(1055) Multi-Cohort Analysis Identifies Consensus Sepsis Clusters With Prognostic and Therapeutic Relevance

Introduction: Several transcriptomic signatures for endotyping sepsis patients have been proposed, but none have been translated into clinic. A significant barrier to their implementation is a need for consensus among these signatures. We investigated whether these signatures identify overlapping endotypes with prognostic and therapeutic relevance that could accelerate clinical translation.

Methods: Our research was made possible by the SUBSPACE consortium, an international collaboration of sepsis investigators who shared whole blood bulk transcriptome data. We co-normalized 2,985 samples across 8 SUBSPACE sepsis cohorts, applied seven sepsis transcriptomic signatures, and identified consensus clusters using a cluster greedy forward algorithm and community detection using Spearman correlation. We integrated 171 samples (474,994 cells) from 3 sepsis peripheral blood single-cell RNA-seq datasets inclusive of the neutrophil compartment to identify the cellular origin of these consensus clusters. We then applied unified signatures to determine their association with severity and response to steroids and anakinra.

Results: Our analysis revealed a consensus among all seven transcriptomic signatures, leading to the identification of four distinct clusters. These clusters were found to separate along myeloid and lymphoid lineages and were termed myeloid protective, myeloid detrimental, lymphoid protective, and lymphoid detrimental. We further derived myeloid and lymphoid dysregulation scores by combining these signatures. Detrimental scores were associated with higher rates of ICU admission and 30-day mortality. Notably, high lymphoid protective scores were linked to increased mortality with steroid treatment in septic shock, while myeloid detrimental scores were associated with reduced mortality with anakinra in COVID-19.

Conclusions: Our findings strongly suggest that published sepsis gene signatures share a common underlying molecular pattern, leading to the identification of four reproducible consensus molecular clusters. These clusters are not only associated with differential severity but also with treatment response to steroids and anakinra. This has significant implications for patient care, as it could guide the use of these treatments based on the patient's molecular profile.