Loss of periostin/OSF-2 in ErbB2/Neu-driven tumors results in androgen receptor-positive molecular apocrine-like tumors with reduced Notch1 activity

Previous reports have shown that Postn-null mice are viable but display impaired periodontal ligament development [14,38,39]. Therefore, prior to the introduction of the Postn-null allele into MMTV-NeuNDL mice, we assessed Postn expression and whether Postn ablation was compatible with normal mammary gland development. Mice carrying a LacZ knock-in [14] into the Postn locus were bred into an FVB/N background and assessed for Postn expression and mammary gland development in virgin and pregnant females. Postn protein isoforms [40] are readily detectable in the mammary glands of 8-week-old virgin wild-type females but absent from Postn-null glands (Figure S1 in Additional file 2). Mammary gland whole mount analyses (Figure S1 in Additional file 2) revealed normal branching and ductal outgrowth in Postn(−/−) nulliparous females, indistinguishable from that of control mice. Similarly, 12-week-old pregnant (14.5dpc) Postn-deficient females displayed extensive outgrowth and arborization that was similar to wild-type or Postn(+/−) littermates (Figure S1 in Additional file 2). Histological analysis by hematoxylin and eosin (H&E) staining of nulliparous mammary gland showed no defects in the ductal epithelial and stromal layers (Figure S1L-O in Additional file 2). Supporting this, we observed that FVB/N Postn^−/− female mice were able to lactate and produce viable litters through multiple rounds of pregnancy (data not shown). However, in contrast to what has been observed in a C57/Bl6 background [14], the viable Postn-null pups displayed a mild growth retardation phenotype that became negligible by 9 weeks of age (data not shown). Together, these results suggest that Postn is not required for normal mammary gland development.

Postn has been shown to be expressed in the stromal and epithelial compartment of the embryonic heart [41]. Similarly, Postn expression was detected in both the stroma and tumor cells of human breast cancer samples [19-21]. Therefore, the cell type-specific expression of Postn in the mouse mammary gland was investigated using immunohistochemical analysis of paraffin-embedded sections. Examination of mammary tissue derived from Postn^+/+ and Postn^+/− mice revealed that Postn is expressed in the stromal cells lining the ducts and lobules, but not in the mammary epithelial cells (Figure S1 in Additional file 2). Postn reactivity was also observed in the intercellular space between the adipocytes.

Postn has been shown to play multiple roles in tumor progression through activation of pathways involved in invasion, cellular survival, angiogenesis, and metastasis [15,42]. Furthermore, Postn expression levels were found to be increased in a high proportion of breast cancers [19-21]. Supporting this, our analysis of human tissue microarray (Figure 1) revealed that about 46% (16/35) of human breast cancers acquire Postn expression irrespective of receptor status (Table 1).

Therefore, to investigate whether Postn expression is required for mammary tumorigenesis in an ErbB2-positive model, the in-frame Neu deletion transgene under the MMTV promoter (MMTV-NeuNDL2-5) [31] was introduced into mice lacking Postn. Postn wild-type, heterozygotes and null NeuNDL virgin females were monitored for tumor development. In all Postn genotypes, hyperplastic lesions or palpable tumors could be detected as early as 16 weeks of age with no differences in the number of observed lesions (Figure 2A,B and Figure S2 in Additional file 3). Similarly, no differences were observed in tumor growth as all the Postn genotypes reached end point with similar kinetics (Figure 2G). Further analysis of NeuNDL-Postn^+/+ and Postn^−/− tumors revealed a mixture of intracystic or encapsulated intraductal papillary carcinomas (Figure 2C-F) that retained Neu expression (Figure 2H,J). In addition, no changes were observed in CD31-positive tumor blood vessels or in the Ki-67 proliferative index (Figure S3 in Additional file 4). These findings suggest that Postn ablation does not prevent or delay primary tumor initiation or growth. Similar findings have been reported in the MMTV-PyMT mouse model [22].

In human breast tumor tissues, Postn was reported to be highly expressed in the stromal cells immediately surrounding the tumor and in late-stage mammary tumors [19,21]. Supporting this, immunohistochemical analysis of Postn in NeuNDL Postn^+/+ and Postn^+/− tumors shows expression in the stromal compartment around the epithelial mammary tumors. Interestingly, Postn was undetectable in the mammary tumor cells (Figure 2K). Similarly, Western blot analysis of mammary glands and tumors shows higher levels of Postn in whole mammary gland containing a higher proportion of mammary fibroblasts (Figure 2L). Together, these results suggest that Postn expression is restricted to the stromal compartment of NDL tumors and is not upregulated in the tumor epithelium in MMTV-Neu-NDL mice.

To further investigate the effect of Postn deletion in mammary tumorigenesis, we surveyed a number of markers activated downstream of Postn treatment. Postn stimulation has been previously shown to activate FAK and Akt through β3 or β4 integrin binding [16,43,44]. Western blot analysis showed an overall 2-fold decrease in the levels of phospho-FAK-Y397 protein kinase (Figure 3A). However, β3 integrin levels remained unchanged in Postn-null tumors (not shown). Interestingly, the levels of phospho-Akt-S473 were not affected, suggesting that the loss of Postn can preferentially affect specific pathways in mammary tumors. Similarly, ErbB3 levels were unaffected (not shown). Supporting our Ki-67 observations, cyclin D1 levels also remained unchanged in tumors derived from Postn-null animals (Figure 3B).

Although Postn has been shown to be overexpressed in a high proportion of human tumors and to play a role in tumor cell growth in vitro [15,19-21,42], Postn deletion has no effect on tumor development in ErbB2- (Figure 2) or PyMT-expressing mice [22]. However, previous data have shown that Postn plays a critical role in the establishment of tumor cell niche at secondary sites [22]. Supporting this, administration of Postn neutralizing antibodies in xenograft models of breast and ovarian cancers also suppressed invasion and metastasis [45,46].

Therefore, in light of these observations we tested whether mammary tumor cells injected at heterotopic sites could form tumors in a Postn-null environment. As skin tissue is a Postn-rich environment [47,48] we tested the requirement for Postn in tumor growth using subcutaneous injections. As PyMT also activates ErbB2 downstream signaling during tumorigenesis [49], we used MMTV-PyMT-derived Met-1 cells [33] in an isograft model.

Whereas large tumors were observed in Postn wild-type females, little or no growth was observed following the subcutaneous injection of Met-1 cells into Postn-deficient animals. Tumor volume measurements showed that Met-1 cells in a wild-type environment grew to about 10 times the size of the tumors collected following injections into Postn-null mice (Figure 3C,D). In the Postn-null background, no significant increase in tumor growth was observed up to 45 days whereas all wild-type mice had to be euthanized beyond 5 weeks due to very large tumors (data not shown). Interestingly, exogenous expression of Postn in Met-1 cells did not rescue their growth defect in a Postn-null environment. As those animals never expressed Postn protein, this is likely due to immune rejection of Postn-expressing cells in null mice. These data strongly suggest that Postn is required for sustaining tumor growth at heterotopic sites.

Postn has been shown to interact with Wnt and to potentiate Wnt signaling to enhance tumor cell seeding [22]. Interestingly, Western blot analysis for β-catenin (or active β-catenin; not shown) levels, nuclear localization or activity showed no difference in tumors derived from Postn-null mice (Figure S4 in Additional file 5), suggesting that the loss of Postn does not affect canonical Wnt signaling in ErbB2-driven tumors. Furthermore, this suggests that the impaired tumor growth observed in Postn(−/−) mice is unlikely due to deficient Wnt signaling.

Previous studies have shown that the lack of Postn during embryonic development results in aortic valve defects [50]. These anomalies are due to increased expression of the Notch1 antagonist Dlk1 and suppression of Notch signaling. Similarly, Postn association with Notch1 has been found to maintain Notch1 levels under stress conditions [51]. As Notch also plays an important role in mammary carcinoma and stem cell function [52-55], we investigated whether Postn-deficient tumors also displayed alterations in the Notch pathway. As Dlk1 mRNA levels were unaffected (not shown), we directly assessed the levels of active Notch and its downstream target Hey1 in mammary tumors derived from wild-type and Postn-deficient animals.

Western blot analysis revealed a 3-fold decrease in the levels of the transcriptionally active Notch intracellular domain fragment (NICD; reviewed in [52]) (Figure 4A). This was accompanied by an approximately 50% downregulation in Hey1 mRNA levels, a NICD target gene (Figure 4B). However, Hes1 levels remained unchanged (not shown), supporting the notion that Notch-mediated activation of Hes1 transcription is context and cell type-dependent [56]. Together our results suggest that the loss of Postn is accompanied by reduced Notch1 activity and a failure to support tumor growth at secondary sites.

Our data suggest that high Notch activity is required for tumor growth at heterotopic sites. Therefore, we tested whether increased Notch activity was sufficient to bypass the Postn requirement in our heterotopic model. Met-1 cells stably expressing NICD were generated and compared to Met-1 expressing Postn or a nuclear localization signal mutant NICD construct (ΔNLS; Figure S5 in Additional file 6). In cell culture, overexpression of Postn in Met-1 cells resulted in a γ-secretase-dependent upregulation of NICD protein levels (Figure 4C), suggesting that Postn signaling in mammary tumor cells can contribute to the regulation of active Notch levels. The upregulation of NICD in Postn-expressing cells was accompanied by a 4-fold increase in Hey1 mRNA levels (Figure 4D). Met-1 cells expressing NICD upregulated Hey1 mRNA by more than 50-fold. In contrast, no induction was observed in control cells expressing the NLS mutant form of NICD. When injected subcutaneously into Postn-deficient FVB/N female mice, Met-1 cells expressing NICD grew as well as in wild-type mice, suggesting that NICD re-expression is sufficient to bypass the Postn requirement observed for control cells (Figure 4E). Interestingly, in wild-type mice, NICD overexpression did not enhance tumor growth beyond what was observed for control cells, suggesting that the levels of downstream NICD effectors might be limiting. Expression of the inactive NICD could not rescue the Met-1 growth defect in Postn-null mice, suggesting that Notch activity is required.

Interestingly, Postn or NICD expression in Met-1 cells did not enhance the activity of a β-catenin reporter gene (Figure S4 in Additional file 5), suggesting that Notch1 does not bypass the Postn requirement by indirectly stimulating the Wnt system. Although Notch1 has been previously shown to play a role in the expansion of premalignant and tumor-initiating cells in NICD-induced mammary tumors [55,57], we did not find any differences in the primary or secondary tumorsphere forming ability of primary tumors derived from Postn-null mice (Figure S6 in Additional file 7), suggesting that the loss of Postn has no effect on the primary tumor stem cell compartment.

To test the requirement for the Notch pathway in a wild-type environment, a dominant-negative form of the Notch co-activator Mastermind-like 1(dnMAML1; [58]) was overexpressed in Met-1 cells. Expression of dnMAML1 in Met-1 cells resulted in Hey1 downregulation and poor tumor growth upon subcutaneous injection (Figure 4F). Furthermore, expression of dnMAML1 resulted in a 50% decrease in tumor take rate when compared to GFP-expressing control cells (GFP: 5/6 vs dnMAML1: 8/16). Together, these data suggest that Postn is necessary to maintain a threshold level of active Notch, required for tumor establishment and growth at heterotopic sites.

Although loss of Postn did not affect overall survival and tumor development, our results show that it impaired tumor growth at heterotopic sites. In addition, careful histological examination of these tumors revealed a molecular apocrine-like phenotype, characterized by abundant granular eosinophilic cytoplasm, vesicular nuclei with prominent nucleoli (Figure 5A,B). The molecular apocrine phenotype is a characteristic of androgen receptor (AR)-positive breast cancers [59,60]. Therefore, we assessed AR protein levels in wild-type and Postn-null NeuNDL tumors by Western blot analysis. Although AR mRNA levels were unchanged (not shown), AR protein levels were found to be upregulated 2-fold in the majority of tumors derived from Postn-null mice (Figure 5H). Furthermore, immunohistochemical analysis for AR expression revealed a significant increase in nuclear localization of AR in tumors from Postn-null animals, suggesting an increase in AR activity (Figure 5C-G). The increase in AR protein levels in Postn-null tumors suggests a posttranscriptional upregulation or stabilization of AR protein levels [61,62]. Interestingly, we have identified one tumor sample where AR levels were unchanged which could be correlated with wild-type levels of active and total Notch. Together these findings demonstrate that the loss of Postn results in Notch downregulation which is accompanied by AR upregulation in the primary tumors. Furthermore, this suggests that the loss of Postn in ErbB2-expressing tumors confers an AR+ apocrine-like phenotype.

To gain further insights into the molecular mechanisms regulating tumor progression in a Postn-null environment, we compared the gene expression profiles of Postn(−/−) and Postn(+/+) tumors using Affymetrix microarrays. Table S2 in Additional file 1 shows the genes that were up- or downregulated by at least 10-fold in Postn-null tumors. Supporting the histological findings, numerous genes that have been previously reported as AR target genes, such as the androgen-binding proteins [63] and PIP [64] were upregulated in Postn-null tumors. These observations suggest that the activation of critical genes in the absence of Postn allows primary tumor growth in the absence of high Notch activity.

In human tumors, PIP (PIP/GCDFP15) expression was shown to be associated with AR+ molecular apocrine-like tumors [65-67]. To confirm the array data, Q-PCR was performed on tumor RNA to validate PIP overexpression in Postn-null tumors. In addition, we assessed the mRNA levels of androgen-binding proteins (Abp), previously shown to be upregulated by testosterone in cultured cells [63]. Quantitative PCR analysis revealed that both PIP and Abp (ε and ζ) were significantly upregulated in Postn-deficient tumors (Figure 6). Whereas PIP was increased by about 20-fold (Figure 6A), Abp ε and ζ were upregulated by more than 1,000-fold (Figure 6B), suggesting transcriptional activation of AR target genes in tumors derived from Postn-deficient animals.

Interestingly, in addition to being an AR-regulated gene [64], PIP/GCDFP-15 has been demonstrated to be expressed in a large proportion of human breast cancers and to enhance the growth and invasion of breast cancer cell lines [68-71]. In addition, PIP was found to activate AR activity, which in turn, upregulates PIP transcription in feed-forward loop [70]. As PIP is an AR target gene that is highly induced in tumors from Postn-null mice, we tested the possibility that PIP expression in wild-type Met-1 cells could compensate for the absence of Postn in vivo. The murine PIP cDNA was cloned and stably expressed in Met-1 cells and pools were injected subcutaneously in Postn-null female mice. Similar to NICD-expressing cells, at the 4-week end point, Met-1 cells overexpressing PIP formed large tumors in Postn-null mice that were comparable in size to those observed in wild-type animals (Figure 6C). Control puromycin-resistant cells grew very poorly in the Postn-deficient environment. These results suggest that PIP is a major downstream effector of AR activation and that its expression in Met-1 cells is sufficient to overcome the Postn deficiency in vivo. However, mechanistically, it is likely that the rescue by NICD and PIP proceeds through distinct pathways.

Further, AR levels did not change in control and NICD expressing subQ tumors in Postn-null mice as compared to subQ tumors grown into wild-type mice (Figure S5B in Additional file 6), suggesting that the NICD growth rescue in null mice does not result in an apocrine phenotype. Together, our results show that the loss of Postn leads to Notch downregulation and selects for AR-positive molecular apocrine-like tumors.