Long-read discovery in cancer cell lines defines a cis-spliced fusion RNA panel and its clinical associations across TCGA.
BACKGROUND Cancer-associated chimeric RNAs represent an important class of molecular alterations with diagnostic and prognostic potential. However, cis-spliced fusion RNAs generated by read-through transcription between adjacent same-strand genes remain insufficiently characterized because conventional short-read RNA sequencing and fusion-calling pipelines often deprioritize adjacent-gene junctions. Defining the cancer-associated cis-spliced fusion RNA landscape may expand the molecular genetic repertoire used for tumor classification, clinical correlation, and biomarker development. METHODS We integrated Oxford Nanopore long-read direct RNA sequencing of five human cancer cell lines with large-scale tumor, normal-tissue, molecular, and clinical datasets. Adjacent same-strand gene pairs were screened for cis-spliced fusion RNA junctions and filtered using mapping-quality and canonical splice-motif criteria. Candidate junctions were then evaluated by short-read reanalysis of 30,562 TCGA and GTEx RNA-seq samples using Snaptron. Tumor enrichment, driver-mutation association, pathway co-expression, pathologic stage, overall survival, biomarker performance, and biofluid detectability were assessed, with parental gene expression a covariate in the driver, pathway, stage and survival models and local splice coverage in the within-patient and stage models. RESULTS We identified 461 quality-controlled cis-spliced fusion RNA junctions, 431 of them absent from four major chimeric-RNA databases. We then tested this long-read panel against a wider short-read universe. Sixty-eight junctions, 27 of them panel members, were enriched in matched TCGA tumor cohorts, and 34 of the 68 were highly tumor-enriched relative to matched normal tissues. Within patients, junction detection remained tumor-associated after adjustment for 5' and 3' parental gene expression, library depth, and local splice coverage (odds ratio 1.59, p = 2.1 × 10⁻⁵). These junctions marked lineage-specific molecular programs, including TP53-mutated basal-like breast cancer and the NFE2L2-KEAP1 cytoprotective axis in lung squamous carcinoma. Fifty-eight candidate-cohort pairs changed monotonically across pathologic stage in cross-section, 35 rising and 23 falling, within a 363-candidate scan, 317 of them panel members. Across a 996-candidate survival screen, 300 of them panel members, six junctions stratified TCGA overall survival at hazard ratios of 2.2 to 4.5, and 2.2 to 4.8 after adjustment for 5' parental gene expression (6/6 significant with the raw log-rank threshold, 1/6 with a Benjamini-Hochberg q below 0.05 across all 1,228 tests). Two of the six are panel members and four showed high specificity against matched normal tissues. Biofluid detection did not distinguish cancer from control plasma cell-free RNA cohorts. CONCLUSIONS This study defines cis-spliced fusion RNAs as a previously underappreciated class of cancer-associated transcript alterations, linked here to driver-associated molecular programs, pathologic stage, and survival across a screening universe wider than the 461-candidate long-read panel. That evidence is consistent with a tumor association not fully explained by measured parental gene expression. Discovery, selection and modeling of the six survival candidates took place within TCGA, and we present them for confirmation in an independent junction-resolved cohort. This work provides a discovery framework and a prioritized set of candidate markers for cancer classification and prognosis, ready for evaluation in independent cohorts.