GDF15 as a biomarker of metabolic dysfunction and mortality in sepsis and ARDS: a cohort study

Abstract

Two inflammatory phenotypes of sepsis and ARDS, hyperinflammatory and hypoinflammatory, have consistently been described with distinct biology and clinical outcomes. Mitochondrial metabolic dysfunction has been identified as a key characteristic of the Hyperinflammatory phenotype. Growth differentiation factor-15 (GDF15) is a stress-responsive protein induced primarily by mitochondrial integrated stress response. In murine sepsis, GDF15 promotes survival through β-adrenergic signaling and hepatic metabolic adaptation, suggesting it may identify treatment-responsive subgroups. We sought to characterize GDF15’s prognostic value in critical illness, determine whether it identifies differential treatment responses, and define its multi-omic signatures. We analyzed plasma GDF15 across four independent cohorts of critically ill adults ( N  = 1,247): FACTT (ARDS, N  = 400), ROSE (severe ARDS, N  = 160), CLOVERS (septic shock, N  = 595), and EARLI (sepsis/ARDS, N  = 92). We tested the association of GDF15 with inflammatory phenotype and mortality, benchmarked it against other protein biomarkers and lactate, tested treatment interactions in the two fluid-strategy trials, and integrated whole-blood transcriptomics and untargeted plasma metabolomics. Hyperinflammatory phenotype was associated with higher GDF15 in all four cohorts ( p  < 0.001). GDF15 was independently associated with mortality in the three trial cohorts but not in EARLI. Among the eight biomarkers measured in all four cohorts, GDF15 had the highest random-forest permutation importance in every trial cohort but ranked seventh of eight in EARLI, where IL-8 was most important. Adding GDF15 modestly improved discrimination beyond severity scores in FACTT and CLOVERS. GDF15 added prognostic value beyond phenotype classification in the three trial cohorts, and its prognostic association did not differ significantly between phenotypes in any cohort (all interaction p  > 0.3). Exploratory treatment interactions with fluid strategy were not significant in FACTT or CLOVERS. Multi-omic integration linked elevated GDF15 to the integrated stress response, OXPHOS and mitochondrial biogenesis, and impaired fatty acid β-oxidation, correlating most strongly with a previously defined mortality-associated metabolic dysfunction signature ( r  = 0.69, p  < 0.001). GDF15 is a biomarker of mitochondrial metabolic dysfunction and is prognostic across three trial populations of sepsis and ARDS, but it was not independently associated with mortality in a smaller observational cohort. It is a candidate for prognostic enrichment in trials of metabolically targeted therapies, though whether it enables predictive enrichment remains uncertain.