Symptom-level transcriptome-wide associations of NMN-nominated KEGG pathways in major depression: a distributed endocytosis–hypersomnia signal, gene-weighted immune and motor cassettes, and limits of pathway-level polarity

Major depressive disorder is clinically heterogeneous. Its diagnostic criteria combine affective, cognitive, sleep, appetite, weight, energy, and psychomotor symptoms that may have partly different genetic and biological correlates. Recent symptom-level genetic studies indicate that directional neurovegetative features, including hypersomnia versus insomnia and weight gain versus weight loss, should not automatically be treated as interchangeable manifestations of a single depression factor. Work on immunometabolic and reversed-neurovegetative depression has likewise linked hypersomnia and weight gain to biological correlates that differ from those associated with insomnia and weight loss. This study examined whether KEGG gene sets nominated from a mouse NMN-aging experiment show symptom-specific associations with genetically predicted human brain expression in major depression. Public S-PrediXcan gene-level Z scores were available for 12 depression symptoms across brain tissues. Gene-level tissue Z scores were combined across tissues, and pathway-level Stouffer statistics were reconstructed from the stored gene-by-symptom matrix. The analysis included enrichment statistics, permutation tests, bootstrap confidence intervals, gene-class decomposition, leave-one-gene-out analysis, tissue sign concordance, size-matched empirical null sampling, library ranking, pathway overlap, and symptom-profile clustering. Primary interpretation was restricted to associations supported by permutation or confidence-interval criteria and by robustness checks. Gene-class and clustering analyses were treated as secondary. Four lead associations remained most defensible. Endocytosis was negatively associated with hypersomnia, with a pipeline Stouffer Z of approximately − 5.52, permutation p = .0025, 194 contributing genes, agreement across six brain tissues, rank 1 of 240 library tests, and a size-matched union-null p = .005. The association was distributed and aligned with the core symptom direction rather than being driven by the polar-divergent shortlist. Antigen processing and presentation was positively associated with hypersomnia, but the result was small and TAPBP-weighted. Peroxisome was positively associated with psychomotor agitation and was supported by IDH2 and HSD17B4 rather than HMGCL. A vasopressin-labelled set was also positively associated with agitation, with leverage concentrated in dynein and CREB3-family genes rather than AVP. HMGCL showed an extreme gene-level weight-gain versus weight-loss contrast but did not drive the peroxisome–agitation result. The findings are library-specific and hypothesis-generating. They support symptom-level prioritization of genes and biological processes for replication and colocalization, not claims about measured RNA, pathway activity in patients, NMN responsiveness in human brain, treatment response, or biological patient subtypes.