Deciphering sepsis molecular subtypes using large-scale data to identify subtype-specific drug repurposing.

Sepsis is a life-threatening dysregulated response to infection, the heterogeneity of which precludes effective precision medicine. Deciphering this heterogeneity requires substantially larger molecular datasets of sepsis patients. Here, we create a transcriptomic atlas of publicly available adult sepsis data. In total, we harmonized data from 3,713 samples across 28 datasets, including 2,251 unique sepsis patients. On this data, we performed molecular subtyping and identified potential subtype-specific drug repurposing opportunities. We identified 4 molecular subtypes of sepsis (C1 - C4) based on expression differences in immune- and lipid-related genes. Pathway analysis of gene signatures unique to each molecular subtype revealed patterns of immune exhaustion and metabolic dysregulation in C1, suggesting potential benefit from corticosteroid treatment. C2 expression patterns were often anti-correlated with those of C1, and C2 had the youngest patient population and the lowest mortality. C3 was enriched for inflammatory and cellular stress pathways, while the highest mortality subtype, C4, showed evidence of immunosuppression and metabolic reprogramming. Gene and pathway-level analysis of our molecular subtypes statistically correlated with results from analysis of 28-day mortality, with the best (C2) and worst subtypes (C4) exhibiting similar molecular dysregulation as survivors and non-survivors, respectively, suggesting that targeted subtype-specific approaches could improve sepsis management. Using a large-scale pharmacogenomics database, ClinPGx, we then evaluated potential targeted therapies for each subtype and assessed the potential clinical impacts of these drugs. We identified several potential subtype-specific candidate therapies, including possible responsiveness to Methylene Blue therapy for patients from our highest mortality subtype, C4. Notably, our drug repurposing analysis revealed a significant representation of anti-inflammatory monoclonal antibody therapies across molecular subtypes. The anti-correlated signatures in C1 and C2 suggest that monoclonal antibody therapies may only be effective for patients in one of the subtypes, which may explain why prior clinical trials have been unsuccessful. Altogether, our detailed molecular subtyping and drug repurposing analysis identified subtype-specific potential drug targets, with implications for future precision medicine for sepsis.