CDβGeo transplants are considered premalignant because they form hyperplastic outgrowths, some of which progress to invasive tumours. Transient TGFβ-treatment of CDβGeo cells in vitro promotes EMT that is sustained after withdrawal (designated pTD cells) and transforms these mammary epithelial cells such that they become mesencymal-like and highly tumorigenic in vivo. The pTD cells, and the tumours that develop from them, are de-differentiated, having lost markers that define both luminal epithelial and myoepithelial cells. Interestingly, there is no comprehensive acquisition of stem cell markers, but rather decreased expression of several key stem cell markers including CD44, CD49f, CD29 and Sox9, with no change in the expression of Nanog or Pou5f1 (Oct4). This is consistent with Nguyen et al., who demonstrate that induction of TGFβ only accelerates tumorigenesis, and that radiation-induced notch signalling is required for expansion of mammary stem cells
. Although EMT has been reported to increase the population of cells with stem-like characteristics
[20, 25, 43], TGFβ-induced persistent EMT in the CDβGeo cells was not accompanied by increases in the stem cell pool. Although CDβGeo cells clearly have mammary progenitors
 the mammosphere forming capacity and transplant capability is similar to primary mouse mammary epithelial cells
. Similar to other reports
[45, 46], these cells do show enrichment of the stem cell pool during TGFβ-treatment in vitro (Additional file
4: Figure S3), but enrichment is transient, and the equilibrium in cell populations is restored upon subsequent passages and may not be essential for tumours.
The cancer stem cell theory proposes that only a small subset of cells, the tumour initiating cells, can seed a new tumour or a metastasis
. Therefore, there is great interest in identifying cancer stem cells in order to identify pathways and targets to reduce the metastatic potential of cancer. However, the defining line between EMT, mesenchymal cells, cancer stem cells and bulk tumour cells is indistinct
[48, 49] with substantial overlap among makers of EMT and profiles to define stem cells
[29, 30]. These signatures also align with human claudin-low and metaplastic breast cancers
, although clearly these are not the only tumours with metastatic potential. The CDβGeo model identifies changes in ECM, MMPs, and transcription factors such as Snai1, Snai2, and Zeb2 as indicative of EMT. Because our model represents EMT without changes in the stem cell population, it suggests that ITGA6, DUSP6, Sox9, and KLF4 are valid markers for stem cells as suggested by Gupta et al.
. Because pTD cells demonstrate persistent EMT without increases in the stem cell pool, this model can be used to separate markers for EMT and consequently refine signatures that define tumour initiating cells.
Previous work has demonstrated that transdifferentiation of mammary epithelium in response to TGFβ-treatment is transient
[19, 24] and that sustained transdifferentiation and tumorigenesis in vivo only occurs with sustained TGFβ exposure or transformation with v-Ha-Ras oncogene
. Deletion of p53 also promotes EMT by releasing the repression of Zeb1, Zeb2 and BMI1
. However, our experiments with TM40A cells show that blocking p53 is not sufficient for TGFβ-mediated EMT. Additionally, even though the CDβGeo cells are p53-deficient, cell growth was repressed by TGFβ. This agrees with other reports that TGFβ-mediated cell cycle arrest is p53-independent
 and that p63/p73 may compensate in TGFβ-mediated pathways
, including possibly those that promote EMT.
Persistent EMT has also been shown to be dependent on sustained TGFβ exposure through an autocrine positive loop
[21, 32, 40, 53]. The pTD cells have elevated TGFβ2 and there is partial rescue, with decreased expression of Snail and increased expression of Sfrp1, when the pTD cells are treated with the TGFβRI inhibitor LY364947. While higher doses of the TGFβRI inhibitor or a longer course of treatment may achieve a more robust rescue, the transcriptional profiles suggest that the transformed pTD cells have undergone epigenetic modifications, affecting multiple pathways, such that targeting TGFβ-pathways alone will not be effective. With extended expansion in culture (>5-10 additional passages), the pTD cells gradually regain a cobblestone epithelial morphology in vitro. This partial MET in vitro may be due to the dilution, during sequential passaging, of TGFβ2 and other factors that support the mesenchymal phenotype. EMT and acquisition of mesenchymal properties are necessary for some metastatic processes including intravasation, transport in circulation and extravasation. Dilution of mesenchymal supporting factors during dissemination may explain the paradox of why secondary tumours often exhibit an epithelial phenotype rather than a mesenchymal phenotype