For this study 48 formalin fixed paraffin embedded (FFPE) skin samples diagnosed with CCH from dogs ≤ 4 years of age were selected from the archive of the Institute of Veterinary Pathology, Leipzig University. All specimens were screened for the extent of secondary inflammation and classified into four groups according to the degree and distribution of infiltrating lymphocytes as previously described (7). The study design comprised four groups in which CCH of 12 animals per group (6 female and 6 male) were examined. The diagnosis of CCH was confirmed via immunohistochemistry using ionized calcium-binding adapter molecule (Iba)-1 and exclusion of other round cell neoplasms like histiocytic sarcoma (CD204), cutaneous T-cell lymphoma (CD3) and mast cell tumor (CD117) antibodies in addition to hematoxylin–eosin (H&E) stain. Further, mitotic count in tumor cells was evaluated per mm² by counting mitotic figures in 10 consecutive and randomly chosen areas of 0.237 mm² (2.37 mm² in total) using a microscope at 400-fold magnification and a field number of 22.

Antibodies used for immunophenotyping are summarized in Table 1. Positive control tissues known to include the target epitope were used as indicated in Table 1. For negative controls the primary antibody was replaced by normal serum from the respective species of the primary antibody in an equivalent concentration as the primary antibody on CCH slides as well as positive control tissues. Consecutive sections were mounted on positively charged glass slides (New Erie Scientific LLC, Portsmouth, NH, United States) followed by deparaffinization and rehydration. Afterwards endogenous peroxidase was inactivated with methanol and 0.5% hydrogen peroxide. Epitope retrieval was performed by incubation in either ethylenediaminetetraacetic acid (EDTA; pH 9.0) or citrate buffer (pH 6.0) at 96°C for 25 min in a water bath. For CD3 a protease-based epitope retrieval using 0.5 g/L protease (derived from Bacillus licheniformis, Sigma-Aldrich, St. Louis, MO, United States) in phosphate buffered saline for 5 min at 37°C was applied. Goat serum (Biozol Diagnostica Vertrieb GmbH, Eching, Germany) at a dilution of 5% was used for 30 min in order to block unspecific antibody binding. This was followed by incubation with the primary antibody. After incubation at 4°C overnight, secondary antibodies were incubated at room temperature for 30 min. For visualization, avidin-biotin-complex (ABC; Vector Laboratories, Newark, CA, United States) with 3,3’-diaminobenzidine (DAB; Sigma-Aldrich, St. Louis, MO, United States) and Mayer’s hematoxylin counterstain was used in all cases.

Whole slide images (WSI) were obtained using an Axioscan 7.0 with Plan-Apochromat 20x/0.8 lens (200x magnification; Zeiss Group, Jena, Germany). Subsequently, WSI were subjected to image analysis using QuPath (version 0.4.2) (34). Staining parameters were manually set by choosing representative regions for hematoxylin and DAB staining. Tumor area for each slide was selected manually as region of interest (ROI). For each slide 10 areas with 500 × 500 μm (equals 2,5 mm² in total per slide) were created with QuPath’s “Tiles” function. The number of positive cells per mm² were then determined in representative areas, with a manually chosen threshold applied to evaluate the DAB signal in either the nuclear or cytoplasmic cell compartment. Cytoplasmic staining was considered for CD3, Iba-1, Survivin, CD80, CD86 and cleaved caspase-3 and nuclear staining for forkhead box protein 3 (FoxP3) and Ki-67. Both the number of positive cells per mm² and the percentage of positive cells in relation to the total number of cells were evaluated. Only tumor cells were evaluated for CD80, CD86, PD-L1, Survivin and Ki-67. A minimum of 5% of immunolabeled tumor cells per case were used as a cut-off value for CD80, CD86, PD-L1 and Survivin. The analysis of Iba-1 and CD3 positive cells was performed using a whole-slide approach including the entire tumor area.

For the evaluation of Ki-67, PD-L1, CD80, CD86, Survivin and FoxP3 immunohistochemistry an algorithm using QuPath’s “Train classifier” tool with “random tree” method to differentiate immune and tumor cells was trained using selected WSI of all CCH groups. Single or few cells were manually annotated as either “immune” or “tumor” identity until the classifier resulted in satisfactory classification in all ROIs. Main discriminatory factors in choosing the phenotype were size, shape, color as well as amount and texture of visible cytoplasm. After this step was repeated for all training slides, the results were visually validated in representative areas of randomly chosen slides. This involved the counting of 100 cells and a cross-check of their phenotype against the classification provided by the algorithm. The resulting classifier was subsequently run on the previously selected 500 × 500 μm fields to exclude immune cells for Ki-67, PD-L1, CD80, CD86 and Survivin and tumor cells for FoxP3 evaluation, respectively. As a last step of validation, the percentage of “immune” cells recognized by the algorithm per group was checked for similarity with the percentage of CD3-immunolabeled cells and examined for changes between the groups.

In-situ hybridization was used to visualize and quantify IFN-γ and mx1 transcripts. Using a biopsy punch with 3 mm diameter, one core of each neoplasm (n = 48) with a representative cell composition was taken. Cores were randomly assigned to one of four blocks consisting of 12 cores each. The RNAscope technology was used for the identification of IFN-γ and mx1 (Advanced Cell Diagnostics, Inc., Minneapolis, MI, United States) according to the manufacturers’ instructions. The probes were designed using a canine sequence for IFN-γ (Entrez Gene ID 4038014) and mx1 (Entrez Gene ID 403745) targeting IFN-γ and mx1, respectively. For mx1 inflamed brain tissue from a canine distemper encephalitis case served as positive control (30). Since IFN-γ expression in CCH is previously described (8) no additional positive control tissue was included. Process controls consist of a negative control (RNAscope® Negative Control Probe—dapB) and a positive control (RNAscope® Probe—Canis lupus familiaris peptidylprolyl isomerase B) according to manufacturers’ instructions. Negative controls were checked for the absence of specific labeling. For identification of the source of IFN-γ and mx1 expression sequential serial slides were stained for CD3 and Iba-1 using immunohistochemistry. Slides were digitalized using the Axioscan’s 40×/0.65 lens.

The quantification of IFN-γ and mx1 transcripts was undertaken as follows: Firstly, positive pixels within the area of each individual core were measured using a pixel-based classifier (“positive pixel”) and expressed in percentage. Next, the number of positive cells with at least 1% positive pixels (“positive cells”) was identified and followed by execution of the cell detection function of QuPath. This resulted in the number of positive pixels per cell (“positive signal”) as a measure of the intensity of the labeling. From all measurements of all cells of one core a median value was calculated as indicator for the positive signal per positive cell per sample. Median values served as variable for the comparison between the investigated neoplasms.

Lastly, to determine the cell’s phenotype, the Interactive Image Alignment tool (v0.3.0) was used, allowing an overlay of two or multiple slides within QuPath. Whenever possible, 50 cells positive for mx1 or IFN-γ signal were manually selected for each individual core. Next, consecutive slides immunostained for CD3 and Iba-1 were used as overlay and cells were labeled according to the immunophenotype as “iba-1” or “cd3,” respectively. If the cell could be identified in both immunohistochemistry slides and displayed neither signal, they were labeled as “other.” Cells that were not consistently shown in all three slides or only shown in one immunohistochemical slide as immunonegative were labeled as “mismatch.” Generally, it was assumed that individual cells either expressing Iba-1 or CD3 do not express the other marker. Overall, for IFN-γ only 16 cores could be evaluated due to difficulties in overlay, resulting in 800 evaluated cells. For mx1 44 cores were suitable for evaluation with 2,179 evaluated cells.

Statistical analysis was made with R 4.2.1 using RStudio and the tidyverse package (35–37). Specimens were grouped according to group and measurements checked for normal distribution using Q-Q-plots. Subsequently, medians and quantiles were calculated and statistical significance between tumor group and sex was evaluated using the Kruskal-Wallis test and Wilcoxon-rank-sum test for post hoc analysis. For the analysis of the in-situ hybridization the median of all cells per core were evaluated. A p-value < 0.05 was chosen as threshold for statistical significance. Possible correlation was analyzed using Spearman’s ρ.

The study included 48 CCH with 12 samples each belonging to groups 1 to 4, respectively (Figure 1). No significant differences were observed between the epitopes investigated with regard to sex or breed. Consequently, specimens were evaluated solely in relation to their assigned group. Mean age at excision was 1.84 years. Neoplastic cells were positive for Iba-1 and negative for CD204, CD117 and CD3.

Iba-1 positive cells in tumor ROI showed a significant decline between groups from 3288.8 cells per mm² (85.26%) to 2523.3 per mm² (53.22%), while CD3 positive cells showed an increase from 1039.9 per mm² (11.3%) to 3225.4 per mm² (53.5%) in group 1 and 4, respectively (p < 0.05; Figure 1). Mitotic count in tumor cells differed between groups with group 1 and 2 (9.3 and 9.5 mitotic figures per mm², respectively) having significantly higher counts than groups 3 and 4 (4.0 and 0.6 per mm², respectively; p < 0.05). Overall, there was a high heterogeneity of the mitotic count between samples with values varying between 0 and 21.1 per mm² and an overall median of 11.5 mitoses per mm².

Immune checkpoint evaluation revealed an expression of CD80 and PD-L1 in all samples (48/48), while 45/48 showed considerable levels of CD86 (Figure 2).

PD-L1 immunolabeling showed a variable degree of staining intensity. Samples showed a group-independent high number of PD-L1 immunolabeled tumor cells (overall median: 8748.0 cells per mm² and 73.89%, respectively).

The density of CD80 immunolabeled tumor cells showed a significant decline from group 1 to groups 3 and 4 (e.g., 5161.6 positive tumor cells per mm² in group 1 and 1941.12 positive tumor cells per mm² in group 4; p < 0.05; Figure 2). Significant differences between the groups with regard to percentage values were identified. These differences were detected between group 1 (43.44%) and group 3 (22.38%) but not between group 1 and group 4 (29.05%). The density of CD86 immunolabeled tumor cells showed no group-related differences. Median expression ranged between 1519.93 positive tumor cells per mm² (21.69%; group 3) and 3846.21 positive tumor cell per mm² (30.28%; group 1).

No correlation was found for CD3 and CD80, CD86 or PD-L1 positive (tumor) cells per mm², respectively. A moderate correlation was found for mitotic count and CD80 (Spearman’s ρ = 0.49, p < 0.05) as well as mitotic count and CD86 positive tumor cells per mm² (ρ = 0.34, p < 0.05).

Regarding cell proliferation, the mitotic count was supplemented by the investigation of Ki-67 and Survivin (Figure 3). For Ki-67 the median number ranged between 521.4 tumor cells per mm² (6.56%) in group 2 and 745.4 tumor cells per mm² (11.55%) in group 1 with no significant differences between groups. The distribution of Ki-67 positive tumor cells was heterogenous within and between samples. Individual samples showed variable numbers of Ki-67 positive tumor cells ranging between 0.4 and 4966.8 tumor cells per mm² (Figure 3).

The density of tumor cells with cytoplasmic Survivin expression ranged between 273.73 tumor cells per mm² (3.4%; group 2) and 431.69 tumor cells per mm² (6.2%; group 4) with no difference between groups and an overall median of 386.28 positive tumor cell per mm² (5.1%; Figure 3). Across all groups 50% of samples showed more than 5% immunolabeled tumor cells, with no differences between groups. Nuclear expression of Survivin was occasionally observed in neoplastic cells (data now shown). A moderate positive correlation was found for positive cell per mm² between Ki-67 and Survivin (ρ = 0.49; p < 0.05) as well as Ki-67 and PD-L1 (ρ = 0.38, p < 0.05).

The rate of apoptosis was determined by assessment of cells positive for cleaved caspase-3. The density of cleaved caspase-3 positive cells was significantly higher in group 4 (55.0 positive cells per mm²; 0.55%) compared to the other groups (e.g., group 1: 15.2 positive cells per mm²; 0.12%; Figure 3). A moderate negative correlation was found between the mitotic count and cleaved caspase-3 positive cells per mm² (ρ = −0.47, p < 0.05).

Lastly FoxP3, as marker for regulatory T-cells, was quantified. FoxP3 positive cells were randomly distributed with a median between 62.58 per mm² (1.78%) in group 1 and 154.55 per mm² (2.60%) in group 4, without significant differences between groups (Figure 3). Overall median was 94.6 cells per mm² (1.86%).

The percentage of IFN-γ positive pixel (with 0.66% and 1.12% positive pixels within the ROI, respectively) increased from group 1 to 4 (p < 0.05). In addition, the number of positive cells was higher in group 3 (34.43%) and 4 (21.08%) compared with group 1 (11.26%; p < 0.05). The median positive signal increased from groups 1 and 2 to group 3 (3.94%, 3.93%, and 5.89% positive pixels per cell, respectively; p < 0.05; Figure 4). IFN-γ was most frequently detected in CD3 immunolabeled cells (68.28%; 409/599 cells), while no tumor cells were positive for IFN-γ. A high number of IFN-γ positive cells were classified as “other” (31.72%; 190/599). There was no difference in IFN-γ expressing cells’ identity between groups (Figure 4).

For mx1 there was no difference in the percentage of positive pixels or the number of positive cells (overall median of 1.88% positive pixel and 15.21%, respectively). For mx1 positive signal showed no statistically significant difference between groups (overall median of 7.57% positive pixels per cell; Figure 4). Mx1 was most frequently colocalized with Iba-1 (63.75%; 925/1,451 cells) with lower frequencies of CD3 positive cells (10.54%; 153/1,451) and “others” (25.71% 373/1,451). The identity of mx1 expressing cells showed no significant variation among CCH groups (Figure 4). The comparison of mx1 on transcript and protein level with respect to distribution and frequency showed similar results with an overall median of 7.85% immunolabeled cells (788.8 positive cells per mm²; Figure 4).