Multi-omics signatures of hemolysis and vesiculation in stored red blood cells

Substantial donor-to-donor heterogeneity shapes red blood cell (RBC) storage biology, but the extent to which integrated donor and molecular measurements capture hemolysis and extracellular-vesicle (EV) phenotypes remains incompletely defined. We analyzed the tail-enriched REDS-III RBC-Omics recalled cohort using one storage-day-42 record from 649 donors. Demographic, complete blood count/ferritin, trace-element, metabolomic, lipidomic, and proteomic predictor blocks were benchmarked for three hemolysis and five EV outcomes. The molecular predictors and most outcomes were measured concurrently at day 42; change in total EVs was day 42 minus day 10. Ridge models used 10 repeats of five-fold outer cross-validation with five-fold inner tuning and fully fold-contained preprocessing and feature screening. Mean Pearson correlations were highest for osmotic hemolysis (r=0.63; full model), oxidative hemolysis (r=0.56; full model), RBC-derived CD235a+CD108− EVs (r=0.47; metabolomics), storage hemolysis (r=0.43; full model), and platelet-derived CD41a+CD62p− EVs (r=0.47; metabolomics). Change in total EVs and RBC-derived CD235a+CD108+ EVs were more modest (r=0.27 and 0.26, respectively), whereas total EV counts were not predicted above chance. Max-block permutation testing supported signal for seven outcomes (Benjamini-Hochberg q=0.006) but not total EVs (q=0.726). Elastic-net sensitivity analysis reproduced the principal ranking. Thus, day-42 storage phenotypes contain endpoint-specific distributed molecular signatures. These internally validated, cross-sectional results from a phenotype-enriched cohort do not establish prospective forecasting or transportability to routine donor populations; external validation is required.