Pathobiology of the 129:Stat1 ^−/− mouse model of human age-related ER-positive breast cancer with an immune infiltrate-excluded phenotype

129:Stat1 ^−/− mice [7, 23] were provided by the Schreiber laboratory (Washington University, St. Louis, MO, USA), and 129S6/SvEvTac mice were purchased from Taconic Farms (Hudson, NY, USA). Mice were housed in a vivarium according to National Institutes of Health guidelines, and all animal experiments were performed following procedures approved by the UC Davis Institutional Animal Care and Use Committee. The procedures for transplantation [7] and the conditions for maintaining mice [24] were described previously.

Most tissues were fixed in 10% neutral buffered formalin (NBF) at room temperature for 24 h, then placed in 70% ethanol until processing, which was normally within 24 h. To validate the immunohistochemical staining with NBF-fixed samples, some tissues were fixed in zinc-based fixative as described elsewhere [24]. Procedures for infiltrating tissue with paraffin, sectioning, and hematoxylin and eosin (H&E) staining were described previously [24]. The preparation of whole mounts of mouse mammary glands was carried out as previously described [7].

Fragments (1 mm³) of freshly dissected tumors were transplanted into uncleared thoracic and inguinal mammary fat pads of 3- to 6-week old nulliparous WT and Stat1-null females under appropriate anesthesia and sterile conditions as previously described [7]. The animals were monitored at least twice per week. Growth of palpable tumors was noted and measured in two dimensions using calipers. The volumes were calculated using the average diameter [4/3 × π × (d1 + d1/4)³].

Immunohistochemistry (IHC) was performed as previously described [7, 24]. Stained slides were scanned on an Aperio AT2 ScanScope (Leica Biosystems, Buffalo Grove, IL, USA), and digital images were viewed using the ImageScope application (Leica Biosystems). Digital images were captured and processed using Photoshop software (Adobe Systems, San Jose, CA, USA). The imaging analysis for counting marker-positive cells was performed with inForm Cell Analysis software (PerkinElmer, Waltham, MA, USA). Multiplex IHC was performed with tyramide signal amplification (TSA)-based fluorescence color visualization [25]. TSA-based multiplex IHC was performed according to the manufacturer’s protocol (PerkinElmer). FoxA1, epithelial cell adhesion molecule (EpCAM), and keratin-14 (KRT14) were detected with TSA Plus fluorescein, cyanine 3 (Cy3), and Cy5, respectively. Visualization of multiplexed images was performed with an LSM710 laser scanning confocal microscope (Carl Zeiss Microscopy, Oberkochen, Germany). Antibodies used were anti-ER-α (clone MC-20; Santa Cruz Biotechnology, Dallas, TX, USA), anti-PR (polyclonal; Dako North America, Carpinteria, CA, USA), anti-Ki67 (clone Ab-4, Lab Vision; Thermo Fisher Scientific, Fremont, CA, USA), anti-CD3 (clone SP7; Abcam, Cambridge, MA, USA), anti-CD4 (clone RM4-5; eBioscience, San Diego, CA, USA), anti-CD8a (clone 4SM15; eBioscience), anti-F4/80 (clone MCA497, AbD Serotec; Bio-Rad Laboratories, Hercules, CA, USA), anti-B220 (clone RA3-6B2; BD Biosciences, San Jose, CA, USA), anti-KRT14 (Covance Poly19053; BioLegend, San Diego, CA, USA), anti-FoxA1 (clone EPR10881; Abcam), anti-EpCAM (EPR677; Abcam), anti-KRT8/18 (CK209; Fitzgerald Industries International, Acton, MA, USA), and anti-type IV collagen (polyclonal; Abcam).

A Stat1-null cell line (SSM2) [2], an MMTV-NDL-derived cell line (NDL) [26], and an MMTV-PyMT-derived cell line (Met1) [22] were cultured in T75 culture flasks at 1 × 10⁶ cells in growth medium (DMEM supplemented with 10% FBS) for 2 days. A phenotypically normal MEC line, EpH4, was cultured as described previously [27]. To prepare conditioned medium (CM) from each cell line, cells were rinsed three times with serum-free medium and then cultured in 10 ml of serum-free medium for 48 h. CM was collected, centrifuged to eliminate cells and debris, and filtered (0.2 μm). To isolate extracellular vesicles (EVs), CM was ultracentrifuged at 100,000 × g for 2 h. The pellet (EV-rich fraction) was resuspended in DMEM with the same volume as the supernatant (soluble fraction).

The green fluorescent protein (GFP)-tagged H2B [28] was transduced into RAW264.7 mouse macrophages (American Type Culture Collection, Manassas, VA, USA) using pLenti-EF1a-Puro bearing a GFP-tagged H2B complementary DNA [27]. Preparation of lentiviral particles and transduction of target cells were performed as previously described [27]. To maintain GFP-H2B-positive cells, growth medium was supplemented with 0.5 μg/ml puromycin. To observe RAW264.7/GFP-H2B cell migration, cells were plated at 2 × 10⁴ cells/well of an eight-well Lab-Tek II chambered coverglass (Thermo Fisher Scientific, Waltham, MA, USA) in growth medium for 2 days before stimulating cell migration with CM. Fluorescence was visualized with an LSM710 confocal microscope equipped with a temperature- and CO₂-controlled chamber [29, 30]. Before cell migration was analyzed, cells were rinsed twice and maintained in 400 μl of serum-free DMEM for 2 h. The action of RAW264.7/GFP-H2B cells was monitored at 5-minute intervals for more than 8 h. Cell migration was evaluated using time-lapse images with Imaris software (Bitplane, South Windsor, CT, USA).

RAW264.7 cells were resuspended in DMEM at a density of 1 × 10⁶ cells/ml, and 100 μl of resuspended cells were placed into the upper chamber of Transwell culture inserts (8-μm pore size) in 24-well plates (Corning, Corning, NY, USA). Quantities of 600 μl of DMEM or CM from each cell line were applied in the bottom chamber for 5 h to test the chemoattractant activity. Cells on the underside of the insert were fixed with 70% ethanol for 10 minutes and then stained with 0.2% crystal violet before rinsing to remove background staining and air-drying, followed by microscopic imaging.

All statistical analyses were done using Prism 7 software (GraphPad Software, La Jolla, CA, USA). Kaplan-Meier plots were generated to compare the tumorigenesis of nulliparous and multiparous 129:Stat1 ^−/− animals (Fig. 1). The difference was detected with the log-rank test. Similarly, Kaplan-Meier plots were created to compare the latency of palpable tumor onset between tumors transplanted to 129:Stat1 ^−/− or 129:WT hosts. The difference was detected with the log-rank test. The asymmetries of primary tumor occurrence between left-right and caudal-cephalad mammary fat pads (Additional file 1: Table S1) were tested with a binomial test.

A total of 24 palpable mammary tumors developed in 20 female 129:Stat1 ^−/− mice. Tumors were detected in two tumor-bearing parous females before 1 year of age (at 32 and 47 weeks). The other 18 tumor-bearing mice developed tumors between the ages of 1 year (52 weeks) and 2.3 years (120 weeks). For comparison, 30 female 129:Stat1 ^−/− mice without palpable tumors were killed at various times beyond 52 weeks of age (Additional file 1: Table S2). The differences in time to tumor onset between parous and nulliparous females in our colony were significant (p = 0.038 by log-rank test), consistent with the initial report about the Washington University in St. Louis colony [2] that had an equally prolonged latency and incomplete (~53%) penetrance. Although the tumor incidence in our cohort remained consistent after 1 year of age (35%), subsequent pathological analysis revealed that surviving nulliparous females had mammary intraepithelial neoplasia (MIN) lesions, suggesting that the penetrance of mammary tumors could approach 100% were animals to age even further.

The 13 tumor-bearing nulliparous females ranged in age at tumor onset from 73 to 120 weeks, with 50% incidence at 91 weeks. The comparable 24 tumor-free nulliparous WT females covered a similar age range (53–120 weeks) and a similar average age when killed (83 weeks). Similar to a previous report [2], pregnancy shortened tumor latency from T ₅₀ = 91 weeks to T ₅₀ = 78 weeks. A comparable cohort of six tumor-free parous KO mice was examined (mean age 83 weeks, range 52–95 weeks). The fact that not all aged 129:Stat1 ^−/− animals were tumor-bearing within 120 weeks also suggests a stochastic component to tumor initiation in addition to the GMM intrinsic susceptibility (Fig. 1).

When these data are compared with the recorded or calculated T ₅₀ from other GMM models that are KOs of tumor suppressor genes or employ a mammary-specific promoter driving the targeted expression of an oncogene, the 129:Stat1 ^−/− model stands out as being unique (Fig. 1). The T ₅₀ for 129:Stat1 ^−/− is over three times the mean T ₅₀ (28 weeks) of these other GMMs and 40 weeks longer than the closest model, MMTV-cMyc [31]. When compared with other GMMs using their respective Kaplan-Meier plots, the 129:Stat1 ^−/− model plot does not overlap with any of those for commonly reported models (Fig. 1).

The palpable mammary tumors showed a notable cephalad (thoracic) to caudal (inguinal) dominance (Additional file 1: Table S1). Eighty-seven percent (21 of 24) of tumors were detected in the thoracic mammary fat pads (p = 0.0003 by binomial test) (Additional file 1: Table S1). These distributions are consistent with the topographical asymmetries previously described in other models and suggest that the local microenvironment influences tumor development [32]. In addition, the majority of palpable tumors were proximate to the nipple rather than at the periphery of the fat pad. These masses around or adjacent to the nipple can also explain the dilation and congestion of some more distal ducts (Additional file 2: Figure S1).

Necropsy of the three subcohorts (129:WT, tumor-free 129:Stat1 ^−/− , and tumor-bearing 129:Stat1 ^−/− ) was performed with attention to microscopic mammary and nonmammary pathologies that may correlate with the tumor-bearing phenotype (Figs. 2 and 3). The older animals in all three cohorts shared sporadic conditions associated with aging, such as bronchioloalveolar adenoma, eosinophilic (crystalline) pneumonitis, atrophic cystic hyperplasia of the endometrium, segmental interstitial nephritis, ovaries with luteinized stroma and fewer Graafian follicles, and low-grade small cell lymphoma [33]. The adrenal glands, thyroid glands, pancreas, and pituitary glands were all disease-free. Significantly, the vaginal mucosa in all three cohorts had evidence of ovarian function at older ages. For example, at 96 weeks of age, the vaginal mucosa of both KO and WT displayed cornification (estrus), inflammation (metestrus), or mucinous differentiation (proestrus). The vaginal mucosa is a standard histological indication of stage of the estrous cycle (see Fig. 3). In contrast, the mammary glands and ovaries had features unique to each cohort, as described below (Figs. 2 and 3).

The early, preinvasive MIN lesions in 129:Stat1 ^−/− females have been mentioned previously without morphologic descriptions [2, 17]. These lesions are atypical foci that stand out from the surrounding normal mammary epithelium but remain confined within the basement membrane (Figs. 2 and 4), thus fitting the morphological criteria for MIN [4, 37]. These MIN could be found contiguous or adjacent to invasive tumors as well as in nontumorous glands and ducts of 129:Stat1 ^−/− tumor-bearing females, as well as in some tumor-free females older than 52 weeks. These MIN never occurred in 129:WT females, including two 97-week-old females, or in disease-free 129:Stat1 ^−/− females younger than 1 year of age (n = 42).

The Stat1-null MIN are small, nonpalpable, and not easily identified by gross inspection. Obvious lesions can be identified at the subgross level in mammary gland whole mounts as discrete, hypercellular foci that stand out from the background, typically as solid ducts, micronodular clusters, or small cysts (Fig. 4, Additional file 2: Figure S2e–g). The smaller MIN were not associated with a significant host response. The larger, presumably more advanced lesions were surrounded by a significant host inflammatory response that had a rich mast cell component but were composed primarily of T cells, macrophages, and increased vascularity (Fig. 4d, Additional file 2: Figure S2h). The distribution of MIN in affected glands was restricted to one or several ducts of the mammary network, whereas adjacent ducts had normal histology and cytology (Fig. 5a, b).

Microscopically, the MIN were characterized by clusters of unique atypical cells that stood out from the adjacent MEC. These dysplastic cells had large, oval, moderately pleomorphic nuclei with an open chromatin pattern and prominent nucleoli with abundant pale cytoplasm (LOP cells) (Fig. 6a, Additional file 2: Figure S3). These LOP cells are large relative to adjacent luminal and basal MECs (Fig. 6, Additional file 2: Figure S3). In addition, LOP cells in MIN and associated malignancies were KRT8/18⁺, with some also being KRT14⁺ but negative for smooth muscle actin (Additional file 2: Figure S4). The LOP cells were further characterized using multiplexed immunofluorescence, which revealed nuclear staining for FoxA1 in neoplastic EpCAM⁺ luminal cells and in sparse populations of KRT14⁺ basal-like cells, suggesting that FoxA1⁺ cells within MIN might be pluripotent (Fig. 6g, g′). The cells in Stat1-null invasive carcinomas (tumors) shared these immunophenotypic characteristics.

Neoplasms in Stat1-null mice contained LOP cells within MIN (Figure 6b, c; Additional file 2: Figure S5). These LOP cells were also interspersed among normal ductal MECs along relatively normal ducts (Fig. 6a, b, and d–f; Additional file 2: Figure S5). In general, the LOP cells appear to be more basally oriented but remained inside the myoepithelial layer (suprabasal). In some presumably more advanced cases, small collections of LOP cells appeared as multicellular nodules along the ducts (Figs. 5 and 6). In other instances, these LOP cell nodules formed outgrowths attached to but bulging out from the main duct (Fig. 5, Additional file 2: Figure S1). These nodules resembled abortive ductal side buds in normal mammary glands.

The distribution of LOP cells and MIN was restricted to one or several ducts of the mammary network in affected glands (Fig. 5), whereas the other adjacent ducts had an entirely normal histology and associated cytology (Fig. 5). The LOP cells in the “diseased” duct were FoxA1⁺, ER⁺, and PR⁺ (Fig. 6, Additional file 2: Figure S6). Further, laser capture-microdissected samples from the diseased duct illustrated in Fig. 5 were extracted, and exon 10 of the PRLR was sequenced. The specific duct (right inguinal mammary gland) contained three separate nontruncating single-nucleotide variants. In contrast, the adjacent ducts and lymph node had a normal WT PRLR sequence, whereas the palpable mammary tumor in the same animal’s left thoracic mammary gland was heterozygous for a truncating PRLR mutation similar to those previously described [17].

The microscopic characteristics of the 129:Stat1 ^−/− tumors have been described previously and display consistently similar morphologic phenotypes from tumors in one mouse to tumors in the next (low intertumoral heterogeneity) [2]. In contrast, spontaneous tumors from Trp53- and Brca1-KO mice, as well as from pregnancy-induced Balb/c:Stat1 ^−/− tumors, have a mixture of tumor types, with basal-like and “Wnt pathway” phenotypes being predominant [5, 21, 38]. Balb/c:Stat1 ^−/− mice crossed with mice expressing other oncogenes develop tumors with phenotypes identical to those of the oncogenic transgene [18‐20]. The signature 129:Stat1 ^−/− tumor phenotype consists of nodular nests of cells with large oval nuclei and abundant pale cytoplasm. These cells are cytologically identical to the LOP cells found in MIN. Eighty percent to 95% of these cells in MIN and tumors were immunopositive for ER, PR, and FoxA1 (Fig. 4, Additional file 2: Figures S5 and S6).

The Stat1 ^−/− tumors exhibited a pronounced local inflammatory response at the interface of tumor cells and surrounding stroma. In contrast to mammary tumors from other GMM models [5], each 129:Stat1 ^−/− tumor was surrounded by a thick mantle of host inflammatory cells and fibrosis (Fig. 7, Additional file 2: Figure S9). The inflammation consisted primarily of a nearly equal mixture of F4/80⁺ macrophages and CD3⁺ T lymphocytes in tumor-adjacent stroma. Quantitation of IHC marker-positive cell density in tumor and tumor-associating stroma in spontaneous Stat1-null tumors (Additional file 2: Figure S9) indicated tumor-infiltrated F4/80⁺ cells were at a significantly (p < 0.0001) lower density (6 cells/mm², SE = 2) than the cellular density in the stroma (2580 cells/mm², SE = 473). CD3⁺ cells were also significantly (p < 0.0001) less densely populated in tumors (262 cells/mm², SE = 59) compared with the stroma (3590 cells/mm², SE = 439). Thus, significantly fewer immune cells penetrated the tumor, with more intratumoral infiltration of T lymphocytes than macrophages. In addition, the density of B-cell lymphocytes was significantly (p < 0.0001) lower in tumors (34 cells/mm², SE = 6) than in the stroma (381 cells/mm², SE = 64), and the total number of B cells was significantly lower than both macrophages (p = 0.0002) and T lymphocytes (p = 0.0001).

The microenvironmental effect was further characterized in the grafts where primary tumor tissue was transplanted into mammary glands in WT host or KO host animals. The IHC results showed a predominance of F4/80⁺ macrophages and CD3⁺ T-lymphocytes in the tumor-associated stroma (Fig. 7c–f). The mean density of F4/80⁺ cells in the tumor interstices from a WT host was 19 cells/mm² (SE = 3). By comparison, there was a significantly higher density of F4/80⁺ cells in the surrounding WT stroma (2472 cells/mm², SE = 243). Similarly, transplants of the same primary tumor into the age-matched KO host also showed a predominance of macrophages in the stroma, with only 5 cells/mm² (SE = 2) in the tumor interstices and 2102 cells/mm² (SE = 233) in the stroma. The density of tumor-infiltrated macrophages was significantly higher in the tumors engrafted into a WT host than either the KO host or the spontaneous Stat1-null tumors (Additional file 2: Figure S9b). Similarly, CD3⁺ T lymphocytes were found in greater density in the stroma of WT hosts (1990 cells/mm², SE = 218) than in the tumors (463 cells/mm², SE = 54) (p < 0.0001). CD3⁺ cell density in KO hosts was significantly lower in both tumors (353 cells/mm², SE = 65) and stroma (1530 cells/mm², SE = 194) than in the WT hosts, but the CD3⁺ cell density was still higher in stroma than in tumors (Additional file 2: Figure S9d). The ratios of tumor-infiltrated immune cells to total cell count in WT, KO host, and spontaneous Stat1-null tumors were 8%, 4.5%, and 2.2%, respectively, for CD3⁺ T lymphocytes and 0.77%, 0.24%, and 0.23% for F4/80 macrophages, respectively. In contrast, the cell density of B lymphocytes in stroma was lower in transplants than spontaneous 129:Stat1 ^−/− tumors. The density of B lymphocytes was higher in the stroma of WT hosts than the tumors, but there was no significant difference between the tumor and stroma in KO hosts (Additional file 2: Figure S9f). Thus, as with the primary tumors, the host immune cells, especially macrophages, rarely invaded the interstices of solid tumors. This pattern of immune cell reaction to tumors has been described as the “excluded infiltrate” phenotype, in contrast to tumors with a sparse or absent immune reaction, as well as those with an intratumoral “inflamed” phenotype [39].

Given the abundance of macrophages in the tumor-associated stroma (Fig. 7, Additional file 2: Figure S9), we reasoned that Stat1-null tumors might secrete macrophage chemoattractants. To test this possibility, migration assays with the RAW264.7 macrophage cell line were performed using CM from (1) the Stat1 ^−/− tumor cell line SSM2, (2) the Stat1 ^+/+ normal MECs (EpH4, EpH3) and the polyoma middle T-induced mammary tumor cell line Met1, and (4) the Neu-induced mammary tumor cell line NDL. Migration of GFP-H2B-transduced macrophages on 2D plastic was monitored using time-lapse imaging to record track length and time of the mobile cell nuclei.

Macrophages exposed to CM from the Stat1-null tumor cell line SSM2, compared with CM from the other three cell lines, had shorter track lengths for each time-lapse interval and moved more slowly (Fig. 8a–f). Transwell migration assays also confirmed there was less migration by RAW264.7 cells exposed to SSM2-CM than by the other three cell lines (Fig. 8g). These experiments suggest that Stat1-null cells either secrete less chemoattractant or secrete factors that inhibit macrophage migration.

To distinguish between these and other possibilities, SSM2-CM was mixed with the markedly stronger chemoattractive Met1-CM, and we compared this mixture with DMEM/Met1 using the cell-tracking technique (Fig. 8h). Whereas the mixture of Met1-CM and DMEM (50%/50%) resulted in significant movement of RAW264.7 cells, the mixture of Met1-CM and SSM2-CM (50%/50%) reduced the track length of Met1-CM to the level of DMEM alone (Fig. 8g). This supports the notion that SSM2-CM inhibits macrophage migration. This finding is also consistent with the observation of macrophage exclusion from the Stat1-KO tumors.

The above results did not, however, explain the dense accumulation of macrophages observed in the stroma surrounding tumors (Fig. 7, Additional file 2: Figure S9), raising the possibility that 129:Stat1 ^−/− cells might secrete factors that can both stimulate and inhibit macrophage migration. Recent studies have shown that EVs/exosomes can stimulate macrophage migration [40], which could explain the attraction of macrophages to the tumor-associated stroma. To test this possibility, Met1-CM and SSM2-CM were fractionated into EV-rich and soluble factor-rich portions using ultracentrifugation. The effects of these fractions on RAW264.7 cell migration were subsequently tested using the Transwell migration assay. Both the supernatant (soluble portion) and EV-rich fraction of Met1-CM induced higher levels of macrophage migration than the same fractions from SSM2-CM (Fig. 8i). Interestingly, the SSM2 EV-rich fraction induced more macrophages to migrate than the soluble fraction of SSM2-CM in the Transwell migration assay (Fig. 8i). The migration is notably comparable to that of the undiluted, noncentrifuged SSM2 media (Fig. 8g). This suggests that factors secreted as EVs from Stat1-null tumors are more attractive to macrophages than factors in the soluble portion that inhibit macrophage migration. These could prevent macrophages from infiltrating the tumor, helping to create the excluded infiltrate phenotype.

The majority of human breast cancers occur in older postmenopausal women. Ironically, this is the first mouse model that directly relates age to mammary tumorigenesis. The prolonged latency and acceleration of tumorigenesis following pregnancy are two key features of tumorigenesis in 129:Stat1 ^−/− females. Because the STAT1 deficiency is a germline mutation, the emergence of tumors requires secondary mutations and/or other adaptations within the microenvironment. The prolonged latency supports the “adaptive oncogenesis” theory, which postulates that changes in the host microenvironment facilitate the expansion of preexisting mutant populations [41]. Altered “fitness” then favors the emergence of specific subsets of mutated cells adapted for the new, aging environment. Although well-documented in the hematopoietic system [42, 43], the aging microenvironment in breast cancer has only recently been reviewed in detail [8].

Although “aging” in mice may vary with strain, animals aged 58 weeks or older can generally be considered “old” [44]. Mice in the present study were aged 30 to 120 weeks (2.3 years), with 38 being older than 70 weeks, which enabled documentation of numerous age-related pathological changes. The most dramatic nonmammary morphological changes in the aging 129:Stat1 ^−/− mice occurred in the uterus and ovaries. Whereas rete cysts of the ovary were found in aging 129:Stat1 ^−/− animals, they were present in both tumor-free and tumor-bearing animals and thus cannot be considered causal. Nonetheless, they provide direct morphological evidence of age-related changes in the ovary reflecting endocrine changes in an age-related environmental milieu.

Pregnancy accelerated tumorigenesis in 129:Stat1 ^−/− females and reduced the median tumor onset from 91 weeks of age in nulliparous females to 78 weeks. Studies of mouse mammary tumors in virus-induced and GMM models have consistently shown that pregnancy accelerates tumorigenesis. This phenomenon has generally been ascribed to the hormone responsiveness of MMTV-LTR and/or other mammary-specific promoters. Our previous mammary development experiments demonstrated that the mammary fat pads of young 129:Stat1 ^−/− mice were deficient in locally derived stromal cytokines and did not support optimal growth of the normal and neoplastic mammary epithelium, but they could be reversed by administering exogenous PRL and progesterone [2]. In the present study, we show that transplant of primary tumors into young nulliparous 129:Stat1 ^−/− female hosts resulted in much slower growth onset than in age-matched WT hosts (Additional file 2: Figure S8).

PRL signaling appears to be a critical molecular pathway in the etiology of impaired mammary gland development and tumorigenesis in 129:Stat1 ^−/− mice, in keeping with the demonstration that heterozygous truncating PRLR mutations arise in invasive tumors and some of the MIN lesions [17]. Our finding that some of the intraductal proliferation of the FoxA1⁺/ER⁺ LOP cells and areas of MIN had nontruncating single-nucleotide variants for the PRLR suggests that these variants might be responsible for a weaker, possibly nonprogressing proliferation.

These data also highlight the potential relationship between PRLR and ER expression in mammary cancer. In the mammary glands of ovariectomized mice, exogenous PRL suppresses estrogen- and progesterone-induced proliferation [45], similar to findings in transgenic mice that overexpress local PRL [46]. Estrogen and PRL synergistically evoke epithelial proliferation in the mammary glands of pigs [47], whereas overexpression of local PRL in the mammary glands of mice leads to ER-positive mammary tumors [48], and PRL induces ER expression in cultured breast cancer cells [49]. However, cooperation between these two hormones to effect breast cancer cell proliferation varies in accordance with the physical properties of the cell’s microenvironment [50], where the combination of estrogen and PRL enhances breast cancer cell proliferation in stiff collagen but not in low-density collagen. Reciprocal ER and PRLR signaling in a fibrotic microenvironment around Stat1-null tumors may drive LOP cell proliferation.

129:Stat1 ^−/− mammary carcinomas are exclusively FoxA1⁺ as well as ER⁺ and PR⁺. Recent evidence indicates that FoxA1, a “pioneer transcription factor” [51‐53], is necessary for ER expression [53‐55] and for branching morphogenesis in the mammary glands by maintaining ductal and alveolar luminal cell lineages from basal stem/progenitor cells [54]. ER⁺, PR⁺, and FoxA1⁺ MECs belong to a class of “hormone-sensing” (HS) MECs thought to be directly involved in mammary differentiation and in some human breast cancers [56]. These cells stand out in H&E-stained mammary ducts as morphologically unique, pale, basaloid or suprabasal cells (Additional file 2: Figure S5). They are abundant in mammary glands from young mice but are sparse in older mice (Additional file 2: Figure S5a–c). They resemble the oval clear cells in the mouse mammary gland described by Smith and Medina as “committed progenitor cells” [57, 58], as well as the suprabasal clear cells described in the human mammary gland [59‐62].

FoxA1 is expressed in ductal progenitor cells [63, 64]. The location and distribution of the morphologically unique FoxA1⁺ LOP cells identified in the present study are consistent with the notion that LOP cells represent a neoplastic form of suprabasal ductal HS-MEC progenitors [56]. The presence of FoxA1⁺/ER⁺/PR⁺ LOP cells in the 129:Stat1 ^−/− mammary ducts and neoplasms is also consistent with a ductal, rather than alveolar, origin of the tumors (Figs. 4, 5, and 6).

The LOP cells in 129:Stat1 ^−/− mammary glands are larger and more pleomorphic than either the FoxA1⁺ cells in normal ducts or the oval clear cells described by Smith and Medina [57, 58]. Individual LOP cells are present in some ducts associated with MIN and their malignant counterparts. The LOP cells also accumulate in tumor-free mammary glands and ducts as small, atypical lesions. As these lesions become larger, the LOP cells form small clumps along the involved mammary duct (Fig. 3), perhaps as abortive side branches (Figs. 4 and 6). This phenomenon is reminiscent of the aberrant branching morphogenesis in the developing 129:Stat1 ^−/− mammary gland [7], where excess abortive side buds are also present in other GMM models [65].

The human terminal duct lobular unit (TDLU) has been identified as the most likely site of origin of most human breast cancers [66‐69]. Computer-assisted 3D reconstructions of the mammary tree of human breasts demonstrated that ductal carcinoma in situ (DCIS) is limited to a single lobe [70, 71], with “multifocality” confined to foci within a single “sick lobe” [72]. The “sick lobe hypothesis” is also supported by the observation that recurrent invasive cancers and DCIS are found at the site of the prior biopsy or excision that showed DCIS [73]. In contrast, the quiescent and nulliparous mouse mammary glands do not have a specific counterpart to the human TDLU, but they do have HS progenitor cells immune-positive for FoxA1 that are involved in ductal development [54]. As demonstrated here, potentially neoplastic Stat1-null progenitor LOP cells, marked by FoxA1, appear to be topographically limited to a single branch of the mouse mammary tree (Fig. 5).

This “branch” (lobe)-restricted oncogenesis has not been described in other GMM or other mouse models of breast cancer. Carcinogenesis in most GMM models uses alveolar MEC-specific promoters and strong oncogenic transgenes that potentially activate oncogenic pathways in all MECs, resulting in scattered neoplastic foci in all branches of any given mammary gland. By contrast, the 129:Stat1 ^−/− mouse appears to model relatively rare oncogenic events at the level of primary branches of the mammary tree originating in the ducts rather than the alveoli/TDLU. Thus, 129:Stat1 ^−/− tumorigenesis is a unique mouse model of human breast cancers originating from ducts within an at-risk clonally related segment/branch of the ductal tree and models the “sick lobe” of humans.

Tumor cells influence their environment by paracrine-secreted and cell surface factors [23]. The host immune response to the primary MIN, tumors, and transplants was more intense and extensive than we observed in other GMM models without specific immunostimulatory interventions. Immune reactions to human cancers have been broadly categorized into “immune desert” with little or no immune cell infiltration, “inflamed” with abundant immune cell infiltration into tumor, and “immune excluded” where immune cells accumulate around the periphery of the tumor without infiltration [39]. The 129:Stat1 ^−/− tumors exhibit this “immune excluded” phenotype, whereas tumors resulting from activation of the ErbB2 or Wnt pathways have a scant host response, more like the “immune desert” phenotype [38, 74, 75]. In addition to connective tissue, a marked cellular response in Stat1 ^−/− neoplasms included macrophage and T-cell invasion and a scattering of granulocytic cells such as neutrophils and mast cells. Notably, macrophages have been implicated in the fibrotic response to cancer [40].

On the basis of the observation of a high tumor “excluded infiltrate” mononuclear density [39], we postulated that the Stat1-null tumor cells might mediate the host immune response through release of cytokines and other growth factors. Therefore, the Stat1-null tumor cell line SSM2 was used to study macrophage migration [7]. Our cell culture experiments suggest that Stat1-null tumor cells secrete both macrophage migration-promoting and macrophage migration-inhibiting factors, though we cannot distinguish between a direct and indirect microenvironmental effect. Nevertheless, our data show that chemokine secretions from Stat1 ^−/− tumor cells have dual effects that also help to explain the attraction but lack of penetrance of macrophages in vivo. Macrophages are known to have both tumor-suppressive and tumor-promoting capacity, depending on the context. For example, macrophages can induce fibrosis, contributing to the stiffened tumor-promoting stroma. Soluble inhibitory activity, meanwhile, prevents tumor-suppressive macrophage migration into the tumor cell interstices. This chemokine secretion from Stat1 ^−/− tumor cells is corroborated by the tissue localization and histopathology.

Our observations suggest that Stat1-null LOP cells, combined with the age-related changes in the tumor microenvironment, contribute to the development of a malignant neoplasm (Fig. 9). In this model, hormonal and microenvironmental factors induce the expansion of a FoxA1⁺ progenitor population. Further expansion and transformation are the result of a rare event, often a somatic truncating PRLR mutation, that is stochastically favored by the increase in the susceptible progenitor population seen with age. The increase in these progenitors is also associated with aberrant side budding and cysts originating midduct. Progression to invasive carcinoma is preceded by a marked stromal response, suggesting that a critical mass of intraductal cells secreting chemokines is required to create a permissive/tumor-promoting microenvironment.