Background Anorexia nervosa and binge-eating behaviour are clinically distinct yet genetically and neurobiologically overlapping phenotypes. The molecular systems that differentiate restrictive from binge-related eating pathology remain incompletely defined, particularly when genetic risk is integrated with age-related neuronal chromatin programs. A companion preprint prioritized synaptic vesicle dynamics, Ras homolog family member A (RHOA)-centered trafficking, and mitochondrial substrate utilization as candidate interfaces between eating-disorder genetic risk and nicotinamide mononucleotide (NMN)-sensitive biology. Objective This study aimed to extend the companion NMN-sensitive framework by applying a Polycomb- and Kyoto Encyclopedia of Genes and Genomes (KEGG)-derived multi-gene-set transcriptome-wide association study (TWAS) to the same brain-tissue summary-statistics-based PrediXcan (S-PrediXcan) results for broad anorexia nervosa (AN broad) and broad binge-eating behaviour (BE broad). The analysis is hypothesis-generating and is not an independent replication of the companion study. Methods Sixteen Polycomb- and KEGG-derived gene sets, constructed from age-associated H3K27me3 programs in neurons, were interrogated. Analytical steps included enrichment testing, competitive permutation analysis, bootstrap confidence intervals, disease-differential testing, tissue-level analyses, profile proximity, concordance testing, leave-one-out influence analysis, pooled Benjamini-Hochberg false discovery rate (BH-FDR) correction across the prespecified family of enrichment, differential, proximity, and pairwise tests, network co-occurrence, module clustering, cross-module candidate prioritization, and robustness-weighted prioritization. Results The synaptic vesicle cycle showed the strongest positive enrichment in BE broad (Stouffer Z = 4.628; competitive permutation p = 0.0179; Wilcoxon p = 0.00229; bootstrap 95% confidence interval (CI): 0.93-8.23). Positive tissue-level statistics were observed in the frontal cortex BA9, anterior cingulate cortex BA24, and caudate basal ganglia, although these tissues are correlated and were not treated as independent replications. AN broad exhibited its strongest meta-level signal in axon guidance, with additional enrichment involving adherens junctions, long-term potentiation, and ribonucleic acid (RNA) polymerase-related genes. Stage 3 integration identified four higher-order module clusters, 28 cross-module prioritized candidates influential across at least three modules and both phenotypes, 577 influential candidate genes, and 576 cross-module network nodes. RHOA emerged as a recurrent, high-influence, and directionally unstable network node and candidate linking adhesion, actin remodeling, trafficking, and signaling. Across pathways, the two phenotypes shared substantial molecular infrastructure yet demonstrated only moderate directional concordance, consistent with a dimensional rather than strictly categorical model. Conclusions The results supply structural and regulatory context for the companion NMN framework and support a dimensional model in which the two phenotypes share substantial molecular infrastructure yet differ in pathway magnitude and regional weighting. All pathway and gene prioritizations require independent replication, colocalization, fine-mapping, and functional validation before any mechanistic or therapeutic inference can be drawn.
