As discussed, the most commonly used representative of normal luminal breast epithelium is MCF10A cells. Culture of these cells in Matrigel^TM has yielded some impressive acini-like structures which are polarised, contain hollow lumens and produce basement membrane proteins [40, 59]. This has also been recapitulated in collagen I matrix [60]. Branching 3D duct-like structures have also been achieved through culture of HB2 cells in collagen I [51]. However, while culture of these cells in 3D matrix has resulted in structures akin to in vivo breast acini and ducts which cannot be achieved through culture in 2D, in vivo breast architecture is much more complex including a variety of other cell types that influence epithelial architecture. This has led to a rise in 3D co-culture with stromal cells with co-culture with fibroblasts most commonly reported [38, 47, 60] and examples of combinations of fibroblasts and adipocytes also demonstrated [61, 62]. It has emerged that fibroblasts regulate epithelial cell invasion [63] secreting various growth factors such as MMPs [64], which have been proven through use of 3D co-cultures. Markedly, some of these effects are only evident upon culture in collagen I [65]. Myoepithelial cells also play a key role in normal in vivo breast. Given their tumour suppressive nature [66, 67] and their capacity to maintain luminal cell polarity [68], it seems apt to include these cells in normal in vitro models of breast; however, to date studies incorporating these cells are lacking.

It stands to reason that luminal epithelial organisation is maintained by a delicate balance between the opposing functions of fibroblasts and myoepithelial cells. It seems apt that both these cell types are included in a 3D in vitro model of normal breast with luminal epithelial cells to accurately reflect the in vivo environment. One such tri-culture model has been achieved previously by Holliday et al. [38] but this used aberrant luminal epithelial cells thus representing DCIS. In our laboratory, we have developed a tri-culture model of normal breast that incorporates HB2 cells, myoepithelial cells and fibroblasts isolated from breast reduction mammoplasty samples in the physiologically relevant matrix collagen I. We have demonstrated that each individual cell type retained an in vivo phenotype and that combined shared morphology and phenotype with normal breast reduction mammoplasty acini. In addition, we have proved through overexpression of HER2 that formation of structures that accurately reflect DCIS in vivo can be achieved. What is more, we have proved that these structures are quantifiable by standard inexpensive laboratory techniques [unpublished].

However, despite better recapitulating the in vivo breast environment, models such as these are still reliant on cell line culture which may already represent aberrant breast epithelium. As discussed previously, the use of new primary cell cultures that represent the morphology, immunoprofile and genetic profile of normal and earlier benign breast lesions would be necessary to accurately model breast cancer initiation in the laboratory. The culture of primary breast epithelial cells in 3D Matrigel^TM has been achieved previously [69] but has been limited to a single cell population. In order to accurately recapitulate the in vivo breast environment, multiple primary cell types would need to be isolated, characterised to ensure they accurately represented their in vivo counterparts, labelled to enable tracking and then cultured together in 3D ECM. The finite life span of primary cell cultures in vitro would make this challenging for researchers with further genetic and proteomic manipulation even more difficult.