Introduction

Mammary gland tumours are the most commonly occurring neoplasm in the female dog and represent a remarkably heterogeneous group in terms of morphology and biological behaviour [32, 43]. About half of canine mammary tumours are considered malignant, and the identification of reliable prognostic factors is essential in order to estimate the individual risk of unfavourable clinical outcome [29, 54].

Several studies have recognised some reliable prognostic factors such as tumour size, histologic type, histologic grade and lymph node status [19, 30, 31]. Moreover, in recent literature, we found an increasing number of investigations searching for suitable prognostic markers for canine mammary cancer [54], including proliferation markers [25], hormone receptors [23], p53 and human epidermal growth factor receptor 2 (HER2) [21, 24] and adhesion molecules [14, 26], among others. The clinical experience is still limited, however, and reliable results of prospective studies are not always available.

Human and canine mammary cancer studies based on single molecular markers probably cannot accurately account for the heterogeneity of this disease [39]. Given the large number of cellular events involved in cell growth, differentiation, proliferation, invasion and metastases [4], the investigation of multiple molecular alterations in concert has been assuming great importance due to the introduction of high-throughput technologies [39]. In fact, recent gene expression profiling studies on human breast tumours have identified distinct molecular subtypes of breast carcinomas, which differ in their pathobiology and clinical outcomes [36, 47, 48]. Sorlie et al. [48] analysed the expression profiles of 115 sporadic breast tumour samples and categorised them into five main groups: luminal A, luminal B, HER2-overexpressing, basal like and normal breast tissue like. Luminal A and B subtypes are based on the expression of estrogen receptor (ER), usually with luminal cytokeratin (CK) expression, whereas the basal-like subtype is characterised by the absence of hormonal receptors and expression of basal cell markers [5, 33].

Given that gene expression profiling is impractical as a routine diagnostic tool, there are immunohistochemistry surrogate panels proposed that can potentially distinguish breast cancer subtypes [27, 33]. In the present study, we sought to identify phenotypical subtypes in canine mammary cancer with possible clinical implications. To accomplish this goal, we have characterised by immunohistochemical analysis 102 canine mammary carcinomas based on the immunohistochemical panel proposed by Matos et al. [27], which involved the evaluation of five molecular markers (ER, HER2, CK5, p63 and P-cadherin).

Materials and methods

Tumour specimens

The present study is based on a series of 102 cases of canine malignant mammary tumours (n = 102) selected from the histopathological files of the University of Trás-os-Montes and Alto Douro, Vila Real and from the Institute of Biomedical Science at the University of Porto, Portugal. The material was fixed in 10% neutral formalin and embedded in paraffin wax. Sections (3 μm) were cut and stained with haematoxylin and eosin for histological examination or used to perform immunohistochemistry.

Follow-up data

Sixty-nine cases (n = 69) had available follow-up data, with a median overall survival time of 15 months (range 5–74 months). Overall survival (OS) was defined as the period between surgery and animal natural death or euthanasia due to cancer. Disease-free survival (DFS) was defined as the period of time between surgery and recurrent or metastatic disease. During the follow-up period, according to the referring surgeons, 35 animals died or were euthanized due to metastatic disease and/or local recurrence.

Histological examination

Tumours were diagnosed according to the WHO criteria for canine mammary neoplasms [30]. Clinicopathological variables included in the present study were age, ovariohysterectomy status, contraceptive administration, tumour size, tumour location, tumour histological type and grade, presence of intra-tumoral necrosis, presence of vascular invasion and presence of lymph node metastasis.

Tumours were evaluated for grade in accordance with the Nottingham method for human breast tumours [11] based on the assessment of three morphological features: tubule formation, nuclear pleomorphism and mitotic counts. Each of these features was scored on a scale of 1 to 3 to indicate whether it was present in slight, moderate or marked degree, giving a putative total of three to nine points. Grade was allocated by an arbitrary division of the total points as follows: grade I (well differentiated), 3, 4 or 5 points; grade II (moderately differentiated), 6 or 7 points; and grade III (poorly differentiated), 8 or 9 points.

Immunohistochemistry

Tissue sections were incubated with primary monoclonal antibodies against ER, HER2, CK5, p63, P-cadherin and Ki67. Table 1 summarises the antibodies used and the staining procedures adopted for each antibody. Antigen retrieval was carried out by microwave treatment in a 10 mM citrate buffer, pH 6.0, with the exception of P-cadherin, which was performed with an EDTA buffer, pH 8.0 (Lab Vision, USA) in a boiling bath, during 20 min. For Ki-67, slides were previously incubated with 0.2 mg/mL trypsin (Merck) in phosphate-buffered saline (PBS) for 10 min at 37°C. After cooling (20 min at room temperature), the sections were immersed in 3% hydrogen peroxide (H2O2) and distillated water during 30 min to block endogenous peroxidase activity. All slides were then incubated with a blocking serum (Lab Vision) for 10 min and then incubated with the specific antibody. After incubation, slides sections were incubated with biotinylated secondary antibody, followed by streptavidin-conjugated peroxidase (Lab Vision), except for ER and HER2. For these antibodies, a polymeric labelling methodology was used as a detection system (Novolink Polymer Detection System, Novocastra, Newcastle, United Kingdom), following the manufacturer’s instructions. Subsequently, the colour was developed with 3,3-diaminobenzidine tetrahydrochloride, and slides were counterstained with Gill’s haematoxylin, dehydrated and mounted for evaluation by light microscopy.

Table 1 Primary monoclonal antibodies and immunostaining protocols used
Full size table

Adjacent normal mammary tissues were used as internal positive controls for CK5, p63, P-cadherin (basal and myoepithelial cells) and Ki67. As positive controls, we also used canine uterus sections for ER and a human breast carcinoma with proved amplification (by FISH) and overexpression for HER2. Negative controls were carried out by replacing the primary antibody with PBS.

Evaluation of the immunohistochemical data

Nuclear ER immunoreactivity was considered positive when more than 10% of the neoplastic cells expressed this marker. To evaluate HER2 expression, Herceptest scoring system was applied (0 = no membrane staining or <10% of cells stained; 1+ = incomplete membrane staining in >10% of cells; 2+ = >10% of cells with weak to moderate complete membrane staining; and 3+ = strong and complete membrane staining in >10% of cells), with 2+ and 3+ cases considered positive. As for CK5 and p63, a semi-quantitative analysis was performed as follows: 0, <10% positive cells; 1, 10–50% positive cells; and 2, >50% positive cells, with a cytoplasmic (CK5) or nuclear (p63) pattern of cellular distribution. Ki-67 and P-cadherin immunostainings were evaluated as previously described in canine tissues [13, 14]. CK5, p63 and P-cadherin were considered positive when more than 50% of the neoplastic cells expressed each marker.

Statistical analysis

For statistical analysis, association between subtype tumour groups and continuous variables (mitotic and Ki-67 indices) was assessed with non-parametric Kruskal–Wallis test. Associations between groups and categorical variables such as tumour size, histological type, histological grade and invasion were performed using the chi-square test. Survival curves were generated by the Kaplan–Meier method, and the survival rates were compared using the log-rank test. All statistical analysis was performed using SPSS 11.5 statistical software. A P value <0.05 was considered statistically significant.

Results

Patients and tumour characteristics

The mean age of dogs at the time of surgical removal of tumours was 9.7 ± 2.5 years (range 4–16 years of age). The mean maximum tumour diameter was 4.21 ± 3.4 cm (range 0.5–18 cm), with tumours more frequently located in caudal mammary glands (n = 36; 59%). In ten (15.2%) out of the 66 female dogs with available clinical information, ovariohysterectomy was performed prior to the removal of mammary tumours. Contraceptive administration was confirmed in eight (13.8%) cases. Histological evaluation yielded 39 (42.4%) simple carcinoma, 41 (44.6%) complex carcinoma and 12 (13%) carcinosarcoma subtypes. According to the Nottingham method, tumours were classified as grade I (n = 14, 15.2%), grade II (n = 33, 35.9%) and grade III (n = 45, 48.9%). Necrosis was present in 87 (94.6%) cases, and vascular invasion was present in 51 (55.4%) cases. Lymph nodes were available in 49 cases, with confirmed metastasis in 26 cases (53.1%).

Immunohistochemistry profiles in canine tumours

The results of the immunohistochemical analysis performed for ER, HER2, CK5, p63 and P-cadherin are shown in Table 2 and Fig. 1. The immunohistochemical detection of ER was reliable in 96 cases: The remaining tumours have lost ER antigenicity (adjacent mammary gland was negative) and were excluded. Immunohistochemical evaluation of HER2 and P-cadherin was available in 100 and 96 cases, respectively.

Fig. 1
figure 1

Immunohistochemical expression of the different proteins studied by IHC in canine mammary carcinomas. a–d ER staining, e–h HER2 staining, i–l CK5 staining, m–p p63 staining, q–t P-cadherin staining. Each column represents a distinct molecular subtype. From left to right, each column represents luminal A, luminal B, basal and HER2 overexpressing subtypes. (Original magnification ×400)

Full size image
Table 2 Immunohistochemical results in the present study
Full size table

ER and p63 positive cases showed the characteristic nuclear staining, whereas CK5-positive ones showed a cytoplasmic pattern of expression. HER2 positive tumours showed a membranous staining, and P-cadherin positive tumours showed a cytoplasmic and/or membranous immunostaining. We observed that 58.3% of canine mammary carcinomas in our series were ER positive, whereas 21% were HER2 positive (2+ and 3+). A positive basal cell marker expression was present in 32.4% tumours for both CK5 and p63 and in 42.8% tumours for P-cadherin.

According to Nielsen et al. [33], we classified each tumour based on its ER and HER2 expression. A total of 96 cases were immunohistochemically interpretable to allow sample characterisation into one of five categories (Table 3). If a tumour was ER positive, it was classified as luminal; moreover, we distinguish luminal A and B on the basis of HER2 overexpression. If a tumour was ER positive and HER2 negative (0 or 1+), it would be classified as luminal A (ER+/HER2−); however, if it was ER and HER2 positive, it would be classified as luminal B (ER+/HER2+). If a tumour was ER negative and HER2 positive (ER−/HER2+), it would be classified as HER2-overexpressing, and if it was both ER− and HER2− negative but positive for at least one basal marker (CK5 and/or p63 and/or P-cadherin), it would be classified as basal (ER−/HER2−). If a tumour did not show expression for any of these markers, it would be classified as negative (null phenotype) and would not be considered in the remaining analyses.

Table 3 Frequencies of immunohistochemically defined subtypes of canine mammary carcinomas (n = 96)
Full size table

Using this definition, we observed that luminal A and B subtypes comprised 44.8% and 13.5% of all tumours, respectively; basal subtype comprised 29.2%; HER2 overexpressing subtype represented 8.3%, and negative/null phenotype accounted for 4.2% in this tumour series (Table 3).

Statistically strong significant differences between the four groups were observed in this study when related with some relevant clinicopathological parameters (Table 4). Basal and HER2 overexpressing subtypes were associated with simple or carcinosarcoma histological types, whereas complex carcinomas were mostly of luminal A subtype (P < 0.0001). In addition, basal subtype tumours presented higher histological grade, representing 55.6% of grade III tumours (P < 0.0001) and were also significantly associated with the presence of vascular invasion (P < 0.0001).

Table 4 Association between tumour subtypes and clinicopathological characteristics
Full size table

Basal marker expression clearly differed across distinct molecular subtypes (Table 5). Basal and HER2-overexpressing tumours demonstrated a higher frequency of the basal cell markers p63 and P-cadherin (P < 0.0001 and P = 0.001), and CK5-positive tumours were frequently basal subtype tumours (P = 0.001). In contrast, luminal pattern was associated with a lower expression of basal markers. In fact, when analysing basal marker expression simultaneously, we found that the majority of luminal tumours were simultaneously negative to CK5, p63 and P-cadherin. All HER2-overexpressing tumours expressed at least one basal marker, and the basal subtype tumours showed frequently the expression of two or all basal markers (P < 0.0001).

Table 5 Association between different subtypes versus basal marker expression and proliferation indices
Full size table

With regard to proliferation indices, luminal A tumours showed lower median mitotic and Ki67 labelling indices (P = 0.001 and P < 0.0001, respectively), whereas all other groups were characterised by higher proliferation rates, with basal subtype showing the highest Ki67 index.

Follow-up data revealed that basal subtype was significantly associated with lower overall (P = 0.002, Fig. 2a) and disease-free (P = 0.01, Fig. 2b) survival rates, whereas the other groups showed higher survival rates, including the HER2-overexpressing group.

Fig. 2
figure 2

Kaplan–Meier overall survival (a) and disease-free survival (b) curves of the different subtype groups

Full size image

Discussion

Recently, gene expression profiling has redefined breast cancer taxonomy and identified five distinct subtypes of carcinomas: luminal A, luminal B, normal breast like, HER2 overexpressing and basal like [36, 47, 48, 53]. These molecular subtypes not only reflect the heterogeneity of breast carcinomas and the possible different cell lineage pathways in breast carcinogenesis but also demonstrate the difference in clinical outcome, with basal-like subtype associated with a more aggressive behaviour [1, 47, 48, 52, 53].

Although gene expression profiling is still considered the “gold standard” for the identification of breast carcinoma subtypes, this technology requires highly sophisticated technical equipment and is not readily available for clinical application or for retrospective studies using formalin-fixed, paraffin-embedded samples [39]. For this reason, immunohistochemistry has been used in several studies, and the evaluation of a limited panel of immunohistochemical cell markers have shown that breast carcinomas can be subdivided into subgroups remarkably similar to the ones defined by gene expression profiling [1, 3, 22, 27, 33, 38, 52].

In the present study, we found in a series of canine mammary tumours similar findings observed in human breast cancer. We have also identified distinct phenotypical subtypes in a series of canine mammary carcinomas by using an immunohistochemical panel, which included five molecular markers (ER, HER2, CK5, p63 and P-cadherin). Based on ER/HER2 molecular classification, we defined four main subgroups: luminal A (ER+/HER2−, 44.8%), luminal B (ER+/HER2+, 13.5%), basal-like (ER−/HER2−, 29.2%) and HER2 overexpressing (ER−/HER2+, 8.3%). In contrast, Sarli et al. [44] have only identified luminal A and B subtypes when studying a series of 39 canine mammary carcinomas. Although using a similar terminology, they used a distinct panel of molecular markers, and the adopted classification was not identical, with luminal subtype defined as CK19 positive tumours, regardless of hormonal status (luminal A, HER2− and luminal B, HER2+), and HER2 overexpressing and basal-like subtypes defined as CK19 negative tumours, HER2+ and HER2−, respectively.

In the current study, we found statistically strong significant differences between the four groups, with ER positive luminal A tumours more frequently associated with complex tumour type, low histological grade, less invasive and low proliferative tumours, whereas basal-like and HER2 overexpressing subtypes were associated with simple and carcinosarcoma tumour types, high histological grade, lymphovascular invasion and high proliferation, features that are in accordance to the ones described in recent human literature for basal-like cancers [20, 22, 27, 40].

CK5, p63 and P-cadherin are proteins that are expressed early in epithelial differentiation and may contribute to a committed stem cell and/or progenitor phenotype [6, 7, 9, 35]. In this study, we demonstrate that these markers are upregulated in the basal subtype, similarly to the previous results of Matos et al. [27]. In fact, the basal subtype rarely expressed just one basal marker but frequently expressed them simultaneously, which suggests a more undifferentiated profile. HER2-overexpressing subtype was also characterised by an up-regulation of basal markers, confirming some human breast studies, which suggested that HER2-overexpressing tumours should be included in a bona fide basal-like subclass [5, 27]. In contrast, the majority of luminal tumours in our series were simultaneously negative for basal cell markers, with some cases showing basal marker expression, which was also described by some authors who reported tumours co-expressing basal CK and hormone receptors or HER2 [40, 50].

Similarly to human breast cancers, in this study, we further demonstrate the molecular heterogeneity of canine mammary cancer. A “hierarchy or stem cell” model of breast cancer oncogenesis has been proposed to elucidate the observed functional heterogeneity of tumours. In this model, transformation occurs in a stem cell or in a progenitor “highly proliferating” cell, and expansion proceeds until various maturation stages, depending on the genomic alterations. Specific genetic alterations would lead to distinct cellular transcriptomic programmes, including the change of hormonal receptors and CK expression pattern, characterising distinct subgroups of breast carcinomas [5, 8, 39].

Survival analysis revealed that distinct subtypes were associated with different clinical outcomes, with basal subtype associated with lower survival rates, similarly to human breast cancer studies [36, 47, 48]. These results also corroborate a previous study in canine mammary cancer performed by Griffey et al. [16], which firstly described basal carcinomas as having poor prognostic features. Despite many different studies associating basal-like tumours with a more aggressive clinical history and shorter survival [3, 33, 37, 47–49, 52], others did not find such a prognostic significance [12, 18]. These variations are probably related to differences between studies in patient cohorts, analytic methods and, most importantly, the immunohistochemical definitions of basal-like breast cancer [39]. Recently, Tang et al. [51], comparing several classifications with similar terminology but different definitions (such as ER/HER and triple negative classification), concluded that these classifications are related but not interchangeable.

In contrast to basal subgroup, luminal and HER2 overexpressing subtypes showed increased survival rates. The fact that luminal tumours were associated with a better prognosis is not surprising since ER positive human breast carcinomas are usually associated with a more favourable clinical outcome. In veterinary pathology, however, the prognostic value of ER in canine mammary cancer is still a matter of debate. Previous studies using biochemical [45] and immunohistochemical [34] methodologies have demonstrated the prognostic value of ER, but others have failed this confirmation [23, 28]. The observed discrepancies between different studies are probably related with sample selection, differences in antibodies, staining procedure and evaluation or sensitivity of the detection system. In our series, luminal tumours were mostly of complex type, which comes in accordance to previous canine studies reporting complex carcinomas as being more likely ER positive [15, 23, 28]. Given that this tumour type is generally associated with a better clinical outcome, its high proportion in luminal subtype groups is probably in part responsible for their favourable prognosis.

Despite HER2 recognition as a prognostic factor in human breast cancer [41, 46], the significance of HER2 overexpression in dogs with mammary carcinoma is still unclear. Some studies have shown that either HER2 amplification or protein overexpression are present in canine mammary carcinomas [2, 42], while others found no gene amplification [24]. Similarly to previous studies [10, 24], HER-2 overexpressing tumours were found usually associated with established indicators of poor prognosis such as large tumour size, high histologic grade, invasion, simple histologic type and high proliferative indices. However, Kaplan–Meier analysis revealed that this subtype was related with a more favourable clinical outcome, findings that are in contrast with human studies, which describe similar survival rates for HER2 overexpressing and basal-like subtypes [36, 47, 48] and are probably related to the small number of cases that comprise the HER2 overexpressing subtype. However, a recent study performed by Hsu et al. [17] revealed that HER2 overexpression in canine malignant mammary tumours is associated with higher survival rates. Additional large-scale studies are warranted to further explore the value of HER2 in canine mammary carcinomas.

In conclusion, as in humans, our study defined distinct molecular phenotypes in canine mammary carcinomas based on immunohistochemical analysis. Moreover, we have identified a basal-like subtype representing almost 30% of our series, which was associated with a more aggressive clinical behaviour. We believe that canine mammary carcinomas would be suitable natural models for the study of this particular subset of carcinomas. However, more studies are needed regarding the prognostic value of these immunohistochemically determined subtypes in canine mammary cancer.