How fruit fly tumors reveal ancient pathways that drive human RAS cancers

RAS genes are the most frequently mutated oncogene family in human cancer, yet the way oncogenic RAS dynamically interfaces with tumor suppressor networks to orchestrate malignant transformation remains poorly understood. A team at Westlake University has now built a systematic platform to close this gap, using the fruit fly Drosophila melanogaster to chart, stage by stage, the conserved molecular trajectories that RAS-driven tumors follow from initial clonal expansion to invasive, wasting disease. The resource, called the Atlas of Ras-driven Tumors in Drosophila (ART-D), is described in Molecular Systems Biology and reveals a four-pathway nexus—coordinated activation of JNK, Notch, and NF-κB/Toll signaling alongside inactivation of the Hippo pathway—that is strikingly mirrored in human KRAS- and NRAS-mutant cancers.

The study began with a large-scale forward genetic screen. Using ethyl methanesulfonate mutagenesis on Drosophila chromosome 3L and the ey-FLP-based mosaic analysis with a repressible cell marker (MACRM) system, the researchers induced homozygous mutant clones in cells already carrying oncogenic Ras (RasV12), then searched for recessive-lethal alleles that synergized with Ras to drive malignant overgrowth of the eye-antennal disc. Eight previously uncharacterized tumor suppressors emerged: the endocytic genes Vacuolar protein sorting 36 (Vps36), Syntaxin 7 (Syx7), Rabaptin-5-associated exchange factor (Rabex-5), and Tumor susceptibility gene 101 (TSG101), plus furry (fry), fiery mountain (fmt), emei, and misshapen (msn). Together with the canonical polarity tumor suppressors scribble (scrib) and lethal (2) giant larvae (l(2)gl), these genes defined a cohort of ten genetically distinct Ras-driven tumor models.

Each of the ten models was then tracked longitudinally through three time points—5.5, 9.5, and 13.5 days after egg laying—using four quantitative malignancy metrics: larval tumor burden (measured by GFP signal and tumor-to-body ratio), tumor overgrowth (volumetric analysis of disc tissue), tumor-mediated cachexia (larval transparency reflecting muscle and fat body degradation), and invasion of the neighboring ventral nerve cord. The resulting atlas revealed three phenotypic stages shared by all tumors: an initial stage marked by cell overgrowth, a promotion stage characterized by uncontrolled proliferation, onset of invasion, and early cachexia, and a progression stage distinguished by severe wasting and aggressive tissue invasion. Notably, the researchers also observed intestinal metastasis in larvae bearing Ras-driven tumors, indicating genuine malignant potential beyond local overgrowth.

Although all ten tumors ultimately developed aggressive phenotypes, their progression trajectories diverged. To capture this heterogeneity, the team applied robust linear regression to six digitized phenotypic parameters, extracting the intercept (initial malignancy), slope (progression rate), and adjusted R-squared (goodness of fit) of each regression model. Principal component analysis projected the ten models onto a two-dimensional map in which PC1 captured features associated with cachexia and PC2 reflected tumor burden. K-means clustering then classified the tumors into three subtypes: Group A (scrib, l(2)gl, Vps36, fry, emei, fmt), Group B (Syx7, msn), and Group C (TSG101, Rabex-5). This phenotype-centric stratification—rather than the gene-centric approach typical of longitudinal cancer studies—became the analytical backbone of the entire platform.

Bulk RNA sequencing at five pseudotime points (T1, wild-type tissue; T2, benign RasV12; T3, initial; T4, promotion; T5, progression) generated longitudinal transcriptomes for each subtype. Weighted gene co-expression network analysis (WGCNA) linked gene modules to the phenotypic traits, Mfuzz soft clustering delineated continuously upregulated and downregulated gene clusters, and PANGEA functional annotation identified the biological processes and signaling pathways significantly associated with malignant progression. All three subtypes shared enrichment for glycolysis, gluconeogenesis, and amino acid metabolism, and for the EGFR/FGFR/PvR MAPK-related pathway. Group B uniquely enriched fatty acid oxidation and the Hippo pathway, while Group C showed inositol phosphate metabolism, O-glycan biosynthesis, and TGFβ/Activin signaling. Transcription factor enrichment identified the Notch-related factor longitudinals lacking (lola) and the AP-1 components Jun-related antigen (Jra) and kayak (kay) as co-enriched across groups, while protein interaction networks pinpointed metabolic hub genes including Got2, Pgi, and Hex-A in at least two subtypes.

Comparing malignant tumors at T3 with benign RasV12 tumors at T2 revealed 247 co-upregulated and 578 co-downregulated genes, annotated primarily to JAK/STAT, NF-κB/Toll, and MAPK signaling. Gene set variation analysis and single-sample gene set enrichment analysis confirmed widespread activation of JAK/STAT and MAPK pathways from T3 onward, and immunofluorescent staining showed increased phosphorylation of the MAPK effectors rolled (ERK) and basket (JNK), elevated JNK targets MMP1 and wingless, and consistently raised levels of the NF-κB/Toll transcription factor Dorsal across all ten models. Notch signaling scores correlated positively with tumor progression, whereas Hippo activation correlated inversely with severity. Staining confirmed robust upregulation of the Notch ligand Delta, the Notch intracellular domain, its target dMyc, and the Hippo target Diap1, together with increased nuclear translocation of the Hippo co-factor Yorkie.

Functional validation followed. Expressing constitutively active Notch (Notchact) synergized with loss of scrib, l(2)gl, Vps36, or fry to drive aggressive overgrowth, but failed to promote expansion in Syx7-, Rabex-5-, or TSG101-deficient cells—a context-dependent interaction the authors attribute to differences among endocytic processes. Activated Yorkie (YkiS168A), by contrast, cooperated with seven of the tumor suppressor losses. The fry gene emerged as a particularly potent suppressor: its 3,479-amino acid product functions with the NDR kinase tricornered and Mob as tumor suppressor (mats) to regulate polarized cell growth, and mammalian studies show FRY depletion promotes nuclear localization of YAP/TAZ. Because fry-deficient cells cooperated with multiple oncogenic inputs (RasV12, Notchact, YkiS168A), the team selected RasV12; fry tumors—a previously uncharacterized model—for deep multi-omics analysis combining bulk RNA-seq and ATAC-seq across tumor stages.

The integrative analysis of RasV12; fry tumors identified dynamic transcriptional modules enriched for MAPK, NF-κB/Toll, Notch, and Hippo signaling, with stage-specific expression peaks: Toll, mastermind, Diap1, frizzled, and wingless peaked at the initial stage, while insulin pathway genes rose at later stages coinciding with severe cachexia and invasion. ATAC-seq confirmed elevated chromatin accessibility at promoters of key effectors in all four pathways, with motif enrichment for AP-1, DREF, and Initiator binding sites. Genetic perturbations then established necessity: RNAi or dominant-negative inhibition of JNK components (Tak1, hemipterous, basket, kayak, Jra), suppression of Toll and Dif, blockade of Notch ligands and receptor, and overexpression of warts to reactivate Hippo each markedly shrank the GFP-marked tumor region and rescued the pupariation delay in tumor-bearing larvae. Under physiological conditions, fry knockdown left JNK-dependent thorax closure unaffected but enlarged the adult eye and rescued the eiger-induced small-eye phenotype, indicating that fry modulates JNK-dependent processes specifically in pathological contexts.

Translational analyses then connected the fly models to human oncology. Using fourteen orthology databases, the ten Drosophila tumor suppressors were mapped to eighteen human orthologs, including SCRIB, LLGL1/LLGL2, VPS36, STX7/STX12, RABGEF1, TSG101, FRY/FRYL, PPP6R1-3, TMEM161A/B, and MINK1/TNIK/MAP4K4. Pan-cancer evaluation of TCGA data showed that SCRIB is mutated in more than 10% of pancreatic, esophageal, endometrial, uterine carcinosarcoma, and melanoma cases; LLGL1/LLGL2 shows the highest mutation rates in endometrial carcinoma and melanoma; endocytic genes are frequently mutated and downregulated in endometrial cancer and melanoma; and FRY/FRYL mutations exceed 5% in most KRAS/NRAS-driven cancers, with FRY consistently downregulated. Single ortholog mutations significantly affected patient survival—SCRIB mutations threatened overall and disease-free survival in pancreatic adenocarcinoma, and FRY mutations adversely affected survival in esophageal carcinoma—while expression levels correlated positively with overall survival across multiple cancers.

Cross-species validation at single-cell resolution sealed the argument. The team integrated publicly available single-cell datasets from colorectal, lung, pancreatic, and melanoma patients, retaining 174,385 high-quality cells, and scored pathway activity in malignant epithelial cells. Within RAS-activated subclusters, RAS signaling correlated positively with MAPK, JNK, Toll-like receptor, and Notch activation and with Hippo inactivation in nearly all cancers—an activation pattern directly mirroring RasV12; fry tumors. A composite fry/FRY-deficient tumor signature, summarizing JNK, TLR, and Notch activation with Hippo inactivation, correlated with cell proliferation, epithelial-mesenchymal transition, and cell migration across cancers, and low FRY expression was associated with altered immune infiltration, including increased CD8 T cells and decreased M2 macrophages. Finally, machine learning models built from the conserved pathway components—117 algorithm combinations evaluated for one-, three-, and five-year survival prediction—performed excellently for lung adenocarcinoma, pancreatic adenocarcinoma, and melanoma (AUC values above 0.85), with a StepCox[both] + random survival forest model effectively stratifying LUAD and melanoma patients into high- and low-risk groups. The authors acknowledge that the fly lacks adaptive immunity and that the precise molecular scaffolds mediating the four-pathway crosstalk remain undefined; future work will use immunoprecipitation-mass spectrometry to identify FRY-interacting proteins and validate the findings in patient-derived organoids. Even so, ART-D demonstrates that evolutionarily ancient oncogenic mechanisms, first dissected in a fruit fly eye disc, can be translated into diagnostic and prognostic tools for some of the deadliest RAS-driven human malignancies.

Subject of Research: Cross-species oncogenic mechanisms uncovered by the Drosophila ART-D atlas of Ras-driven tumors

Article Title: Cross-species insights from ART-D to uncover evolutionarily conserved oncogenic mechanisms

Article References: Guo, Y., Zheng, J., Wu, Y., Zhao, K., Lv, D., Liu, W., Wang, M., Lu, J.-Y., Xu, W., Wang, X., & Ma, X. (2026). Cross-species insights from ART-D to uncover evolutionarily conserved oncogenic mechanisms. Molecular Systems Biology, 22(8), 1214-1251. https://doi.org/10.1038/s44320-026-00215-8

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00215-8

Keywords: Drosophila, RAS oncogene, ART-D, tumor suppressors, JNK signaling, Notch pathway, NF-κB/Toll, Hippo pathway, furry, cross-species cancer research, machine learning prognosis, KRAS