Abstract Background Molecular subphenotyping has the potential to enable precision medicine in critically ill patients. However, a thorough analysis of protein biomarkers used to define these subphenotypes has been largely limited to associative conclusions. Measuring and understanding interactions between these markers can give insight into causal relationships with respect to critical care outcomes. Recent work in ARDS has highlighted network analysis as a useful tool for analyzing relationships between plasma biomarkers, exposing the roles of known and novel proteins in complex pathogenesis. However, the interplay between subphenotyping and correlative biomarker relationships has not been thoroughly investigated. Methods Using a large multiplex panel (NULISA, Alamar Biosciences), we analyzed the plasma proteome at baseline using samples from the CAF-PINT study of pediatric heart and lung failure. Age- and sex-matched samples were grouped by previously defined LCA subphenotype (hypo- or hyper-inflammatory) using undersampling to produce evenly-split subsets (N = 42 per subphenotype group). Unsigned Spearman correlation matrices were computed separately in each group between all 250 protein markers, and these matrices were subtracted to compute changes in correlation. Significant changes in correlation were identified using permutation testing at FDR0.05. Network analysis of these differences was performed using Gephi v0.10.1. Protein markers were colored as protective or detrimental based on their mortality effect size, and their network statistics were computed including eigenvector centrality. Results A large panel of inflammatory markers identified n = 155 differentially expressed proteins between hypo- and hyper-inflammatory subphenotypes. Among these markers, there were 353 significant changes in correlation represented in the network, most of which indicated a significant loss of correlation (n = 309/353 edges). Protein markers that were significantly elevated in nonsurvivors were central to the network as measured by eigenvector centrality (Wilcoxon P = 0.01). Conclusions Network analysis of a large proteomic panel demonstrates substantial decoupling between plasma biomarkers in hyper-inflammatory versus hypo-inflammatory subphenotypes. Further, the markers central to this decoupling are differentially elevated in nonsurvivors. Notably, the most central markers in the network, such as TNFSF15 and CD274, represent outsize influence in subphenotype-specific protein interactions but are largely underrepresented in the current literature. ICAM1 also emerged as a high-influence marker, consistent with prior results in pediatric ARDS. This work suggests that molecular subphenotypes can be characterized by substantial restructuring in protein interactions through under-studied pathways, and offers a number of novel targets to consider in future analysis and mechanistic studies. This abstract is funded by: None
Taylor et al. (Fri,) studied this question.