Here, we have analysed histological diagnoses from an archive of 129,258 records of submissions of samples from bitches, including 13,401 mammary epithelial neoplasms, sent for histological assessment to a single histopathology laboratory between 2008 and 2021. In multivariable analysis, we found a significant association between breed, age and neuter status and the odds that a sample submitted for histological diagnosis would be diagnosed as a mammary epithelial tumour, as opposed to any other diagnosis. We also found a significant association in multivariable analysis between breed, age and neuter status and odds of diagnosis of a malignant versus benign mammary epithelial neoplasm and also the odds of being diagnosed with multiple versus single mammary epithelial neoplasms. Our study was not designed to determine how breed, age and neuter status affect odds of mammary tumour diagnosis relative to the general canine population as this requires comparison to the entire population of canines in a particular region as a denominator. We (D. Varney, D. O'Neill, M. O'Neill, D. Church, A. Stell, S. Beck, M. Smalley and D. Brodbelt: The epidemiology of mammary tumours in bitches under veterinary care in the UK in 2016, submitted) and others [3, 7, 9, 24,25,26,27] have, however, addressed this question elsewhere. Our study was primarily designed to address how these variables might influence clinical features of tumours once histologically diagnosed; however, we also took advantage of our dataset to understand how they affect the likelihood that any sample undergoing histological assessment was diagnosed as a mammary epithelial neoplasm rather than any other histological diagnosis.

An increased likelihood of a sample submitted being histologically diagnosed as a mammary epithelial tumour in older animals reflects mammary neoplasia (and in particular adenoma/ adenocarcinoma) as a disease of older dogs [29, 43]. Similarly, an increased likelihood in entire animals would be consistent with studies showing that neutered animals are protected against mammary neoplasia [11,12,13, 44,45,46,47] (Varney D, O'Neill D, O'Neill M, Church D, Stell A, Beck S, Smalley M, Brodbelt D: The epidemiology of mammary tumours in bitches under veterinary care in the UK in 2016, submitted). However, caution should be exercised in the interpretation of our results, as our findings specifically relate to samples submitted for histological examination and sample submission bias is likely to be a factor. This is discussed further below.

Using the large dataset, of 13,401 cases submitted with one or more confirmed epithelial-origin mammary neoplastic lesions (sometimes in addition to other pathologies) available to us, we next asked whether the age, neuter status and breed variables were also associated with the clinical behaviour of the mammary neoplasms in these cases. We examined the odds of developing a malignant (as opposed to benign) lesion and the odds of being diagnosed with multiple (as opposed to single) lesions. In all cases, our analysis was based on odds compared to a baseline group (age < 3 years; ‘entire’ neuter status and Crossbreed animals) (summarised in Table S9).

A number of studies have assessed the relationship between age at diagnosis and the risk of being diagnosed with benign or malignant mammary lesions [7, 20, 26, 27, 48,49,50] and of being diagnosed with single or multiple mammary lesions [26] (although the proportion of dogs with single or multiple lesions varies widely between reports, potentially a result of differing proportions of breeds within the study populations) [49, 51]. We find in multivariable analysis that older dogs are more likely to be diagnosed with malignant disease and with multiple lesions. This is consistent with previous reports that dogs with malignant tumours are more likely to be older than dogs with benign tumours [26, 48, 49], that age is an independent prognostic factor correlating with poor survival [20] and that older animals tend to have higher grade ER negative tumours [27]. A number of mechanisms may drive this increased risk of malignancy in older dogs. For example, it may simply be a stochastic process, with more mutations required to generate a malignant tumour than a benign one. An example of this may be carcinoma arising in complex adenoma/benign mixed tumour. Cellular aging may also be a factor as a result of telomere shortening, which would lead to increased risk of genomic abnormalities [52]. Finally, older individuals tend to have a dysfunctional immune system [53], with potential for reduced immunosurveillance permitting tumour growth and metastasis [54].

Neuter status is also well established as a risk factor for development of mammary epithelial neoplasia. Multiple studies, including our own, have demonstrated that neutered bitches have a reduced risk of neoplastic mammary disease compared to entire animals [10, 11, 44,45,46,47] (summarised in [13]). Here, we have now shown in multivariable analysis that neutered animals are also at reduced odds of developing multiple lesions; however, if they do develop neoplastic mammary disease, they are at increased odds that this will be malignant. It is notable that ER negative canine mammary tumours (CMT) are more likely to be malignant than ER positive CMT [16,17,18]. Furthermore, a study which examined the link between neutering, serum estrogen levels and CMT hormone receptor status demonstrated that ER negative tumours in entire animals with high serum hormone levels had a longer time to metastasis than such tumours in neutered animals [15]. A protective effect of estrogen via non-receptor mechanisms was suggested.

Some studies have found no links between breed and predisposition to develop malignant as opposed to benign lesions [50] and no difference in malignancy between cases presenting with single or multiple masses [55]. However, others have suggested particular breeds are more likely to develop malignant mammary tumours (Samoyed, Dobermann, Schnauzer and Yorkshire Terrier) [56] (although notably when cancer of all sites was considered Aupperle-Lellbach and colleagues found Yorkshire Terriers among the breeds more likely to develop benign tumours [24]). There is little, if any, information on the link between breed and risk of developing single as opposed to multiple lesions. We find a significant association between breed and both of these aspects of tumour biology. The patterns of odds of developing malignant disease, or multiple lesions, across individual breeds showed that no breed was found to be at increased odds of both outcomes. They were either at increased odds of malignant disease and decreased odds of multiple lesions (e.g. King Charles Cavalier Spaniel, Golden Retriever, Husky, Staffordshire Bull Terrier) or vice versa (e.g. Cocker Spaniel, Setter, Yorkshire Terrier) or they were at significantly altered odds of one outcome with no change in the other outcome.

The simultaneous appearance of multiple mammary epithelial neoplasms in cases presenting with multiple lesions suggests a ‘field cancerisation’ model. Field cancerisation was first proposed by Slaughter and colleagues in 1953, to describe the development, in oral squamous cell carcinoma, of regions of the oral epithelium which are clinically apparently normal but in which multiple independent primary squamous cell carcinomas continually arise [57]. It is now considered that field cancerisation is the product of an underlying preneoplastic stem cell which has acquired mutations giving it a competitive advantage overing neighbouring cells, allowing clonal progeny of the original preneoplastic cell to spread. This then creates a large target population requiring fewer mutational events for full transformation [57,58,59,60,61]. We suggest that the presence of multiple independent tumours, a high proportion of which are non-simple tumours composed of multiple cell lineages, supports a model that in some dogs the entire mammary epithelium is a field of preneoplastic (stem) cells. Furthermore, we suggest that there can be genetic factors predisposing to the development of a preneoplastic mammary field. A number of candidate genes have been identified which affect cell competition and have known roles in cancer, for example p53 [60], so these would be an excellent starting point for future studies.

Interestingly, Gunnes and colleagues [49] reported that in bitches presenting with multiple tumours, the chance that two tumours would have the same diagnosis and level of malignancy was greater than would be expected by chance alone, leading them to conclude the existence of a hormone-driven field cancerisation effect. They also suggest that there might be breed variations in predisposition to malignant (as opposed to benign) tumours but their study was not sufficiently powered to identify specific at-risk breeds.

Links between canine mammary tumour histological subtype and prognosis are well established [19,20,21,22] and the prevalence of different histological subtypes within canine mammary neoplasia has also been assessed in a number of studies, for example [7, 49, 50]. Results differ widely from study to study, likely as a result of differing study populations and differing histological interpretations. Salas and colleagues report similar frequencies of benign and malignant tumours in their study, with more epithelial-type than mixed neoplasms [7]; Gunnes and colleagues reported 61% of examined tumours were benign, 39% were malignant, with complex adenoma and complex carcinoma the most frequent diagnosis in each category [49]; Ariyarathna and colleagues report 56% of examined tumours as malignant (simple carcinomas being most common) and 44% benign (benign mixed tumours being most common). We find that in cases with single lesions being diagnosed, the most common lesions were the benign mixed tumour (23.6%), adenoma, complex (14.8%), carcinoma, simple (subtype not otherwise specified (11.4%), carcinoma, complex (10.7%) and carcinoma, ductal (8.0%).

While a diagnosis of mixed and complex (non-simple) mammary neoplasms is common in the dog, no studies have yet reported a link between a diagnosis of non-simple tumours and development of multiple lesions. The link between complex/mixed lesions and presentation with multiple lesions (with a potential field cancerisation effect) is of particular interest as the origin of complex/mixed lesions remains unclear. Complex tumours contain distinct proliferating luminal epithelial and myoepithelial populations [35] Mixed tumours contain, in addition, mesenchymal elements [35]. It is possible that complex/mixed tumours are polyclonal in origin, with separate transformed luminal and myoepithelial cells (in the case of a complex tumour) or luminal, myoepithelial and mesenchymal stem cells (in the case of mixed tumours) all contributing to a neoplasm presumably in response to a highly localised tumour promoting factor, such as an inflammatory signal. Alternatively, these tumours might arise monoclonally from a mammary epithelial stem cell capable of undergoing both luminal and myoepithelial differentiation as well as metaplastic potential. Previous studies have addressed this question by analysis of mitochondrial DNA mutations in the epithelial and mesenchymal elements [62] or by analysis of immunohistochemical staining patterns and DNA ploidy of the different components of the tumour [63]. These analyses have suggested some tumours may be polyclonal while others may be monoclonal, but many of the analyses have not been informative. A definitive answer on the aetiology of these tumours awaits further study.

Our analysis supports the hypothesis that neuter status, age and intrinsic biological and genetic factors all influence the heterogeneity of clinical presentation of canine mammary neoplasia. The term ‘heterogeneity’ includes both ‘intra-tumour heterogeneity’, the genetic, epigenetic, phenotypic and/or behavioural differences in cells within a tumour (including both neoplastic cells and non-neoplastic cells such as tumour-associated fibroblasts and macrophages), and inter-tumour heterogeneity, the classification of tumours as different histotypes which may have different clinical behaviours (e.g. benign or malignant disease) and approaches to therapy [64]. Tumour heterogeneity may be considered to arise from two processes. In the early phase of tumour development, it arises from the interaction between cell of origin of the tumour and the initiating genetic lesions occurring in that tumour [65,66,67]. Then, as the tumour progresses, the random generation of progressively more mutated clones, combined with selective pressures on these variant clones, lead to tumour heterogeneity through a Darwinian evolutionary process [68]. We would argue that the former process is the main driver of inter-tumour heterogeneity while the latter is the main driver of intra-tumour heterogeneity. Of course, these are not clear-cut divisions, considering that a change in the proportion of different cell types within a tumour (for example, cells that express hormone receptors in a mammary neoplasm) could lead to a breast cancer being reclassified as progressing from ER positive to ER negative disease i.e. a change in intra-tumour heterogeneity leads to a change in inter-tumour heterogeneity.

Modelling the interaction between the cell of origin and genetic lesion as a determinant of mammary tumour heterogeneity in genetically modified mice has confirmed the principles that tumour histotype is driven by the interactions between cell of tumour origin, initiating genetic lesion and in some cases developmental history of the gland [65,66,67, 69]. However, this is an artificial system in which candidate genes are bred into mice in conditional knockout/overexpression scenarios and in which a limited number of tumour histotypes develop with (for the most part) little clinical relevance either to human or veterinary medicine. In contrast, canine mammary tumours combined with the power of dog genetics offer a system in which, rather than choosing the genes of interest and working forward to understand what, if any, effect they have on tumour phenotype, one can work backwards from tumour phenotype to elucidate the underlying genetics. The first stage of this is to use an epidemiological approach to establish associations between tumour biology and breed and develop hypotheses. These can then be tested in case–control genomic studies within and between breeds [70] to identify loci associated with the particular aspect of biology (e.g. presentation with multiple tumours or a diagnosis of malignant disease) which can be taken forward into mechanistic studies. The results we present here represent the first stage of this process.

Our study has limitations and caveats, the principle one being selection bias (i.e. the dogs in the current study may not be representative of the general caseload of canine mammary tumours as a whole) [71]. There are many factors which affect clinical decision making when deciding whether to submit lesions for histological analysis and these could all bias the population of dogs that appear in the current study ( a ‘biopsy-only’ dataset). For example, dogs with very severe disease or poorly resectable lesions may undergo euthanasia or palliative care rather than surgical resection and histological analysis. In contrast, an entire bitch which presents at a primary care practice with a mammary mass may be more likely to have that mass sent for histology than a neutered bitch, simply because of the previously postulated links between neuter status and mammary tumour risk. Furthermore, financial limitations regarding the cost of surgery and biopsy submission may mean that dogs owned by owners of lower socio-economic status or uninsured dogs may be less likely to appear in the current dataset. Previous studies have shown that both breed and neutering status in dogs are associated with owner socioeconomic status, therefore if owner demographics are biased in the dataset, the distribution of breeds or neuter status may be altered [4]. The socio-economic circumstances or age of owners may also have resulted in an underrepresentation of benign lesions, for example, if individuals with lower economic means or who face challenges accessing veterinary care were less inclined to take a dog with a mammary mass for veterinary attention unless it shows obvious signs of malignancy, such as extremely rapid growth or obvious morbidity. Furthermore, in certain breeds a single benign lesion may be more difficult for an owner to detect owing to body shape or conformation, whereas multiple lesions may be more likely to lead to an owner seeking attention. We found an inverse correlation between average breed body weight and likelihood of a histopathological sample being diagnosed as a mammary epithelial tumour. However, while a predisposition for mammary tumour development in small dogs has been reported [7, 29] population-based case–control studies have identified both small and large breeds as being at higher odds of developing mammary epithelial neoplasia [3, 7, 9, 24,25,26]. It may be that smaller dogs are more likely to be picked up and have mammary masses discovered by owners. Another alternative is that small dogs live longer than large dogs, and mammary neoplasia is a disease of older animals. Overall, selection bias could have affected the observed associations between breed, age and neuter status and mammary tumour number or histotype.

The study may also have been affected by missing data because in veterinary clinical practice, there is not a standardised approach to lumpectomy vs mammectomy vs full mammary strip removal. While full mammary strips were frequently submitted (and indeed such submissions were encouraged), we cannot exclude that some practices did not submit all tissue or all lesions present in a dog for analysis, leading to reporting of cases with more than one lesion as single lesion cases. However, it is more likely that in cases which present with only one, or a small number of, palpable lesions, all will be sent for analysis, whereas in cases presenting with many such lesions only a representative sample might be examined. Such an approach would still result in the case being correctly categorised as having multiple lesions. It is more likely that cases could be incorrectly categorised as having single lesions, if only a single palpable lesion is detected and sampled, but small, clinically undetectable, lesions are already present.

Once tissue has been submitted, given the extensive experience of VPG Histopathology, it is highly unlikely that cases would be incorrectly categorised as ‘benign’ rather than ‘malignant’, even during histological analysis of large pieces of tissue (such as whole mammary strips) owing to the meticulous nature of the analysis. However, we cannot definitively exclude that occult malignant cell clusters may be present in an otherwise benign tissue and, if small, these cases could theoretically be miscategorised as benign-only.

We did not carry out Bonferroni corrections for multiple comparisons, which may have caused some Type-1 errors. However, this is usually considered overly stringent where there may be correlation between variables [72]. Furthermore, although unlikely, we cannot exclude the possibility that there may be multiple samples submitted for histological examination from the same animal on different occasions. To definitively exclude this would require access to information that would identify owners and therefore such data was withheld from the current study.

We also did not account for the expected breed lifespan when considering risk factors for cancer. However, breed and neuter status may affect the years-at-risk of dogs, and thus certain breeds with longer lifespans, or neutered dogs (shown to live longer on average) may appear to be more at risk simply because they experience more years-at risk of disease. The inclusion of this complex variable was beyond the scope of this study because a reliable published lifespan could not be sourced for all of the included breeds, however future analyses considering years-at-risk would be valuable for validation of the conclusions presented here.

Importantly, our study represents only a snapshot in the clinical pathway of each animal diagnosed. The diagnostic records available to us are not linked to clinical outcome, so we do not know whether a diagnosis of single or multiple lesions affects prognosis although it is clear that malignant disease has a worse prognosis than benign disease (the prognostic significance of histological subtypes is also established) [19, 20]. Both benign and malignant disease was seen in cases diagnosed with both single and multiple lesions, and previous findings have also shown no relationship between the presence of single and multiple masses and a malignant diagnosis [55]. It is unclear whether having multiple lesions is an independent prognostic factor. To test this would require a prospective study in order to ensure that potential confounding factors are controlled for (e.g. the lack of a uniform approach to treating mammary masses). Testing this would be an important follow-up to the current study.

Our study also only represents a snapshot of the underlying biology of the disease. We cannot exclude, for example, that a bitch diagnosed with a single malignant lesion might have, in the future, gone on to develop multiple lesions, or that an animal with multiple lesions that has a full mammary strip, and all those lesions are diagnosed as benign after surgery, might have developed a malignant tumour at a later date if no interventions were performed. To draw definitive conclusions that a case with a single lesion diagnosis would not have later been diagnosed with multiple lesions, or that a case with one or more benign lesions would not have later been diagnosed with one or more malignant lesions would require surgeons to perform only individual lumpectomies on dogs presenting with mammary masses, no matter how many masses they present with and no matter how many times they return to clinic. This is not consistent with the best welfare of the animals involved. However, as our study sample overall captures a very large population over a period of years (canine lifespans), generalisations seem reasonable, as the specific sampling time in the course of the disease becomes less important for each individual.

Therefore, although the caveats above must be kept in mind, we have found significant associations between the breed, age and neuter status of a bitch, and whether a presentation for mammary neoplasia is likely to be for single lesions or multiple lesions, and whether those lesions are likely to be benign or malignant. Furthermore, non-simple lesions are enriched in cases presented with multiple neoplasms. We therefore suggest that underlying genetic factors can affect tumour heterogeneity, by influencing clinical behaviour (the development of benign or malignant disease), tumour number and cellular composition. Environmental influences such as aging are also likely to play a role. The mammary epithelium of breeds at higher risk of presenting with multiple neoplastic mammary lesions may be a pre-neoplastic field genetically primed for tumour development. Case–control genomic studies, and mechanistic evaluation, have the potential to identify, in an unbiased manner, genes driving mammary tumour behaviours and thus such studies could substantially advance our understanding of the drivers of mammary tumour formation and tumour heterogeneity and, ultimately, identify new targets for therapy.