Single-cell map reveals lncRNA networks shaping cancer outcomes

Long noncoding RNAs are often described as genetic switches, but a new pan-cancer analysis shows that their importance may depend less on any single molecule than on the cellular neighborhoods in which they operate. Researchers have introduced SIGLNCOS, a computational framework that combines single-cell RNA sequencing with functional enrichment analysis to identify cell-specific lncRNAs and the pairs that appear to regulate related biological programs. The study analyzed more than half a million cells from 31 cancer types, resolving 39 cell populations and uncovering 667 cellular signature long noncoding RNAs, or clncRNAs, alongside 3,381 functionally co-associated lncRNA pairs. The results are available through the new sCo-Lnc web resource, which allows researchers to browse, search, visualize and download the networks.

SIGLNCOS is designed for a practical problem in single-cell cancer research. Most workflows are optimized around protein-coding genes, whereas lncRNAs are often scarce in individual cells, strongly restricted to particular cell types and difficult to connect to biological functions. The new framework first standardizes lncRNA expression across tumors, then compares each transcript with the pooled cells of other types. A transcript qualifies as a clncRNA if it is detected in at least 10% of cells in a target population, shows a log2 fold change greater than 0.25 and passes a Bonferroni-adjusted significance test. The analysis also accounted for doublets, high mitochondrial content and batch effects, while checking whether the main patterns survived stricter filtering and regression of cell-cycle, stress, apoptosis and mitochondrial signals.

The resulting atlas confirms that lncRNAs carry information about cellular identity that is often obscured in bulk-tissue measurements. Their Gini indices, a measure of expression inequality across cell types, were consistently higher than those of messenger RNAs, indicating more restricted expression. Many lncRNAs were concentrated in malignant, immune and stromal compartments, and strong signals appeared in neutrophils, mast cells, megakaryocytes and common myeloid progenitors. Individual transcripts also showed striking context dependence: MEG3 was enriched in fibroblasts in several tumors, while NEAT1 was associated with myeloid cells, including macrophages and monocytes, in selected cancers. Conversely, DHRS4-AS1 and PSMA-AS1 were prominent in papillary thyroid carcinoma but suppressed in esophageal squamous cell carcinoma, illustrating how the same lncRNA landscape can differ sharply between tumor types.

Expression specificity was only the first layer of the analysis. For each clncRNA, SIGLNCOS ranked protein-coding genes according to their single-cell correlation with that transcript and used those rankings for preranked gene-set enrichment analysis across Gene Ontology, KEGG and immune-related collections. The resulting normalized enrichment scores indicated whether a lncRNA was associated with activation or suppression of a pathway. This approach linked clncRNAs to processes including differentiation, intercellular communication, metabolic adaptation and immune evasion. Their strongest cancer-hallmark associations involved invasion and metastasis, deregulated cellular energetics and sustained proliferative signaling. Among the 667 clncRNAs, 115 were classified as hypervariable: 44 were cancer-specific, seven recurred broadly and 64 showed intermediate recurrence. Some shared markers, including MALAT1, NEAT1 and MEG3, appeared across many tumors, while others were restricted to particular diseases.

NEAT1 illustrated why cellular context cannot be ignored. The lncRNA showed different functional relationships in macrophages from pancreatic neuroendocrine tumors and rhabdomyosarcoma. In pancreatic neuroendocrine tumors, its enrichment pattern was consistent with reduced apoptosis and innate immune activity and with a more immunosuppressive, M2-like macrophage state. In rhabdomyosarcoma, the relationship was reversed, aligning with a more inflammatory, M1-like state. Pseudotime analysis also indicated that NEAT1 expression changed nonlinearly along macrophage-state trajectories, with high-expressing cells concentrated at early and late stages. These findings do not prove that NEAT1 causes the state changes, but they demonstrate that a recurrent lncRNA can be associated with opposite immune programs in different tumor environments.

The framework’s main advance is the conversion of individual markers into pair-level regulatory networks. Two clncRNAs were considered potential partners when they shared more than half of their enriched functions. The researchers then compared the direction and strength of each shared enrichment signal, down-weighting functions common to many lncRNAs so that more specific relationships received greater influence. A weighted intensity score separated directionally concordant pairs, whose shared functions moved in the same direction, from discordant pairs, whose functions moved in opposite directions. This produced networks that varied substantially among tumors and cell populations. The largest included 452 pairs among CD8-positive T cells in papillary thyroid carcinoma and 416 pairs among fibroblasts in ovarian clear cell carcinoma. By contrast, CD8-positive T-cell networks in papillary and anaplastic thyroid carcinoma differed in both connectivity and the direction of selected relationships.

Most networks were highly cancer-specific, but a compact recurrent core emerged. SNHG8 and ZFAS1 formed the most frequently recurring pair in tumor-cell networks, appearing in at least four cancers and remaining detectable under stricter overlap criteria. Across cancers, the pair shared associations with ribosome- and translation-related functions, with additional links to inflammatory response, epithelial differentiation or transcriptional programs depending on the cancer type. Perturbing SNHG8 or ZFAS1 individually in pancreatic and thyroid cancer cell lines reduced colony formation, wound closure and migration, whereas overexpression produced the opposite effects. Independent metacell, spatial-transcriptomic and public perturbation analyses reproduced their shared spatial and functional patterns. The experiments support the biological relevance of each RNA, but they do not yet establish that the two molecules physically interact or act cooperatively through a single molecular mechanism.

Clinical data suggested that lncRNA pairs may contain information that individual transcripts miss. The researchers integrated statistically supported networks with TCGA clinical records and estimated the abundance of each cell compartment in bulk tumors. Of 233 pair–cell-type contexts tested across ten cancer cohorts, 78 were associated with overall survival after adjustment for the matched cell fraction and available clinical factors; 50 remained significant after correction for multiple testing. The signals were concentrated in particular cellular compartments rather than distributed uniformly: all 17 supported bladder-cancer associations came from lymphoid populations, whereas all 28 associations in uveal melanoma were assigned to tumor cells. In bladder cancer, for example, the LINC01138–PCED1B-AS1 pair in CD8-positive T cells was associated with improved overall survival, with a hazard ratio of 0.668 and a false-discovery-rate-adjusted P value of 0.001. In several contexts, the combined pair score had a stronger prognostic association than either RNA alone.

Fibroblast results added another level of resolution. Cancer-associated fibroblasts in bladder, ovarian and pancreatic cancers were classified as myCAF-like, inflammatory or antigen-presenting states, and different lncRNA pairs mapped preferentially to different programs. Bladder-cancer pairs were associated mainly with myCAF-like and inflammatory states, ovarian pairs were dominated by myCAF-like activity, and pancreatic pairs favored inflammatory CAFs. Adding CAF-state scores to survival models changed the statistical support of individual pairs in a cancer- and state-dependent manner, showing that the signals reflect cellular composition as well as overall fibroblast abundance. The study nevertheless has important limits: functional co-association is inferred from shared enrichment profiles rather than direct interaction, single-cell sparsity can affect low-abundance lncRNAs, and survival analyses use bulk cohorts with computationally estimated cell fractions. Prospective cohorts, spatial measurements and cell-type-specific perturbation studies will be needed before the prioritized pairs can guide clinical decisions.

The sCo-Lnc platform makes the atlas publicly accessible for further testing. Researchers can search by cancer type, cell type, lncRNA symbol or Ensembl identifier, compare pan-cancer patterns and use tools for clustering, co-regulation, spatial co-expression and cancer-hallmark enrichment. Standardized node and edge tables include identifiers, expression fractions, fold changes, Jaccard indices, intensity scores, shared functions, association directions and empirical significance statistics. A fixed release of the SIGLNCOS code is also available with its configuration files, random-seed settings and example inputs and outputs. As single-cell multiomics and spatial assays add new information about where regulatory RNAs operate, the authors say the resource will be reviewed annually and maintained for at least five years, with suitable epigenomic, spatial and disease datasets incorporated through versioned releases.

Subject of Research: Single-cell identification of cell-specific lncRNA networks and prognostic associations across 31 cancer types

Article Title: A novel framework for identifying cellular signature LncRNAs and their functional coregulatory networks with prognostic implications at single-cell resolution

Article References: Bai, J., Li, X., Hu, C., Peng, C., Chen, J., Liang, W., Li, X., & Zhang, Y. (2026). A novel framework for identifying cellular signature LncRNAs and their functional coregulatory networks with prognostic implications at single-cell resolution. Journal of Advanced Research. https://doi.org/10.1016/j.jare.2026.10.034

Image Credits: AI Generated

DOI: 10.1016/j.jare.2026.10.034

Keywords: single-cell RNA sequencing, long noncoding RNAs, cancer genomics, tumor microenvironment, functional co-regulation, prognostic biomarkers, pan-cancer analysis, cellular heterogeneity, novel, framework, identifying, cellular