Most shared microRNA family coexpression in extracellular vesicle RNA cargo is binding degree and the residual is neither family specific nor vesicle specific

Abstract Extracellular vesicle (EV) transcriptomes are widely used to build competing endogenous RNA (ceRNA) networks. In these networks one RNA is thought to relieve another from microRNA repression by competing for the same microRNAs. They rest on one observation that transcript pairs sharing more microRNA families are more strongly coexpressed, read as competition. Kinetic theory predicts the same pattern but also shows that a shared microRNA pool correlates two transcripts without any competition. Which of the two drives the pattern had not been measured in EV cargo. Methods adjusting ceRNA inference for confounders exist, although none of the ninety one indexed EV or exosomal ceRNA network studies reports a permutation or null model test of its edges. This study therefore measured how much of the shared family gradient is binding degree and how much survives strict matching, which has not previously been done in EV cargo. We used exoRBase profiles of 2,125 blood EV and 125 urine EV samples plus an independent neurodegenerative plasma EV cohort. Pairs sharing at least K microRNA families were compared against control pairs matched for expression level and binding degree. Coexpression rose with K, but most of that rise disappeared once each pair was compared against control pairs binding a similar total number of microRNAs. This removed about 70% of the effect, leaving roughly a third of the original size at the highest sharing level. The remaining signal replicated in blood and urine EV but not in the smaller neurodegenerative cohort. It did not map to any single microRNA family, since none of the 166 sufficiently powered families passed false discovery correction. The gradient was also not vesicle-specific in either biofluid. In blood the vesicle fraction showed the weakest gradient against cell-free plasma and whole blood. In urine it ranked below kidney cortex and prostate tissue. The gradient itself is therefore not the finding. What is new is measuring what drives it. A shared family gradient is not a diagnostic signature of competition and reporting one without such a comparison does not establish a ceRNA network. Predicted sponge interactions should be treated as hypotheses requiring these controls and direct validation.