Adenomyoepithelial tumours are a widely accepted, yet poorly understood diagnostic entity of breast neoplasms, the definition of which is based on cytological and architectural features. According to the most recent WHO classification, the spectrum of tumours included under this heading ranges from organoid, bi-phasic tumors with a combination of easily recognisable tubular formations and multi-layered spindle-celled myoepithelial elements to monophasic lesions with "mesenchymal" appearance [1–3].

In tumours of monophasic "mesenchymal" appearance, a potentially glandular differentiation can sometimes only be identified by staining for low-molecular-weight cytokeratins (Ck) Ck8/18. Malignant tumours without obvious morphological and immunophenotypic glandular features are currently classified as malignant myoepithelioma or myoepithelial carcinoma. These malignancies are characterized by infiltrating proliferations of atypical spindle cells with high mitotic activity, often displaying a dot-like cytoplasmic expression of smooth muscle α-actin (SMA), vimentin and Ck14. Because of this phenotypic heterogeneity, a specific position within the spectrum of proliferative breast lesions has not been assigned to adenomyoepithelial neoplasms, and unequivocal criteria for prognostic assessment have not yet been established [4, 5].

We studied a series of 27 adenomyoepithelial tumours by comparative genomic hybridisation (CGH), double immunofluorescence and immunostaining for Ck5, Ck14, Ck8/18 and Ck19, as well as for vimentin, p63, SMA and the proliferation antigen detected by the monoclonal antibody MIB-1. Our data indicates that adenomyoepithelial tumours represent a spectrum of lesions that constantly express Ck5 and alternately show glandular and myoepithelial elements.

Tumours with an adenomyoepithelial differentiation span a broad and morphologically diverse spectrum of neoplasms ranging from patterns with biphasic glandular/myoepithelial architecture to monophasic pure spindle-cell variants. The biology of these rare tumours is poorly understood and their complex nature often gives rise to diagnostic uncertainty, especially when it comes to accurately assessing the prognosis of a given case. The hallmark of these tumours are SMA-positive neoplastic cells co-expressing cytokeratins in variable amounts. Furthermore, all of these tumours share a high proportion of cells that are reactive for basal-type Ck5/6 and Ck14 and p63 and most, even the monophasic ones, also contain tumour cells with expression of glandular type Ck8/18. This is in glaring contrast to myoepithelial carcinomas which are defined by a complete absence of glandular features. Nevertheless only one (case 16) of the spindle-cell tumours included in this series met the criteria of this category, all other tumours showed at least a small positive reaction with glandular-type cytokeratins. This would indicate that pure myoepithelial carcinomas are extremely rare. This admixture of glandular and myoepithelial components was also described in metaplastic spindle cell carcinoma, "fibromatosis-like" which has to be taken into account in differential diagnosis [9]. In view of our present morphological and immunohistochemical data, some of these tumours might be classified as adenomyoepithelial tumours.

Not unexpectedly, biphasic adenomyoepithelial lesions displayed a Ck8/18 expression that was confined to glandular structures and therefore similar to patterns present in normal breast tissue. Lesions with a monophasic "mesenchymal" appearance, however, displayed a Ck8/18 and Ck19-positive immunophenotype in spindle-cells often arranged in a reticular pattern (Figs 1 and 2. B, D, F). In addition, double immunofluorescence experiments showed co-expression of basal type keratins either in combination with glandular type keratins or SMA as a marker of the myoepithelial lineage. Part of the tumour cells solely expressed Ck5/6. Within the context of a recently published model of differentiation [6, 10, 11] in normal breast tissue this data would further point towards the existence of Ck5-positive progenitor cells with multi-lineage differentiation potential [12–16]. The potential importance of these presumed progenitor cells has already been shown in mouse tumours with both myoepithelial and glandular differentiation [17].

The assumption that adenomyoepithelial neoplasms are progenitor cell-derived lesions would help to explain the mixed glandular and myoepithelial architecture in the biphasic tumours as well as the bilinear immunophenotype in monophasic lesions. It seems that the tumour cells retain at least some functional remnants of their physiological counterparts. These include basement membrane formation of hemidesmosomes and laminin-binding sites and a strong polarity [18–21]. In the context of this hypothesis, the typical architectural element of adenomyoepithelial tumours, namely the predominant multilayered myoepithelial component with its swarming aspect and enclosed tubules or purely reticular pattern, can be attributed to an immature myoepithelial phenotype and function. Especially the latter may trigger a loss of polarity which leads to an inappropriate sorting of the myoepithelial cells.

From a clinical point of view, no clear criteria for the assessment of patient prognosis have been defined yet. Because these tumours are so rare, available data concerning the prognosis of these tumours is scarce. In accordance with reports in the literature [4, 23], we applied criteria commonly used for assessing malignancy of these tumours which unfortunately were shown to be lacking in precision. Mitotic rates of up to 16 mitoses per 10 HPF were reported in lesions categorised as benign, whereas in contrast tumours with less than 3 mitoses per 10 HPF were diagnosed as malignant. Additionally, ill-defined margins and infiltrative growth patterns are observed in tumours of this entity, although it was never conclusively decided whether such features are indicative of a malignant outcome or a remnant of normal behaviour of myoepithelial cells as already documented in vitro [21].

Given this information, we propose a third category of tumours of uncertain prognosis in addition to a benign and a malignant category. The CGH analyses of our series of adenomyoepithelial neoplasms support this idea, because a stepwise increase in the average number of genetic alterations was observed. Growing numbers of genetic alterations are also associated with an increasing growth fraction (>10 %) However, this suggests a continuum in the progression of these tumours and is hardly helpful in the assessment of individual cases. For example, in our series 4 of 14 lesions with features of malignancy did not display any gross genetic alterations. This lack of practical applicability was painfully obvious in the case of an 80-year-old woman (no. 8 and 24) with an incompletely removed lesion described as benign adenomyoepithelial tumour. This apparently benign tumour recurred two years later as an overtly malignant adenomyoepithelioma. The initial tumour showed gains of genetic material in three distinct loci including 8q, whereas the malignant recurrence showed a further amplification on the long arm of chromosome 8 in addition to a loss of genetic material on the long arm of chromosomes 11 and 12. This suggests a clonal progression of changes already present in the initial lesion. Yet, similar to myoepithelial tumours of the salivary glands [24], a specific alteration pattern could not be defined for their counterparts of the breast, a result which is nevertheless in line with most published work in the field.

Our data suggests the use of a two-tiered terminology according to which adenomyoepithelial tumours are classified as biphasic and monophasic in terms of their deviation from the normal architecture of the myoepithelial-epithelial formation. Furthermore, our data revealed hints for Ck5/6-positive cells as a key component in the histiogenesis of these neoplasms. Nevertheless, any prognostic assessment has to be made on morphological grounds with special attention to proliferative activity.