Breast cancer is the most common invasive cancer in women. Sex steroid hormones estrogen and progesterone are key drivers in the carcinogenesis through their actions on estrogen receptor alpha (ER) and progesterone receptor (PR). In daily practice, breast cancer molecular classification is based on the immunohistochemical expression of these receptors (ER and PR) and of Human Epidermal Growth Factor Receptor 2 (HER2), a member of the epidermal growth factor receptor family. However, the androgen receptor (AR), another member of the steroid receptor family, is also largely expressed in more than 70% of breast cancers and is now clearly implicated in the pathogenesis of breast cancer 1. Although largely co-expressed with ER, AR can also be overexpressed in ER(-) breast tumors 2. The ER(-) tumors represent 30% of breast cancers and are highly heterogeneous, including at least basal-like (BL) tumors and part of the HER2 positive tumors. Yet, among these ER(-) tumors, several teams have identified the molecular apocrine breast cancer (MA) subtype, characterized by AR expression and AR pathway activation on genome-wide expression analyses, paradoxical expression of genes known to be ER targets or expressed in ER(+) tumors and HER2 overexpression in around 50% of cases 34. The existence of this MA subgroup suggests a new molecular classification for breast cancers, including luminal, MA and BL breast cancer subgroups 5. AR overexpression may provide a new therapeutic target for breast cancer 6, especially in patients with ER(-) tumors that do not benefit from endocrine or HER2 targeted therapies. A potential therapeutic effect of AR inhibition in MA subtype has already been shown using in vitro models 4. However, there is no clear consensus yet to define the MA subgroup, except by transcriptomic analysis. An easy and reproducible method to identify MA breast cancers is needed to better understand the behavior of these tumors and to enable their inclusion in specific trials.

Here, we used a molecular apocrine qRT-PCR signature initially defined on a set of breast cancer samples annotated with their transcriptional profiles. We retrospectively identified a group of MA tumors based on this signature. We described their clinical, molecular and pathological features and we identified a new simplified immunohistochemical and molecular signature leading to an easy to use and reproducible diagnostic tool for these tumors.

Breast cancers are highly hormone-dependent tumors. The prominent role of the estrogen and progesterone receptors as prognostic factors and treatment targets is well established. Despite the fact that androgen expression in breast cancer has been known for a long time, interest for this type of hormone nuclear receptor as a possible biomarker is more recent. The description of the MA breast cancer subgroup raises new questions about AR implication in prognosis or prediction of response to hormone blockade in an ER-negative setting. Moreover, easy and robust identification of this new tumor subgroup is needed for proper evaluation. We report here a series of 58 MA breast tumors and compare protein and mRNA expression of genes included in the transcriptomic MA signature 4. We propose a sensitive and specific composite immunohistochemical and molecular diagnostic signature.

MA tumors were originally described by Farmer et al. 3 who identified, with microarray data, a new subgroup of breast tumors which often displayed apocrine morphology (five out of six cases), and they, therefore, suggested that these tumors be called "MA breast carcinomas". To date, data have been somewhat confusing regarding the relationship between MA breast tumors and pathologically-defined apocrine tumors. On breast tissue sections, histological apocrine changes are defined by eosinophilic cytoplasms with fine granularity, large nuclei with occasional large nucleoli and frequent apical snouts 2122. The term "apocrine carcinoma", as defined by microscopical examination, should be used for neoplasms composed of at least 90% apocrine type epithelium 23. These tumors account for 0.5% to 4% of breast carcinomas and mostly occur at advanced ages. Their prognosis is similar to that of other breast carcinomas 24. In our study, we found only four morphological apocrine tumors among 58 MA cases, strongly suggesting that the MA subtype could in fact be much broader than initially reported by Farmer et al. 3.

A frequently reported feature of apocrine differentiation, whatever being defined morphologically or molecularly, is a strong androgen receptor expression in an otherwise ER(-) background. Indeed, although sometimes reported to belong to several molecular subtypes of breast cancer 25, morphological apocrine carcinomas are almost always ER(-) tumors 26 and, among 19 cases, all were ER(-), PR(-) and strongly AR(+) by IHC 27. Currently, the MA subgroup is characterized by AR gene overexpression and Niemeier et al. 28 suggested that the ER(-)/PR(-)/HER2(+)/AR(+) or ER(-)/PR(-)/HER2(-)/AR(+) IHC signature could include the "MA group" described previously by Farmer et al. 3. However, our results raise the question whether AR, as assessed by IHC, is a good marker to define this tumor subgroup. Indeed, although this marker was pretty specific for MA carcinomas, since it was absent in BL cancers, we found that only 58% of MA tumors were IHC AR(+), thus making this marker not sensitive enough for clinical use. Conversely, AR mRNA, evaluated by qRT-PCR, was highly expressed, whatever the level of IHC expression. Doane et al. 4 also found that only 50% of MA tumors expressed AR by IHC, despite high AR expression levels in microarrays data. This dissociation between AR mRNA and protein expression may be due to a lower sensitivity of immunohistochemistry or a faster AR degradation due to activation, leading to undetectable protein.

AR related genes AGR2, ALCAM SPDEF and TTF3 were found in the top 50 significantly overexpressed genes in the transcriptomic profiles of MA tumors 4. Some of them, like AR itself or SPDEF, have putative androgene response elements (ARE). Moreover, AGR2 (androgen-inducible gene anterior gradient-2) is highly induced by androgen in an androgen receptor (AR)-dependent manner. After all, the main trait of MA tumors is the activation of the AR pathway, and AR target genes could help to identify this tumor subgroup.

FOXA1 (Forkhead box protein A1), a member of the forkhead class of DNA-binding proteins, is a pioneer factor facilitating the recruitment of ER and AR to their response elements on the genome. In breast cancer, FOXA1 expression is highly correlated with ER(+), PR(+) and endocrine signaling, and may have a prognostic value in ER(-) tumors 29. Doane et al. reported that FOXA1 is overexpressed in the MA subtype 4. In our study, according to the transcriptomic definition of our MA group (ER(-) AR(+) FOXA1(+)), up to 90% of MA tumors were also FOXA1 positive by IHC. However, 31% of BL tumors were positive as well, thus making FOXA1 immunostaining not specific enough to identify MA tumors.

GCDFP15, the product of the AR-target gene PIP (Prolactin Induced Protein) 30, has been initially described as an apocrine marker. It is present at high concentrations in breast cyst fluid 24 but also expressed in ER(+) or ER(-) breast carcinomas. To date, it is routinely used for the diagnosis of metastasis of unknown origin to identify tumors coming from the breast, with 95% specificity and 74% sensitivity 2431. GCDFP15 mRNA are also overexpressed in MA tumors 4. However, in our study, GCDFP15 was not more sensitive than AR to identify MA tumors when analyzed alone by IHC (57% IHC GCDFP15(+) in the MA group). Our results also showed that GCDFP15 negativity was constant in BL tumors. Therefore, in ER(-) tumors, GCDFP15 positive IHC staining seems to be highly specific of MA cancers but not very sensitive of this subtype.

We then analyzed the sensitivity and specificity of the potentially relevant following IHC signatures: AR alone, AR AND FOXA1, HER2 OR GCDFP15. Doane et al. 4 also proposed one IHC signature without AR: ER(-) PR(-) ALCAM(+) and SPDEF(+), but this signature was validated on only 10 MA tumors. Besides that, ALCAM/SPDEF immunostainings are not widely used and have not been standardized so far. Niemeier et al. 28 reported that the tumors with the IHC profiles ER(-) HER2(3+) AR(+) or ER(-) HER2(-) AR(+) may be MA tumors but it is likely that other MA tumors will be missed with this definition centered on AR IHC. We propose here, in ER(-) breast tumors, a composite signature including AR gene expression pathway analysis by qRT-PCR in addition to "HER2(3+) or GCDFP15(+)" protein expression by IHC. Our IHC signature is composed of validated immunostainings, with international guidelines, used on a daily basis in pathology laboratories. The qRT-PCR profile can be easily obtained on frozen tissue, provided that a frozen-tissue workflow is organized. The validation of quantification on formalin-fixed paraffin-embedded tissues is under way and will allow qRT-PCR analysis on almost all samples.

We also analyzed other molecular alterations in these MA tumors. The phosphatidylinositol-3 kinase (PI3K) signaling pathway is crucial for cell growth and cell survival. Mutations in the gene encoding the p110a (PIK3CA) subunit of PI3K are commonly found in breast cancer. Gonzales-Angulo et al. 32 reported a possible association between the level of AR and PIK3CA mutations. In our study, although we found PIK3CA mutations only in MA tumors, the difference was not statistically significant; however, the number of patients was low. Also, no difference in AR mRNA expression level could either be detected in the PIK3CA mutated group of MA tumors compared to the PIK3CA wild type counterpart (data not shown).

TP53 mutations are commonly found in ER(-) breast cancer, and nearly 90% in the triple negative subgroup, but were present in only 50% of our MA tumors. In prostate cancer cells, it was described that loss of p53 function contributes to increased AR expression 33. In our study, no differences in AR expression could be detected in the p53 nonfunctional subgroup of MA tumors compared to the functional p53 group (data not shown). This potential link between AR and p53 should be further evaluated.

In breast cancers, AR is overexpressed in up to 70% of cases 34, mainly in luminal and low grade tumors 635. This expression has been associated with a favorable outcome in ER(+) tumors 3637 as well as in ER(-) tumors 353839

Little is known about the clinical outcome of MA carcinomas, as very few tumors were available in previous transcriptomic studies (six specimens in the works of Farmer et al. 3, and nine in Doane et al. 4). In our study cohort, demographic or clinical presentation did not show specific features. We did not find any association with Cowden disease, as reported by some authors 40. Though limited by small sample size and retrospective review, we found that MA tumor phenotypes appeared to be rather aggressive, with a high proportion of poor prognosis factors (grade SBR3, lymphovascular invasion, node involvement). They were also associated with a poor clinical outcome despite the fact that the wide majority of patients received chemotherapy.

Recently, Lehmann et al. 41 re-analyzed 21 breast data sets, including 587 triple negative breast carcinomas (TNBC), and identified a new TNBC subtype called "luminal androgen receptor" (LAR) type. Though not named "molecular apocrine" and all being HER2(-), these 62 LAR tumors seemed to be included in the MA group. Relapse free survival was significantly decreased in the LAR subtype compared with other TNBC subtypes.

Finally, our study demonstrates that we could accurately identify MA breast tumors by AR pathway analysis by qRT-PCR, in addition to HER2 or GCDFP15 protein overexpression by IHC. This composite tool may be sensitive, specific and useful to differentiate ER negative tumors in MA or BL, in daily practice. Accurate identification of MA tumors could help to include these patients with rather aggressive tumors in specific "AR-pathway" therapeutic trials or to identify other therapeutic targets.

AGR2: androgen-inducible gene anterior gradient-2; AR: androgen receptor; ARE: androgene response element; AUC: area under the receiver operating characteristics curve; BL: basal-like; DFS: disease free survival; ER: estrogen receptor alpha; FASAY: functional assay in yeast; HER2: Human Epidermal Growth Factor Receptor 2; IHC: immunohistochemistry; LAR: luminal androgen receptor; MA: molecular apocrine; MCA: multiple correspondence analysis; OS: overall survival; PI3K: phosphatidylinositol-3 kinase; q-RT PCR: quantitative reverse transcription PCR; TNBC: triple negative breast cancer

The authors declare that they have no competing interests.

JLC, ASH and PB designed the study, collected and analyzed the data, and wrote the manuscript. JLC and ASH contributed equally to this work. RP carried out all the statistical analyses. MB carried out the biological data collection, the data management and contributed to data analysis. FB carried out all immunohistochemical stainings.

HH and SLD carried out the clinical data collection with review of the medical charts and survival follow-up. PB and AdR carried out all diagnoses and pathological interpretations. LCD, EB, CdB and MA obtained fresh frozen tissues sections under optimal conditions for all the biological studies. SG and CC enrolled patients. ME contributed to the design of the study, enrolled patients and contributed to data collection. PdC, AJ and HdT read and critically revised the manuscript. All authors read and approved the final manuscript.