Canine mammary carcinoma (CMC) is a frequent neoplasm in unspayed dogs with substantial mortality and has been proposed as a comparative model for human breast cancer. This study performed an integrated RNA‑Seq bioinformatics analysis combining a newly generated dataset (CPA‑UN) with three publicly available GEO RNA‑Seq datasets to identify recurrent differentially expressed genes (DEGs) between CMC tissues and matched adjacent healthy mammary tissue. The primary aim was to detect candidate genes with potential diagnostic or prognostic value and to characterize associated biological pathways.
The integrated analysis included transcriptomic data from 88 matched CMC and adjacent healthy mammary tissues drawn from four datasets: CPA‑UN (newly generated in this study), GSE119810, GSE136197, and GSE135183. The CPA‑UN cohort initially recruited 12 suspected CMC cases; 10 were confirmed as carcinoma after histopathological review and processed for RNA extraction. Tumor and adjacent normal tissue samples were collected intraoperatively, stabilized in RNAlater, and stored at −70°C until processing. Public datasets comprised 47 cases (GSE119810), 16 cases (GSE136197), and 15 cases (GSE135183); GSE135183 uniquely derived from laser‑capture microdissected stromal FFPE tissues, while the others used fresh‑frozen tissue.
Total RNA extraction, quality assessment, library preparation with TruSeq Stranded Total RNA kit, and sequencing on Illumina platforms (including NovaSeq X Plus) were conducted for CPA‑UN. Raw reads were quality‑checked with FastQC and processed with Trimmomatic for trimming. Reads were aligned to the canFam4 reference genome using HISAT2 and quantified with HTSeq‑count using strand‑specific parameters. Public raw datasets were reprocessed with the same pipeline to preserve comparability. CPA‑UN raw FASTQ files and project accession details were deposited in the NCBI SRA.
Each dataset underwent independent differential expression analysis using DESeq2 after filtering out invariant genes and appropriate normalization. Significantly differentially expressed genes were defined by log2 fold change ≥1 or ≤−1 with adjusted p‑value (padj) <0.05. An integrative Venn‑based approach identified recurrent DEGs consistently observed across the four datasets. Eight genes were common to all datasets: five consistently upregulated in CMC (ACAN, COL11A1, EDIL3, NDUFA4L2, IGFBP5) and three consistently downregulated in CMC (TNNC1, PCK1, METTL24).
Functional enrichment was performed initially on the CPA‑UN dataset and then on the combined gene lists across datasets using ClusterProfiler and Pathview. Upregulated DEGs were predominantly associated with extracellular matrix (ECM) remodeling, cell adhesion, tumor microenvironment interactions, immune and inflammatory responses, and signaling pathways implicated in tumor progression and metastasis. Downregulated DEGs were enriched for cytoskeletal organization and tissue structural integrity processes, consistent with loss of normal mammary gland architecture and myoepithelial functions during tumorigenesis. Enrichment significance was assessed with hypergeometric tests and FDR correction; terms with padj <0.05 were considered significant.
GSEA was conducted using normalized CPA‑UN counts and also on the combined batch‑corrected matrix. The analysis employed the MSigDB Hallmark collection and gene set permutation mode with 1,000 permutations. Tumor‑associated hallmark signatures enriched in the CMC phenotype included programs related to cell‑cycle dysregulation, proliferation, epithelial–mesenchymal transition (EMT), metabolic adaptation, inflammatory signaling, and stromal remodeling. Upregulated hallmark gene sets were reported using FDR thresholds (reported criteria: FDR <25% and p <0.05), supporting coordinated transcriptional programs involved in tumor progression and tumor microenvironment remodeling.
Co‑expression network analysis identified a highly modular architecture with densely interconnected clusters of co‑expressed genes. These modules likely reflect coordinated biological processes and regulatory programs active in CMC. The modular organization suggests that subsets of genes act in concert to drive specific tumor‑related processes; however, specific hub genes or causal regulators were not claimed without further functional validation.
Immune deconvolution using CIBERSORTx estimated the relative abundances of 22 immune cell types in tumor and matched normal samples. After correction for multiple comparisons, no statistically significant differences in immune cell proportions were detected between CMC and paired healthy mammary tissues. Dataset‑specific trends were observed for regulatory T cells, M1 macrophages, activated CD4 memory T cells, plasma cells, dendritic cells, and mast cells, but these trends did not survive multiple‑testing correction and were reported as exploratory findings requiring cautious interpretation.
An exploratory overall survival analysis was performed using the GSE119810 dataset. From 1,759 DEGs screened, 53 genes showed nominal associations with overall survival; however, none retained significance after correction for multiple testing. These results are explicitly exploratory, emphasizing the need for validation in larger cohorts before any prognostic claims can be made.
The study presents a preliminary transcriptomic framework for CMC by integrating independent RNA‑Seq datasets and identifying recurrent DEGs and enriched pathways. The authors emphasize that findings are exploratory: limitations include cohort sizes, potential variability in tissue collection and processing across studies, and the secondary analysis nature of public datasets. The identified candidate genes and pathways merit functional validation and testing in larger, independent cohorts to determine diagnostic or prognostic utility. The study supports comparative oncology approaches and suggests avenues for future mechanistic and translational work using the canine model to inform both veterinary and human breast cancer research.