Introduction
Pre-clinical studies of breast cancer often include experimental rodent models; however, there are important biological differences between humans and rodents with regard to mammary gland tumor development and progression, as well as hormone dependency that complicate the transferability of pre-clinical findings into clinical practice for human cancer [
1,
2]. There is, therefore, a need for animal models that mimic human breast cancer more precisely. The dog has emerged as a potential candidate and has been demonstrated to be an efficient model for human cancers [
3‐
6], including breast cancer [
7‐
9].
Mammary gland tumors are common in dogs with an incidence rate twice as high as in humans, but with large differences between breeds [
10‐
12]. In contrast to many rodent models, mammary gland tumors in dogs occur spontaneously without intentional genetic or chemical manipulation, and importantly, they occur in an intact immune environment. There are many similarities between canine and human mammary gland cancers, such as age of onset, hormonal influence on tumor development, disease course, clinical outcome parameters (such as tumor size, clinical stage, and lymph node invasion), mode of metastatic spread, and numerous molecular markers and genetic risk factors [
7,
8,
13,
14]. Importantly, dogs often live together with humans and the two species are therefore exposed to many of the same environmental factors that can influence cancer development. There are also differences between the two species; while human breast cancer is dominated by epithelial tumors, canine mammary gland tumors (CMGTs) frequently also contain myoepithelial components [
15,
16]. Tumor mutational burden (TMB) is generally lower in canine tumors, and found to be comparable to TMB in pediatric tumors [
17,
18]. In addition, diagnosis generally happens later in the progression course in dogs than in humans, since it depends on discovering palpable lesions as opposed to detection by screening mammography in humans.
In human breast cancer, tumors belong to one of several intrinsic molecular subtypes that have different prognostic and predictive impacts [
19,
20].
Basal-like tumors are predominantly triple negative i.e., they lack expression of estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor 2 (HER2). Basal-like tumors are usually highly proliferative and are associated with a poor prognosis.
HER2-enriched tumors often show high expression of HER2 (also known as erb-b2 receptor tyrosine kinase 2 and encoded by the
ERBB2 gene) and are also poor prognosis tumors; however, these may benefit from HER2-targeted therapy.
Luminal A tumors are usually ER-positive tumors that proliferate slowly and have a relatively good prognosis, while
luminal B tumors are also ER-positive, but are more proliferative, and have a significantly worse prognosis than luminal A tumors [
21].
The intrinsic subtypes can be determined by the PAM50 method, a nearest centroid classifier in which gene expression of 50 genes is correlated to previously defined subtype centroids [
21‐
23]. The PAM50 method, including the 50 genes and the associated centroids were identified from studies of human breast cancer, however, this method captures core biological characteristics of tumors that are relevant in both humans and dogs. There are few studies on the applicability and relevance of PAM50 subtyping in canine mammary gland tumors; however, in a study by Graim et al., PAM50 subtyping was performed on CMGTs from 16 canine patients [
9]. They showed considerable resemblance between canine and human tumors of the same subtype at both the transcriptional and mutational levels.
In a comprehensive study by Kim et al. including gene expression analyses of 158 CMGTs, clustering was performed across the PAM50 genes, but without calling the PAM50 subtypes [
18,
24]. Their analysis revealed a strong resemblance between CMGTs and human breast tumors including the well-known, clear distinction between basal-like and luminal-like tumors. In the current study, we utilize this unique data resource that includes both exome DNA and whole-transcriptome RNA sequencing data to thoroughly explore the molecular subtype landscape in canine mammary gland tumors. We use the well-established gene expression-based subtyping method PAM50 and assess its applicability as a tool for stratification of tumors when performing canine/human comparative analyses. Using gene expression, mutation and copy number data we performed a comprehensive subtype-specific comparison of CMGT and human breast tumors from The Cancer Genome Atlas (TCGA) and confirmed that the intrinsic subtypes also represent distinct biology in CMGT. We find that luminal A and basal-like are the two main subtypes in CMGT and underline the necessity of stratifying tumors by molecular subtypes when using CMGTs as models for human breast cancer.
Discussion
The dog is presumed to be a valuable comparative model for human breast cancer, but there is a need for increased knowledge about the molecular similarities as well as differences between mammary gland tumors in the two species. The PAM50 subtypes are defined for human breast carcinomas; however, in a comparative setting when exploring dogs as models, it is important to determine the relevance of these subtypes also in canine mammary gland tumors using the same method in both species. Here, we have studied the incidence and characteristics of the PAM50 subtypes in a cohort of canine mammary gland tumors. Overall, the PAM50 subtypes captured molecular subtype characteristics in dogs that are well known from human breast cancer. As in human breast cancer, there was a distinct dichotomy between basal-like and luminal canine tumors that could not solely be explained by hormone receptor status.
Nearly 60% of the CMGTs were of the luminal A subtype. These showed high correlation to the luminal A subtype centroid at a level comparable to human luminal A tumors and relatively low correlation to all other subtype centroids. Canine luminal A tumors were, in general, similar to human luminal A tumors with high expression of luminal epithelial genes such as
ESR1,
PGR,
FOXA1 and
MLPH and low expression of genes involved in cell proliferation.
PIK3CA mutations were more common in canine luminal A tumors than in other subtypes as is also the case in human tumors. The luminal A tumors carried markedly fewer mutations than basal-like tumors. This, combined with fewer copy number aberrations, indicates a more stable genome in canine luminal A tumors compared to basal-like tumors. The CMGT cohort includes both benign and malignant tumors and most benign tumors were of the luminal A subtype. The benign tumors displayed characteristics in line with their PAM50 subtype, both with regards to proliferation, ESR1 expression, mutations and copy number aberrations. This result is in line with the findings of Sørenmo et al. who proposed benign tumors as precursors of malignant tumors in canines [
69].
Tumors of the basal-like subtype constituted 27% of the canine cohort and showed gene expression characteristics well known from analyses of human basal-like tumors. They expressed low levels of luminal epithelial genes and high levels of multiple proliferation-associated genes indicating that these are highly proliferative tumors. The basal-like CMGTs were more often high-grade tumors and of the simple carcinoma histology type compared to luminal A tumors, similar to human breast cancer [
70]. However, in contrast to human breast cancer, where basal-like tumors paradoxically are associated with less lymph node invasion than luminal tumors [
71], there was a higher number of lymph node-positive cases among canine basal-like tumors compared to the other subtypes. This finding was also evident in one of the independent validation datasets that consisted of approx. 50% locally metastasized tumors. Hence, together with noticeably lower expression of gene signatures related to immune cells in canine basal-like tumors compared to human, this indicates intrinsic differences between canine and human basal-like tumors that might be explained by different immune responses. Exploring species-specific differences in the immune response towards tumors could generate valuable knowledge, especially in view of the emergence of immune therapy in breast cancer treatment [
72]. In addition, the basal-like canine tumors had markedly more copy number aberrations than the luminal A tumors in particular, which fits well with what is known about human breast cancer [
28]. Canine basal-like tumors also encompassed slightly higher mutation frequencies compared to other subtypes and carried several subtype specific aberrations such as mutations and deletions of
TP53, amplifications of
MYC and deletions of
CDKN2A, all known features of bona-fide basal-like human tumors [
28].
In our study, 8.9% of the CMGTs were characterized as HER2-enriched, however, correlation to the HER2-centroid was low for these tumors.
ERBB2-amplifications in canine tumors were only low level, and were not restricted to tumors of the HER2-enriched subtype. This indicates that the HER2-enriched subtype does not have the same prevalence and relevance in CMGTs, confirming results from several previous studies [
18,
66,
73]. Similar to the HER2-enriched subtype, luminal B did not emerge as a definite subtype in the canine cohort. Luminal B tumors were characterized by high expression of genes involved in cell proliferation, low expression of basal keratins and a high copy number aberration count, but displayed large variation in the expression of luminal epithelial genes,
ERBB2, as well as in TMB.
The canine samples in this study include mammary gland tumors from predominantly small companion dogs. To obtain a more comprehensive and nuanced overview of the potential of dogs as models for human breast cancer, these findings should be validated in an RNA sequencing dataset including tumors from larger dogs and working breeds; however, such datasets of sufficient size do not yet exist. Nevertheless, we were able to validate our main findings in two smaller independent datasets. For this study, we merged the canine and human cohorts and performed subtyping on the merged datasets. Additionally, subtyping was performed using 44 out of the original 50 genes. These factors, however, do not seem to have affected the subtyping results significantly. Different chromosomal composition in dogs and humans complicates comparative copy number analyses. Nevertheless, our study shows that such analyses need to be performed stratified by subtype.
Conclusions
CMGTs are highly heterogeneous biologically, and represent an unmet potential for modeling human breast cancer. Our study identifies many similarities between mammary gland tumors of dogs and humans, but also discovers important differences and emphasizes that the molecular subtypes should be taken into account when considering dogs as models for breast cancer. Generally, we found a high degree of similarity between canine and human tumors across the four main intrinsic subtypes, but we also pinpointed differences that are important when considering dogs as comparative oncology models. Knowledge obtained from canine/human comparative studies may contribute towards facilitating individualized treatment in dogs suffering from mammary gland tumors and such studies could therefore be of relevance and interest to both the veterinary and human medical communities.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.