Introduction

Human breast cancer cells are generally believed to originate from the luminal epithelial lineage of terminal duct lobular units [1]. Accordingly, the phenotype of most breast cancer cell lines is luminal, and they express sialomucin and keratin K19 [2,3] as well as other epithelial markers. In recent years, however, there has been an increasing appreciation of human breast carcinoma cells being more flexible in their differentiation program. Intra-tumor and inter-tumor heterogeneity of human breast cancer should no longer be viewed as a consequence of phenotypic drifting due to genetic instability, but also from the point of view of different differentiation repertoires available to the neoplastic cells in response to the tumor microenvironment, including reversion to a 'normal' phenotype (for an overview, see [4]). This is important because it allows for possible strategies to influence the breast cancer cells towards a more differentiated state (for a review, see [5]). We recently showed that the myoepithelial lineage is derived directly from the normal luminal epithelial lineage [6]. It has been observed previously that the luminal epithelial lineage of breast cancer was defective in its ability to differentiate along the myoepithelial differentiation program [7]. The growing number of myoepithelial markers is similarly providing evidence that neoplastic breast cancer cells frequently exhibit at least a partial myoepithelial differentiation program [8,9,10]. If the normal myoepithelial differentiation program is indeed tumor suppressive [5,11,12], then defining these traits becomes rather important for diagnosis and therapy. Furthermore, recent data based on loss of heterozygosity suggest that the neoplastic and mesenchymal compartments, of which the latter contain the myofibroblasts, are directly evolutionarily connected [13,14,15]. It has also been shown that loss of differentiation in breast cancer, invasion and metastasis both in vivo and in culture concur with epithelial to mesenchymal transition (EMT) of the tumor cells [16,17,18,19,20,21]. The fact that EMT-derived tumor cells and the myoepithelial lineage are both defined by expression of the same mesenchymal marker (i.e. vimentin), and furthermore that the myoepithelial lineage and peritumoral myofibroblasts share almost the same myodifferentiation program, warrants a more ambitious set of markers to define these cells and reveal the true implications of their presence in vivo.

Myoepithelial differentiation in breast neoplasia

Myoepithelial cells are ectodermally derived, as are luminal epithelial cells, and they rest between the basal side of the luminal cells and the basement membrane that they manufacture themselves [22]. They nevertheless express mesenchymal vimentin and Thy-1 along with certain 'basal' keratins including K5, K14 and K17 [1].

Over the years, the literature has accumulated the fact that the myoepithelial differentiation pathway is essentially absent in breast carcinomas. If the criteria used for identification of myoepithelial cells in breast cancer are the expression of a complete differentiation program, then it is correct that myoepithelial differentiation in breast cancer is an exception. Many studies have, however, focused exclusively on the expression of the basement membrane molecule laminin, which is usually absent in high-grade neoplastic breast cancer cells [23,24]. But if we look at some of the other myoepithelial markers, the picture changes radically. Both keratins K14 and K17 as well as vimentin have been reported present in 20-33% of invasive breast carcinomas and often at the epithelial-stromal junction, which is also the location of myoepithelial cells in the normal breast [25,26]. Remarkably, similar numbers have been reported for the frequency of ultrastructurally identified myoepithelial cells in breast cancer and for the expression of other markers such as oxytocin receptors, metallothionine, and connexin 43 [8,9,10,27]. This suggests that, even though breast cancer cells first and foremost express luminal epithelial keratins and sialomucin, some of the cells still retain their intrinsic ability to switch to a myoepithelial differentiation program [6], albeit expressing only partial changes towards a myoepithelial phenotype.

What is the potential consequence of myoepithelial differentiation in breast cancer? Basically, all myoepithelial-specific proteins tested in experimental tumor assays point in the same direction; namely, that they are tumor suppressive. The genes qualify as type II tumor suppressor genes [28], and include maspin [12], α6-integrin [29], cyto-keratin 5 [30], connexin 43 [31], caveolin-1 [32], α-smooth muscle actin [33], and myoepithelium-derived serine proteinase inhibitor [34]. The soluble factors from normal myoepithelial cells such as activin [35] and relaxin [36] also inhibit growth of breast cancer cells and facilitate their differentiation. Furthermore, some factors (e.g. myoepithelial-specific CD44) were shown to be anti-angiogenic [37]. It has been experimentally difficult, however, to extrapolate from expression analyses in culture to the function of cancer-derived myoepithelial cells in vivo. To this end, precise assays for the functional phenotype of normal and abnormal breast myoepithelial cells are highly warranted.

Myoepithelial cells and myofibroblasts have many traits in common including vimentin and α-smooth muscle actin, and occasionally smooth muscle heavy-chain myosin [1]. It could thus be asked whether the plasticity of breast carcinoma could ever go further to result in the formation of true mesenchymal cells in an epithelial→myoepithelial→mesenchymal transition process. If this were the case, stromal myofibroblasts should occasionally express remnants of the myoepithelial phenotype, such as keratins. We have not so far been able to document such a differentiation program [38]. Others have, however, reported the occurrence of myofibroblasts with remnants of K14 expression (described as 'converted myoepithelial cells') in the central acellular zone of desmoplastic reactions [39]. Also, in a rat tumor progression series and in canine mixed mammary tumors, one of the steps in the evolution of mesenchymal cells involves the expression of typical myoepithelial traits [40]. Further studies are needed to elucidate how widespread this phenomenon is and to determine the possible consequences of such a conversion. It is clear that defining all these lineages based exclusively on mesenchymal markers such as vimentin may no longer be sufficient (see later).

Mesenchymal differentiation

Although vimentin is distinctly a part of the myoepithelial phenotype as already described, its presence in human carcinoma cells has mainly been attributed to a direct EMT without myoepithelial intermediates (for example, see [19]). Originally, EMT was described in morphogenic remodelings during embryonic development (for reviews, see [41,42]). Later, it was shown to occur in cultured mammary epithelial cells [43] and in a bladder carcinoma cell line, NBT-II cells [44]. In culture, the definition is limited to the escape of single cells from epithelial sheets, increased motility, and a modification of the differentiation program so that the migrating cells no longer express epithelial characteristics, but acquire a mesenchymal phenotype [44]. More specifically, in NBT-II cells, the mesenchymal phenotype is — apart from the morphological change to spindle cells — defined by the acquisition of vimentin, while the loss of epithelial phenotype is defined by a decline in keratin and desmoglein expression; there is no change in the expression of E-cadherin or catenins. The epithelial phenotype is thus not completely lost, and a single marker defines the mesenchymal phenotype. This definition of EMT has been adopted in the field of breast cancer research.

Vimentin expression was originally found in hormone-independent cell lines [45]. This was later expanded to cell lines that were drug resistant [46]. This transition was monitored by a decline in keratin K19 and loss of E-cadherin, and a reduced expression of desmoplakins and occludins [46]. Similar criteria have been used in other laboratories as well as in one of ours, also with vimentin as the only molecular marker of a mesenchymal phenotype [16,47]. This definition of the mesenchymal phenotype is, however, similar to the myoepithelial phenotype. The definition of EMT in breast cancer, therefore, is much more demanding to address than in, for example, bladder cancer or colon cancer, where the myoepithelial program is not an option. Another possible definition of EMT in breast epithelial cells is the spread of cells in collagen gels [48]. We have previously shown that a subset of cells within the luminal epithelial lineage may convert to myoepithelial cells, and these have been shown to be very active in cell spread in collagen gels [6,49]. This definition is therefore also not exclusive. These observations underscore the general importance of further lineage characterization of spindle cells derived from cuboidal cells in the normal breast and in breast cancer. A simple keratin profiling of the mesenchymal-like cells should reveal whether they are the product of a direct EMT from the luminal lineage or whether they represent cells converted via the myoepithelial lineage. In the first case, remnants of the luminal phenotype should be present in the form of keratins K7, K8, K18 and possibly K19. In the latter case, basal keratins K5, K14 and K17 would be expected to dominate the profile. Such distinction is important because, as already stated, the biological and clinical implications of direct luminal epithelial EMT is very different from that deduced from the tumor suppressive myoepithelial phenotype.

Clinically, the concept of direct EMT is used to explain the phenotype of very aggressive metastatic cells to the bone marrow [50]. These cells express vimentin in conjunction with luminal epithelial markers [50]. Their aggressive phenotype concurs well with observations that vimentin-positive luminal breast carcinoma cells are invasive in cell culture assays [19]. For comparison, the vimentin-positive spindle cells of pure myoepitheliomas are considered essentially non-metastatic [51]. The complications in deciphering the mesenchymal phenotype become even more compelling when focus is turned from metastatic cells in the bone marrow to heterogeneous primary tumors that consist of many cell types. In primary tumors, there is no unequivocal correlation between gain of vimentin or loss of E-cadherin expression and poor prognosis [16,52]. One explanation for this could be that some of the cases with vimentin-positive cells and E-cadherin-negative cells reflect myoepithelial differentiation rather than direct EMT from luminal epithelial cells.

A stromal condition that has been overwhelmingly associated with poor prognosis in breast cancer is the desmoplastic reaction [53]. Desmoplasia is believed to be the result of excessive extracellular matrix formation as manufactured by peritumoral myofibroblasts [54]. As seen from the aforementioned considerations, however, the current definition of EMT does not incorporate the complete loss of keratins or gain of other mesenchymal markers other than vimentin (such as, for example, α-smooth muscle actin). Indeed, we have some evidence (Petersen et al., submitted for publication) that breast cancer cells, given the right conditions, may convert all the way to myofibroblasts. Also, it has been shown that typical and presumably myofibroblastic stroma of breast carcinomas frequently share the same loss of heterozygosity, with the epithelial neoplasia indicating a common cellular origin [13,14,55]. Stromal cells have also been shown to share p53 mutations with the carcinoma cells [14]. Moreover, a number of phenotypic traits such as IGFII, several transcription factors and protein kinases in myofibroblasts appear to be permanently overexpressed as if they were the result of typical genomic amplifications [56,57]. A consequence of a possible evolutionary relationship between the stroma and the neoplastic lesion remains to be seen.

Concluding remarks

Evidence for phenotypic plasticity of luminal epithelial cells to the other two lineages present in the normal breast is now compelling. Understanding breast carcinoma plasticity may be instrumental in attempts to take advantage of the tumor suppressiveness of myoepithelial differentiation or to interrupt EMT-dependent or desmoplasia-mediated progression. Thorough characterization of the epithelial and stromal compartments of normal breast and breast cancer need to be continued along with further emphasis on development of culture model systems for epithelial-stromal interaction and conversions.