Introduction
Classic epidemiological studies on the increase in cancer incidence with age predicted that from three to six independent events would be required to convert a normal cell into a tumour cell [
1]. Experimental studies have now proven that it is indeed possible to transform a cell in culture by modifying the activity of only a few critical genes. Cell lines quantitatively transformed by expressing oncogenes or inactivating tumour suppressor genes have been produced from normal fibroblasts, embryonic kidney cells and human mammary epithelial cells (HMECs) [
2,
3]. The genes initially chosen for these studies were those encoding simian virus 40 large T and small t antigen, activated Ras and telomerase. Subsequently it was shown that viral oncogenes can be replaced by activated
MYC and genes targeting the retinoblastoma pathway [
4]. This combination will transform HMECs, but the resulting tumours do not express oestrogen receptor α (ERα); this is an important weakness of current models, because about 70% of human breast tumours are ERα-positive.
ERα behaves quite differently in ERα-positive cell lines derived from human breast cancer and in normal human mammary epithelium
in vivo. Oestradiol is a direct mitogen for ERα-positive cancer cell lines, but in normal human breast tissue the ERα-positive cells do not themselves divide in response to oestrogen [
5]. Instead, they relay a proliferative signal to neighbouring ERα-negative cells. The barrier to proliferation of ERα-positive normal cells may explain why HMECs rapidly lose ERα expression in culture and why the transformation studies performed so far have produced ERα-negative tumour cell lines.
The target cell of oncogenic mutations in the breast is probably a stem cell or bipotent progenitor cell [
6,
7]. It is possible to enrich for these cells by growing primary HMECs in the non-adherent conditions previously developed for culture of neural stem cells, leading to the formation of so-called floating mammospheres [
8]. The Polycomb-group gene
BMI1 can suppress activation of the p53 and Rb pathways by silencing the expression of p14
ARF and p16
CDKN2A [
9] and it has been shown to increase the rate of self-renewal of mammospheres in response to Wnt, Hedgehog and Notch signals [
10]. Since Polycomb-group genes are overexpressed in breast cancer [
11,
12],
BMI1 is a relevant candidate to test in HMEC transformation assays.
BMI1 was originally identified as an oncogene that cooperates with
MYC to induce lymphomas in mice [
13], and
MYC is commonly amplified in breast cancer, so it is reasonable to use
MYC in a transformation protocol that includes
BMI1. We show here that lentiviral transduction of HMECs with
ERα,
BMI1,
MYC and
TERT leads to the formation of ERα-positive tumours whose growth is dependent on oestrogen.
Discussion
We have developed a model for ERα-positive breast cancer by transformation of normal HMECs with
ERα,
BMI1,
MYC and
TERT. Metastasis occurred in 38% of the mice after 90 days. Previous attempts to make an ERα-positive model probably failed because ERα induces growth arrest and differentiation. We have not addressed the mechanism in this study, but transforming growth factor-β is known to restrain the proliferation of ERα-positive murine mammary epithelial cells [
34]. Expression of BMI1 prevents differentiation and relieves the growth arrest, allowing the expansion of oestrogen-dependent HMECs in culture.
There are several differences between our culture system and those used previously. We grew the cells in floating mammosphere conditions before the first passage for several reasons. At a practical level, the final step after tissue digestion is a single-cell straining step; this facilitates efficient infection of the cells with lentiviral vectors less than 24 hours after the cells are removed from the patient. A second practical advantage is that fibroblasts do not survive in suspension, so floating mammosphere culture is an efficient way to eliminate fibroblasts. More importantly, it is based on techniques developed initially for propagation of neural stem cells [
35] and later adapted for culture of HMECs [
8]. The mammosphere approach enriches for bipotent progenitor cells that are capable of differentiating to myoepithelial and luminal cells, with production of milk proteins by the latter after treatment with prolactin in three-dimensional Matrigel culture [
8]. The main difference between our approach and that of Dontu and colleagues [
8] is that we used the same medium for suspension and adherent cell culture, and we omitted basic fibroblast growth factor. The medium is based on B27 [
15], a serum-free medium supplement that is known to preserve the phenotype of human tumour cells in culture better than serum-containing media [
36]. In our study, the relative importance of the medium versus the suspension culture is unclear but we have preliminary evidence that suspension culture may not be strictly necessary.
In the absence of a stem cell assay for HMECs it is not possible to state definitively whether mammospheres contain true human mammary epithelial stem cells (MaSCs). It is possible that the mammosphere approach enriches for mammary colony-forming cells (Ma-CFCs) rather than mammary repopulating units (MRUs) [
37,
38]. The nature of the cell initially infected with lentiviruses in our protocol is unknown because the infections were performed on the mixed population of cells present in reduction mammoplasty tissue. An intriguing question is whether cells expressing the recently identified murine MaSC markers would be more sensitive to transformation. Given the uncertainty surrounding the identity of human MaSCs, our main aim was to reduce the duration of growth
in vitro to limit the potential for selection of adaptations to culture
in vitro. The present study used 10
6 cells per fat pad injection, for which we needed to expand the cultures
in vitro for a total of 28 days. We have preliminary results indicating that 5,000 cells are sufficient to form tumours, so it should be possible to greatly reduce the duration of culture
in vitro.
Polycomb-group genes such as
BMI1,
EZH2 and
SUZ12 have repeatedly been identified as adverse prognostic factors in breast cancer [
11,
12]. BMI1 is required for proliferation and renewal of stem cells in the brain and hematopoietic system. In mammospheres,
BMI1 is thought to act as a point of convergence of the Wnt, Notch and Hedgehog signals that promote stem cell renewal [
10]. BMI1 probably has at least a dual role, allowing cell proliferation by suppressing p14
ARF and p16
CDKN2A expression, and preventing differentiation through a more complex mechanism. Both processes are clearly visible in the microarray data reported here (Figure
2). ERα-positive tumours typically contain wild-type p53 and have fewer genomic changes than ERα-negative tumours [
39]. The ERα-positive tumour model we have produced matches the human disease in this respect. The most likely explanation for the tumours to have retained wild-type p53 is that BMI1 suppresses p14
ARF expression [
9]. Previous quantitative transformation models included genes such as those encoding simian virus 40 T antigen and p53
DD to inactivate p53 [
3,
4]. The MCF10A and MCF15 HMEC-derived cell lines show large differences in their DNA damage response despite both retaining wild-type p53 [
40], and ERα-positive human breast tumours respond poorly to chemotherapy despite having wild-type p53. It is therefore important to note that although we have shown that the p53 cDNA is wild type, we have not shown that the p53 pathway is functional in our cells. The genes suppressed by BMI1 in ERα-expressing cells include many associated with neural and squamous differentiation. Suppression of these genes presumably favours proliferation by avoiding entry into a terminal differentiation program. We found that
BMI1 itself was one of the genes suppressed by exogenous BMI1 expression. Bracken and colleagues showed by chromatin immunoprecipitation (ChIP) that the Polycomb-repressive complex 1 (PRC1) component CDX8 and the PRC2 component SUZ12 were present at the
BMI1 promoter [
27]. Suppression of PRC function by RNA-mediated interference (RNAi) led to derepression of genes with PRC proteins at the promoter [
27]. On the basis of the ChIP data and the transcriptional response to RNAi against
BMI1,
EED,
SUZ12 and
EZH2, Bracken and colleagues suggested that PcG proteins autoregulate their own synthesis [
27]. Autoregulation of BMI1 itself would by definition not have been detectable in their RNAi experiment, but when taken together with our results it is plausible that BMI1 suppresses its own expression through binding to its own promoter.
Wild-type
ERα is not normally considered to be an oncogene, but behaves like one in our protocol. It is well known that ERα expression is rapidly lost from HMECs in culture. This is not solely a consequence of growth inhibition by ERα, because expression is still lost when cells are forced to express exogenous BMI1. In comparison with previous studies, the combination of genes we used to transform the cells seems rather gentle. In particular, we see no need to activate Ras signalling [
2‐
4,
41].
TERT was essential for successful transformation of HMECs in previous studies [
3], but
BMI1 and
MYC can both activate TERT expression [
42,
43], so it is possible that
TERT may not be required in our protocol. We included
MYC in the protocol because
BMI1 was originally identified as an oncogene that cooperates with
MYC in lymphoma production in mice [
13],
MYC is commonly amplified in human breast cancer, and several groups have reported that
MYC is required for HMEC transformation. Indeed, when Elenbaas and colleagues [
3] used an HMEC transformation protocol lacking
MYC, the cells spontaneously amplified
MYC during culture
in vitro. We used a wild-type
MYC (c-myc) clone for our studies, rather than the activated form of
MYC (T58A) used by Kendall and colleagues [
4]. Despite the strong evidence that
MYC is important, the requirements may differ when the selection conditions are changed, and we have preliminary evidence that
MYC may not be required in
ERα/BMI/TERT-transduced HMECs, at least for the initial stages of tumour formation.
It is intriguing that the transformed HMECs in our model can form polarised epithelial structures
in vivo that express the correct luminal and basal keratins (Figure
6). This indicates that the double-positive keratin staining pattern
in vitro is more likely to reflect a specific progenitor state than a loss of control of lineage-specific gene expression. Unlike the primary tumours in the mammary gland, the metastases were K18-negative. This suggests that the cells differentiate in response to signals from their local environment and that the mammary fat pad supplies specific signals that promote luminal keratin expression. Although the tumours contained regions of invasive adenocarcinoma, the predominant pathology was squamous carcinoma. Squamous differentiation is uncommon in human breast tumours. The squamous differentiation we observed in the NOD/SCID mice may reflect a general property of the mouse mammary fat pad model, a specific property of the target cell of the
in vitro transformation protocol, or a defect in the transactivation of critical ERα target genes in the transformed cells. In comparison with the human breast, the mouse mammary gland contains much less fibrous connective tissue [
44]. To promote engraftment of HMECs in the mouse mammary gland, human fibroblasts are commonly injected either at the same time as the HMECs or a few days earlier, to 'humanise' the stroma [
32]. Human cancer-associated fibroblasts (CAFs) are similarly used to promote engraftment of human tumour cells in mice [
45]. It is possible that the HMFs we injected together with the HMECs may have contributed to the squamous phenotype, but injection of our HMECs without HMFs led to the formation of tumours with similar kinetics and histology (data not shown). As mentioned above, we do not know the identity of the target cell of the transformation protocol. The mammosphere protocol enriches for mammary epithelial progenitor cells, but it is possible that expression of BMI1 promotes the survival of more differentiated cells or, conversely, forces progenitors to adopt a more stem-cell-like phenotype. HMEC protocols commonly give rise to squamous tumours in mice [
3,
46,
47], so we consider it unlikely that the mammosphere protocol has led to the expansion of cells that are unrelated to mammary epithelium. Another possibility is that, despite expressing ERα, the cells are unable to respond appropriately to it. The gene expression profile of breast tumours is dominated by genes that are tightly associated with ERα status [
48]. Many of these genes are direct targets of ERα but others are thought to represent markers of cell type. Some classic ERα target genes, such as the progesterone and prolactin receptor genes, were induced by oestradiol in the
ERα/BMI1 HMECs, but others, such as
TFF1 and
XBP1, were not (Figure
2; the full data set is available in the GEO database under accession number GSE6548). The uninduced group includes many of the genes shown by RNA interference to require co-activation by FOXA1 [
49]. Because FOXA1 is expressed only weakly in the transformed cells, transduction with a
FOXA1 vector might lead to the activation of a broader range of ERα target genes and suppression of the squamous phenotype. More generally, there may be regulators of the differentiation programme of mammary epithelial cells, such as
GATA3 or
TP73L, that are not correctly expressed in the xenografts. We are currently testing these models to develop a transformation protocol that more faithfully reproduces the histology of human breast tumours.
It has long been known that ERα-positive breast tumours metastasise early, leading to distant relapse many years after excision of the primary tumour. In comparison with ERα-negative tumours, they have a better initial prognosis but this is followed by a relentless increase in breast cancer-specific mortality that continues even 15 years after treatment of the primary tumour [
50]. It is tempting to speculate that the correct paradigm for these tumours is a low-grade lymphoma: a systemic disease characterised by few genetic changes and poor response to therapy. Intriguingly, there is a class of ERα-positive breast tumours that have only a single change on genomic profiling: an unbalanced translocation leading to gain of chromosome 1q and loss of chromosome 16q [
51]. For this model to be correct, metastasis would have to occur early. Metastasis has not previously been reported in studies with HMECs transformed with defined oncogenes [
3]. At early time points the transformed HMECs reported here invaded the fat pad rapidly, forming a prominent duct-like structure that arose either from cells deposited in the needle track or by the migration of cells out from the main mass (Figure
4g–i). We have not tested the invasive properties of the cells
in vitro, but the specific combination of genes used to transform the cells certainly triggers an invasive and metastatic program
in vivo. Because the key difference between this model and previous HMEC models is ERα expression, it is tempting to speculate that ERα itself has a critical role in the early metastasis of ERα-positive human breast tumours.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
SD carried out the lentiviral vector construction, tissue culture and mouse studies. ALN and SA carried out the microarray studies. MF participated in the pathological studies. SD, HB, CB and RI participated in the design and coordination of the study. SD and RI drafted the manuscript. All authors read and approved the final manuscript.