Fig 1.
The immune environment across canine cancer types.
(A) Normalized median abundances of immune checkpoints (ICs). Notably, ICs such as TIM-3 and GAL-9 are universally expressed, while others, such as GITR, B7-H4, and NECTIN4, are variable. The gray accessory plots summarize the gene expression per cancer and per IC, with lymphomas showing the strongest total IC expression and SIRPA/CD47 exhibiting the highest total expression. (B) This heatmap displays the abundance scores for the IC categories. The bar plot above illustrates the ratio between inhibitory and stimulatory scores, with a ratio of 1 indicating an equal balance. Notably, MC, UC, LC, GLM, and HSA (see abbreviations) had high ratios, suggesting a potentially more inhibitory environment. (C) A heatmap of standardized scores for various immune cell populations showing the diversity in immune infiltration across different cancer types.
Fig 2.
Dendrogram representing the hierarchical clustering of a distance matrix that compares canine and human cancers with respect to their immune checkpoint (IC) expression levels.
Cancers grouped predominantly within their respective species (bold font - canine, normal font - human). However, notable interspecies similarities emerged, particularly between canine and human gliomas and between canine osteosarcoma and human sarcoma. The figure also illustrates that human cancers tend to cluster according to their histological subtype and primary cancer site. Colors distinguish 14 different clusters. For detailed distance values, refer to S4 Table.
Fig 3.
Principal component analysis (PCA) of immune checkpoint (IC) signatures in human and canine cancers.
(A) Optimal dimensionality determination for PCA using the elbow method. (B) Visualization of the top five principal components (PCs). (C) Clear species separation along the second principal component (PC2). (D) Component loadings for PC1 to PC3, highlighting four influential ICs (CD160, A2AR, NKG2A, and OX40) contributing to PC2. (E) StringDB-based analysis showing that immunoregulatory function was the only feature shared among the four influential ICs. (F) Dendrogram resulting from clustering of signatures consisting solely of the 4 identified genes, demonstrating nearly complete separation of cancers by species.
Table 1.
Median-based differential IC expression between human cancer type clusters. Genes that were found to be significant in both the median- and mean-based analyses are marked in bold to indicate increased confidence. Fc - fold change, log2_fc - Log of fc, pval - p value, HCC - hepatocellular carcinoma, MEL - melanoma, RCC - renal cell carcinoma, SCC - squamous cell carcinoma, TCC - transitional cell carcinoma.
Table 2.
Median-based differential IC expression comparing human vs canine cancer types. Gene that was found to be significant in both the median- and mean-based analyses is marked in bold to indicate increased confidence. Fc - fold change, log2_fc - Log of fc, pval - p value.
Fig 4.
Individual IC abundance signatures via UMAP representation underscore cancer type-specific patterns and intrinsic heterogeneity.
(A) Global overview of the signatures distinguishes two primary clusters: canine and human, highlighting species-dependent variation. (B) Glioma cases from both species are closely situated, signifying a shared IC signature landscape. (C) Inspection of brain cancer cases from both species reveals not only mutual proximity but also marked uniformity within each cancer type. (D) Examination of sarcoma cases unveils a proximity between human sarcoma and canine osteosarcoma, while canine hemangiosarcoma exhibits a more distant relationship. (E) Human cholangiocarcinoma displays notable case dispersion, indicative of a high degree of heterogeneity in the IC signature. (F) A cross-species comparison of prostate cancer evidences a lack of coclustering, suggesting distinct IC signatures for these species-specific prostate cancers.
Fig 5.
PCA representation of individual IC signatures validates and augments the insights from UMAP analysis, revealing unexpected similarities between human chronic lymphocytic leukemia (CCL) and canine cancers.
(A) Consistent with the UMAP findings, the global overview reveals two primary clusters: canine and human. (B) Notably, the PCA revealed a surprising finding missed by UMAP: patient signatures of human CCL, a B-cell blood cancer, exhibit a striking similarity to canine cancers along the PC2 axis, which accounts for between-species differences. This finding suggests a resemblance between human CCL and canine malignancies in the context of the identified key immune checkpoint genes. (C) Further examination of brain cancer cases reaffirms the mutual proximity observed in UMAP, which is particularly evident in gliomas, and the overall uniformity of signatures within each cancer type. (D) Human sarcoma and canine osteosarcoma, while displaying variability along PC-1, demonstrate alignment with each other, further supporting the concordance between species. (E) Similar to UMAP, human cholangiocarcinoma exhibits notable case dispersion, reflecting the high degree of heterogeneity in IC signatures. (F) Consistent with previous observations, PCA confirmed that human and dog prostate carcinomas display distinct IC signatures, as they do not align along PC-1.
Fig 6.
Immune Checkpoint Expression Patterns Differ Between Cancers and Their Corresponding Normal Samples.
UMAP visualizations of IC gene expression patterns in individual samples of five different cancer types along with corresponding healthy tissues. (A) Renal cell carcinoma, (B) chromophobe renal cell carcinoma, (C) lung adenocarcinoma, (D) cholangiocarcinoma, and (E) hepatocellular carcinoma. Distributions provide insights into differential IC landscapes between cancerous and normal tissues, with a notable shift in the case of renal carcinomas and lung adenocarcinoma.
Fig 7.
Differential Immune Checkpoint Expression Patterns Between Cancers and Normal Samples Unveiled Through PCA and Component Loadings Analysis.
PCA biplots (PC-1 vs PC-2) present the IC expression distributions for the same cancer types and corresponding healthy tissues as in Fig 6. (A) Renal cell carcinoma, (B) chromophobe renal cell carcinoma, (C) lung adenocarcinoma, (D) cholangiocarcinoma, and (E) hepatocellular carcinoma. (F) A plot of component loadings for PC-1 to PC-3, emphasizing the genes CD86, TIGIT, ICOS, and BTLA, which show the most substantial contributions to the PC-1 component. These genes deserve further scrutiny for their potential roles in cancer immunotherapy, particularly in renal cell carcinomas.