N- and T-genes framework for pathway deregulation: Taxonomy, progression, and protection in cancer
En palabras de los autores
Background. Pathway analysis is widely used to interpret gene expression, but conventional enrichment depends on arbitrary thresholds and does not quantify deregulation magnitude. Recent reformulations based on average log-fold changes remove the threshold but still average over all annotated genes, diluting the signal from the genes that actually drive the normal-to-tumor transition. The N/T framework, which identifies genes with exclusive expression intervals in normal (N) or tumor (T) samples, offers a selective alternative, but it had not been extended to pathway- level quantification, taxonomy, or cross-cancer comparison. Methods. We define effective activation frequencies for Reactome pathways based on N- and T-genes: for each sample and pathway, the effective N (T) frequency is the fraction of annotated N (T) genes active in a normal (tumor) sample. The analysis proceeds in three explicit steps: (i) cohort averages identify the pathways important in each tissue; (ii) averages over hierarchical-clustering groups, ordered by the number of active genes, reveal sequences of pathway activation and deactivation along progression; (iii) a minimum spanning tree (MST) built on pathway-frequency vectors -- independent of the number of active genes -- defines the molecular taxonomy, and branch averages identify subtype-specific programs. We apply the framework to the full set of Reactome pathways in prostate adenocarcinoma (PRAD), lung adenocarcinoma (LUAD), liver hepatocellular carcinoma (LIHC), and glioblastoma (GBM), analyzing both tumor and normal compartments. Results. In PRAD tumors (n=499), hierarchical clustering identifies four ordered progression groups and the MST resolves six main molecular subtypes comprising 406 tumors (81.4%): B1 (42.1%; MET/NFE2L2/energy metabolism), B2 (13.8%; RAS/apoptosis/telomere), and four smaller subtypes defined by PTEN (B3), DNA repair and lipid metabolism (B4), inflammation and LXR-mediated lipid metabolism (B5), and neuronal development and base excision repair (B6). An eight-gene perfect T panel reproduces the MST subtypes at gene level. Analysis of normal prostatic tissue (n=52) reveals a hierarchy of protection: immune response, apoptosis, and senescence are lost or partially lost early, DNA repair at an intermediate stage, and cell cycle regulation and receptor tyrosine kinase signaling last; an eleven-gene N panel reproduces this hierarchy at gene level. The cross-cancer comparison shows that this architecture is a general property: DNA repair, cell cycle, and RTK protection are maintained in normal tissue across PRAD, LUAD, LIHC, and GBM, while the specific pathways lost in normals and gained in tumors are tissue-specific (MET/NFE2L2 in PRAD; NTRK/LXR/phagosome in LIHC; surfactant/FGFR/glycosylation in LUAD; TP53loss/lipid metabolism/glycosylation in GBM). Taxonomic heterogeneity decreases from PRAD (six subtypes) to LIHC (two) to LUAD and GBM (one dominant program), paralleling an increase in the number of activated T-genes. Conclusions. The framework eliminates dilution by non-nformative genes, quantifies pathway deregulation sample by sample, and unifies tumor taxonomy with a molecular indicator of progression. Its most distinctive contribution is the joint analysis of normal and tumor compartments across cancer types, which links loss of protection in normal tissue to gain of progression programs in tumors and reveals both shared and tissue-specific cancer mechanisms. The method is general and applicable across cancer types, although quantitative results depend on the underlying cohort.
Apareció: miércoles, 18 de noviembre. bioRxiv. Preprint, todavía sin revisión por pares.