Advances in genomic signatures have begun to dissect breast cancer heterogeneity and application of these signatures will allow the prediction of which pathways are important in tumor development. Here we used genomic signatures to predict involvement of specific E2F transcription factors in Myc-induced tumors. We genetically tested this prediction by interbreeding Myc transgenics with mice lacking various activator E2F alleles. Tumor latency decreased in the E2F1 mutant background and significantly increased in both the E2F2 and E2F3 mutants. Investigating the mechanism behind these changes revealed a reduction in apoptosis in the E2F1 knockout strain. E2F2 and E2F3 mutant backgrounds alleviated Myc proliferative effects on the pregnant mammary gland, reducing the susceptible tumor target population. Gene expression data from tumors revealed that the E2F2 knockout background resulted in fewer tumors with EMT, corresponding with a reduction in probability of Ras activation. In human breast cancer we found that a low probability of E2F2 pathway activation was associated with increased relapse-free survival time. Together these data illustrate the predictive utility of genomic signatures in deciphering the heterogeneity within breast cancer and illustrate the unique genetic requirements for individual E2Fs in mediating tumorigenesis in both mouse models and human breast cancer. Cancer Res; 71(5); 1924–32. ©2011 AACR.
Data at the genomic scale have been used to characterize the heterogeneity in human tumors and in mouse model systems. Given that most cellular events result in perturbations in gene expression, the characteristic changes in gene expression reflect a measure of pathway activation. Together, this suggests that the higher order present within gene expression data can be analyzed to reveal the individual involvement of various pathways that cumulatively contribute to tumor formation. In order to interpret these patterns embedded within gene expression data, an approach has been developed to query the data for the probability of discreet pathways and signaling modules being activated. This approach is built upon established statistical methods (1–3). Subsequently, the approach has been used to characterize human tumors, mouse mammary development, and recently to characterize the heterogeneity present within Myc-induced tumors (2, 4, 5). This method uses controlled training data for given pathways to develop a signature that is then applied to tumor gene expression data, enabling the prediction of probability of activation of various signaling pathways.
When Myc tumors were examined with this pathway signature technique, it was seen that there were a number of distinct histological classes of Myc-induced tumors that had characteristic gene expression patterns (4). Moreover, these various classes of Myc tumors, including papillary, microacinar, and EMT/ squamous were associated with unique patterns of signaling pathway activation. Together, this allowed a characterization of the heterogeneity present within a mouse model system. The idea of heterogeneity in mouse models is critical when considering comparing the mouse models to the human condition. In similar work, Met has also been shown to induce heterogeneous tumors, including tumors with EMT characteristics and the associated gene expression patterns (6). These studies have illustrated the utility of examining large numbers of samples in mouse models systems and have refined the view of heterogeneity within mouse models induced by overexpression of oncogenes.
In examining the Myc model tumors, it is critical to recall the established role of the activator E2F transcription factors in mediating Myc effects. Indeed, E2F1, E2F2, and E2F3 are categorized as transcriptional activators (7). Overexpression of these E2Fs is sufficient to drive cells into S phase (8–10) and conditional loss of these transcriptional activators results in a block in proliferation (11). Microarray analysis has illustrated the overlapping and distinct roles of these E2Fs in mediating proliferation (12). In addition to the importance of E2Fs in proliferation, additional work has illustrated that E2F1 has a unique role in the induction of apoptosis in vitro (13–16). In the context of Myc activation in vitro, E2F1 has been shown to regulate Myc-induced apoptosis whereas E2F2 and E2F3 are required for mediating S-phase entry (17). Considered together, these proteins are critical in mediating transcriptional regulatory events resulting in activation of both proliferative and apoptotic pathways.
In order to consider the role of the E2Fs in Myc-induced breast cancer, it is also vital to examine the role of the E2Fs in normal mammary gland development. Previous work has illustrated that the E2Fs are involved in the development of the mammary gland (5). Of the activator E2Fs, E2F1, and E2F3 were seen to be required during outgrowth of the mammary gland and during branching. Interestingly, E2F1 was found to have no role in apoptosis associated with involution whereas E2F3 was found to be involved in this process. Together, these results indicate that the E2Fs, and in particular E2F1 and E2F3, are involved in the normal development of the mammary gland. Any impact on tumor development in the absence of an E2F should be considered in the light of these findings. Interestingly, a recent study on the role of the E2Fs in Myc-induced tumors in the lymphatic system revealed individual roles for the E2Fs, largely dependent upon the context of expression (18). In short, it was concluded that loss of E2F2 accelerated tumor formation in that tissue. Similar experiments in skin revealed that E2F2 had an integral role in tumor development (19).
In this study, we have predicted E2F activity in microarray data from Myc-induced mammary tumors in a mouse models. We then tested these predictions through genetic crosses between E2F mutant mice and MMTV-Myc transgenics, revealing the importance of the E2Fs in tumor latency and incidence. These results also correlated with similar findings in human breast cancer. Together, our data have illustrated a context specific role for the E2Fs in mediating breast cancer formation.
Materials and Methods
Animal use and husbandry was in accordance with institutional and federal guidelines. Mice used were previous described including MMTV-Myc (Leder Myc mice), E2F1 knockouts, E2F2 knockouts, and E2F3 heterozygous mice. Upon generation of mice in the desired background, the females were maintained in a constant breeding program. Mice were monitored twice weekly for tumor development by palpation and visible tumors were measured twice weekly. Mice were euthanized when the primary tumor reached 20 mm in the largest dimension and tumor material was taken for both flash frozen samples and histological analysis. Kaplan–Meier curves for tumor latency were generated using GraphPad Prism. Tumor volume calculations were made twice weekly using the formula:
tumor volume = length × width × height × 0.5236
Tumors were flash frozen in liquid nitrogen, with a piece fixed in 10% formalin for histological analysis. Thin sections from involuting mammary glands were also used to assess cell death. Apoptosis was detected by labeling the nicked ends of DNA through a TUNEL analysis using the In Situ cell death detection kid POD (Roche Applied Science). In addition to histology and TUNEL analysis, sections were also used for immunohistochemistry using mouse monoclonal Proliferating Cell Nuclear Antigen (PCNA) (PC10) (Santa Cruz Biotechnology) (1:300 dilution) with the M.O.M. (Mouse on Mouse) staining procedures (Vector Laboratories). Signal detection was achieved through DAB staining (Vector Laboratories). Sections were briefly counterstained in hematoxylin and were mounted in Permount.
Gene expression analysis
RNA from 80 tumors, 20 each from MMTV-Myc and MMTV-Myc in E2F1 knockout, E2F2 knockout and E2F3 heterozgous backrounds was prepared using the RNeasy Midi kit from Qiagen. Sample RNA was submitted to the Duke Microarray Facility for quality control and processing using Affymetrix 430A 2.0 microarrays. Expression values were submitted to the Gene Expression Omnibus (GEO) with reference number GEO24594. Analysis of the data was completed as previously described (4), using established methods for clustering and for pathway signature analysis (2, 3). The generation of E2F signatures has been previously described. Briefly, the signatures are generated from training gene expression data obtained from cell lines engineered to be controls or expressing individual E2Fs. Using a binary regression algorithm we then generate a signature for each E2F. The probability of signature activation is then calculated for additional tumor microarray data sets. The algorithm, code and a step by step code narrative for this procedure is found in the supplementary information in a previous publication (20). In addition, the Matlab script and a tutorial are found online at; http://www.duke.edu/∼dinbarry/BINREG/ (21). The E2F1 signature is described in great detail in the supplemental methods from two prior publications (3, 20). The generation of the E2F2 signature was also described by Huang et al. (3). The E2F3 and Ras signatures have also been previously described (2). Human breast cancer microarray data and clinical annotation were downloaded from the GEO data sets GSE7390 and GSE2990(22, 23), merged using Bayesian Factor Regression Modeling (20) and E2F probabilities were examined in the data. Relapse-Free Survival times were plotted for the highest and lowest quartiles using GraphPad Prism.
In order to use the predictive capacity of genomic analysis, we began by generating pathway activation predictions in Myc-induced tumors. Interestingly, our previous work indicated significant heterogeneity in Myc-induced mammary tumors and we focused on predictions generated from these tumors (4). In order to determine in which tumors E2F regulation may be critical, our previous Myc tumor microarray data were analyzed solely for activator E2F activity. After predicting pathway activation status, the data were clustered to identify phenotypic patterns. The probability of pathway activation is shown in the heat map in Figure 1 and probability values are listed in Supplementary Table 1. In Figure 1 we note that MMTV-Myc–induced tumors clustered into groups based on patterns of E2F activation. For the 2 major histological groups, microacinar and papillary, there are a number of tumor samples that share a high probability of activation of all E2Fs. However, there are also populations within these tumor types that have activation of a single E2F, or dual E2Fs. Interestingly, when examining the EMT / Squamous and the mixed phenotype tumor categories, it seems that E2F2 is the primary E2F that is activated. These results clearly illustrate that there are variable patterns of E2F activation in the subtypes of Myc-induced tumors. We therefore hypothesized that there may be specific roles for the activator E2Fs in the development of the heterogenous Myc-induced tumors.
To directly test whether the E2Fs were involved in Myc-induced tumor formation, mice lacking E2F1 (24), E2F2 (25) or mice heterozygous for E2F3 (26) were interbred with MMTV-Myc transgenic mice (27). Although the E2F3 knockout mice suffer from early lethality, our previous work has illustrated that the E2F3 heterozygous mice have significant defects in mammary gland development (5). Upon generation of 20 female MMTV-Myc mice in each of the various mutant backgrounds and on a wild-type control background, mice were maintained in constant breeding programs and tumor development was monitored. Fertility in the various strains was not notably affected. Striking differences in tumor latency were observed for Myc-induced tumors in the various E2F backgrounds (Fig. 2A). In the absence of E2F1, tumors latency was decreased (P = 0.0042), tumors incidence was complete and tumors grew more rapidly (Fig. 2B) (P = 0.0101). Conversely, in both the E2F2 knockout and E2F3 heterozygous backgrounds, tumor latency was increased significantly (P < 0.0001 for both) and was associated with a significant decrease in tumor incidence with censored mice lacking tumors shown as a data point. In addition, in the E2F3 heterozgous background, the tumors grow slightly more slowly, reaching 2500 mm3 on average 9.5 days more slowly than controls (P = 0.0482).
To begin to elucidate the mechanism by which the E2Fs regulate Myc-induced tumors, we considered the previous findings that E2F1 is specifically involved in mediating Myc-induced apoptosis (17). To determine if this role for E2F1 is part of the mechanism accelerating tumor development, we carried out a TUNEL analysis on tumors from the MMTV-Myc and MMTV-Myc E2F1 knockout mice. As seen in the histological sections, a visible difference was immediately apparent (Fig. 3). When three sections from three tumors for each strain were quantitated, it revealed that there was a greater than a 2-fold reduction in the number of apoptotic cells. Clearly, the reduction in apoptosis in the E2F1 knockout line would contribute to the rate of tumor growth and quite possibly tumor detection times. It is also well established that the activator E2Fs are required to mediate cell-cycle progression for proliferation (17). To measure whether alterations in rate of tumor growth in the E2F mutant background Myc-induced tumors was owing to reduced proliferation, PCNA immunohistochemistry was conducted. However, no difference was noted in PCNA staining in any of the mutant E2F backgrounds. However, this was not a surprising finding given that the tumors were actively growing at the endpoint.
In order to understand the mechanism behind the E2F2 and E2F3 effects, we considered our previous work on mammary gland development in E2F mutant mice (5) and in Myc transgenics (28), and extended it with an additional analysis of pregnancy and lactation in the various E2F mutant backgrounds. Routine staining revealed that during lactation there were significant changes in the E2F1 null and E2F3 heterozygous mammary glands with overgrowth toward the center of the acini. Immunohistochemistry for PCNA confirmed that this phenotype was due in part to aberrant proliferation on the fifth day of lactation in the E2F1 and E2F3 mutants (Supplementary Figure 1). Given previous work showing that Myc induced early lactation owing to aberrant proliferation (28), we then sought to examine the effects of mutant E2F backgrounds on MMTV-Myc transgenics. Interestingly, the loss of E2F1 accentuated the Myc transgene effects, resulting in accelerated precocious lactation compared with a E2F wild-type background (Fig. 4). Unlike the MMTV-Myc transgenics which successfully nursed over 20% of litters, the MMTV-Myc E2F1 null mice were completely unable to nurse their pups (E2F WT litters, n = 48, E2F1 knockout litters, n = 62). The E2F2 and E2F3 mutant backgrounds reduced the early proliferation effect of the Myc transgene, potentially reducing the population of cells susceptible to tumors (Fig. 4). The E2F2 and E2F3 mutant backgrounds had significant effect and allowing the normal lactation response, resulting in pup survival in 50% and 78% of litters respectively (n = 59 and 79). In light of the overgrowth in the E2F1 null and E2F3 heterozygous mice during lactation (Supplemental Figure 1), it was surprising that the MMTV-Myc E2F3 heterozygous mice had such marked phenotypic differences in comparison to the E2F1 mutants. However, these data illustrate that the E2F2 and E2F3 mutants reduced the Myc proliferative effects during pregnancy. Reducing the proliferative capacity may have contributed to the delay in tumor formation by reducing numbers of potential susceptible initiating cells.
We then reasoned that loss of the E2Fs in Myc-initiated tumors may impact the heterogeneity of tumor development. Accordingly, the genetic changes implicit within tumor development were analyzed at the genome wide scale for the MMTV-Myc–induced tumors in a wild type E2F background and in the 3 mutant E2F backgrounds. Eighty samples composed of 20 tumor samples from each strain were arrayed to determine changes in gene expression. After normalization and distance-weighted discrimination (DWD) (29), the data were merged with our previous MMTV-Myc tumor data for a combined data set of over 200 samples. Unsupervised hierarchical clustering for this merged data revealed that the Myc-induced tumors in a wild-type E2F background clustered into 3 distinct groups that we previously characterized as having distinct phenotypic characteristics; EMT, papillary and (illustrated in the dendrogram with red, blue, and green lines). Interestingly, Myc-induced tumors with E2F1 and E2F3 mutant backgrounds followed a similar distribution. However, loss of E2F2 reduced the number of tumors clustered with the EMT classification (Fig. 5A and 5B). Given the previous association we noted with the Ras signature and EMT in Myc-induced tumors, we examined the Ras signature in the various E2F backgrounds and noted a significant decrease in activated Ras probability in the E2F2 knockout Myc-induced tumors relative to the wild-type E2F tumors (P = 0.0056). Importantly, this Ras signature has previously been validated by examining predictions in tumors with and without activating mutations in KRas (4). However, this reduction in Ras pathway activity was not reflected in mutation status in this study. Indeed, when examining the frequency of KRas mutations across the various Myc/E2F strains, no statistically significant difference for the various E2F mutants relative to the control was noted (data not shown). However, based on the predictive results, histology for the various types of tumors was closely examined, and tumors with any EMT patterns were quantitated. The histological patterns in the tumors confirmed that there was a significant reduction in the percentage of Myc-induced tumors containing an epithelial to mesenchymal transition when comparing wild type and E2F2 knockout backgrounds of 36.5% to 16.3% respectively (P = 0.0258) (total tumors, n = 52 for Myc and n = 49 for Myc E2F2 knockout). Together, these data illustrate that loss of E2F2 alters the morphologicial pattern of the resulting tumors. In addition, examining pathways such as Ras allows for quantification of the underlying biology responsible for these morphological changes.
In order to test whether the role of the activator E2Fs in the mouse model system apply to human breast cancer, we downloaded breast cancer microarray data that also had associated relapse-free survival data (GSE7390 and GSE2990). These data sets were merged using Bayesian Factor Regression Modeling (20) and probability of E2F1, E2F2 or E2F3 pathway activation was calculated as described above. The probability of E2F activation is shown in Supplementary Figure 2 and these probability values are listed in Supplementary Table 1. The relapse-free survival times for the top and bottom quartiles for probabilities from each E2F signature were then compared in a Kaplan–Meier plot for the first 6,000 days. This illustrated that although E2F1 and E2F3 had no significant effects, a low probability of E2F2 activation was associated with a significant increase in relapse free survival times (Fig. 6) (p = 0.0007). These predictions were also completed on the individual data sets where the trends remained the same as shown in Supplemental Figure 3. In an attempt to determine whether the E2F2 signature was associated with a specific subtype of human breast cancer, we compared the signature with the intrinsic classifier grouping (30, 31). To do this, we examined whether the E2F2 signature predictions correlated with specific subtypes of breast cancer. We used the list of intrinsic genes (32) to cluster the gene expression data in the combined human data set where we predicted E2F2 status. This revealed that a low probability of E2F2 was predominantly associated with the Luminal A subtype (Supplemental Figure 4). Conversely, a high probability for E2F2 was associated with Luminal B, Basal and ErbB2 tumor types. Together this illustrates that the E2F predictions offer an important refinement to classification of human breast cancer. Considered with the mouse model data, these results clearly indicate the importance of the E2Fs in mediating tumorigenesis.
One of the pressing problems in understanding and treating human cancer is deciphering the heterogeneity of the disease. In the data presented here, we have used genomic signatures to predict how individual pathways may be activated in specific tumor types to dissect this tumor heterogeneity. Indeed, we predicted that the E2Fs would have varied roles in Myc-induced tumors. This prediction was then directly tested genetically. Through interbreeding the E2F mutant mice with the MMTV-Myc transgenics, E2F1 was found to accelerate tumor growth and reduce latency with a reduction in apoptosis. Moreover, E2F2 and E2F3 mutants delayed tumor latency and reduced tumor incidence. Consistent with genomic predictions, ablation of E2F2 also reduced the number of EMT containing tumors. Together these data illustrate how we have used genomic signatures to direct a genetic examination of the E2Fs in Myc-induced tumors. The advantage of combining genomic predictions with signaling pathways is that we can use the predictions to narrow the focus of our genetic tests. Indeed, using this approach we have already identified and initiated further genetic crosses to explore the role of the E2Fs, and other pathways, in additional tumor models. Through studies to combine data sets from a wide spectrum of mouse mammary tumors, we will be able to expand these studies to a wide array of model systems. Importantly, we have also extended the results of these studies of model systems to human breast cancer. Herein we have shown that a low probability of E2F2 activation is associated with longer relapse free survival in human breast cancer, analogous to the delay noted in the mouse model system.
In examining the role of the E2Fs in Myc driven murine mammary tumors, we are struck by a comparison to several recent manuscripts outlining the role of the E2Fs in Myc driven tumors. Specifically, we noted a role for E2Fs in an Eμ-Myc lymphomagenesis model (18) and in a K5-Myc skin epithelial model (19). Briefly, in the Eμ-Myc model, tumors were generated on various E2F mutant backgrounds, illustrating that of the activator E2Fs, only E2F2 altered tumor latency and in opposition to what we observed, loss of E2F2 accelerated tumor formation. In agreement with this, tumors were noted with greater incidence in the K5-Myc model in an E2F2 mutant background. Although both of these models illustrate results that are strikingly different from our results, this is most likely simply an issue of tissue context. An additional study revealed that K5-Myc in an E2F1 null background resulted in tumor acceleration (33), in agreement with our studies. However, the decrease in apoptosis we noted in the E2F1 null Myc tumors are in contrast to this report where they found that loss of E2F1 enhanced apoptosis and caused the tumors to grow more slowly. The variety of observed results is likely an indication of requirement and use of E2Fs in distinct tissues. These results clearly illustrate that defining effects of the E2Fs is dependent upon the context. In the lymphomagenesis model, the relative level of E2F expression was determined across a number of tissues. Interestingly, E2F2 levels were low in the mammary gland relative to the lymphatic system, where a significant role for E2F2 was noted. Despite these lower levels, we noted a significant reduction in tumor incidence with the loss of E2F2. Together these data suggest that the role of the E2Fs will need to be carefully examined in each tissue context.
In examining the effects of the E2Fs on Myc-induced tumors, we noted that E2F1 loss accelerated tumor formation and resulted in tumors in all susceptible mice. Investigation into the effect of E2F1 loss illustrated the aberrant proliferation in the mammary gland, which in turn accelerated the Myc-induced proliferation and early lactation. As a consequence, the E2F1 null MMTV-Myc mice were completely unable to nurture pups and were observed to develop tumors more rapidly. This decreased latency is also partially due to the decreased apoptotic levels in the Myc-induced tumors. Given this lack of apoptosis, the tumors grew more quickly and are also detected more quickly. Together, the accelerated mammary proliferation and decreased apoptosis resulted in a reduction in tumor latency owing to E2F1 loss.
In contrast to the E2F1 results, loss of E2F2 or E2F3 resulted in a significant reduction in tumor incidence. This reduction in incidence suggests that these E2Fs have effects on the population of cells that are susceptible to tumorigenesis. Examining the mammary gland of these mice reveals the Myc transgene effects are reduced and that the mice are lactationally competent. The lactation defect in Myc transgenic mice has previously been seen to be a result of aberrant proliferation to accelerate lobuloalveolar development (28). In this manuscript the authors also hypothesized that altering this effect of Myc on proliferation would impact tumorigenesis. Our results fully support this hypothesis given that the E2F2 and E2F3 mutant Myc mice do not have accelerated lobuloalveolar development or the associated precocious lactation and have a reduction in tumor incidence. We therefore hypothesize that by reducing Myc proliferation, E2F2/3 loss would also reduce potential for transformation. This reduction of Myc effects may have the result of reducing the target population of susceptible cells during the late stages of pregnancy. For E2F3, this ability to lactate and reduction of Myc effects may be due in part to our previously observed delay in involution mediated apoptosis (5). However, given the requirement for MMTV-Myc mice to undergo repeated rounds of pregnancy for tumor formation (34), the E2F2 and E2F3 altered mammary effects and reduced tumor incidence suggest that the tumor initiation events may occur in the late stages of pregnancy when Myc proliferative effects are most profound. In addition, we noted a significant increase in the ability of Myc transgenics to nurse their pups in the E2F2 and E2F3 mutant background. Given that the MMTV promoter directs high levels of expression through lactation, the E2F2 and E2F3 mutant background would then have higher levels of the Myc transgene for an extended period of time. The decrease in tumor incidence in these strains suggests that Myc effects on proliferation during pregnancy are a critical event in tumorigenesis.
In our previous work, we elucidated the heterogeneity within MMTV-Myc–induced tumors in both morphology and in gene expression patterns (4). Upon placing Myc-induced tumors from the various E2F mutant backgrounds on array we used a similar strategy. Interestingly, this revealed that the E2F2 mutant background resulted in fewer tumors clustering into the EMT component. This was also associated with a significant decrease in the probability of an activated Ras signature. Examining histological sections revealed that these computational predictions were reflected in the observed morphology with a significant reduction in EMT with loss of E2F2. Interestingly, this phenotype was only observed with E2F2, no EMT differences were noted in the E2F1 or E2F3 mutant background. The specificity for E2F2 in EMT loss mirrors the signature predictions where only E2F2 had an elevated probability in EMT samples. Together, these computational predictions coupled with observations of the tumors suggest that E2F2 is involved in the transition from an epithelial to mesenchymal phenotype.
Finally, these data also illustrate the utility of employing genomic signatures in a discovery based approach to determining which cell signaling pathways may be essential in tumorigenesis. The identification of unique roles for the activator E2F transcription factors through gene expression data that were then validated genetically provides an important glimpse into how heterogenous tumor biology may be deciphered. In the future, we plan to extend this work by predicting and genetically testing roles for the E2Fs, and other key pathways, in a variety of mouse models. It will also be critical to extend this work to human breast cancer. Our studies have illustrated that the activator E2Fs are critical in mediating tumor progression and are involved in a number of aspects of tumor development in both the MMTV-Myc mouse model system and human breast cancer.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Received June 30, 2010.
- Revision received December 7, 2010.
- Accepted January 3, 2011.
- ©2011 American Association for Cancer Research.