Gene expression profiling of murine tumors and their comparison to human tumors identified characteristics relevant to individual murine models, to murine models in general, and to cancers of both species. First was the discovery that some murine models developed highly similar tumors within models, while others showed heterogeneity in expression and histological phenotypes. For the homogenous models, the study of progression or response to therapy is simplified because confounding variation across individuals is low. An example of this consistency even extended to secondary events that occurred within the TgC3(1)-Tag model, where many tumors shared the amplification and high expression of Kras2 (Figure ) - a feature also evident in a subset of human basal-like tumors.
In contrast to the 'homogenous' models are models such as TgWAP-T121
, DMBA-induced and Brca1Co/Co
, where individual tumors within a given model often showed different gene expression profiles and histologies. It is likely that these models fall into one of three scenarios that could explain their heterogeneity: the first, represented by the TgWAP-T121
], is that the transgene is responsible only for initiating tumorigenesis, leaving progression events to evolve stochastically and with longer latency periods. Such a model would likely give rise to different tumor subtypes depending on the subsequent pathways that are disrupted during tumor progression. A second possibility is that the initiating event generates genomic instability such that multiple distinct pathways can be affected by the experimental causal event, which may be the mechanism in the Brca1
-inactivation tumors. The third scenario is that the target cell of transformation is a multi-potent progenitor with the ability to undergo differentiation into multiple epithelial lineages, or even mesenchymal lineages (for example, DMBA-induced and Brca1Co/Co
); support for this hypothesis comes from Keratin IF analyses in which, even within a histologically homogenous tumor, two types of epithelial cells are present (Figures ). The presence of subsets of individual cells positive for markers of two epithelial cell types also supports this possibility (Figure ). Alternative hypotheses include the possibility that multiple cell types sustain transforming events, and also that extensive non-cell-autonomous tissue responses occur. Regardless of the paradigm of transformation for these heterogeneous models, the study of progression or therapeutic response will best be accomplished by first sub-setting by subtype, and then focusing on biological phenotypes.
There are at least two major applications for genomic comparisons between human tumors and their potential murine counterparts. First, such studies should identify those models that contain individual and/or global characteristics of a particular class of human tumors. Examples of important global characteristics identified here include the classification of murine and human tumors into basal and luminal groups. It appears as if four murine models developed potential luminal-like tumors (TgMMTV-Neu
, and TgWAP-Int3
), which is not surprising since both MMTV and WAP are thought to direct expression in differentiated alveolar/luminal cells [38
]; however, it should be noted that the luminal profile across species was not statistically significant, likely due to the lack of ER and ER-regulated genes in the murine luminal tumors. Several murine models did show expression features consistent with human basal-like tumors, including the TgC3(1)-Tag
-deficient models. The SV40 T-antigen used in the TgC3(1)-Tag
models inactivates p53 and RB, which also appear to be two likely events that occur in human basal-like tumors because these tumors are known to harbor p53
], have high mitotic grade and the highest expression of proliferation genes (Figure ) [2
], which are known E2F targets [40
]. The proliferation signature in human breast cancers is itself prognostic [41
], and is also predictive of response to chemotherapy [42
]. These data suggest that human basal-like tumors might have impairment of RB function and highlight an important shared feature of murine and human mammary carcinomas.
The finding that Brca1
loss (coincident with p53
mutation) in mice gives rise to tumors with a basal-like phenotype is notable because humans carrying BRCA1
germline mutations also develop basal-like tumors [3
], and most human BRCA1
mutant tumors are p53-deficient [43
]. These data suggest a conserved predisposition of the basal-like cell type, or its progenitor cell, to transform as a result of BRCA1
, and RB
-pathway loss. Most DMBA-induced carcinomas also showed basal-like cell lineage features, suggesting that this cell type is also susceptible to DMBA-mediated tumorigenesis. Finally, some TgMMTV-Wnt1
tumors showed a combination of basal-like and luminal characteristics by gene expression, which is consistent with the observation that tumors of this model generally contain cells from both mammary epithelial lineages [45
The second major purpose of comparative studies is to determine the extent to which analyses of murine models can inform the human disease and guide further discovery. An example of murine models informing the human disease is encompassed by the analysis of the new potential human subtype discovered here (that is, claudin-low subtype). Further analysis will be necessary to confirm whether this is a bona fide subtype; however, the statistically significant gene overlap with a histologically distinct subset of murine tumors suggests it is a distinct biological entity. A second example of the murine tumors guiding discovery in humans was the common association of a K-Ras containing amplicon in a subset of human basal-like tumors and in the murine basal-like TgC3(1)-Tag strain tumors.
An important caveat to all comparative studies is that there are clear biological differences between mice and humans, which may or may not directly impact disease mechanisms. A potential example of inherent species difference could be the aforementioned biology associated with ER and its downstream pathway. In humans, ER is highly expressed in luminal tumors [1
], with the luminal phenotype being characterized by the high expression of some genes that are ER-regulated like PR
], and other luminal genes that are likely GATA3-regulated, including AGR2
]. In mice, ER expression is low to absent in all the tumors we tested, as is the expression of most human ER-responsive genes. This finding is consistent with previous reports that most late-stage murine mammary tumors are ER-negative ([47
] and references within). However, it should be noted that two human luminal tumor-defining genes (XBP1
], were both highly expressed in murine luminal tumors (Additional data file 2). Taken together, these data suggest that the human 'luminal' profile may actually be a combination of at least two profiles, one of which is ER-regulated and another of which is GATA3-regulated; support for a link between GATA3
and luminal cell origins comes from GATA3
loss studies in mice where the selective loss of GATA3
in the mammary gland resulted in either a lack of luminal cells, or a significant decrease in the number and/or maturation of luminal cells [48
]. These results suggest that, in the mouse models tested here, the ER-regulated gene cassette that is present in human luminal tumors is missing, and that the GATA3-mediated luminal signature remains. Due to the partial luminal tumor signature in mice, we believe that the murine luminal models, including TgMMTV-Neu
profiled here, best resemble human luminal tumors and more specifically possibly luminal B tumors, which are luminal tumors that express low amounts of ER and show a poor outcome [2
]. While human HER2+/ER- subtype tumors and the murine TgMMTV-Neu
, and TgWAP-Myc
fall next to each other in the intrinsic-shared cluster (Figure ), all of the other data argue against this association. A few murine ER-positive mammary tumor models have been developed [50
]; however, none of these models were analyzed here.
Of note, many expression patterns detected in this study were observed in only one species (Additional data file 5), and it is possible that some of these differences may arise from technical limitations rather than reflect important biological differences. Comparison between two expression datasets, especially when derived from different species, remains a technical challenge. Thus, we acknowledge the possibility that artifacts may have been introduced depending on the data analysis methodology. However, we are confident that the analyses described here identified many common and biologically relevant clusters, including a proliferation, basal epithelial, interferon-regulated and fibroblast signature, thus showing that the act of data combining across species did retain important features present within the individual datasets. There are many murine models of breast cancer that we did not look at in this study and many more will be developed. Like the 13 models we discussed here, we would expect that some of these models will have overlapping gene expression patterns with human subtypes while others will not. We believe that additional studies with larger numbers of samples, including more diversity from each species, is warranted. These analyses do confirm the notion that there is not a single murine model that perfectly represents a human breast cancer subtype; however, the murine models do show shared features with specific human subtypes and it is these commonalties that will lay the groundwork for many future studies.