Integrating multi-omics with neuroimaging and behavior: A preliminary model of dysfunction in football athletes

Contact sports affect measures at multiple scales such as transcriptomics, metabolomics, brain function, and behavior, but studies have not yet studied the statistical structure of how they are integrated. This preliminary study, examining collegiate American football players, integrated across-season changes (Δ) from transcriptomic and metabolomic variables (neuroinflammatory miRNAs and metabolites), neuroimaging (resting-state fMRI network fingerprint similarity), and virtual reality (VR)-based motor control. These findings were then assessed against head acceleration events (HAE). Using permutation-based moderation analysis (all $p_F^{perm},p_{{\beta }_3}^{perm}$ $\le$ 0.05 $)$ , we observed that (1) Δtridecenedioate, a mono-unsaturated fatty acid, interacted with ΔmiR-505 to predict default mode network (DMN) fingerprint similarity, meaning the interaction between two molecular biology measures predicted a neuroimaging measure. Further, (2) Δtridecenedioate and DMN fingerprint similarity interacted to predict motor control, indicating the interaction of a molecular and a neuroimaging measure predicted behavior. ΔmiR-505 was positively related to HAE and DMN fingerprint similarity was negatively related to HAE and reduced relative to non-athlete subjects. These multi-scale, moderating relationships between a potential ROS scavenger, neuroinflammatory miRNA, reduced brain connectivity, and diminished motor control argue that seemingly healthy athletes with frequent HAE may experience chronic neuroinflammation. This imaging-omics framework using permutation-based mediation/moderation analysis has general applicability for human-animal translational studies.