STAT1-deficient mice spontaneously develop estrogen receptor α-positive luminal mammary carcinomas

Formalin-fixed paraffin-embedded tissue blocks were retrieved from the archive of the Department of Pathology in Spedali Civili di Brescia, Brescia, Italy. This retrospective study was conducted in compliance with the Declaration of Helsinki and with policies approved by the Ethics Board of Spedali Civili di Brescia. For this large-scale retrospective and exclusively observational study on archival materials, patient consent was not needed, as established by Italian regulations (Delibera del garante n. 52 del 24/7/2008 and DL 193/2003). The cohort consisted of 161 primary breast carcinomas selected from a series of routinely examined cases collected between 2006 and 2008, adjacent normal breast tissues from 11 patients with cancer, and normal breast tissues from five healthy individuals (tissues kindly provided by Monica Guaragni, Clinica S. Anna, Brescia, Italy). Breast carcinoma samples were characterized on the basis of histology, ERα (clone SP1; Thermo Fisher Scientific, Waltham, MA, USA) and PR (clone PgR 636; Dako, Glostrup, Denmark) expression, and HER2 amplification by immunohistochemistry (Herceptest; Dako) and fluorescence in situ hybridization (PathVision HER-2 DNA probe kit; Abbott Laboratories, Abbott Park, IL, USA) (summarized in Table 1). STAT1 protein expression was evaluated on four-micron (4-μm) tissue sections by using a rabbit polyclonal antibody against STAT1 (sc-346, 1:400; Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA). Antigen retrieval was performed by microwaving in citrate buffer (pH 6.0). Positive signal was revealed by the Super Sensitive polymer-horseradish peroxidase immunohistochemistry detection system in accordance with the instructions of the manufacturer (BioGenex, San Ramon, CA, USA). Mouse spleens obtained from WT and STAT1^-/- mice were used as positive and negative controls, respectively, to confirm the specificity of the STAT1 antibody (Figure 1A). STAT1 expression of the human normal breast tissues and breast tumor samples was assessed according to the percentage of STAT1⁺ tumor or stromal cells (percentage score: 1 = fewer than 5% of positive cells; 2 = 5% to 25% of positive cells; 3 = 25% to 75% of positive cells; and 4 = greater than 75% of positive cells) and to the intensity of the staining (intensity score: 1 = low; 2 = intermediate, and 3 = high) (Table 1 and Figure 1B), similar to the determination of ERα expression by using the Allred score [27]. The percentage score was then added to the intensity score to produce the final STAT1 score.

129S6/SvEvTac-Stat1 ^tm1Rds (referred to here as STAT1-deficient, or STAT1^-/-), 129S6/SvEvTac-Stat1 ^tm1Rds -Rag2 ^tm1Fwa (referred to here as STAT1^-/- × RAG2^-/-), and B6.129S-Stat1 ^tm1Dlv (referred to here as STAT1-null, or S1N) mice were generated and previously characterized in our laboratories [10, 11, 13]. RAG2^-/- mice were generated by Frederick Alt [28]. Wild-type (WT) 129S6/SvEv and STAT1^-/- (129S6/SvEv) mice were purchased from Taconic Farms (Hudson, NY, USA), and STAT1-null mice (mixed C57BL/6-129/SvEv and pure C57BL/6) were maintained and housed in our laboratories. Mice were scored as tumor-bearing when they reached 10 by 10 mm in size. All animal experiments were carried out in accordance with the guidelines of the American Association for Laboratory Animal Science as an approved protocol by the Animal Studies Committees in both the Washington University School of Medicine and New York University. These experiments were performed in specific pathogen-free facilities that were accredited by the Association for Assessment and Accreditation of Laboratory Animal Care and that were located at the Washington University School of Medicine and New York University School of Medicine.

STAT1^-/- mammary tumor cell lines, SSM1, SSM2, and SSM3, were originated from three individual STAT1^-/- tumor-bearing mice and were created by mechanical disaggregation of tissue before an overnight incubation with collagenase type IA (200 U/mL; Sigma-Aldrich, St. Louis, MO, USA) and insulin (10 μg/mL; Sigma-Aldrich) in DMEM/F12/10% FBS/1% L-glutamine/1% penicillin-streptomycin. Clusters of epithelial cells were enriched by sedimentation through five washes of 40 mL of DMEM/F12/5% FBS each. Stromal fibroblasts were eliminated by differential trypsinization by using 0.05% trypsin once a week over the next 2 months. The absence of fibroblasts was determined by immunofluoresence by using an antibody specific for vimentin (sc-7557; Santa Cruz Biotechnology, Inc.). The SSM cell lines are maintained in DMEM/F12/10% FBS/1% L-glutamine/1% penicillin-streptomycin/50 μM 2-mercaptoethanol/0.3 μM hydrocortisone/5 μg/mL insulin/10 ng/mL transferrin. MCF7 and NMuMG were purchased from American Type Culture Collection (Manassas, VA, USA) and are cultured in DMEM/10% FBS/1% L-glutamine/1% penicillin-streptomycin. The MCF7 cell line is a human ERα⁺ breast cancer cell line. NMuMG is an immortalized nontransformed murine mammary gland epithelial cell line.

Southern blot analysis was performed as described previously [29]. The probe was prepared by digesting a plasmid containing the MMTV long terminal repeat with BamHI (the MMTV-LTR plasmid was a generous gift from Elena Buetti of the Swiss Institute for Experimental Cancer Research in Switzerland) [30].

MCF7, SSM1, SSM2, or SSM3 was plated in 96-well plates in phenol red-free media with 10% charcoal-treated FBS (HyClone, Logan, UT, USA) supplemented with hydrocortisone, insulin, and transferrin and with or without 10 nM 17-β-estradiol (Sigma-Aldrich). Cell proliferation was determined by using 3-(4,5-dimethyl-2-thiazlyl)-2,5-diphenyl-2H-tetrasolium bromide (MTT) assays in accordance with the instructions of the manufacturer (Promega Corporation, Madison, WI, USA).

SSM1, SSM2, SSM3, and NMuMG cells were lysed in RIPA buffer (R0278; Sigma-Aldrich) with 2 mM sodium vanadate, protease inhibitor cocktail (P8340, 1:500; Sigma-Aldrich), and phosphatase inhibitor cocktail (P5726, 1:100; Sigma-Aldrich). Cleared lysate (200 μg) was resolved on SDS-PAGE and transferred onto nitrocellulose membranes. ERα expression was detected by incubating blots with the monoclonal antibody 6F11 against ERα (VP-E613, 1:500; Vector Laboratories, Burlingame, CA, USA). PR expression was examined with a PR antibody against the C terminus of both PR-A and PR-B (sc-538, 1:500; Santa Cruz Biotechnology, Inc.).

Female WT or STAT1^-/- mice were either sham-operated or ovariectomized at 6 to 8 weeks of age under general anesthesia. Two weeks after the surgery, 10⁵ SSM1, SSM2, or SSM3 mammary tumor cells in 10 μL were injected into the inguinal fat pads of the mice. Alternatively, tumor fragments of about 1 mm in size were transplanted into the inguinal fat pads. Tumor growth was monitored by palpation once every 3 to 6 days and measured at two perpendicular diameters. The average of the two perpendicular measurements was plotted. In experiments in which exogenous E2 was supplemented to ovariectomized mice, 60-day release E2 pellets at 0.5 mg per pellet were used (Innovative Research of America, Sarasota, FL, USA). In the endocrine treatment experiment, nu/nu mice or STAT1^-/- mice were transplanted with 10⁶ SSM3 tumor cells or tumor fragments, respectively. When the established tumors reached 5 mm in diameter, the animals were either sham-operated or ovariectomized. Tumor growth was monitored as described above.

Mammary glands were harvested from 10- to 14-month-old STAT1^-/- female mice and digested in collagenase and hyaluronidase solution, as described in [31]. Single-cell suspension was prepared after dissociated tissues were treated with trypsin/DNase for 1 minute and dispase/DNase for 2 minutes and passed through 40-μm cell strainers. Cells were first blocked with anti-CD16 and anti-CD32 Fcγ receptors and normal rat serum and then stained with anti-TER119-PE/Cy7 (BioLegend, San Diego, CA, USA), anti-CD31-PE/Cy7 (BioLegend), anti-CD45-PE/Cy7 (BioLegend), anti-CD24-APC (BioLegend), or anti-CD49f-biotin (BioLegend) for 20 minutes at 4°C. Streptavidin-APC/Cy7 (BioLegend) was added, and cells were incubated for 20 minutes at 4°C. Stained cells were collected by using an LSRII flow cytometer (BD Biosciences, San Jose, CA, USA). Dead cells were gated out by using DAPI (Invitrogen Corporation). Cells depleted of CD31, CD45, and TER119 were further analyzed on the basis of their CD49f and CD24 surface expression. Myoepithelial cells were defined as CD49f^hi CD24^int, whereas luminal epithelial cells were CD49f^int CD24^hi, as established previously [31‐33]. Flow cytometry profiles were analyzed by using FloJo software (TreeStar Inc., Ashland, OR, USA).

Retrovirus expressing GFP alone, STAT1.IRES.GFP, and STAT1 mutants Y701F.IRES.GFP and S727A.IRES.GFP was prepared by co-transfecting Phoenix cells with the retrovirus plasmid and pCMV.VSVg by using FuGENE HD (Roche, Basel, Switzerland). Supernatant was harvested 48 and 72 hours after transfection and overlayed on NMuMG, SSM1, SSM2, or SSM3 cells in the presence of 8 μg/mL of polybrene for 6 to 8 hours. Infection was carried out for 2 consecutive days. The infected cells were harvested on day 2 after infection for Western blotting to confirm expression and on day 3 for flow cytometry to quantitate the percentages of cells undergoing early apoptosis. Apoptosis was measured by a flow cytometry-based annexin V-binding assay in accordance with the instructions of the manufacturer (BD Biosciences). Only the early apoptotic cells (annexin V-positive, 7AAD-negative) were analyzed.

Total RNAs were isolated from normal mammary glands of primary STAT1^-/- mammary tumors by using Trizol in accordance with the procedure of the manufacturer (Invitrogen Corporation). RNA integrity was confirmed by using an Agilent bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA, USA). Labeled target cRNAs were synthesized and hybridized to Affymetrix GeneChip MOE 430 2.0 arrays in accordance with the instructions of the manufacturer. Raw data were modeled and normalized by using dChip [34]. Well-annotated mouse and human orthologs were identified by using the Mouse Genome Informatics database [35], and medians were used in cases of redundant probesets. The Herschkowitz data matrix contains 232 human breast cancer datasets and 122 murine datasets from 13 different mammary tumor models, which were analyzed by using 106 intrinsic genes common to the two species [36]. To combine our STAT1^-/- mammary tumor datasets with the Herschkowitz datasets, 96 of the 106 intrinsic genes were used because of platform differences. Gene-wise normalization was carried out separately with each dataset such that each gene has median zero and unit standard deviation. Distance-weighted discrimination (DWD) was used to merge our datasets with the Herschkowitz datasets to eliminate large systematic biases arising from different RNA purification procedures and distinct microarray platforms [37, 38]. For unsupervised cluster analysis, the average linkage hierarchical clustering algorithm was then applied to the merged datasets by using XCluster [39] with the centered correlation coefficient as the similarity/dissimilarity metric. The gene expression heatmap and dendrogram were generated in Java TreeView [40] to visualize the relationship between the human breast cancer subtypes and murine mammary tumor models according to the gene expression intensities of the 96 genes. Gene expression profiling data have been deposited in the National Center for Biotechnology Information's Gene Expression Omnibus (GEO) under accession number GEO:GSE31942 [41].

sigClust examines the significance of a given clustering by testing the null hypothesis that the datasets are from the same Gaussian distribution [42]. We applied sigClust to five STAT1^-/- primary tumor samples and 63 human luminal breast cancer samples by using the 96-gene intrinsic gene list. The resulting P value is 0.99998, which indicates that the STAT1^-/- mammary tumors are highly likely to be in the same cluster with the human luminal breast cancer datasets. In addition, we applied sigClust on the MMTV_Neu and MMTV_PyMT datasets from the Herschkowitz study [36]. To avoid potential bias due to different sample sizes, we drew a set of five samples out of these two sets of data through 200 iterations and implemented the test on the set of five samples and the human luminal breast cancer datasets. The average P values were 0.9110 (MMTV_Neu), 0.9630 (MMTV_PyMT), and 0.9665 (both mouse models combined). These results indicate that the STAT1^-/- mammary tumor model exhibits a higher degree of resemblance to human luminal breast cancers at the gene expression level than the MMTV-Neu and MMTV-PyMT models do.

To investigate the stability of clustering between STAT1^-/- mammary tumors and human luminal breast cancer, consensus clustering, which is a re-sampling-based technique that uses perturbation to simulate a set of new samples from the original merged dataset, was employed [43]. Consensus index on empirical clustering results across all perturbed datasets was then summarized by the normalized proportion of times that two samples were assigned to the same cluster. The underlying assumption is that the induced cluster composition is more trustworthy if the clustering is robust to sampling variability. We re-sampled 1,000 times and considered the number of clusters ranging from two to 15 for assessment. Using the 96-gene intrinsic gene list, we found that the five STAT1^-/- mammary tumors belong to the same cluster 95% of the time after 1,000 re-samplings with an interquartile range (IQR) of 0.9128 to 0.9775 (range = 0.8964 to 1.0), which indicates that the STAT1^-/- mammary tumors are molecularly homogeneous. The STAT1^-/- mammary tumors and human luminal breast cancers cluster together 62% out of 1,000 re-samplings with an IQR of 0.57 to 0.67 (range = 0.32 to 0.78). In contrast, the MMTV-Neu and MMTV-PyMT mouse models cluster with the human luminal breast cancer datasets only 42% out of 1,000 re-samplings with an IQR of 0.44 to 0.48 (range = 0.01 to0.64). These results further demonstrate that the molecular signature of the STAT1^-/- mammary tumors significantly overlaps with that of human luminal breast cancers.

Time to onset was analyzed by the Kaplan-Meier product limit method, which generated the Kaplan-Meier survival curves. P values were reported by log-rank test. All numerical results are presented as mean and standard error of mean and represent data from a minimum of three independent experiments unless otherwise stated. Tumor growth curves were analyzed by a distribution-free test [44]. The unpaired t test for two independent samples was used to determine the statistical significance between the experimental groups and control groups. Wilcoxon signed rank test was used to compare STAT1 intensity levels in tumor samples and adjacent normal breast tissues. Association between clinicopathological characteristics and ERα status was tested by chi-squared test or Fisher exact test, whichever was appropriate. All tests were two-sided, and a P value of not more than 0.05 was considered significant. GraphPad Prism (GraphPad Software, Inc., La Jolla, CA, USA), SAS 9.2 (SAS Institute Inc., Cary, NC, USA), and R 2.11.1 [45] were used for all statistical analyses.

To explore the role of STAT1 in breast tumor development, the relative cellular levels of STAT1 protein were immunohistochemically assessed in a cohort of 161 primary breast cancer samples (78 ERα^- and 83 ERα⁺ cases) (Table 1) by using a STAT1-specific polyclonal IgG antibody that recognizes a STAT1 epitope shared by the human and mouse proteins (Figure 1A). Normal breast tissues from 11 patients with cancer and five healthy individuals were used as nontransformed controls (Table 2). STAT1 expression in epithelial cells and infiltrating stromal cells was quantified on the basis of the percentage of STAT1⁺ cells (percentage score) and the intensity of the positive signal by using a three-tiered scale (intensity score) (Figure 1B and Materials and methods).

In normal human breast tissue from healthy individuals, STAT1 was detected in the cytoplasm of luminal epithelial cells, in the nuclei of the spindle cells in the surrounding stroma, and sporadically in myoepithelial cells (Figure 2A). In contrast, STAT1 expression was highly variable among tumor samples (Figures 1B, 2B, D, and 2F), consistent with previous reports [46, 47]. Specifically, in 11% of cases (17 out of 161), STAT1 expression was detectable in less than 5% of the neoplastic cells, whereas in 37% of cases (59 out of 161), more than 75% of the tumor cells expressed STAT1 (percentage score in Table 1). The remaining 52% of cases displayed an intermediate phenotype in which STAT1 staining was observed in 5% to 75% of the neoplastic cells. In addition to documenting the percentage of STAT1⁺ cells, we recorded the relative intensity of the positive signal (intensity score). Thirty-four percent of the tumor cases (54 out of 161) exhibited low STAT1 staining in the neoplastic cells, whereas 37% and 29% showed intermediate and high staining, respectively (Table 1). Although the percentages of STAT1⁺ neoplastic cells in ERα^- and ERα⁺ tumors were comparable (percentage score in Table 1), the intensity of the staining was significantly lower in ERα⁺ breast cancers than in ERα^- breast cancers (intensity score in Table 1 and summarized in Figure 2G). Specifically, 45% of the ERα⁺ tumors exhibited low levels of STAT1 staining in neoplastic cells, in contrast to 22% of the ERα^- cases, demonstrating that reduced STAT1 expression is associated with ERα⁺ tumors (Table 1 and Figure 2G). A similar trend was observed in the stroma between ERα^- and ERα⁺ cases (Figure 2G). However, since STAT1 was universally elevated in stromal cells compared with neoplastic cells within the same tumors (Figure 2D, F, and quantified in Figure 2G), the difference in stromal STAT1 intensity between ERα^- and ERα⁺ cases is unlikely to be functionally meaningful. These findings thus indicate that STAT1 expression is differentially regulated in epithelial versus stromal compartments.

In contrast to the observation that lower STAT1 expression levels were associated with ERα⁺ breast cancers versus ERα^- cancers, there was no significant association between low STAT1 expression and HER2 status (Table 3). HER2^- and HER2⁺ breast cancers displayed similar percentages of STAT1⁺ cells and levels of STAT1 intensity in the stromal and neoplastic compartments (Table 3). When these data are taken together, the lowest level of STAT1 expression is preferentially associated with the ERα⁺ breast cancer subtype.

Since STAT1 is uniformly expressed in the luminal epithelial cells of breast tissues from normal healthy individuals, the observation that a large subset of breast cancer cases exhibit low levels of STAT1 expression suggests that STAT1 may be downregulated specifically in the neoplastic cells. To further investigate this possibility, STAT1 expression was also assessed in morphologically normal breast tissues of patients with breast cancer. In all of the 11 STAT1-low breast cancer cases in which adjacent normal breast tissues were available, normal breast tissues had significantly more STAT1⁺ epithelial cells and exhibited stronger staining intensity than matched tumor tissues (Figure 2C-F and Table 2). This resulted in an overall higher STAT1 score in normal epithelial cells than in breast tumor cells (Figure 2H). These findings thus demonstrate that STAT1 expression is dramatically diminished or lost in a significant proportion of breast cancer cells during tumor progression.

The findings that diminished STAT1 expression is associated with breast cancer progression prompted us to investigate whether loss of STAT1 was a cause, not merely a consequence, of mammary tumorigenesis. We monitored WT and STAT1^-/- 129S6/SvEv-strain female mice for tumor development and found that 65% (15 out of 23) of the STAT1^-/- mice developed spontaneous mammary adenocarcinomas (median tumor onset of 23 months) but that none of the WT mice developed the disease (Figure 3A). In agreement with our previous report [13], we observed mammary tumor development in a larger cohort of STAT1^-/- × RAG2^-/- female mice in similar disease incidences (Figure 3A). STAT1^-/- mice developed mammary tumors only, whereas STAT1^-/- × RAG2^-/- mice developed mammary and intestinal tumors. Mice lacking RAG2 alone did not develop mammary tumors (Figure 3A), demonstrating that this phenotype is specifically associated with STAT1 deficiency. Cells from our STAT1^-/- mice express low levels of a nonfunctional truncated STAT1 protein and are incapable of responding to IFN either in vitro or in vivo [10, 48]. Nevertheless, by examining STAT1-null (S1N) mice generated by using a different targeting strategy and different embryonic stem cells that were maintained on a mixed C57BL/6-129/SvEv background, we ruled out the possibility that this truncated protein is involved in disease development [11]. Female S1N mice also developed spontaneous mammary tumors with a median tumor onset of 14.5 months (Figure 3A). Although the S1N mice exhibited a shorter latency than the 129S6/SvEv STAT1^-/- mice, the penetrance of the two cohorts was indistinguishable and the difference in the overall tumor incidences was not statistically significant (P = 0.26). These results demonstrate the generality of the phenotype to mice lacking functional STAT1, regardless of targeting strategy or mouse strain. STAT1^-/- mice that were sterilely re-derived and housed exclusively in a commercial gnotobiotic facility also developed mammary tumors, a result suggesting that the disease was not of infectious origin. This conclusion was substantiated by viral microarrays [49] that failed to detect either known or novel viruses in mammary tumors or other tissues of STAT1^-/- mice and by the lack of evidence for translocation and additional chromosomal integration of endogenous mammary tumor proviruses in these tumors (Figure S1 in Additional file 1).

Since tumor development did not reach a complete penetrance in the STAT1^-/- and S1N mice, we examined whether parity might influence tumorigenesis. Strikingly, multiparous STAT1^-/- mice developed mammary tumors sooner and at a higher frequency than nulliparous STAT1^-/- mice (Figure 3B). Specifically, multiparous mice had a median tumor onset of 14.8 months versus 24.6 months of nulliparous mice. Most importantly, the incidence of tumor development was 91% (10 out of 11) in mice that had undergone multiple rounds of pregnancy and lactation compared with 62% of nulliparous mice. Taken together, these data suggest that pregnancy-associated hormones might accelerate mammary tumor formation in STAT1^-/- mice.

Examination of whole mounts of mammary gland tissues from 12- to 24-month-old female STAT1^-/- mice without palpable masses revealed focal atypias in about 50% of the cohort (n = 19) (Figure 4A, B). This abnormality was not observed in mammary glands of age-matched WT female mice (n = 12). The early lesions in STAT1^-/- mammary glands varied from distended ducts to small cystically dilated clusters of alveoli (Figure 4B). Analysis of the H&E-stained tissue sections revealed that these abnormal foci contained atypical cells fulfilling the criteria for mammary intraepithelial neoplasia (MIN) (Figure 4C, D) [50]. Surprisingly, atypical nuclei in MIN lesions stained positively for ERα and PR (Figure 4E-H). Similar analyses of overt tumors revealed the presence of solid nests of neoplastic ERα⁺ or PR⁺ cells with frequent central necroses (Figure 4I-N). Over 90% of all tumor cells expressed ERα (Figure 4K, L). Some tumors had invasive nests of neoplastic cells with areas of fibrosis and inflammation (Figure 4I, J). All primary STAT1^-/- mammary tumors examined, regardless of parity, showed similar histopathological characteristics and patterns of hormone receptor expression. None of the primary tumors stained positively for HER2 (data not shown). Thus, mammary tumors in STAT1^-/- mice progress developmentally from intraepithelial neoplasia to carcinoma in a manner that is remarkably similar to the progression from ductal carcinoma in situ to invasive ductal carcinoma seen in the human disease [51]. Since spontaneous ERα⁺/PR⁺ mammary tumors are rarely observed in mice, our findings also suggested that STAT1^-/- mice may represent a relatively novel model for human ERα⁺/PR⁺ luminal breast cancer.

To facilitate biological and biochemical characterization of the STAT1^-/- mammary tumors, three tumor cell lines were established and designated spontaneous STAT1^-/- mammary (SSM) epithelial tumor cell lines (SSM1, SSM2, and SSM3). All expressed CK but not vimentin in vitro, documenting their epithelial origin (Figure S2 in Additional file 2). SSM2 and SSM3 expressed nuclear ERα, which is similar to the human ERα⁺ breast cancer cell line, MCF7 (Figure 5A). SSM2 and SSM3 also expressed the two PR isoforms, PR-A and PR-B (Figure 5C), suggesting that ERα is functional in these cells. In contrast, SSM1 expressed a very low level of ERα (Figure 5B) and no detectable PR-A/PR-B expression (Figure 5C), indicating a lack of ERα signaling in the SSM1 cells. Whereas SSM2 and SSM3 required the presence of estrogen to proliferate in vitro, SSM1 did not (Figure 6A). Finally, when transplanted into the fat pads of WT or STAT1^-/- mice, SSM2 and SSM3 grew only in recipients with intact ovaries or in ovariectomized recipients that received subcutaneous estrogen pellet implants (Figure 6B). In contrast, SSM1 did not require ovarian hormones for in vivo growth (Figure 6B), revealing that not all STAT1^-/- mammary tumor cells display hormone dependency. This result corresponds with the finding that most, but not all, tumor cells in the primary STAT1^-/- mammary carcinomas display ERα and PR positivity (Figure 4). However, these hormone-independent cells are extremely rare in the primary STAT1^-/- mammary tumors as demonstrated by the fact that tumor fragments from two independent primary tumors failed to engraft in ovariectomized mice (Figure 6C). Together, these results demonstrate that SSM2 and SSM3 are ovarian hormone-responsive and -dependent in vitro and in vivo and that SSM1 arises from a very rare subset of ERα^- cells and is used in this study as a hormone-nonresponsive cell line.

Having demonstrated that the STAT1^-/- ERα⁺ mammary tumor cells required ovarian hormones for establishment of tumor growth, we next examined whether these tumors also depended on ovarian hormones to maintain tumor progression. We used ovariectomy as the treatment modality since ovarian ablation is commonly used to treat premenopausal patients with ERα⁺/PR⁺ breast cancers [52]. Established SSM3 tumors continued to grow in sham-treated mice (Figure 6D). However, established tumors failed to progress after ovariectomy, indicating that the STAT1^-/- ERα⁺/PR⁺ tumor cells require ovarian hormones not only for in vivo engraftment but also for maintenance of growth. Tumors that were established from transplanted primary tumor fragments were also highly sensitive to ovarian hormone deprivation and became unpalpable upon ovariectomy (Figure 6E). Thus, the STAT1^-/- mammary tumors are functionally similar to human ERα⁺/PR⁺ breast cancers.

Hormone receptor expression is one of the critical parameters used to determine the suitable treatments for human patients with breast cancer [2]. In addition to hormone receptors, mammary tumors can be classified by biomarkers that are expressed on the tumor cell surface. Myoepithelial and luminal epithelial cells can be identified on the basis of the well-established differential expression of murine mammary epithelial cell population markers, CD49f and CD24 (that is, myoepithelial cells = CD49f^hi CD24^int and luminal epithelial cells = CD49f^int CD24^hi) [31‐33]. These two epithelial cell populations could be readily differentiated in nontransformed mammary glands of STAT1^-/- mice ('myo' and 'lum' in Figure 7B) by flow cytometry upon elimination of dead cells and lineage-positive cells (Figure 7A). Marked expansion of the luminal epithelial cell subset was evident in MIN, the earliest stage of neoplastic changes identifiable by histology, in comparison with nontransformed mammary glands (Figure 7B). Further expansion of these luminal epithelial cells continued as MIN progressed to carcinomas (Figure 7B). Neoplastic cells in carcinomas homogeneously displayed a CD49f^int CD24^hi phenotype. Consistent with this result, primary STAT1^-/- mammary tumors are strongly positive for the luminal epithelial markers, CK19 and CK8/18 (Figure 7C). Occasional CK5⁺, CK14⁺, or p63⁺ myoepithelial cells were observed (Figure 7C and data not shown). Together, these results indicate that the STAT1^-/- mammary tumor cells exhibit a marker phenotype that is characteristic of luminal epithelial cells.

Perou and colleagues [3, 53] reported previously that human breast cancers could be differentiated into five molecular subtypes (luminal A, luminal B, HER2, basal, and normal-like) on the basis of their gene expression profiles. Strikingly, when gene expression patterns of mammary tumors from 13 different pre-existing mouse breast cancer models were compared with those of 232 human breast cancers, none showed significant overlap with human ERα⁺ luminal breast cancers [36]. To extend these findings, we compared the gene expression patterns of the aforementioned datasets (primary data generously provided by Dr. Charles M Perou, University of North Carolina at Chapel Hill) with those of five primary STAT1^-/- ERα⁺ mammary tumors. Hierarchical clustering analysis revealed that the five primary STAT1^-/- ERα⁺ mammary tumors not only resembled one another but also closely resembled human luminal breast cancers (Figure 8A). STAT1^-/- ERα⁺ mammary tumors also showed elevated transcription of genes characteristic of luminal breast cancers (for example, keratin 8, keratin 18, XBP1, GATA3, MYB, AREG, and FOXA1) (Figure 8B). In agreement with the previous report [36], samples from other mouse mammary tumor models consistently clustered away from human luminal breast cancers (Figure 8A). Finally, analyses using two other statistical methods (sigClust [42] and consensus clustering [43]) also supported the conclusion that STAT1^-/- ERα⁺ mammary tumors grouped more consistently and reproducibly with human luminal breast cancers than any other mouse mammary tumor model (see Materials and methods). Specifically, results from the consensus clustering analysis indicated that the STAT1^-/- mammary tumors and human luminal breast cancers clustered 62% of the time upon 1,000 re-samplings but that the MMTV-Neu and MMTV-PyMT clustered with human luminal breast cancers only 42% of the time upon 1,000 re-samplings. Together, the STAT1^-/- ERα⁺ mammary tumors display high molecular homogeneity and a striking similarity to human luminal breast cancers.

The above observations led us to conclude that STAT1 suppresses tumor development in the mammary gland. It is possible that STAT1 confers this role by mediating the elimination phase of immunoediting. However, since the immunodeficient RAG2^-/- mice never developed mammary tumors and the loss of STAT1 expression was observed only in the neoplastic cells of human breast cancers, we hypothesized that STAT1 might act as a cell-intrinsic tumor suppressor in mammary epithelium. We first investigated whether enforced expression of STAT1 in STAT1^-/- mammary tumor cells affected their tumorigenic phenotype. This approach has been used in the past to define several other tumor suppressors [54]. Retroviral transduction of WT STAT1 into SSM1, SSM2, and SSM3 resulted in expression of STAT1 protein levels in each cell line comparable to that in unmanipulated NMuMG (Figure 9A). Enforced expression of WT STAT1 led to apoptosis of a substantial percentage of SSM2 and SSM3 but not of SSM1 or NMuMG (Figure 9B). Therefore, the tumor suppression function of STAT1 is cell-autonomous.

Since phosphorylation at the tyrosine residue at position 701 synergizes with that at the serine residue at position 727 of STAT1 to effect the maximal transcriptional activity [53, 55], two STAT1 mutants, one lacking the Tyr residue (Y701F) and one lacking the Ser residue (S727A), were tested for their ability to induce tumor cell death. Even when expressed at levels comparable to those achieved with WT STAT1, the Y701F mutant was unable to induce apoptosis when expressed in SSM2 and SSM3 (Figure 9C). In contrast, the S727A mutant displayed an indistinguishable phenotype as the WT STAT1. Thus, the tumor suppressor action of STAT1 requires tyrosine phosphorylation, but not serine phosphorylation, of STAT1.