High-resolution single-cell RNA sequencing using canFam4 reveals novel immune subsets and checkpoint programs in healthy dogs

Abstract

Introduction

Single-cell RNA sequencing (scRNA-seq) enables high-resolution profiling of immune heterogeneity. Although previous studies have mapped the single-cell transcriptomic atlases of peripheral leukocytes in healthy dogs, the identification and functional characterization of distinct immune subsets remain incomplete.

Methods

We constructed a single-cell atlas of peripheral leukocytes from six healthy small-breed dogs using the 10x Genomics platform and the updated canFam4 genome.

Results and discussion

Analysis of 30,040 high-quality transcriptomes revealed 51 distinct immune subsets, including CD14⁺CD33⁺ monocytes, XCR1⁺CD1D⁺ dendritic cells, CEACAM1⁺CD24⁺ neutrophils, and IL32⁺BATF⁺ regulatory T cells, which were underrepresented in canFam3.1-based studies. Interferon-enriched CD14⁺ monocytes and CD4⁺ T subsets associated with myxomatous mitral valve disease were also identified. Functional enrichment analyses suggested that PDCD1 is associated with attenuated TCR signaling, whereas LAG3 was associated with malate metabolism pathways in CD4⁺ T cells and reduced TBX21 expression in CD8⁺ T cells linked to antiviral responses. CD274, which encodes PD-L1 was linked to IL-10 production in neutrophils, and CTLA4 represented an initial activation of double-negative T subsets. T cell exhaustion scores and proliferative fractions varied across cohorts, reflecting differences in environmental antigenic exposures.

Conclusion

To our knowledge, this study represents the first comprehensive, gene-resolved single-cell analysis to reveal immunoregulatory checkpoint mechanisms underlying immune homeostasis in healthy dogs. Our dataset will serve as a valuable resource for future comparative and translational immunology research in dogs.

Introduction

Humans and dogs inextricably share not only genetic traits but also environments, lifestyles, stressors, and microbial exposures (1, 2). Moreover, dogs have many of the same naturally occurring diseases as humans (3), including cancer, inflammatory bowel disease, and cardiomyopathy (2, 4). Consequently, companion dogs serve as valuable spontaneous models for translating basic science into clinical applications (5); however, cross-fertilization between veterinary and human medicine is most advanced in oncology (6).

Single-cell RNA sequencing (scRNA-seq) dissects dynamic and heterogeneous tissue microenvironments by characterizing the transcriptome at the single-cell level, thereby improving our understanding of cell identity, fate, and function in the context of both normal biology and pathology (7). To date, scRNA-seq has begun to unlock the secrets of veterinary diseases with application to canine cells derived from blood (8–13), lymphoid tissues (14–16), bronchoalveolar lavage (17), hippocampus (18), liver (16), lung (19), adipose tissues (20), inflamed tissues (12, 21–23), and cancers (10, 24–27). More recently, the single-cell transcriptome atlas has elucidated canine hematopoiesis (28) and pathogenesis of peri-implantitis via pseudotime and interactome analyses (29). Among various sample origins, peripheral blood mononuclear cells (PBMCs) have served as an easily accessible and valuable source for gaining insight into the tumor microenvironment (8, 13), lymphocyte clonal expansion (10), inflammatory disease etiology (23), innate immunity (16), aging (20), and preclinical immunotherapy optimization (12).

Although the canine genome has been sequenced, the commonly used reference genome (canFam3.1) remains incomplete (30), with gaps in transcript annotation, the absence of key immune-related genes, e.g. CD14, FCGR3A, CD1 family, and a portion of scRNA-seq reads that remain unmapped due to deficient or partial genomic regions (8–12). Recently, new canine genome references, such as GSD_1.0 (canFam4), have been released, offering potential for improved transcriptional resolution with high contiguity (30). Therefore, evaluating canFam4 in the context of scRNA-seq analysis may enhance single-cell recovery and improve immune cell-type identification, yet this has not been systematically examined to date.

Prior studies have profiled canine leukocytes using scRNA-seq (8–10, 23). However, the identification and functional characterization of distinct immune subsets remain incomplete, and it is unclear whether all clinically relevant cell states have been fully captured. To address these gaps, we performed scRNA-seq on PBMCs from clinicopathologically healthy, client-owned, indoor small-breed dogs. To our knowledge, this study demonstrates, for the first time, that canFam4 substantially increases cell recovery and enables the detection of previously unannotated markers, such as CD14. In addition, we provide immune subsets potentially involved in disease and regulatory mechanisms of immune checkpoint genes, such as PDCD1, CTLA4, LAG3, and CD274 in dogs. To our knowledge, this is the first scRNA-seq study to provide molecular evidence of how canine immune subsets maintain homeostasis. Our dataset and methodology establish a foundational resource for future investigations into canine cancers and other immune-related disorders.

Materials and methods

Study subject enrollment and inclusion criteria

Client-owned adult to geriatric dogs (7 to 12 years old) were enrolled from the Veterinary Medical Teaching Hospital at Jeju National University (Jeju-si, South Korea), comprising two recognized breeds, Maltese (n = 2) and Poodle (n = 1), as well as mixed-breed dogs (n = 3). All dogs were housed indoors and confirmed by their owners to have no preexisting disease conditions. Inclusion criteria included absence of clinical signs of disease on physical examination, normal clinicopathologic results, and no vaccinations or treatments within four weeks before blood sampling. Written informed consent was obtained from each owner before enrollment. The study protocol was reviewed and approved by the Jeju National University Institutional Animal Care and Use Committee (IACUC No. 20230072).

Clinicopathologic examinations

Resident veterinarians conducted thorough physical examinations and clinical assessments. Hematological and biochemical analyses were performed using the ProCyte Dx and Catalyst One systems (IDEXX Laboratories, MA, USA), respectively. To screen for Babesia infection, we employed both a point-of-care antibody rapid test (Canine Babesia Antibody Rapid Kit, BioNote Inc., Gyeonggi-do, South Korea) and real-time PCR (CareDx™ Canine Babesia Real-Time PCR Kit, Carevet Inc., Gyeonggi-do, South Korea). Additionally, Anigen Rapid CaniV-4 (BioNote Inc.) and SNAP 4Dx Plus (IDEXX Laboratories) assays were used to detect other protozoal pathogens. Only dogs testing negative for all infectious agents were included in the study, and six healthy dogs were selected for study.

Isolation of peripheral blood mononuclear cells

Blood was collected via jugular venipuncture and processed immediately. PBMCs were isolated by density gradient centrifugation using SepMate-15 tubes (Stemcell Technologies, Vancouver, Canada) and Ficoll-Paque PLUS (GE Healthcare, Chicago, IL, USA). The PBMC fraction was washed twice with Dulbecco’s phosphate-buffered saline (DPBS; Thermo Fisher Scientific, Waltham, MA, USA) containing 2% heat-inactivated fetal bovine serum (FBS; Life Technologies, Pleasanton, CA, USA). A small aliquot of the suspension was cytospun onto glass slides and stained with Diff-Quik to confirm the purity and composition of isolated cells. The remaining cells were cryopreserved in a cytoprotective medium (Cellbanker 1, Zenogen Pharma, Koriyama, Japan) and stored in liquid nitrogen for further analysis.

Fluorescence-activated cell sorting

Thawed PBMCs were washed and incubated with an anti-dog Fc receptor blocking reagent (Invitrogen, eBioscience, San Diego, CA, USA) on ice for 10 minutes. After washing with ice-cold DPBS containing 2% FBS, cells were stained for 30 minutes at 37 °C, protected from light, with Fixable Viability Dye eFluor 780 (Invitrogen, eBioscience), anti-dog CD45 PE (clone YKIX 716.13; Bio-Rad Laboratories, Hercules, CA, USA), and anti-dog CD3 FITC (clone CA17.2A12; Bio-Rad Laboratories). Live single cells were then sorted on a FACSAria III (BD Biosciences Pharmingen, San Diego, CA, USA), gating on CD45⁺CD3⁺ T cells and CD45⁺CD3^- non-T leukocytes. Post-sort viability and cell counts were assessed using a LUNA-FL™ Automated Fluorescence Cell Counter (Logos Biosystems, Anyang, South Korea). Equal numbers of CD45⁺CD3⁺ and CD45⁺CD3^- cells from each separate dog were pooled within the same animal into a new tube, which immediately processed for single-cell library construction on the 10x Genomics Chromium platform in a single run.

10X genomics single-cell RNA sequencing library construction

Single-cell libraries were prepared using the Chromium Next GEM Single Cell 5′ Kit v2 (10x Genomics) following the manufacturer’s protocol (31, 32). Libraries were sequenced on an Illumina NovaSeq 6000 with 2 × 150 bp paired-end reads. Base calling and FASTQ generation were performed with bcl2fastq (Illumina) and cellranger mkfastq (Cell Ranger v8.0.0, 10x Genomics). Reads were aligned and quantified using cellranger count by using two canine genome references: CanFam3.1 (GCA_000002285.3) and GSD_1.0/CanFam4 (GCF_011100685.1). Genome references were constructed with cellranger mkref using filtered FASTA and GTF files containing only protein-coding genes.

single-cell RNA sequencing data integration, initial preprocessing, and sub-clustering

All single-cell data were processed and analyzed in Seurat v5.1.0 (R v4.2). After loading each sample, low-quality cells were filtered out if they met any of the following criteria: fewer than 200 or more than 4000 unique features and a mitochondrial gene content of 10%. Filtered datasets were normalized using Seurat’s default workflow. Next, all six samples were merged and integrated into a single Seurat object using 3,000 anchor features. Principal component analysis (PCA) was performed for dimensionality reduction, and principal components (PCs) 1–30, which were selected based on JackStraw significance, p-value < 1e-5, and percentage variance explained, were used for downstream clustering. We applied graph-based clustering on both t-distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) embeddings, using a final resolution of 3.7 that was empirically determined through iterative evaluation of cluster separability to preserve biologically meaningful immune subsets. The simultaneous use of UMAP and t-SNE follows previous immune profiling strategy that combined both algorithms to provide complementary views of global and subset-level cellular structures (33). To determine cohort-specific phenotypes or validate our findings, we incorporated publicly available canine scRNA-seq datasets, including peripheral blood TCR αβ T cells (GSE218355) (9) and PBMCs (GSE225599) (8). FASTQ files from GSE144730—canine PBMCs in atopic dermatitis (23)—were not included due to unsupported library chemistry. Available external datasets were processed and integrated alongside our own using the same quality control, normalization, and anchor-based integration pipeline. Harmony integration (v1.2.0) was performed on the PCA-reduced data to correct batch effects across studies. The first 30 Harmony dimensions were subsequently used for neighbor finding, clustering, UMAP, and t-SNE visualization. Doublets were detected and removed with scDblFinder v1.18.0 (31). Single-cell clusters were annotated by canonical lineage markers, published signatures for rare populations, and unbiased cell-type inference via SingleR v2.6.0 (34) using celldex v1.14.0 reference panels. Finally, the escape v2.0.0 package (35) was used to calculate enrichment scores for specific immune gene signatures derived from original canine studies (8, 22). Additionally, scRNA-seq datasets GSE225599 and GSE252470, generated from circulating leukocytes and tumor-infiltrating immune cells in canine osteosarcoma, were analyzed independently as an external cohort to validate our findings.

Differentially expressed gene analysis

Differential expression was assessed using Seurat’s likelihood-ratio test, comparing each cluster to all other cells. Marker genes for individual clusters were identified with FindAllMarkers, applying cutoffs of an absolute log₂-fold change > 0.25 and ≥ 25% of cells in the cluster expressing the gene. To compare cluster markers and differentially expressed genes (DEGs) across experimental groups, we used FindMarkers with a more stringent threshold (absolute log₂ fold change > 0.5 and p < 0.05).

Gene set enrichment analysis and gene ontology analysis

We performed gene set enrichment analysis (GSEA) using the escape R package, employing Hallmark gene sets from the Molecular Signatures Database and those from previous canine studies (36–40). DEGs were further analyzed for Gene Ontology (GO) enrichment using both the PANTHER annotation (for Canis lupus familiaris) and the ShinyGO v0.80 web tool with species set to dog. GO terms were considered significant at p < 0.05 and false discovery rate (FDR) < 0.05. Pathway and gene set visualizations were generated with DittoSeq v1.4.4 and pheatmap v1.0.12.

Module score calculation

Module scores were calculated using the AddModuleScore function in Seurat. Group differences were assessed using the Wilcoxon rank-sum test, and p-values were adjusted by the Benjamini–Hochberg method. Gene sets for chemokine (CCL2, CCL3, CCL4, CCL7, CCL20, CCL23, CXCL8), extracellular matrix (ECM) remodeling (S100A4, MMP9, ITGA4, ITGB2, LGALS3, COL1A1, COL1A2), TNF-superfamily (TNF, NFKB1, RELA, NFKBIA, TNFAIP3, TNFSF13), and exhaustion (PDCD1, LAG3, HAVCR2, TIGIT, TOX, ENTPD1) pathways were used.

Trajectory analysis

Pseudotime analysis was performed using Slingshot v2.12.0 based on UMAP embeddings and Seurat-derived clusters, as previously described (35). RNA velocity analysis was performed using Velocyto (v0.17.16) and scVelo (v0.3.3) on loom files generated from CellRanger BAMs aligned to the CanFam4. Spliced/unspliced transcripts were quantified, normalized in Scanpy, and velocities were estimated with the dynamical model on the Seurat-derived UMAP embedding. Naïve T and B cells were specified as the trajectory root. Multiple lineages were inferred, and gene expression dynamics were analyzed along lineage-specific pseudotime axes. The statistical significance of the gene expression trends along pseudotime was assessed using generalized additive models (GAMs), and FDR-adjusted p-values < 0.05 were considered significant.

Cell cycle analysis

Cell cycle phase was assigned using Seurat’s CellCycleScoring function with the cc.genes.updated.2019 gene set, following established protocols (31, 32).

Cell-to-cell interaction analysis

Intercellular communication was inferred from scRNA-seq data using the CellChat R package v2.1.2 (41). Group-specific CellChat objects were merged into a single master object for comparative analysis. Overall, signaling networks were visualized with the netVisual_heatmap function. To identify and display significant ligand–receptor pairs, we applied the subsetCommunication and netVisual _bubble functions using thresholds of ligand log₂-fold change > 0.2, as well as a receptor log₂-fold change > 0.1, and p < 0.01.

Statistical methods and reproducibility

Statistical tests were performed primarily within Seurat v5.1.0. For single-cell differential expression analyses, only the non-parametric Wilcoxon rank-sum test was used via FindAllMarkers and FindMarkers. One-way ANOVA and two-tailed unpaired Student’s t-tests (with Welch’s correction when appropriate) were used exclusively to compare log-normalized summary metrics, such as quality control metrics, module scores, and cell proportions, after confirming approximate normality. To evaluate the correlation between two gene signature enrichments, both Pearson’s correlation and simple linear regression analyses were performed using R. Pearson’s correlation coefficient (r) and its statistical significance were computed using the cor.test() function. A simple linear regression analysis was performed using the lm() function. Results were considered statistically significant at p < 0.05.

Data and code availability

The raw and processed scRNA-seq data generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE301630. All custom scripts and pipelines used in this study are publicly available in a GitHub repository at https://github.com/raiora881030-svg/canFam4-canine-PBMC-scRNAseq-scripts. This repository contains the complete R script used for preprocessing, analysis, and visualization of the scRNA-seq data.

Results

Clinical and laboratory evaluations of the health status of the dogs

Six healthy dogs (median age, 9.3 years), including Maltese (n = 2), Poodle (n = 1), and mixed-breed dogs (n = 3), were enrolled following normal physical, hematologic, and biochemical examinations (Supplementary Data 1, 2). All dogs tested negative for infectious diseases and remained clinically healthy for at least six months under veterinary follow-up.

The study workflow and the quality control of the data

An overview of the study design is provided in Figure 1A. A summary of scRNA-seq library construction and multiplexing statistics is provided in Supplementary Data 3. Single-cell transcriptomes from six dogs yielded 30,040 high-quality immune cells after standard preprocessing (Supplementary Figure 1A–E). Dimensional reduction confirmed comparable transcriptomic structures across breed, age, and sex, ensuring the reliability of pooled analysis. Approximately 15.6% of potential doublets were identified and removed to ensure data integrity for downstream analyses (Supplementary Figures 1F, G).

Figure 1.

Study design and identification of major immune cell clusters by integrated scRNA-seq analysis in dogs. (A) Schematic overview of the study. Live, single CD3⁺ and CD3^- peripheral immune cells were isolated from the jugular vein of 6 clinically healthy dogs, flow-sorted, pooled, and subject to single-cell library preparation and downstream bioinformatic analysis. A representative photomicrograph of a hematoxylin and eosin-stained PBMC smear is shown at 400× magnification (scale bar = 20 μm). Scientific illustrations were created using BioRender under an academic license. Please note that the cellular proportions shown in the figures reflect enrichment and do not physiologically represent the actual proportions of immune cells in canine peripheral blood. (B) A total of 51 transcriptionally distinct clusters were identified in the integrated Seurat object and visualized on a tSNE plot. (C) Canonical lineage markers used to classify major immune cell types are shown on the tSNE feature plot. (D) Representative genes among the top 10 DEGs for each major lineage are visualized on a heatmap. The number of cells analyzed per lineage is indicated below. (E) The t-SNE density plots of the representative canonical marker genes used to identify functionally distinct immune subsets corresponding to the clusters shown in Figure 1B. All marker genes were annotated only in the canFam4 genome reference.

Identification of major immune subsets in circulating leukocytes of healthy dogs

After singlet selection, we identified 25,355 high-quality cells across 51 clusters (Figure 1B). All immune subsets were assigned to each biological replicate (Supplementary Figure 1H). Clustering revealed distinct T, B, and myeloid lineages (Figure 1C), with representative marker genes defining each major subset (Figure 1D). Representative clustering genes used to define functionally distinct immune subsets were present (Supplementary Figure 2 and Supplementary Data 4). The unbiased cell type recognition corroborated our cell type annotation through an enrichment analysis using signatures that define canine and human immune subsets (Supplementary Figures 3A, B). Overall, cell type and subset identification in this study supported previous scRNA-seq results applied to circulating leukocytes of healthy dogs (8–10, 23), identifying major functional subsets, such as regulatory T cells (Tregs) (IL2RA⁺FOXP3⁺), gamma delta (γδ) T cells (RHEX⁺SCART1⁺), plasmacytoid dendritic cells (pDC) (IL3RA⁺TCF4⁺), plasma cells (MZB1⁺JCHAIN⁺), and cycling T cells (cluster 27), characterized by high MKI67 and PCLAF expression and a high proportion of cells in S and G2/M phase (Supplementary Figure 4). It is worth noting that adopting the updated canFam4 enables the discovery of features not reported in previous scRNA-seq studies in dogs, including CD14, CCL23, CEACAM1, and NKG7, which improved annotation of monocyte, dendritic, and lymphocytic populations (Figure 1E).

We then conducted a more comprehensive analysis to investigate reference-specific highly variable features across genome builds. Using the same preprocessing, quality control, and integration pipelines, we applied canFam3.1 to generate a new master Seurat object. Interestingly, we identified 1,490 highly variable features specific to canFam4 (Supplementary Data 5), which were expressed across diverse immune subsets (Supplementary Figure 5A). In addition to revealing novel genes, canFam4 improved scRNA-seq quality control metrics (Supplementary Figure 5B). When applied to the Seurat object, canFam4 significantly increased the number of assigned cells and decreased the proportion of non-human homologous genes. The number of RNA features and the proportion of mitochondrial genes remained consistent across references. In summary, we generated scRNA-seq libraries representing functionally distinct immune subsets in healthy dogs and improved transcriptional resolution and single-cell assignment accuracy by utilizing the canFam4 reference.

Identification and characterization of myeloid subsets

Contrary to humans, granulocytes, polymorphonuclear cells (PMNs), or myeloid-derived suppressor cells (MDSCs) are collected during gradient-based isolation of canine peripheral blood (8, 10, 42). Canine myeloid population is highly heterogeneous; however, our understanding of this population at the single-cell level remains incomplete (8–10, 23). Therefore, we first focused on identifying and characterizing functionally distinct myeloid subsets. Sub-clustering of DPYD⁺ myeloid cells revealed 23 clusters in dogs (Figure 2A). Representative marker genes used for functional classification are summarized in Supplementary Figure 6 and Supplementary Data 6.

Figure 2.

Identification and characterization of myeloid subpopulations by scRNA-seq. (A) UMAP visualization of 23 myeloid subsets classified into classical (CD14⁺) and non-classical (CD14^-) monocytes, granulocytes (PMNs), and dendritic cells (DCs). Representative subsets are labeled. (B) Heatmap of the representative genes among the top 7 DEGs defining each cell type. (C) Density plots of the selected marker genes distinguishing functionally distinct myeloid subsets. (D) Violin plots of the clusters 18 and 20, characterized by the upregulation of the IFN-related genes and CD177, respectively. (E) Dot plot of the canonical markers for functional dendritic cell subsets. (F) GSEA. Immune-related pathways are enriched in DC/classical monocyte clusters (red box; e.g., M1/M2 macrophage, IFN signatures) and in PMN/non-classical monocyte clusters (blue box; e.g., migration, Dog_IMHA, Dog_MMVD signatures). (G) Volcano plot of the DEGs in CD177⁺ PMNs compared to other PMN subsets. Genes associated with Dog_MMVD are highlighted. (H) Scatter plot of the correlation between type I IFN and MMVD gene signature scores in cluster 18. (I) Feature plot of the distribution of S100A4⁺CCL23⁺TNFSF13⁺ monocytes and their negative counterparts. (J) Module score analysis of the TNF/NF-κB, chemokine, and ECM-remodeling gene programs between triple-positive (S100A4⁺CCL23⁺TNFSF13⁺) and triple-negative subsets. (K) Feature plot of the distribution of CD274⁺ PMNs and monocytes. (L) Volcano plot of the representative DEGs in CD274⁺ PMNs compared to CD274^- PMNs. CD274 is highlighted with a log₂FC of 16.4 and p = 0. (M) GO analysis of CD274⁺ PMN-specific DEGs reveals enrichment in biological processes associated with T cell proliferation, glucose starvation, phagocytosis, and IL-10 production. The p-values were derived using one-way analysis of variance implemented in the ggpubr R package, along with Wilcoxon rank-sum tests for pairwise comparisons. * p < 0.05, ** p < 0.01, and *** p < 0.001. Abbreviations: PMNs, polymorphonuclear cells; FC, fold change; DEGs, differentially expressed genes; C, cluster; Mo, monocytes; DCs, dendritic cells; pDCs, plasmacytoid DCs; GO, Gene Ontology; GSEA, gene set enrichment analysis.

Canine myeloid populations comprised granulocytes, monocytes, and dendritic cell (DC) subsets, each exhibiting distinct transcriptional profiles defined by canonical markers such as CD4, CD24, CR2 (CD21/LOC490269), S100A4, CD83, CCL23, TNFSF13, and DLA-DOA (Figure 2B). Among monocytes, we identified S100A4⁺CCL23⁺TNFSF13⁺CD14⁺ clusters representing classical monocytes (clusters 5, 8, 9, 12, 16, and 18) (Figure 2C), which were not previously reported in canine scRNA-seq datasets (8–10, 23). In contrast, clusters 4, 10, and 13 expressing CD83 and CD91 (LRP1) likely corresponded to non-classical or transitional subsets (12, 43). The CD16 or FCGR3A transcript was found to be unannotated in canFam4. A cluster 8 characterized by high expression of interferon (IFN)-stimulated genes (ISG), such as ISG15, MX1, MX2, IFI44, OAS1, and XAF1, represented IFN-stimulated monocytes (Figure 2D) (31). A majority of the IFN-related monocytes showed LGALS9 expression (Supplementary Figure 6B). DC subsets (clusters 15, 21, and 22) expressed GPR183, DLA-DOA, CD86, and PLD4, with all CD1 family genes (CD1A-E) differentially expressed across DC subtypes (Figures 2C–E). Similar to humans, dogs exhibited XCR1⁺ terminally differentiated conventional type 1 DCs (cDC1; cluster 22), TCF4⁺IL3RA⁺ pDC (cluster 21), and ITGAX⁺FCER1A⁺ cDC2 (cluster 15) subsets (44). The XCR1⁺ DC was characterized by significant MIF upregulation (Supplementary Figure 6C). PMNs were marked by CD21 (CR2/LOC490269) and CD4, while CD177, ITGAX, CEACAM1, and CD24 delineated transcriptionally distinct neutrophil clusters (Figures 2C, E). Meanwhile, cluster 14 co-expressed both CD14 and CD3E, meaning that this cluster was likely a doublet.

Next, we functionally characterized how these distinct myeloid subsets contribute to immune homeostasis in healthy dogs (Figure 2F). First, CD14⁺ monocytes and DC subsets were preferentially enriched in inflammatory gene signatures, such as M1/M2 macrophage markers, innate/adaptive immunity, and IFN signaling pathways (Figure 2F, red box). In contrast, neutrophil subsets were preferentially enriched for gene sets related to leukocyte migration involved in the inflammatory response (Figure 2F, blue box). Second, we found that certain myeloid subsets may be associated with immune dysregulation of canine diseases. For instance, CD177⁺ neutrophils showed strong enrichment in gene signatures associated with immune-mediated hemolytic anemia (IMHA) (36) and myxomatous mitral valvular disease (MMVD) (45). There was also a significant positive correlation between IMHA and MMVD signatures in the CD177⁺ neutrophils (Supplementary Figure 6E). Genes such as CD177, CAMP, and MMP9 were significantly upregulated in these neutrophils (Figure 2G). IFN-related monocytes also exhibited MMVD signature enrichment, with notable upregulation of ISG15 and MX1 (Supplementary Figure 6F). Additionally, a moderately strong, significant positive correlation was observed between type I IFN and MMVD gene signatures (Figure 2H). Finally, we further characterized previously unreported CD14⁺ monocytes defined by S100A4⁺CCL23⁺TNFSF13⁺expression. Among CD14⁺ monocytes, the triple-positive and -negative subsets were separated and subjected to module score analyses for chemokine, ECM remodeling, and TNF superfamily programs (Figure 2I). Interestingly, despite high expression of TNFSF13, the triple-positive CD14⁺ monocytes lacked a global TNF/NF-κB activation signature, while exhibiting upregulated chemokine and ECM-remodeling gene programs (Figure 2J).

Anti-PD-L1 antibodies have shown anti-tumor activity in dogs with naturally occurring cancers (46, 47); however, their immunological mechanisms of action in dogs remain insufficiently characterized. We investigated the contribution of CD274, which encodes PD-L1 protein, in myeloid-mediated immune regulation. At steady-state, CD274 was expressed in neutrophils and monocytes but not in DCs (Figure 2K, Supplementary Figure 6G). We separated myeloid cells based on CD274 expression and performed differential gene expression analysis. CD274⁺ neutrophils exhibited 18 differentially expressed genes compared to CD274^- neutrophils (Figure 2L, Supplementary Data 7). GO analysis of these DEGs revealed enrichment in pathways such as αβ T cell proliferation, IL-10 production, phagocytosis, and the inflammatory response, supporting the functional phenotype of PMN-MDSCs (48) (Figure 2M, Supplementary Data 8). Notably, CD47, a part of the IL-10 signature, was significantly upregulated in CD274⁺ neutrophils. Similarly, 172 DEGs defined CD274⁺ monocytes (Supplementary Figure 6G), which were enriched in PD-L1 checkpoint, toll-like receptor, NF-κB, and C-type lectin receptor pathways (Supplementary Figure 6I and Supplementary Datas 7, 8). Taken together, we identified functionally distinct myeloid subsets, demonstrating their potential clinical relevance in regulating immune homeostasis in dogs.

Identification and characterization of CD4⁺ T subsets

T helper subsets play major roles in canine health and disease (49). We next identified and characterized CD4⁺ T cell subsets in healthy dogs (Figure 3A). Cell identities were assigned using canonical markers and canine CD4⁺ T cell references (8, 22, 24), with representative marker genes summarized in Supplementary Figure 7A and Supplementary Data 9. Naïve (SELL⁺CD44^-) (clusters 0, 1, 11, 13, and 15) and effector/memory (SELL^-CD44⁺) (clusters 2, 3, 4, 5, 6, 7, 8, 12, 14, and 17) populations were clearly separated (Figure 3B). Functionally distinct Th1 (cluster 3), Th2 (cluster 6), and Th17 (cluster 5) subsets were identified, exhibiting partial transcriptional overlap (Figure 3C). A transcriptionally unique IFN-related subset (cluster 14) displayed strong enrichment of ISGs, including XAF1, OAS1, MX1, and ISG15 (Figures 3C, D). PDCD1^high cluster 12 represented a putative exhausted phenotype (Figure 3E). Tregs (clusters 7 and 17) expressed canonical FOXP3, IL2RA, CCR4, and CTLA4 alongside BATF and IL32 (Supplementary Figures 7A, B). CTLA4 was broadly expressed across multiple CD4⁺ subsets (Figure 3E). A small CD8A⁺ cluster 9 indicated double-positive T cells, while clusters 8, 11, and 15 remained transcriptionally unclassified but expressed CCDC3, LOC119876465, and STAT3, respectively. Cluster 10, co-expressing CSF3R, DPYD, and S100A8, was likely a doublet population (Supplementary Figure 7B).

Figure 3.

Identification and characterization of CD4⁺ T subpopulations by scRNA-seq. (A) UMAP plot of the 18 distinct CD4⁺ T cell subsets. (B) Cell type recognition using the escape R package. Feature plots show enrichment of canine Th1, Th2, and Th17 gene signatures along with FOXP3 expression. (C) Expression patterns of IC genes are shown on the feature plot. (D) Violin plots of the distinct gene expression in clusters 14 (IFN-related) and 12 (PDCD1⁺CTLA4⁺). (E) Feature and density plots of the expression patterns of PDCD1 and CTLA4 in CD4⁺ T cells. (F) GSEA of the preferential pathway enrichment in effector CD4⁺ T phenotypes (red boxed). IFN-related cluster 14 is strongly enriched for IFN signaling and Dog_MMVD signatures. Treg and PDCD1^high clusters are enriched for terminal differentiation pathways. (G) Volcano plot of the DEGs in IFN-related CD4⁺ T cells compared to other clusters. Dog_MMVD-associated genes are highlighted in bold and boxed. (H) Feature scatter plots of the positive correlation between two gene signatures in the IFN-related CD4⁺ T cells. (I) GSEA reveals that CTLA4⁺ or LAG3⁺PDCD1⁺CD4⁺ T cells are enriched for anti-inflammatory, anergy, and exhaustion signatures. (J) Violin plots of the enrichment pattern of immune-associated gene sets in CD4⁺ T cells with or without PDCD1. (K) GO analysis of LAG3-associated DEGs in PDCD1⁺CD4⁺ T cells reveals enrichment in malate and dicarboxylic acid metabolism, and viral genome replication. (L) Violin plots demonstrates upregulation of LAG3, CTLA4, and TIGIT-specific key genes involved in malate metabolism, Treg differentiation, and IL-27-mediated signaling pathways in PDCD1⁺CD4⁺ T cells, respectively. (M) GO analysis of CTLA4- and TIGIT-specific DEGs in PDCD1⁺CD4⁺ T cells reveals enrichment in Treg differentiation and IL-27-mediated signaling pathways, respectively. Statistical significance was determined by comparing two groups of interest using the non-parametric Wilcoxon rank-sum test. * p < 0.05, ** p < 0.01, and *** p < 0.001. FC, fold change; DEGs, differentially expressed genes; C, cluster; Tregs, regulatory T cells; GSEA, gene set enrichment analysis; GO, Gene Ontology.

We next explored the functional and clinical relevance of these subsets. First, similar to myeloid cells, IFN-related CD4⁺ T cells exhibited marked enrichment of canine MMVD gene signature (blue boxed in Figure 3F), characterized by the upregulation of MX1 and ISG15 (Figure 3G) and a positive correlation between type I IFN and MMVD signatures (Figure 3H, upper panel). A similar pattern was observed in the IFN-related subset regarding mammary complex carcinoma signature, showing upregulation of ISG20 and SAMHD1, along with a corresponding gene set correlation (Figures 3F–H, lower panel). Second, Treg clusters were highly enriched in anti-inflammatory signatures, consistent with their immunosuppressive roles in healthy dogs (50) (red boxed in Figure 3F). Interestingly, upon subclustering of Tregs based on CCR4 expression (Supplementary Figure 7C), CCR4⁺ Tregs were more enriched with gene sets associated with tumor immune evasion and response to immune checkpoint (IC) blockade (Supplementary Figure 7D), suggesting that CCR4⁺ Tregs possess a regulatory program that could become clinically relevant under pathological conditions such as cancer (51, 52). DEGs defining CCR4⁺ Tregs compared to CCR4^- ones were obtained and subjected to GO analysis (Supplementary Data 10), which revealed significant enrichment with the gene set associated with tRNA wobble base modification (Supplementary Data 11). Among genes listed in the signature, CCR4⁺ Tregs showed a significant upregulation in NSUN3, ELP2, GTPBP3, ELP3, and CTU1 but ADAT2 downregulation. Third, PDCD1^highCD4⁺ T cells showed modest enrichment of gene signatures associated with terminal T cell differentiation and exhaustion (Figure 3F), suggesting that healthy dogs have CD4⁺ T cells undergoing transition toward an exhausted phenotype. Finally, CD4⁺ T subsets enriched for mammary gland tumor-associated gene signatures included several naïve, Th1-like, and STAT3^high proinflammatory populations (black boxed in Figure 3F).

In healthy dogs, PD1 blockade has been shown to reverse CD4⁺ T cell suppression induced by tumor-derived PD-L1 (53), but the underlying molecular mechanisms are not well understood. We therefore analyzed the transcriptional profiles of CD4⁺ T cells expressing IC genes. CD4⁺ T cells with or without PDCD1 expression exhibited different enrichment patterns for gene signatures associated with immune activation and exhaustion (Figure 3I). Notably, compared to PDCD1^-CD4⁺ T cells, PDCD1⁺CD4⁺ T cells showed significant enrichment of exhaustion-related gene signatures, suggesting an inhibitory effect of PDCD1 on CD4⁺ T cell immunity (Figure 3J). Similarly, PDCD1⁺CD4⁺ T cells showed downregulation of T cell receptor (TCR)-related genes (ID3, CCR7, FOXP1, and TRAT1) but also showed upregulation of genes such as BATF, ISG20, IFI30, ITGB1, and IDH2 (Supplementary Data 12). Recently, a dog with anti-PD-L1-resistant melanoma exhibited a complete remission of an oral neoplastic lesion following anti-CTLA4 therapy (54). To further elucidate the mechanism of action, we performed differential gene expression and gene set enrichment analyses, which revealed that CTLA4-specific genes, such as FOXP3, TIGIT, HAVCR2, IL10, ADORA2A, and ADA (Supplementary Figure 7E and Supplementary Data 13), were significantly upregulated and enriched in pathways related to negative regulation of the immune system (Supplementary Figure 7F). We next investigated the biological implications of additional IC genes expressed in PDCD1⁺CD4⁺ T cells. Among this population, distinct subgroups co-expressed additional IC genes, including CTLA4 (n = 95), LAG3 (n = 20), TIGIT (n = 11), and HAVCR2 (n = 4). Notably, PDCD1⁺CD4⁺ T cells co-expressing TIGIT, HAVCR2, or LAG3 showed a tendency toward enrichment of an exhaustion signature (Supplementary Figure 7G). DEGs associated with each IC gene are provided in Supplementary Data 14. Given the robustness of differential expression analysis in scRNA-seq datasets (55), GO enrichment analysis of LAG3-, CTLA4-, and TIGIT-specific DEGs was performed. Interestingly, LAG3-specific genes were significantly enriched in biological processes related to malate and dicarboxylic acid metabolism, viral genome replication, and RNA splicing (Figure 3K). Within the metabolic gene set, MDH1 and MDH2 were significantly upregulated in LAG3⁺PDCD1⁺CD4⁺ T cells compared to LAG3^- counterparts (Figure 3L). Similarly, CTLA4- and TIGIT-specific genes were enriched in pathways associated with Treg differentiation and IL-27 signaling, respectively (Figure 3M). In these gene sets, FUT7, OASL, and STAT1 were notably upregulated in IC⁺PDCD1⁺CD4⁺ T cells (Figure 3L). Collectively, our results show the functional diversity and IC heterogeneity of CD4⁺ T cell subsets in dogs, highlighting their potential clinical relevance in immune homeostasis.

Identification and characterization of CD8⁺ and other T subsets

We next performed subclustering of non-CD4⁺ T cells, identifying ten transcriptionally distinct CD8⁺ T clusters (0, 1, 2, 3, 4, 5, 6, 7, 8, and 14), three γδ T clusters (15, 17, 18), two cycling T clusters (9, 12), and one CD4^-CD8A^- double-negative (DN) cluster (13) (Figure 4A). Representative canonical and cluster markers are summarized in Supplementary Figure 8A and Supplementary Data 15. Based on established canine CD8⁺ T cell signatures (8, 22, 24), naïve CD8⁺ T cells (clusters 3 and 5) expressed CCR7, SELL, LEF1, and TCF7. In contrast, effector CD8⁺ T cells (clusters 0, 1, 2, 4, 6, 7, 8, 14) displayed high expression of cytotoxic genes, including GZMB, GZMH (LOC490629), GZMK, and PRF1 (Figures 4B, C, Supplementary Figures 8A, B). Innate-like subsets (clusters 2 and 6) co-expressed FCER1G and PI3 (9), with cluster 6 further enriched in CXCR6 and effector-memory markers (PRDM1, ZNF683, STAT3), consistent with a mucosal-associated invariant T cell-like phenotype. PDCD1⁺LAG3⁺ effector T cells (cluster 7) and cycling/FOXP3⁺CTLA4⁺ subsets (clusters 9, 12, 13) were also identified, reflecting activation or regulatory states (Figure 4D). Interestingly, the CTLA4 expression was largely confined to cycling T (51.6% and 59.7% in clusters 9 and 12, respectively) and DN T cells (54.4% in cluster 13). γδ T cells expressed RHEX and SCART1 (LOC491694), while unclassified clusters 10 and 11 displayed CD44, TOX, and LOC102155278. Cluster 16, co-expressing CD8A, CSF3R, and DPYD, was likely a doublet.

Figure 4.

Identification and characterization of CD8⁺ and miscellaneous T subpopulations by scRNA-seq. (A) UMAP plot of the ten CD8⁺ T, three γδ T, two cycling T, and three unclassified T cell subsets. (B) Feature plots of the enrichment of canine gene signatures associated with naïve, effector, memory, cycling, γδ, and T DN T cells. (C) Feature plot of representative marker gene expression. (D) GSEA of the enrichment of immune-associated pathways across CD8⁺ T subsets. Effector phenotypes (red box) and cycling T subsets (blue box) are enriched in proliferation and exhaustion-related signatures. γδ T cells exhibit mild enrichment in naïve, exhausted, and proliferative signatures. Clusters 2, 4, and 7 show enrichment in cytolytic, proinflammatory, memory, and terminal differentiation pathways. (E) Heatmap of the genes associated with T cell terminal differentiation. PDCD1 and LAG3 are highly expressed in cluster 7. (F) Feature plot of the PDCD1⁺ T cell distribution. (G) GO analysis of PDCD1-specific DEGs reveals enrichment of effector function and cytotoxicity-related biological processes. Feature (H) and violin (I) plots of the LAG3⁺PDCD1⁺CD8⁺ T cells. (J) Volcano plot of LAG3-specific DEGs in PDCD1⁺CD8⁺ T cells. (K) GO analysis of the LAG3-specific DEGs in PDCD1⁺CD8⁺ T cells shows enrichment in antiviral response processes. (L) Violin plots of the representative genes in the antiviral response pathway. All genes except TBX21 are significantly upregulated in PDCD1⁺CD8⁺ T cells. Statistical significance was assessed using the Wilcoxon rank-sum test. * p < 0.05; ** p < 0.01. PMNs, polymorphonuclear cells; FC, fold change; DEGs, differentially expressed genes; C, cluster; T DN, double-negative T; GO, Gene Ontology.

Functional characterization using GSEA revealed several distinct patterns. First, effector CD8⁺ T cells were mainly enriched in proinflammatory, cytotoxic, and response to IC blockade gene sets (red boxed in Figure 4D), indicating functional competence, such as anti-tumor immunity in dogs (56). Second, T cell activation-associated gene signatures were markedly enriched in cycling T cells (blue boxed in Figure 4D), suggesting that cycling T cells may represent a state of steady-state antigenic stimulation (57). Third, although previous studies reported exhausted CD8⁺ T subsets in healthy dogs (9), our data showed that PDCD1⁺CD8⁺ T cells, despite expressing co-inhibitory receptors and being enriched for the T cell terminal differentiation signature (Figure 4E), did not exhibit remarkable enrichment of an exhaustion signature (Supplementary Figure 8C). Fourth, γδ T subsets also showed enrichment in T cell activation along with tumor immune escape signatures, suggesting a potential role in tumor immune surveillance. Lastly, although positioned adjacent to naïve T subsets on the dimensional plot, the DN T subset was enriched for anti-inflammatory and T cell terminal differentiation gene signatures, along with remarkable CTLA4 and TIGIT expression. Additional pseudotime analysis revealed that the DN T subset was localized near the root of the trajectory and was not associated with any terminal lineage (Supplementary Figures 8D). Gene expression dynamics along eight terminal lineages revealed that CTLA4 was transiently upregulated during early pseudotime in multiple lineages, particularly in cell fate 1 and 2 (Supplementary Figure 8F), suggesting DN T subset as an initial activation phase before differentiation.

To further explore the immunoregulatory role of PDCD1 and co-inhibitory receptors in CD8⁺ T cells, we performed DEG analysis between PDCD1⁺CD8⁺ and PDCD1^-CD8⁺ T cells (Figure 4F). PDCD1-specific DEGs (Supplementary Data 16) were enriched in immune effector pathways such as cytotoxicity, cell killing, and defense response (Figure 4G and Supplementary Data 17). Genes, such as BATF, SH2D1A, CORO1A, PRF1, and GZMB were commonly upregulated, while IL7R was downregulated. Meanwhile, similar to CD4⁺ T cells, PDCD1 expression was significantly associated with an inhibitory effect on CD8⁺ T cell immunity (Supplementary Figure 8F). To further investigate functional implications of IC genes, PDCD1⁺CD8⁺ T cells were sub-grouped based on the expression of LAG3 (n = 20) (Figures 4H, I), HAVCR2 (n = 8), CTLA4 (n = 3), and TIGIT (n = 4). Considering robustness of differential expression analysis in scRNA-seq datasets, we focused on LAG3-specific DEGs for downstream functional analysis (Figure 4J, Supplementary Data 18). GO enrichment revealed a significant association with viral response pathways (Figure 4K). Among the enriched genes, PMAIP1, ANKRD17, IRF5, and OASL were significantly upregulated, whereas TBX21 was downregulated (Figure 4L). Meanwhile, PDCD1⁺CD8⁺ T cells with CTLA4 and LAG3 expression, showed a tendency for the enrichment of an exhaustion signature (Supplementary Figure 8G). We next validated our findings using publicly available scRNA-seq datasets from external cohorts of canine osteosarcoma. Using the same procedures for preprocessing, quality control, data integration, and doublet removal as applied to our own datasets, we subclustered CD8⁺ T cells and classified them according to PDCD1 and subsequently LAG3 expression (Supplementary Figures 9A, B). PDCD1⁺CD8⁺ T cells (n = 36) exhibited higher module scores of cytotoxicity, proliferation, and exhaustion compared with their PDCD1^- counterparts (n = 4240) (Supplementary Figure 9C). Notably, LAG3⁺PDCD1⁺CD8⁺ T cells (n = 24) showed elevated module scores of both exhaustion and response to viral programs, further supporting our findings (Supplementary Figure 9D). Taken together, our results reveal transcriptionally distinct CD8⁺ T cell subsets in healthy dogs, with functional profiles supporting roles in immune homeostasis.

Identification and characterization of B subsets

We performed subclustering of CD79B⁺ B cells and MZB1⁺ plasma cells, identifying ten transcriptionally distinct clusters (Figure 5A). Canonical and cluster-defining genes are presented in Figures 5B, C, and representative markers are summarized in Supplementary Data 16. Unbiased annotation using canine B cell references (8, 22, 24) classified subsets into naïve (clusters 0, 1, 2, 3, 6, 9) and plasma (clusters 4, 5, 8) cells (Figure 5D). Naïve B cells expressed LOC100685971, CD79B, BTG1, MS4A1, and MHC class II genes (HLA-DRB1, DLA-DRA), whereas clusters 2, 6, and 9 showed additional expression of GNG2, VPREB3, and MX1, consistent with type I IFN-related B cells (Figures 5D, E). Plasma subsets were characterized by JCHAIN, MZB1, PRDM1, and XBP1 but lacked MHC class II expression (DLA-DMB, DLA-DOA). Distinct transcriptional identities were observed in UBE2C⁺ and OSBPL10⁺ plasma subsets, with the UBE2C⁺ subset displaying a high G2/M cell-cycle score, suggesting proliferative activity under steady-state conditions (Figure 5F). Cluster 7, co-expressing CD3E and DPYD, was excluded as a potential doublet.

Figure 5.

Identification and characterization of B subpopulations by scRNA-seq. (A) UMAP plot of the six B cell and three plasma cell subsets. (B) Heatmap of the top five DEGs defining each subset. (C) Feature plots of the enrichment of naïve and plasma cell gene signatures, along with representative marker expression. (D) Dot plot of the canonical markers used to identify functionally distinct B cell subsets. (E) Cell cycle phase distribution across B cell subsets. (F) GSEA of the enrichment patterns of immune-related pathways across B cell subsets. (G) Volcano plot of the DEGs in IFN-related B cells compared to other B cell subsets. MX1, listed in the Dog_MMVD signature, is highlighted in bold and boxed. (I) Pseudotime ordering of B-cell subsets. (J) Trajectory inference showing bifurcated differentiation paths from naïve B cells toward IFN-related or plasma fates. (K) RNA velocity analysis showing directional flow from naïve B cells toward IFN-related and plasma subsets. Streamlines indicate the inferred transcriptional dynamics based on spliced and unspliced mRNA ratios. (L) Distribution of pseudotime values across B cell subtypes. (M) Expression dynamics of plasma-associated regulators (IRF4, PRDM1, XBP1) along the naïve–plasma trajectory. All three genes showed significant non-linear associations with pseudotime by GAM analysis (FDR < 0.001). (N) IFN module scores plotted along pseudotime, showing selective activation in IFN-related B cells (GAM, FDR < 0.05).

For functional characterization, GSEA revealed significant enrichment in plasma and IFN-related B subsets (Figure 5G, red box). Plasma subsets were enriched with gene sets associated with anti-inflammatory responses and canine immune/metabolic conditions, including IMHA, precursor-targeted immune-mediated anemia (PIMA), obesity, blood sugar, and hyperlipidemia. The IFN-related B subset was highly enriched in canine MMVD signatures, with significant upregulation of MX1 (Figure 5H). A weak positive correlation was observed between type I IFN and MMVD signatures (r = 0.309, p = 0.1723), but it did not reach statistical significance. Given these biologically divergent enrichment patterns, we next examined whether the plasma and IFN-related B subsets represent discrete lineages or instead reflect transcriptionally connected states within a shared differentiation continuum. To address this, these subsets were subjected to pseudotime ordering. Trajectory analysis revealed a continuous transition from naïve B cells toward either IFN-related or plasma fates (Figure 5I, J), which was further supported by RNA velocity analysis (Figure 5K). Naïve and plasma subsets were positioned at opposite ends of the trajectory, while IFN-related B cells occupied an intermediate branch, suggesting bifurcated differentiation (Figure 5L). Expression of plasma cell-associated regulators (IRF4, PRDM1, XBP1) increased significantly and progressively along the naïve-to-plasma lineage (Figure 5M), whereas IFN-related B cells diverged early from the naïve state, selectively upregulating ISG signatures (Figure 5N). Collectively, these analyses demonstrate that canine B cells follow distinct differentiation trajectories leading either to antibody-secreting plasma cells or IFN-driven effector states, underscoring substantial transcriptional heterogeneity and functional diversification within the peripheral B cell compartment.

Cell–cell communication identifies key signaling pathways with ligand and receptor genes in regulating normal homeostasis in healthy dogs

Coordinated crosstalk among immune subsets is essential for maintaining immune homeostasis (31). To explore this, we investigated the cell-to-cell interactions involved in modulating immune equilibrium in healthy dogs. The overall signaling landscape across immune subsets is shown in Supplementary Figure 10A. In healthy dogs, we identified three major patterns of cell-cell interaction (Figures 6A, B).

Figure 6.

Identification of immune cell interactions by scRNA-seq using CellChat analysis. (A) Inferred intercellular communication network highlighting dominant signal senders (red and blue dashed lines) and receivers (black dashed line) visualized on a scatter plot. (B) Global outgoing signaling patterns across identified immune cell clusters. (C–F) Bubble plots of the significant ligand–receptor interactions involved in representative signaling pathways among indicated immune subsets. (G) Feature plots show the expression patterns of representative ligands and receptors involved in MIF, CD99, MHC-II, Selplg, FN1, APP, and galectin signaling pathways.

First, autocrine signaling was strongly inferred among myeloid subsets, including DC (cluster 38), pDC (cluster 49), and CD14⁺ monocytes (clusters 18, 19, 23, and 35) (Figure 6A, red dashed lines, Supplementary Figure 10A). These myeloid subsets also transmitted signals to Treg (cluster 21), cycling T cells (cluster 27), and γδ T cells (cluster 48), suggesting a role in modulating T cell activity. The interactions involved secretory pathways (MIF, RESISTIN, CypA), cell-cell contact (CD99, APP), and extracellular matrix receptor interactions (FN1), through key ligand-receptor (LR) pairs such as RETN:CAP1, FN1:CD44, MIF:(CD74+CD44), and CD99: CD99 (Figure 6C).

Second, neutrophils acted as primary signal recipients, prominently involved in the MHC-II signaling pathway (Figure 6A, black dashed lines, Supplementary Figure 10B). In this context, CD4 was predicted to function as a key receptor, with HLA-DRB1⁺ subsets serving as ligand-bearing partners (Figure 6D). Third, effector immune subsets, including CD4⁺ T cells, CD8⁺ T cells, and plasma cells (clusters 0, 1, 2, 4, 5, 15, 16, 22, 24, 28, 29, 31, and 32), were predominantly engaged in MIF and to a lesser extent CD99 signaling. These involved the LR pairs MIF:(CD74+CD44) and CD99:CD99 (Figures 6A, blue dashed lines, 6E, Supplementary Figures 10C, 10D). Lastly, naïve and DN lymphocytes participated moderately in cell-cell contact (via SELPLG, LCK, and MHC-II) and secretory (GALECTIN, MIF) signaling pathways. Key LR genes included SELL, CD44, CD74, and CD99 (Figures 6A, F, Supplementary Figure 10A). All identified LR genes are listed in Figure 6G. These interactome analyses highlight key ligand-receptor axes and signaling pathways critical for immune crosstalk and homeostasis in healthy dogs.

Integrated scRNA-seq analysis of circulating leukocytes in healthy dogs

Healthy dogs immunologically experience diverse and continuous antigenic stimulation (58), exhibiting a broad and dynamic immune repertoire shaped by environmental exposures (9, 59, 60). To explore cohort-specific immune variation, we performed an integrated analysis of publicly available scRNA-seq datasets of peripheral leukocytes from two independent cohorts of healthy dogs (8, 9). Using a standardized pipeline for preprocessing, quality control, data integration, and removal of potential doublets and dataset-origin bias (Supplementary Figure 11A), we analyzed 90,992 canine immune cells, ultimately identifying 35 transcriptionally distinct clusters in the integrated Seurat object (Supplementary Figure 11B). Cluster-defining marker genes are listed in Supplementary Data 19.

Integration revealed that immune cells from both cohorts generally retained a shared transcriptomic structure (Supplementary Figure 11C). However, proportional differences in immune subsets were observed between cohorts (Supplementary Figure 11D), such as clusters 16, 29, and 33 characterized by gene expression associated with platelets (PPBP), erythrocytes (LOC100855558, a hemoglobin subunit alpha-like gene), and B cells (CD79B), respectively (Supplementary Figure 11E). One particularly study-specific subset, cluster 25, exhibited high expression of cell cycle-related genes (PCLAF, MKI67, and SPC24) and was enriched for a T cell cycling signature (Supplementary Figures 11F, G), suggesting a population of proliferating T cells potentially responding to steady-state antigenic stimuli (57). We further examined cohort-dependent expression of IC genes. Across studies, the proportions of CTLA4⁺ and LAG3⁺ T cells significantly differed (Supplementary Figures 11H, I, Table 1). Consistent with prior observations (61), CTLA4⁺ T cells comprised approximately 25% of circulating T cells in healthy dogs. Additionally, proportions of PDCD1⁺, HAVCR2⁺, TIGIT⁺ T cells, and CD274⁺ myeloid cells differed by cohort. For example, Ammons et al. (8) rarely observed PDCD1⁺ T cells in the cohort; however, PDCD1⁺ T cells represented 4.1% of the overall T cell count in the cohort by Eschke et al. (9). These PDCD1⁺ T cells, associated with IC gene expression, were linked to immune activation, whereas LAG3⁺ and TIM3⁺PDCD1⁺ T cells were enriched for T cell exhaustion signatures (Supplementary Figure 11J). Importantly, T cell exhaustion enrichment scores differed significantly between studies (Supplementary Figure 11K), highlighting distinct immunological imprints shaped by environment, lifestyle, or microbial exposure. In summary, this integrated scRNA-seq analysis uncovers cohort-specific immune features in healthy dogs, providing evidence on how environmental context contributes to immune diversity and checkpoint-related immune phenotypes.

Table 1.

Proportions of immune subset with IC genes across cohorts.

Cell type	Genes	Proportion of immune subset in this study (%)	Ammons et al., 2023	Escheke et al., 2023
T cells	PDCD1	5.7 ± 2.5	0.2 ± 0.1	4.1 ± 1.6
	CTLA4	15.9 ± 3.8	25.2 ± 5.0	32.3 ± 6.2
	LAG3	3.4 ± 0.8	2.5 ± 1.0	6.4 ± 2.6
	TIGIT	0.7 ± 0.6	1.7 ± 0.3	1.3 ± 0.4
	HAVCR2	1.4 ± 0.7	3.1 ± 1.6	1.9 ± 0.3
	CD274	0.7 ± 0.4	1.5 ± 0.6	1.5 ± 0.5
Myeloid cells	CD274	1.8 ± 1.7	5.2 ± 3.2

Discussion

Companion dogs are recognized as valuable translational animal models for studying human diseases. Although emerging studies have applied scRNA-seq to canine peripheral leukocytes (8–10, 13, 16, 20, 23), the identification of functionally distinct and clinically relevant immune subsets, along with the elucidation of their roles and underlying molecular mechanisms in immune homeostasis, remains largely unexplored. To address the existing knowledge gap, we constructed a single-cell atlas of peripheral leukocytes from clinically healthy dogs adopting the canFam4 genome reference for scRNA-seq analysis for the first time. We demonstrate that the canFam4 provides a unique advantage over canFam3.1 by improving single-cell recovery, enabling the discovery of previously uncharacterized immune subsets with potential clinical relevance. To our knowledge, this is the first study to outline transcriptional signatures associated with IC genes, such as PDCD1, CD274, and LAG3, in healthy dogs, providing initial insights into their potential roles in normal immune regulation. We also delineated core transcriptional features of normal B cell differentiation, providing additional reference points for future immunological and clinical interpretations. Finally, we demonstrate that environmental context has great potential to shape cohort-dependent immune phenotypes and diversity. Collectively, our dataset and analytical framework offer a valuable resource for translational immunology in dogs, with implications for cancer research and beyond.

Notably, canFam3.1 has traditionally served as the primary reference for canine transcriptomics; however, recent studies have begun to adopt canFam4 owing to its superior contiguity and improved transcriptional resolution (30, 62, 63). In our dataset, canFam4 increased the number of assignable cells and reduced reads mapping to non-canine homologs, while enabling detection of previously unannotated gene features. These advantages allowed clearer delineation of transcriptionally heterogeneous peripheral immune subsets in dogs. For example, CD14⁺ monocytes have been proposed to resemble CD4⁺ monocytes in dogs (8), yet our data showed minimal CD4 expression in CD14⁺ monocytes, with CD4 instead primarily detected in neutrophils, consistent with prior flow cytometric findings (64). Although CD16 mRNA and protein are clearly detectable in canine immune cells (43, 65), the CD16 gene itself is not annotated in the current canFam4 genome assembly. Consequently, classification of CD16⁺ subsets, such as NK cells and CD14⁺CD16⁺ monocytes, using scRNA-seq mapped to either canFam4 or 3.1 may be limited, meaning the related functional pathway interpretations should, therefore, be approached with caution. In addition to known markers such as CD1c, CD86, CD83, and IL3RA, we also identified features including CD11c (ITGAX), CD1 family genes (CD1A, CD1B, CD1D, CD1E), HLA- DRB1, DLA-DQB1, DLA-64, and CD33, which may refine annotation of DC and granulocyte/MDSC-like subsets in dogs. Likewise, genes such as CD21, CD24, and CEACAM1 may aid in identifying low-density granulocytes or MDSC-like populations. For T cells, GLNY, NKG7, GZMH, IL32, and BATF may improve delineation of effector CD8⁺ T cells and CD4⁺ Treg subsets.

Given that current canine genome references remain incompletely annotated (10), updated assemblies, such as canFam4, may be essential for uncovering the clinical relevance of previously uncharacterized immune subsets (8, 10, 66) and for improving T cell annotation in single-cell datasets. Notably, IL32 and BATF have been implicated in the immunosuppressive functions of tumor-infiltrating Tregs (67, 68), suggesting that IL32⁺BATF⁺ Tregs in dogs may represent a regulatory subset relevant in cancer. Similarly, CEACAM1 and CD24 associated with neutrophil-mediated immunosuppression and T cell tolerance in other species (69–71) were detected in canine granulocytes (72). These findings highlight the value of canFam4-based profiling in identifying immune regulatory programs and suggest that CEACAM1⁺CD24⁺ granulocytes may warrant further investigation in canine models of infection, inflammation, and cancer. By focusing on CD14⁺ monocytes, we further identified a previously unrecognized S100A4⁺CCL23⁺TNFSF13⁺ subset, which may function as tissue-recruiting and matrix-modifying cells rather than classical cytokine-driven inflammatory effectors. Finally, we report the presence of peripheral XCR1⁺ DCs in dogs consistent with findings in humans, mice, and cattle (73–75), which highlights the conserved cDC1 biology across species. Together, these insights underscore the utility of updated genome assemblies, such as canFam4, for revealing clinically and mechanistically relevant immune subsets in canine immunology and translational research (8, 10, 66).

Several immune subsets, including CD14⁺ monocytes, CD4⁺ T cells, and B cells, displayed type I IFN-associated transcriptional programs, with all three populations sharing MMVD-related IFN signatures. These findings indicate a conserved IFN-driven activation axis, suggesting that these subsets might represent a pre-activated immune landscape under steady-state conditions. Similar IFN pathway activation has been reported in canine MMVD, where peripheral leukocytes and valvular tissues exhibit elevated MX1 and ISG15 expression (76, 77). Comparative analysis with published canine datasets confirmed MX1 upregulation in MMVD-affected valves (77) and downregulation of related ISGs in early-stage disease (45), consistent with IFN-mediated antiviral and tissue-protective responses also described in human cardiac tissue (78). IFN-enriched CD4⁺ T and CD14⁺ monocyte subsets, previously characterized in dogs as tissue-patrolling or inflammatory responders (8, 22, 24), may constitute a surveillance network capable of rapid activation under stress or infection. In humans, immune subsets with constitutive expression of IFN-stimulated genes, such as MX1, are considered pre-activated population that can mount rapid responses to malignant cells, pathogens, or autoantigens (31, 79–81). Although the specific role of IFN-related B cells in canine MMVD remains unclear, their transcriptional similarity to type I IFN and MMVD signatures may suggest possible immunological relevance, potentially reflecting a steady-state antiviral-primed B cell state emerging along one of the early activation trajectories. Meanwhile, the CD177⁺ neutrophil subset also overlapped transcriptionally with neutrophil-enriched MMVD and IMHA datasets in dogs (36, 76), both characterized by MMP9 and PTX3 upregulation involved in tissue remodeling and immune activation. Given that our dataset derives from healthy dogs, these shared transcriptional patterns are more plausibly interpreted as primed innate immune phenotypes rather than active pathology. Collectively, our results suggest that IFN-related and CD177⁺ subsets represent pre-activated immune states with transcriptomic overlap to canine MMVD and IMHA profiles. Future validation using disease-cohort single-cell and proteomic datasets will be essential to clarify whether these subsets act as early mediators of immune dysregulation or serve as components of steady-state immune surveillance in dogs.

As in humans, many naturally occurring cancers in dogs exhibit features of immunogenicity, although the magnitude and characteristics of these responses vary across tumor types (82, 83). Recent clinical studies have shown that targeting immune-checkpoint pathways in dogs is feasible and biologically active, and cross-species evidence indicates that canine PD1/PD-L1 signaling is structurally and functionally comparable to that of humans (46, 47, 56, 84–86). Consistent with this, the fundamental inhibitory roles of PD1 and PD-L1 are well established in humans, including PD1-mediated regulation of T cell immunity (87) and PD-L1-induced T cell apoptosis as a tumor immune evasion mechanism (88). However, despite these well-characterized pathways in humans, the canine-specific biological roles and regulatory mechanisms of key checkpoint genes, such as PDCD1, CD274, LAG3, and CTLA4, remain comparatively underexplored. In this study, we leveraged scRNA-seq to begin addressing this gap in the canine immune system. First, consistent with a previous finding (53), our data show that PDCD1 exerts inhibitory effects on canine T cells. Although activation-associated genes were upregulated in PDCD1⁺CD4⁺ and CD8⁺ T cells, several TCR signaling genes (ID3, CCR7, FOXP1, TRAT1, IL7R) were downregulated. These results support the notion that PDCD1 can reflect both activation and early exhaustion; however, in healthy dogs, PDCD1 expression most likely indicates recent activation, with exhaustion becoming relevant only under chronic inflammatory contexts when additional inhibitory receptors co-occur. Second, LAG3 contributed to exhaustion-related programs, particularly in LAG3⁺PDCD1⁺CD8⁺ T cells, which showed reduced TBX21 expression, consistent with human data indicating that T-bet represses LAG3 to maintain functional PD1⁺CD8⁺ T cell responses (89, 90). The transcriptional signatures of these cells were also observed within the tumor microenvironment of canine osteosarcoma, suggesting that similar regulatory mechanisms may influence tumor-associated T cell dysfunction. Additionally, LAG3 expression correlated with upregulation of MDH1 and MDH2, suggesting that malate metabolism plays a role in sustaining PD1⁺CD4⁺ T cell function, as described previously in chronic viral infection models (91). Future studies are warranted to determine whether malate metabolism directly modulates the functional stability of CD4⁺ T cells and prevents the transition of these cells into exhaustion—a phenomenon that remains to be demonstrated in both humans and dogs. Third, we found evidence that CD274⁺ neutrophils and monocytes may contribute to IL-10-mediated immunosuppression. In dogs, IL-10 impairs neutrophil function during babesiosis (92), and in humans, PD-L1⁺ myeloid cells are major IL-10 producers in several cancers (93, 94). Therefore, our findings suggest that PD-L1⁺ neutrophils or PMN-MDSCs may also contribute to shaping an immunosuppressive microenvironment in dogs. Finally, consistent with previous reports (54, 61, 95), CTLA4 expression was observed in regulatory T subsets and notably in DN T cells. Although DN T cells lacked canonical regulatory markers (FOXP3, IL10, IL2RA, GATA3) (96, 97), their CTLA4 expression, naïve-like origin, and absence of exhaustion-related signals based on trajectory analysis suggest that CTLA4 may act as an early regulatory cue guiding DN T cell differentiation. While direct evidence of DN T cells in canine tumors is limited, their immunoregulatory properties in tissues (97, 98) and the clinical activity of CTLA4-targeted therapies in dogs (54) support a potential functional role within the tumor microenvironment.

PBMCs reflect peripheral immunity and are influenced by diverse factors, including breed, lifestyle, and environmental exposures (99–101). In healthy dogs, circulating T and myeloid cells express PD1, PD-L1, and CTLA4 (54, 102, 103). Herein, we investigated the cohort-specific effects on PBMC immune profiles by integrating publicly available scRNA-seq datasets from healthy dogs. For the first time, our integrated analysis delineated distinct cohort-dependent immune features, reflected by differences in the proportions of cycling T cells, checkpoint-expressing subsets, and T cell exhaustion states. In dogs, cycling T cells have been detected in the periphery, duodenum, Peyer’s patches, and mesenteric lymph nodes, indicating ongoing antigenic stimulation (8, 14, 15, 22). This is further supported by our interactome analysis, which reveals that cycling T cells interact with CD14⁺ monocytes, DC, or pDC mediated through MIF and CD99 signaling pathways. Additionally, each cohort exhibited distinct patterns of IC gene expression, such as CTLA4 and LAG3, along with variable of T cell exhaustion states, underscoring the immune heterogeneity among individuals. These differences likely reflect cohort-specific antigen exposure. PBMC datasets have served as essential resources for studies of naturally occurring diseases (8, 104, 105) and for investigating the impacts of environmental and lifestyle factors on immunity (106–109). Accordingly, it may be preferable to adopt immune cell reference datasets that are tailored to the local context, such as domestic conditions in South Korea, and the specific research purpose.

Finally, we investigated homeostatic cell–cell interactions among peripheral immune subsets in healthy dogs. Our interactome analysis indicated that MIF, CD99, MHC-II, and SELPLG constitute key intrinsic pathways governing immune homeostasis. Interestingly, Moore et al. (110) previously proposed that CD4 expressed on canine neutrophils interacts with receptors expressed on leukocytes. Consistent with this model, our analysis predicted interactions between CD4 on neutrophils and HLA- DRB1 on antigen-presenting cells. Notably, human and canine neutrophils endogenously express surface CD4 (111), which can bind certain viral proteins that enhance inflammatory responses. CD4 expression has also been associated with increased neutrophil migratory capacity (112). In agreement with these findings, CD4⁺ neutrophils in our dataset were strongly enriched in inflammatory transmigration pathways, suggesting that MHC-II-mediated interactions may regulate the biodistribution of these cells in vivo. Furthermore, CD99 and MIF signaling emerged as key components of myeloid-myeloid communication. Canine macrophages infiltrating local or metastatic oral melanoma have been reported to significantly upregulate both molecules compared with normal mucosal tissues (113). Meanwhile, elevated circulating levels have also been described in atopic dermatitis and mammary neoplasia (114, 115). In our analysis, CD14⁺ monocytes and DCs displayed homotypic interactions through CD99 and MIF, and XCR1⁺ DCs showed specific expression of MIF. Similar homotypic myeloid interactions have been implicated in shaping anti-tumor immunity in human melanoma patients (116). Collectively, these findings suggest that CD99- and MIF-mediated signaling may represent important components of canine immune homeostasis that may extend beyond the steady state to inflammatory or tumor-associated contexts. These predictions warrant experimental validation using single-cell and spatial transcriptomic approaches.

We have several limitations to acknowledge. First, the sample size was small, limiting representation of the full biological diversity of companion dogs. This includes potential variation associated with breed, age, and individual immune histories, which could not be fully assessed in our cohort. Second, we intentionally enriched CD3⁺ T cells to ensure adequate representation of lymphocyte populations, which may not fully reflect the physiological leukocyte distribution in peripheral blood. However, this approach would provide a more appropriate reference for future T cell-focused studies, particularly in contexts where stress leukograms with neutrophilia and lymphopenia are marked. Third, as with all scRNA-seq studies, our conclusions are based on transcriptional profiles rather than functional assays, and computational inferences, such as module scoring or trajectory analysis, cannot fully substitute for direct experimental validation. In addition, despite using the most recent canFam4 assembly, incomplete annotation of canine genes still poses inherent limitations for functional interpretation. Continued efforts to further refine the canine genome and its transcriptomic annotation, similar to the mature resources available for human and mouse studies, will greatly enhance future immunological and comparative transcriptomic research.

In conclusion, to our knowledge, this is the first study to map the single-cell transcriptomic landscape of circulating leukocytes in South Korean dogs. Our results provide a valuable resource to support future scRNA-seq studies on immune cells in dogs affected by various immunological disorders. Methodology used in this study may help pave the way for investigating potential roles of IC genes in translational research, bridging canine and human immunology.

Funding Statement

The author(s) declare that financial support was received for the research and/or publication of this article. MK was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (No. 202300241779). HK was supported by the Bio & Medical Technology Development Program of the NRF funded by the Korean government (No. 2022M3A9D3016848). WZ was supported by National Institutes of Health grants: CA269661 (W. Zhang), CA260239 (W. Zhang; D. Zhou; G. Zheng), and CA290792 (W. Zhang; G. Zheng; K.S.M. Smalley). WZ was also supported by an endowment fund from the Dr. and Mrs. James Robert Spenser Family and shared resources from the UF Health Cancer Center, which is supported in part by state appropriations provided in Fla. Stat. section 381.915 and the National Cancer Institute (Grant No. P30CA247796).

Data availability statement

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.

Ethics statement

The animal studies were approved by Jeju National University Institutional Animal Care and Use Committee (IACUC No. 20230072). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent was obtained from the owners for the participation of their animals in this study.

Author contributions

MK: Conceptualization, Methodology, Data curation, Formal analysis, Investigation, Writing – original draft, Writing – review & editing, Supervision, Project administration. TG: Investigation, Formal analysis, Writing – review & editing. HS: Investigation, Writing – review & editing. YM: Investigation, Writing – review & editing. HK: Investigation, Writing – review & editing. YK: Investigation, Formal analysis, Writing – review & editing. NB: Data curation, Formal analysis, Writing – review & editing. RK: Formal analysis, Writing – review & editing. WZ: Formal analysis, Supervision, Writing – review & editing. YY: Writing – review & editing. WS: Writing – review & editing. CL: Writing – review & editing.

Conflict of interest

NB was previously employed by Santa Ana Bio and Omniscope and is currently a scientific advisor to Epana Bio and a technical consultant to Columbus Instruments. The work presented does not pertain to any commercial endeavors in the companies listed above.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that no Generative AI was used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2025.1680437/full#supplementary-material