of genes associated with SLIT- ROBO signalling. SLIT-ROBO signalling is essential for maintaining pulmonary endothelial integrity and failure of this mechanism has been shown to cause alveolar oedema in murine models of sepsis and influenza infection. These patients had significantly lower albumin levels than the adaptive endotype, and 48.5% of them required mechanical ventilation. Despite the greater need for mechanical ventilation, the outcomes of these patients were similar to those of patients with the adaptive endotype, of whom only 20.4% required mechanical ventilation. Cluster analysis of patient biomarker concentrations from the HARP-2 study identified three endotypes. Two of these endotypes had elevated serum IL-6 and sTNFR-1 concentrations, consistent with a hyper-inflammatory profile. Patients with one of the hyper-inflammatory endotypes, which we termed MMP-8 dominant, demonstrated a strong therapeutic response to simvastatin compared with placebo (28-day survival, adjusted HR = 0.35, 95% CI 0.18- 0.71; p = 0.003). Patients with this endotype, who received simvastatin, had a similar 28-day survival profile to patients with a hypo-inflammatory endotype, characterised by globally depressed biomarker levels. Patients with the other hyper-inflammatory endotype, which we termed sRAGE dominant, did not show any therapeutic response to simvastatin. The endotypes described are temporally stable, and some relate to novel mechanisms not previously recognised in patients with ARDS. The endotypes are all biologically plausible, amenable to the development of further mechanistic insights using laboratory-based tech- niques, and may influence patient outcomes and response to treatments. Further development and prospective validation of these endotypes are required. If validated, they may offer the opportunity to stratify patients in future clinical trials to treatments that are more likely to improve their outcomes, whilst avoiding treatments that might cause adverse effects. The methods described in this thesis could be applied to other heterogeneous and poorly understood clinical syndromes.