RNA Profiling Splits Deadly Dog Spleen Cancer into Three Actionable Subtypes

One of the most lethal cancers in dogs has long frustrated veterinary oncologists with its stubborn one-size-fits-all treatment approach. Now, a research team at the University of São Paulo has shown that canine splenic hemangiosarcoma—a rapidly progressive cancer of blood vessel cells—is not a single disease at the molecular level. By measuring the expression of 45 cancer-related genes in archived tumor samples from 27 affected dogs, the researchers identified three biologically distinct subtypes, each defined by a different constellation of active signaling pathways and, crucially, each pointing toward different candidate drugs.

Hemangiosarcoma is the most common splenic malignancy in dogs, arising from endothelial cells that line blood vessels. It is notorious for its genetic heterogeneity, high metastatic rate and grim prognosis. The standard of care—surgical removal of the spleen followed by chemotherapy with agents such as doxorubicin or carboplatin—produces only modest improvements in survival, because many tumors resist conventional protocols. Earlier genomic studies, including work by Gorden and colleagues in 2014 and by Wang and colleagues in 2020, had already hinted that canine hemangiosarcoma comprises multiple molecular subtypes with striking similarities to human angiosarcoma. What remained missing was a practical, cost-effective way to detect those subtypes in the formalin-fixed, paraffin-embedded tissue samples that veterinary pathologists routinely produce.

The São Paulo team, led by Heidge Fukumasu and first author Sonia Prince Fiebig, addressed that gap using the Oncomapa qPCR panel, a targeted reverse-transcription quantitative PCR assay designed for clinical molecular profiling. The retrospective study drew on diagnostic tests performed between January 2021 and July 2023 on samples fixed in buffered formalin for at least 48 hours. For each spleen, pathologists sampled at least seven distinct regions histologically—a deliberate strategy to reduce the risk of misclassification in a tumor notorious for focal variation. Two senior pathologists confirmed every diagnosis, evaluating cellular pleomorphism, necrosis, capsular invasion and mitotic counts.

Technically, the workflow was built for robustness on degraded material. RNA extracted from the FFPE specimens was reverse-transcribed and quantified in technical duplicates for 45 cancer genes plus three housekeeping controls: GAPDH, RPL13A and RPS5. RefFinder analysis, which integrates GeNorm, NormFinder, BestKeeper and the ΔΔCT method, identified GAPDH as the most stable reference gene, and all targets were normalized to the geometric mean of the three controls using the 2^-ΔΔCT method. Genes were flagged as highly or lowly expressed when their log2 fold-change exceeded ±2 relative to the cohort average—equivalent to at least fourfold deviation. Missing cycle-threshold values, a common nuisance with fragmented FFPE RNA, were imputed with Multiple Imputation by Chained Equations, and the data were Z-score normalized before clustering. Notably, DHFR was undetectable in 23 of 27 samples, a finding with direct clinical relevance because DHFR expression is associated with methotrexate resistance.

With the expression matrix in hand, the team deployed an unsupervised machine-learning pipeline. Principal component analysis showed that the first two components explained 21.5 and 15.7 percent of the variance respectively, with the first three components accounting for roughly half of the total variation. K-means clustering, validated by the elbow method on within-cluster sum of squares and by an average silhouette score of 0.17, pointed to three clusters. Ward’s hierarchical clustering with Ward.D2 linkage independently reproduced the same three-group structure, with a cophenetic correlation coefficient of 0.68 confirming a good fit between the dendrogram and the underlying distance matrix. Kruskal-Wallis tests with false discovery rate correction then isolated the genes driving the separation.

The three signatures that emerged are strikingly coherent. Cluster 1, comprising eight tumors, was marked by elevated expression of MET, MAP2K1/2, FLT3, SRC, AKT2, TP53, PDGFRA, SETD2 and NR3C1—a profile indicating activation of the MAPK and PI3K/AKT cascades, engagement of specific tyrosine kinase receptors including MET, FLT3 and PDGFRA, and stress and immune regulation mediated by NR3C1 and TP53. This aligns with prior evidence that PIK3CA pathway alterations occur in roughly 46 percent of canine hemangiosarcomas, and with in vitro work showing that the PI3K inhibitor alpelisib is active against PIK3CA-mutated canine HSA cell lines. Significantly, this group also showed low expression of the targets of conventional chemotherapy, suggesting a built-in resistance to standard protocols.

Cluster 2, the largest group with ten tumors, overexpressed BCL2, NOTCH1, RET and KIT. This signature points to evasion of apoptosis and enhanced cell survival, and it highlights some of the most druggable targets in the entire study. The authors propose that RET and KIT signaling could be attacked with veterinary-approved tyrosine kinase inhibitors such as toceranib and masitinib, while elevated BCL2—an anti-death protein exploited by many cancers—makes the tumors rational candidates for venetoclax, a BH3 mimetic that has already shown activity against canine hematological cancers in recent studies. Cluster 3, with nine tumors, displayed a hyperproliferative phenotype: overexpression of MYC, TOP2A, RRM2, TYMS and BRCA1 signaled runaway cell-cycle activity, DNA synthesis and DNA repair. Consistent with the molecular data, these tumors carried significantly more mitotic figures than the other groups—24.43 versus roughly 11 per high-power field, a statistically significant difference. Their gene signature also suggests vulnerabilities to existing cytotoxics: doxorubicin targets TOP2A, hydroxyurea and gemcitabine hit RRM2, and 5-fluorouracil or capecitabine exploit TYMS dependence.

The authors are careful to frame these therapeutic links as hypotheses rather than prescriptions. Because the study was retrospective, the dogs were not treated according to cluster assignment, and standardized survival data were not available, the work cannot yet demonstrate that cluster-guided therapy improves outcomes. Bulk RNA from FFPE blocks also averages the signal across heterogeneous cell populations, meaning intratumoral diversity—well documented in solid tumors—could blur the boundaries between subtypes. The cohort of 27 dogs was small and not validated against an independent external dataset, and protein-level confirmation through immunohistochemistry was not possible with the available archived material. The team argues that the natural next step is prospective validation in larger, independently annotated cohorts, following the established trajectory of translational biomarker development that transformed human oncology after landmark expression-profiling studies in breast cancer and diffuse large B-cell lymphoma.

Even with those caveats, the study carries real weight for the emerging field of veterinary precision oncology. It demonstrates that clinically actionable molecular stratification is feasible using routine FFPE tissue and a comparatively inexpensive qPCR panel, bypassing the cost and infrastructure demands of next-generation sequencing. Expression profiling captures functional pathway activity that DNA mutation panels often miss, and it can in principle track therapy resistance dynamically. Because off-label use of human drugs is already common in veterinary medicine, and because agents such as venetoclax and sorafenib are increasingly deployed against canine tumors, the distance between a molecular signature and a treatment trial is shorter than ever. Given the documented molecular parallels between canine hemangiosarcoma and human angiosarcoma, a better molecular map of the canine disease may ultimately inform the human one as well—turning one of veterinary medicine’s most feared diagnoses into a test bed for personalized cancer therapy.

Subject of Research: RNA expression profiling reveals three molecular subtypes of canine splenic hemangiosarcoma for precision oncology

Article Title: Molecular profiling by RNA expression reveals distinct subtypes of canine splenic hemangiosarcoma

Article References: Fiebig, S. P., Ferrero, A. T., Nunes, A. T., Xavier, P. L. P., Qazi, T. J., Strefezzi, R. F., Sueiro, F. A., Rodrigues, L., & Fukumasu, H. (2026). Molecular profiling by RNA expression reveals distinct subtypes of canine splenic hemangiosarcoma. Veterinary Oncology, 3(1), Article 15. https://doi.org/10.1186/s44356-026-00064-5

Image Credits: AI Generated

DOI: 10.1186/s44356-026-00064-5

Keywords: canine hemangiosarcoma, gene expression profiling, molecular subtypes, precision veterinary oncology, qPCR, FFPE, PI3K/AKT pathway, tyrosine kinase inhibitors, comparative oncology, targeted therapy, dog cancer, transcriptomics