Canine Skin Cancer and the Art of Classifications

Classifications are the basis of morphologic pathology. Pathologists use classifications to give the correct name to what is beneath their eyes. As pathologists analyzing a tissue section, we are faced with a tree of decisions aiming to assign the correct name and modifiers to the process viewed under the scope. This tree in diagnostic pathology often starts with the type of lesion (tumor, degeneration, inflammation), followed by the identification of tumor origin, subtype, and grade. Our attitude toward this process historically assigns us as “lumpers” or “splitters.” The well-known lumper-splitter dichotomy may be found in any discipline that places entities into defined categories and may match with individuals’ interests or motivations.

The splitters are like insect collectors, who are fascinated by diversity and want to have one of each in their collection and understand each as a discrete entity. The lumpers are like entomologists, who want to understand the biology behind the diversity. Similarly, these pathologists tend to see the continuum among diagnoses; the diagnosis might even be irrelevant because the focus is on the underlying biology. The lumpers have a general view and assign terms broadly, assuming that certain differences are not relevant. These pathologists focus on underlying biology that groups the processes according to pathogenesis and how they correlate with outcome.

Attracted by the diversity of discrete entities, a splitter takes precise definitions and creates new categories to classify samples that differ in key ways. Splitting entities according to their morphologic appearance, cell of origin, or differentiation is necessary to find clinical, pathologic, and pathogenic correlates. Despite this, splitting tumors into various categories, even if they bear similar biological behaviors, may seem a pointless exercise. Still, this may be of relevance since, even if different biological entities have a similar prognosis, they may bear diverse parameters useful in the elaboration of grading systems, clinical outcome groups, or specific targeted therapies, thereby providing a way of exploring pathogenesis that might later become relevant to clinically important outcomes. Moreover, following the division, we can opt to group again once excessive splitting is revealed to be pointless.

In pathology, these extremes are both necessary and useful when put into context.

Over the years, diagnostic pathology has evolved from pure morphology, which should still be considered one of our most powerful tools, to the integration of several techniques in a multiprofessional approach to diagnosis, classification, grading, and prognostication of tumors. Immunohistochemistry is utilized routinely for diagnostic purposes and to incorporate objective parameters into the process of tumor diagnosis and tumor grading (eg, Ki-67 labeling index) and for the choice of targeted therapies (eg, imatinib in human gastrointestinal stromal tumors and canine mast cell tumors). Diagnostic molecular tools such as clonality assessment have also been widely implemented and been made largely available, while molecular genetics for the identification of specific entities—as occurs for human sarcomas—have not yet emerged in routine veterinary pathology. The era of “omics” sciences is producing an exponential amount of data that still need practical interpretation at the diagnostic and clinical levels. Despite the complexity of our work, we should bear in mind that diagnostic classifications applied for clinical purposes should identify and validate new biological entities and/or confirm the relevance of old ones.

The relevance of correct classifications and grading systems lays in their role in clinical decision making. The end point should be to supply clinicians with simple, meaningful classifications providing practical prognostic stratification of tumoral entities with a closer link between the diagnostic label and the corresponding prognosis and treatment options. For pathologists, this means effective grading systems and more informative prognostic tools. Thus, new classifications/categorizations should be validated by studies demonstrating intra- and interobserver reproducibility and verified in the setting of epidemiologic studies, prior to the final integration of the novel categories into the diagnostic process. ² Therefore, we need to promote the cooperation of large groups of pathologists and oncologists to integrate our classifications with clinical relevance. Lumper and splitters need to remember that classifications fail when they become too complicated or nonreproducible and/or do not correlate with response to therapy or prediction of survival. One example of a useful information is the prognostic value of the mitotic count cutoffs for differentiating between cutaneous melanoma and melanocytoma. In contrast, identification of the different subtypes of perivascular wall tumors, although they correlate well with the anatomic stratigraphy of the vessel wall, does not bear major clinical or prognostic relevance for most entities. For hair follicle tumors, a true way to predict behavior is still not readily available.

In the current issue, 2 articles address some aspects of the work that we should program in a more coordinated fashion. They investigate the epidemiology of skin tumors in Swiss dogs ³ and the classification of canine cutaneous tumors through hierarchical cluster analysis of cytokeratin expression. ⁵

Epidemiologic studies provide the most direct and relevant evidence for an association between a suspected risk factor (eg, breed) and disease. Fundamental to the assessment of cancer risk are the incidence rates, often presented per 100 000 patients per year in a population. Incidence is a measure of disease occurrence, as it describes the manifestation of new cancer cases. In this instance, epidemiologic studies may identify breed-specific issues that could help breeders and clinicians change selection strategies, or they may provide evidence on country-level mortality data (related to pollution, exposure to ultraviolet light, or other environmental factors). In this issue, Graf and coworkers provide a detailed population-based study of canine skin tumors with a comprehensive statistical evaluation applied to >11 000 dogs from the Swiss Canine Cancer Registry spanning the years 2008–2013. ³ The study expands information on incidence and shows previous non-identified breed and tumor predispositions. An incidence rate of 372 skin tumors per 100 000 dogs per year included a higher incidence observed in purebred dogs. These data are comparable to those of Italian canine populations ¹ and higher than the latest publications from the United Kingdom. The highest overall tumor incidence rate was reported in Giant and Standard Schnauzers, which had a 4-fold greater incidence of skin tumors, with Boxers surprisingly ranking as the sixth breed most frequently affected by skin tumors. ³ These data contrast those of 2 other recent canine cancer epidemiologic studies reporting Boxers as one of the breeds ¹ or the breed ⁴ with the highest cancer rate and incidence. One of these studies was also based on the Swiss canine population. ⁴ Interestingly, cutaneous benign tumors prevailed, and the most common tumor types were mast cell tumors, lipomas, and hair follicle tumors. ³ The high frequency of particular tumors in certain breeds opens more questions on the role of the genetic background in favoring the development of specific cancers, an aspect that needs to be investigated in more depth.

The second report expands the knowledge and diagnostic tools for canine cutaneous epithelial tumors with respect to cytokeratin and stem cell marker expression profiles, analyzing their expression by hierarchical cluster analysis. In the report by Kok and coworkers, immunohistochemistry was utilized for the recognition of specific neoplastic entities to assist in their categorization and classification. ⁵ Immunohistochemistry complements skin tumor morphologic diagnosis and is seen as an aid in its more precise classification, providing a useful diagnostic dendrogram. Interestingly, morphology correlates distinctively with cytokeratin expression patterns, supporting once more the relevance of keratin intermediate filament expression in tumor diagnosis and probably in tumor biology and differentiation.

Also, this work provides a stem cell expression profile especially aimed to the study of the complexity of hair follicle tumors. Moreover, on the basis of stem cell marker expression, the authors hypothesize that tumorigenesis for certain canine skin tumors may originate from alterations in the process of regeneration of epithelial stem cells, as in following exposure to carcinogenic agents. Some hypotheses regarding hair follicle tumor pathogenesis are still open. Two fascinating opposing hypotheses could be engaged from this work—namely, that tumors possibly derive from the different stem cell pools in various anatomic parts of the hair follicle or, on the contrary, that they emerge from a single stem cell pool but are driven by differential expression of oncogenes to varied patterns of differentiation.

These studies represent useful knowledge on occurrence and frequency, assist in classification, and provide interesting insights about predisposition and pathogenesis of canine skin tumors. These new insights give some answers and open more questions that could be explored in canine cancer development from many perspectives.