Prolactin-induced mouse mammary carcinomas model estrogen resistant luminal breast cancer

5-bromo-2-deoxyuridine (BrdU) was obtained from Sigma Chemical Co. (St. Louis, MO, USA), and 17β-estradiol (E2) was purchased from Steraloids, Inc. (Newport, RI, USA. The following antibodies were used for immunohistochemical analyses: BrdU (MAS-250) from Accurate Scientific (Westbury, NY, USA), proliferating cell nuclear antigen (PCNA; PC 10) and progesterone receptor (PR; A0098) from DAKO Cytomation (Carpinteria, CA, USA), c-Fos (SC-52) and estrogen receptor alpha (ERα; SC-542) from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA, USA), pERK 1/2 (Thr202/Tyr204; 9102), pAkt (S473; 3787), and c-Jun (9162) from Cell Signaling Technology (Beverly, MA, USA), pStat5 (AX1) from Advantex BioReagents, LLP (El Paso, TX, USA).

NRL-PRL mice (line 1655-8, TgN(Nrl-Prl)24EPS) were generated and maintained in the FVB/N strain background as described [14]. Mice were housed and handled in accordance with the Guide for Care and Use of Laboratory Animals in AAALAC-accredited facilities. All procedures were approved by the University of Wisconsin-Madison Animal Care and Use Committee. p53^+/- heterozygotes [26] that were backcrossed more than 10 generations into the FVB/N strain [27] were used to generate p53^-/- donors. p53^-/- mammary epithelial cells (MECs) were transplanted into FVB/N nontransgenic female recipients, and the resulting tumors harvested (manuscript in preparation, O'Leary and Schuler). Mammary tumors from other genetically modified murine models (FVB/N) were generously provided by Drs. Kim and Alexander (University of Wisconsin) (MMTV-neu [28], and Drs. Green and Zi-Yao Liu (NIH-NCI) (C3(1)-SV40 Tag tumors [29].

For some studies, mice were injected with 200 mg/kg body weight BrdU 1 h prior to sacrifice to label cells undergoing DNA synthesis. For immunohistochemistry (IHC), deparaffinized slides were exposed to 0.5% H₂O₂ in methanol to block endogenous peroxidase activity, boiled for 15 minutes in 0.1 M citrate (pH = 6.0) or 0.1 M Tris (pH = 9.0) buffer for antigen retrieval, then blocked, and incubated with the primary antibody overnight (for individual antibodies, see Table S1 in Additional File 1). Slides were incubated with the secondary antibody (BioGenex, San Ramon, CA, USA), followed by peroxidase strepavidin and 3,3' diaminobenzidine, and counterstained with hematoxylin. Cells were labeled for PCNA using Vector M.O.M. Basic kit (BMK-2202; Vector Laboratories, Burlingame, CA, USA). pStat 5 was detected as described [30]. Serial sections of tumors were examined for the different endpoints. For all IHC, control sections were stained in parallel (negative controls: omission of primary antibody, irrelevant antibody; positive controls, appropriate for the antibody) as described [14]. To determine labeling indices, 1,000 total cells in at least five randomly chosen microscopic fields in divergent regions of each carcinoma were counted by an investigator blind to the experimental status.

In a preliminary study to compare the tumors that develop in NRL-PRL to other mouse models of mammary cancer, cDNA prepared from tumors was hybridized to Agilent Mouse Oligo Microarrays (Agilent Technologies Inc., Santa Clara, CA, USA) using total RNA from equal numbers of C57Bl6/J and 129 male and female pups as the reference sample, as previously described [31]. Data were processed and hierarchical clustering performed, as described [31].

To compare NRL-PRL tumors identified as ERα positive or negative by IHC, total RNA was purified using RNeasy Midi Kits (Qiagen Inc., Valencia, CA, USA) according to the manufacturer's instructions. RNA integrity was assessed using the RNA 6000 Nano Assay and Agilent 2100 Bioanalyzer (Agilent Technologies Inc.). Total RNA (10 μg) was reverse transcribed and labeled with Cy5 using the Array 50 kit (Genisphere Inc., Hatfield, PA, USA). MECs were isolated as described [32] from 12-week-old nontransgenic nonparous FVB/N females, for common reference RNA. Reference RNA pooled from 15 mice was reverse transcribed and labeled with Cy3. Tumor and reference samples were co-hybridized overnight to Agilent Mouse cDNA Microarrays (G4104A, 8,614 features), washed, and scanned using Agilent's dual-laser Microarray Scanner (G2565BA) with Feature Extraction Software.

RNA from tumors and MECs was examined using quantitative real time PCR (qRT-PCR). cDNA was synthesized from 1 μg RNA using random hexamers (Amersham Biosciences, Piscataway, NJ, USA) and MMLV Reverse Transcriptase (Promega, Madison, WI, USA), and qRT-PCR performed as described [17]. Negative (no cDNA) and positive controls were included with each plate. Specific primer sequences are listed in Table S2 in Additional File 1. Results were calculated using the comparative C_T method and normalized to 18S RNA, and data were analyzed for statistical significance as described in the figure legends.

To examine the responsiveness of tumors to circulating ovarian steroids, tumors were allowed to develop until they reached 1.5 cm in diameter. About one-third of the tumor mass was then removed, concomitant with ovx or sham surgery, or ovx and subcutaneous treatment with 0.1 mg pellets of 17β-estradiol (E2; Innovative Research of America, Sarasota, FL, USA), which result in plasma levels of 50 to 75 pg/ml [33]. Tumors were measured twice weekly with calipers, and tumor volume was calculated (the largest diameter x (the smallest diameter)² × 0.4). Mice were humanely euthanized and tissue collected when the tumors reached 1.5 cm in diameter. Histological comparisons of the tumor biopsies and end stage tumors were carried out as above.

For some experiments, 1 mm³ NRL-PRL tumor fragments were transplanted bilaterally into fourth inguinal mammary glands of 12-week-old nontransgenic syngeneic FVB/N hosts. After the tumors grew to 0.75 cm in diameter, recipient females were treated with ovx, sham surgery, or sham surgery and weekly subcutaneous injections of 5 mg Faslodex (ICI 182,780; AstraZeneca, Wilmington, DE, USA). Tumors were measured and collected as described above.

NRL-PRL nonparous females develop diverse invasive mammary carcinomas with a long latency [14]. In order to examine the breadth of tumor phenotypes promoted by this prolonged PRL exposure, we examined a panel of 39 of these tumors (mean latency +/- standard deviation (S.D.), 18.2+/-3.6 months). The majority were adenocarcinomas of several histotypes; carcinosarcomas constituted a minor population (Figure 1, Table 1). These carcinomas varied widely in the proportion of cells that expressed detectable ERα (0 to 79%). Nineteen of the 39 tumors examined exhibited greater than 10% ERα+ cells, a common clinically used threshold. The majority did not express detectable progesterone receptor (PR); only four tumors demonstrated PR immunostaining in more than 5% of the tumor cells. The morphological heterogeneity suggests cooperation between PRL and other oncogenic factors during the long period of development of these tumors, similar to that which may occur in women.

PRL can activate multiple signaling cascades, which may play distinct roles in its physiologic and pathogenic actions in the mammary gland. In order to determine the relationship of these signals to tumor characteristics, we examined the activation of these pathways in the panel of PRL-induced tumors by IHC. Signal transducer and activator of transcription 5 (Stat5) is one of the best characterized mediators of PRL activity, and is critical for PRL-induced alveolar development during pregnancy (for review, [1, 34]). The tumors that developed in NRL-PRL females contained variable proportions of cells with nuclear pStat5 (range: 4.2 to 93.3%), which was not related to ERα levels (Figure 2a, d, P > 0.05). Although many glandular and solid tumors exhibited relatively high proportions of cells containing nuclear pStat5, evidence for activation of this pathway varied substantially even among these histotypes. Many adenocarcinomas demonstrated low nuclear pStat5, as did all of the poorly differentiated carcinosarcomas (Table 1).

PRL can also initiate other signals, including several MAP kinases, which are elevated in mammary lysates of NRL-PRL females [17], and are particularly strongly activated in some breast cancer cells lines [35]. PRL-activated MAP kinases can increase synthesis and phosphorylation of multiple Activating Protein-1 (AP-1) components, activating the AP-1 transcriptional enhancer [36]. AP-1 target genes have been shown to enhance cellular proliferation, survival, and invasion (for review, [37‐39]). Both c-Fos and c-Jun were variably expressed in PRL-induced tumors (Figure 2b, c, e). Their expression was directly related to each other (data not shown, P = 0.005, R = 0.83). Although adenosquamous carcinomas and carcinosarcomas had the highest levels of c-Fos expression, a large subset of other histotypes also displayed significant staining (Table 1). Interestingly, nuclear pStat 5 levels were negatively correlated with both c-Fos (Figure 2e; P < 0.0001, R = -0.63) and c-Jun (data not shown; P = 0.0058, R = -0.82). The inverse relationship of these signals in primary NRL-PRL tumors is consistent with our previous observation of PRL signals to Stat5 and AP-1 in several breast cancer cell lines [35].

The higher level of nuclear pStat5 in well-differentiated PRL-induced carcinoma histotypes resembles the association of pStat5 with features of more differentiated clinical breast tumors (for review, [34]). In order to examine the association of this pathway with molecular phenotype in adenocarcinomas, we examined levels of two transcripts associated with distinct mammary behaviors. The 5 PRL-induced adenocarcinomas with highest nuclear pStat5 (70 to 93%) and low c-Fos had significantly elevated transcripts for the proteinase inhibitor, Expi, compared to those adenocarcinomas with lowest pStat5 (4 to 10%, Figure 2f, P = 0.003). Expi expression increases in early pregnancy [40], and is reduced in PRLR^-/- mammary glands [41]. In contrast, the adenocarcinomas with lowest nuclear pStat5 contained elevated mRNA for the metalloproteinase, Mmp9 (Figure 2g; P = 0.03), which in women is associated with high tumor grade, metastasis, and reduced survival [42], consistent with a more aggressive phenotype.

Elevation of phosphorylated extracellular signal regulated kinases 1/2 (pERK1/2) is found in some clinical tumors in women [43], and is increased in lysates of NRL-PRL mammary glands [17]. The location of pERK1/2 varied substantially among PRL-induced tumors. The majority of the adenocarcinomas (68%) demonstrated strong stromal staining with rare nuclear staining in epithelial cells (Table 1, Figure 3a). In contrast, all of the carcinosarcomas in the panel displayed both strong epithelial nuclear and cytoplasmic labeling (Figure 3b, c). This histotype had significantly higher levels of Vim (Figure 3h), but similar levels of Krt8 transcripts compared to the adenocarcinomas (Figure 3i). The spindle shaped cell morphology, loss of cell polarity, and expression of Vim mRNA within these tumors suggest the epithelial to mesenchymal transition (EMT, for review, [44, 45]). Significantly enhanced levels of Mmp9 transcripts were also detected in these tumors (Figure 3j).

The PI3K/AKT pathway also has been associated with proliferation, survival, and EMT in human breast tumors, as well as resistance to endocrine therapy [46‐48]. pAKT is also elevated in mammary glands of NRL-PRL females [17]. The majority of the carcinomas in this panel displayed strong pAKT labeling in both the nuclei and cytoplasm of tumor epithelium (Table 1, Figure 3d). Other tumors exhibited pAKT labeled cells in both the stroma and epithelium, and a minority in the stromal compartment alone (Figure 3e). Unlike pERK1/2, pAKT expression was not correlated with a specific histotype; carcinosarcomas demonstrated variable expression levels of nuclear and cytoplasmic pAKT (Figure 3f, g).

Together, these observations suggest that a broad spectrum of signals may contribute to progression of these carcinomas. Nuclear pStat5 is elevated in hyperplastic lesions in NRL-PRL glands [17], and early ablation of Jak2 protects against PRL-induced tumors [49], consistent with an important role for the Jak2-Stat5 pathway in their genesis. However, the data presented here indicate that high pStat5 persists in only a minor subset of established PRL-induced carcinomas, suggesting that other signals drive progression of most of these tumors. Elevated activities of AP-1 [50, 51], AKT [52] and MAP kinases, including ERK1/2 [43], have been implicated in resistance to conventional chemotherapies and anti-estrogens. PRL can activate these signals alone, as well as potently cooperate with growth factors to enhance ERK1/2 and AKT activation in vitro and in vivo [53‐55]. In light of the variable levels of Prlr transcripts in these tumors (see below), it is clear that PRL itself may play different roles in ongoing activation of these pathways as they influence lesion progression, tumor phenotype, and treatment sensitivity.

To evaluate shared characteristics of PRL-induced carcinomas, in a preliminary study we examined the transcript profiles of six NRL-PRL carcinomas, including three each of the most prevalent histotypes (glandular and papillary), expressing variable levels of ERα (three positive, three negative by IHC, not associated with histotype). Unsupervised hierarchical clustering based on an intrinsic gene list developed for murine models [31] showed that the NRL-PRL tumors formed a distinct subgroup neighboring MMTV-neu tumors.

To compare the phenotype of NRL-PRL tumors more closely to other well-studied genetically modified murine models of breast cancer subtypes in the FVB/N strain background, we compared levels of transcripts for ERα-associated genes in the "luminal" signature [20] and those enriched in various mammary epithelial lineages. MMTV-neu tumors model clinical HER2+ tumors. Like both human HER2+ tumors [25] and other murine tumors that develop in response to MMTV-driven oncogenes [31], MMTV-neu tumors display some characteristics of the "luminal" tumor subtype in women. Tumors in this extensively studied mouse model do not express ERα or FoxA1, a transcription factor which co-regulates many targets [56, 57]. However, they do express other ERα-associated genes, including Xbp1 and Gata 3 [58‐60]. In contrast, tumors that develop in both C3(1)-SV40 Tag females and transplanted p53^-/- models have many features of "basal" tumors in women ([31, 61], O'Leary and Schuler, in prep).

Unlike the morphologically homogeneous MMTV-neu tumors, levels of transcripts that mark luminal breast cancers varied in NRL-PRL carcinomas, consistent with the diverse tumors in this model (Figure 4, Figure S1 in Additional File 1). In contrast to the majority of murine models, PRL-induced adenocarcinomas expressed relatively high levels of Esr1 mRNA, although as predicted from the IHC, levels varied considerably (Figure 4a). PRL-induced papillary carcinomas also contained comparatively high levels of the ERα-associated transcripts in the luminal signature, including FoxA1 and Xbp1 (Figure 4b, c). Gata3 mRNA was readily detectably in NRL-PRL tumors, but levels were lower than in MMTV-neu tumors (Figure 4d). Although Gata3 specifies ductal/alveolar cell fate and is markedly reduced in Prlr^-/- glands [1], this transcription factor also maintains luminal differentiation and is reduced in invasive breast cancer [62], consistent with the less differentiated phenotype of the PRL-induced carcinomas.

In light of the luminal phenotype of many tumors promoted by PRL, we examined levels of transcripts associated with distinct mammary subpopulations and differentiation processes [24, 63]. PRL-induced tumors displayed relatively high levels of Itga6 (encoding integrin α6, CD49f) mRNA, a marker for stem cell and progenitor populations (Figure 5a). They expressed relatively low levels of transcripts for both Itgb1 (encoding integrin β1, CD29), which marks stem cell and basal but not luminal subpopulations (Figure 5b), and also Itgb3 (encoding integrin β3, CD61), a marker for luminal progenitors (Figure 5c). In contrast, three of the four NRL-PRL adenocarcinomas displayed relatively high levels of CD44 (Figure 5d), an adhesion molecule associated with stem cells and metastasis [64, 65]. NRL-PRL adenocarcinomas also contained elevated mRNA for several transcription factors, which direct luminal/alveolar cell expansion and differentiation, are co-expressed with PRLR in many normal MECs, and are linked to PRL in genetic models [1, 66]. PRL-induced adenocarcinomas contained relatively high levels of mRNA for Elf5, a PRL-induced factor that directs the alveolar lineage [67], as well as CEBPβ, a transcriptional regulator which has been implicated in stem cell and luminal progenitor repopulation [68] (Figure 5e, f). Consistent with their roles in luminal cell commitment, individual genetic ablations of Gata3, Elf5, C/EBPβ and also Stat5 have been shown to alter the size of the MEC subpopulation expressing integrin β3 (CD61+), which is enriched in luminal precursors. However, the variable direction and timing of the observed changes reflect their complex actions at more than one stage of this process [24, 68, 69]. Future studies will elucidate the effect of PRL on mammary epithelial subpopulations and roles of these transcriptional regulators in its actions, and illuminate differences in cellular origin and/or pathways of progression from MMTV-neu tumors and among the different histotypes of PRL-induced carcinomas.

Transcripts for the cdk4/6 inhibitor, Cdkn2c (p18^INK4c), were very low (Figure 5g). Genetic ablation of this tumor suppressor induces well-differentiated non-invasive luminal tumors in mouse models [70]. Similar results were observed for Cdkn2a (p16^INK4a, data not shown). These results are consistent with the similar mammary phenotypes of Prlr^-/- and Ccnd1^-/- mice [1], and ability of PRL to increase cyclin D1 transcription [71], and support a key role for cdk4/6 in PRL-induced tumorigenesis.

PRL also may contribute to the tumor phenotype of other models; some C3(1)-SV40 Tag as well as MMTV-neu tumors contained appreciable Prlr mRNA (Figure 5h). The CD61+ progenitor population has recently been shown to be the origin of basal tumors [72]; the size of this MEC subpopulation is strongly modulated by Stat5. These observations invite speculation on potential therapeutic or preventative approaches directed at PRL action in multiple tumor types.

These marked differences in the molecular phenotype of PRL-induced tumors compared to other genetically modified mouse models of breast cancer suggest that PRL acts on its physiological target cells to promote diverse luminal-type tumors, but raises questions about the identity of these cells, the role of mediators associated with "normal" function in tumor progression, and factors contributing to tumor heterogeneity. Development of additional markers for the various mammary epithelial subpopulations, and antibodies that recognize the native conformation of the murine PRLR will assist in determining the characteristics of the precursors of luminal tumors, and identifying the target cells of PRL in this process.

In our preliminary studies comparing PRL-induced tumors to other murine models of breast cancer, PRL-induced ERα- and ERα+ tumors clustered together, indicating shared features suggesting a common origin. In order to determine characteristics of PRL-induced tumors that are associated with differences in ERα protein expression, we compared transcript profiles from four ERα+ and four ERα- adenocarcinomas (determined by IHC) of divergent histotypes that did not segregate with ERα status (three papillary, two glandular, one adenosquamous, two solid). Genes that were differentially expressed among the classes were determined, and hierarchical clustering analyses were performed by comparing expression profiles both across the set of samples and across the set of genes. As shown in Figure 6a, ERα+ and ERα- tumors clustered separately. Consistent with the reports of ERα+ tumors in women, NRL-PRL ERα+ tumors expressed higher levels of Gata-3 than ERα- tumors, and markers of differentiation, such as the milk proteins, β-casein and whey acidic protein, and CD24a [20, 73‐75]. Levels of mRNA for the PRL transgene and PRLR varied markedly among both ERα+ and ERα- carcinomas, suggesting that PRL activity itself was not responsible for the differences observed (Figure S2 in Additional File 1).

Activity of the NRL-PRL promoter is not altered by estrogen [15], enabling us to examine the role of this steroid in the disease process. Proliferation of both morphologically normal MECs and preneoplastic epithelial hyperplasias in NRL-PRL females is very sensitive to ovariectomy (ovx) and supplemental estrogen [17, 54]. In order to examine estrogen responsiveness in established carcinomas, NRL-PRL females bearing primary mammary adenocarcinomas were subjected to sham surgery or ovx. At the time of surgery, a biopsy was removed for examination of tumor histology, rate of proliferation, and steroid hormone receptor expression. The growth of the remaining tumors was monitored until they reached 1.5 cm in diameter (15.3 ± 8.5 days; mean ± s.d.), as described in the Materials and methods. Like the tumors in Figure 1, histotypes in this experimental panel were diverse, and ERα/PR expression varied widely. They exhibited highly variable rates of growth (Figure 6b); rates of proliferation of ERα- tumors prior to the surgery tended to be higher than ERα+ tumors (P = 0.07). However, ovx did not alter rates of proliferation of either ERα+ or ERα- tumor cells. Consistently, ovx followed by administration of E2 failed to alter proliferation of an independent set of primary tumors (PCNA staining: 30.1% ± 12 vs 31.19% ± 7.9, mean ± s.d., P > 0.05, N = 5).

Since primary PRL-induced tumors exhibit highly variable histotype and ERα expression, and ovarian function is likely to be faltering in the aged females, we employed tumor transplantation to further investigate the role of ERα-dependent signals in established tumors. We serially passaged NRL-PRL glandular adenocarcinoma fragments that demonstrated high expression of both ERα and PR (48% and 25% labeled cells, respectively) to bilateral inguinal mammary glands of nonparous 12-week-old syngeneic nontransgenic recipients. Although the tumors that grew from these fragments (about eight weeks per passage) retained the morphology and high ERα expression of the original tumors, PR was no longer detectable after passage (not shown), suggesting that this process selects for more aggressive tumor cells. After three passages, recipient females were treated with ovx, sham surgery, or sham surgery and the ERα-selective antagonist, Faslodex. None of these treatments altered rates of proliferation of the tumors, as measured by BrdU incorporation (Figure 6c), indicating that relatively well-differentiated ERα+ tumors display the estrogen independence observed in the more diverse group of primary tumors.

These findings raise interesting questions regarding the mechanism(s) whereby PRL interacts with ERα-mediated signals in the development and progression of ERα+ tumors. The striking estrogen independence of established PRL-induced tumors shown herein contrasts with the cooperation between PRL and estrogen in proliferation of morphologically normal MECs and early lesions [17, 54], suggesting acquisition of estrogen insensitivity. Many primary clinical ERα+ tumors acquire resistance to estrogen-directed therapies and most display marked therapeutic insensitivity after metastasis. A vast literature implicates many mechanisms, including aberrations not only of ERα itself, but alterations in other molecules and signaling pathways consistent with heightened growth factor activity [76‐78]. The ability of PRL to potentiate growth factor signals [53‐55, 79] and high activation of AKT in the tumors shown here are consistent with a role in the latter mechanism. Clinically, high circulating PRL has been associated with failure of ERα+ tumors to respond to tamoxifen and aromatase inhibitors [80‐82], and PRLR transcripts were elevated with recurrence after tamoxifen treatment in some studies [13, 83]. The histological heterogeneity and variable activation of signaling pathways of PRL-induced tumors observed here suggest that multiple mechanisms may lead to loss of estrogen sensitivity. Together, these features suggest that the NRL-PRL model may provide insight into the origin of the diversity of clinical ERα tumors, and the characteristics of the tumor subpopulation that resists therapy.