Background
Metastatic estrogen receptor α positive (ERα+) breast cancer is the leading cause of breast cancer mortality [
1‐
3]. However, the mechanisms that drive progression of these cancers are poorly understood, in part because there are few animal models of ERα + breast cancer. The extracellular matrix (ECM) is increasingly recognized as an important contributor to tumor behavior. Aggressive tumors frequently display desmoplasia, one component of which is increased deposition of fibrillar collagens such as collagen-I [
4‐
6]. This increased matrix deposition frequently increases the stiffness of the tumor and adjacent tissue [
7]. Stiff ECM environments drive tumor-progressive characteristics both in vitro [
8] and in mouse models [
9‐
11]. Moreover, tumors can actively remodel the surrounding ECM. Aligned collagen-I fibers increase ECM stiffness [
12], and collagen fibers aligned perpendicularly to the boundary of larger tumors predict poor outcomes, particularly of ERα + cancers [
13]. Collagen-I density/stiffness increases pro-tumorigenic signaling cascades in tumor epithelia, such as focal adhesion kinase (FAK), src family kinases (SFKs), and extracellular regulated kinase (ERK)1/2 [
14]
. The effects of these changes on hormonal signals and consequences for their roles in the progression of ERα + tumors are not well-understood.
Large prospective epidemiologic studies have linked the hormone, prolactin (PRL), to increased risk of development of aggressive ERα + cancers, and smaller-scale studies also suggest that it contributes to their progression [
15‐
18]. However, activation of STAT5, the primary physiological effector of prolactin (PRL), is associated with favorable clinical outcomes [
19‐
21], and reduces invasion of breast cancer cells in vitro [
22,
23]. Interestingly, FAK, SFKs, and ERK1/2 are also activated by PRL [
24‐
26], and the ability of PRL to activate STAT5 is inversely related to its ability to activate AP-1 via mitogen-activated protein (MAP) kinases and augment invasiveness [
27]. We recently reported that collagen-I density/stiffness is a major determinant of the signaling pathways that are available to the PRL receptor (PRLR). Whereas ERα + breast cancer cells cultured in low density/compliant three-dimensional collagen I matrices respond to PRL predominantly by activating physiological JAK2/STAT5 signals, high density/stiff matrices shift PRL responses to pathological ERK1/2 signals and increase invasiveness [
28]. Under these latter conditions, PRL crosstalk with estrogen increases alignment of the matrix perpendicular to the tumor edge [
29], similar to that correlated with decreased survival of patients with ERα + tumors [
13,
30]. These data indicate that PRL and the ECM cooperate to drive processes leading to progression of breast cancer. However, examination of this interplay in vivo is necessary to confirm its importance and investigate clinical applications.
In order to examine the interaction between PRL and increased collagen-I deposition in an immunocompetent environment in vivo, we took advantage of well-characterized genetically modified mouse models. Hormonally responsive mouse models of breast cancer are rare [
31,
32]. The neu-related lipocalin-prolactin (NRL-PRL) transgenic mouse mimics the local PRL synthesis in the mammary glands of women. Nulliparous female mice spontaneously develop aggressive mammary tumors, about 75% of which are ERα + [
33]. ERα + tumor cell lines derived from these adenocarcinomas are readily transplantable to syngeneic recipients [
34]. To model increased collagen I, we utilized the
Col1a1
tmJae1
mouse [
35]. Mutation of the MT1-MMP cleavage site in Col1a1 reduces collagen-I turnover, leading to its accumulation, without a need for additional fibrotic factors and any associated confounding activities. We have previously shown that this increases metastasis from experimental MMTV-PyMT-induced mammary tumors, and results in a collagen fiber signature that predicts poor survival in patients with breast cancer [
13,
14,
36].
Here, we orthotopically transplanted clonal green fluorescent protein (GFP)-labeled PRL-induced ERα + mammary tumor cell lines into syngeneic wild-type (WT) or heterozygous mutant collagen-I female mice (Col1a1
tmJae/+
, mCol1a1). Tumors that developed in the mCol1a1 environment had similar rates of growth, morphology, ERα, and PRLR expression to those in the WT collagen environment. However, the mCol1a1 environment increased circulating tumor cells (CTCs), and the number and size of lung metastases. It also altered the pattern of activated signaling cascades in the primary tumors: tumors in mCol1a1 female mice had lower pSTAT5 and increased pERK1/2 and pAKT expression, consistent with predictions from in vitro studies. Moreover, the alignment of intratumoral collagen fibers near the tumor boundary in mCol1a1 recipients was more perpendicular to the tumor edge and oriented in parallel to protrusions invading into the adjacent fat pad. These data indicate that extracellular matrix can potently interact with hormonal signals to drive the development of metastasis from ERα + breast cancers.
Methods
Reagents
Antibodies used in these studies were purchased from the following vendors: ERK1/2 (#9102), pERK1/2 (#9101), AKT (#9272), pAKT S473 (#9271), pAKT S473 (#3787S, for immunohistochemical analysis) from Cell Signaling Technology (Danvers, MA, USA); pSTAT5 (#71-6900) from Invitrogen (Grand Island, NY, USA); ERα (#sc-542), PRLR (#sc-20992), STAT5 (#sc-835x) from Santa Cruz Biotechnology (Santa Cruz, CA, USA); eGFP (#AB6658) from AbCam (Cambridge, MA, USA); PR (#A0098) from Dako (Carpinteria, CA, USA); biotinylated goat anti-rabbit (#BA-100) from Vector Labs (Burlingame, CA, USA); pan-actin (#125-ACT) from Phosphosolutions (Aurora, CO, USA); APC-conjugated CD31 (#551262) and CD45 (#559864) from BD Biosystems (San Jose, CA, USA). Avidin-biotin complex (ABC) (#PK-4000) and ImmPACT DAB (#SK-4105) were purchased from Vector Labs (Burlingame, CA, USA). All other reagents were obtained from Fisher Scientific or Sigma-Aldrich.
Cell lines and culture
ERα + mouse mammary tumor cell lines were derived from a NRL-PRL mammary tumor [
34,
37]. Two independently derived cell lines were stably transfected with eGFP, TC2GR12 (TC2) and TC4GR5 (TC4), and clonal sublines were maintained on tissue culture plastic in Roswell Park Memorial Institute medium (RPMI) 1640 supplemented with 10% FBS, 1% penicillin/streptomycin, and 1 mg/ml (TC2GR12) or 400 μg/ml (TC4GR5) puromycin as a selection factor. As shown in Additional file
1, TC2GR12 (TC2) and TC4GR5 (TC4) both express the rPRL transgene and exhibit somewhat differing levels of hormone receptor and signal effector proteins.
Animals
Mice heterozygous for
Col1a1
tmJae
[
35] (
Col1a1
tmJae/+
; mCol1a1) were backcrossed onto the FVB/N strain background for 10 generations. Mice were housed and cared for 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. For some experiments, 2.5 × 10
4 (TC2GR12) or 7.5 × 10
4 (TC4GR5) cells in 50 μl of sterile PBS were orthotopically injected into the left caudal mammary fat pads of 8 to 10-week-old FVB/N WT or mCol1a1 female mice and allowed to progress to end stage (tumor 1.5 cm in diameter). All recipients survived to end stage. For analysis of early-stage tumors, cell lines were injected bilaterally into the caudal mammary fat pads of 8 to 10-week-old WT or heterozygous mCol1a1 female mice, and tumors were allowed to progress for 17 days (TC2) or 24 days (TC4), the time of peak CTCs, respectively, before collection. Each animal was palpated biweekly to assess tumor development, and tumor diameter was measured using electronic calipers. Tumor volume was calculated as the largest diameter * (smallest diameter
2) * 0.4.
Flow cytometry
Peripheral blood (100 μl) was collected from each animal weekly from a maxillary vein in 6 U heparin sulphate, starting 3 days after tumor cell transplantation. Red blood cells were lysed in 0.15 M NH
4Cl + 1.2 mM EDTA for 10 minutes with gentle agitation. Cleared blood was centrifuged at 300 × g for 5 minutes, washed in PBS, and stained with anti-CD31/APC (1:100) and anti-CD45/APC (1:100) for 30 minutes on ice to label hematopoietic cells. The labeled cell suspension was then washed, fixed in 2% paraformaldehyde, and analyzed for endogenous GFP and CD31/45+ hematopoietic cells on a BD Fortessa flow cytometer, using the gating strategy presented in Additional file
2. Seven blood samples from non-injected female mice (five WT, two mCol1a1) served as controls for each experiment, and average background auto-fluorescence was subtracted from the experimental results.
Immunohistochemical analysis and immunofluorescence
Immunohistochemical analyses were performed as previously described [
38]. Briefly, tissues were fixed in 10% neutral buffered formalin, embedded in paraffin, and serial-sectioned. Deparaffinized tissues were hydrated in decreasing concentrations of ethyl alcohol (EtOH) and endogenous peroxidase activity quenched in 3% H
2O
2 in methanol (MeOH). Dilutions of the primary antibodies, antigen retrieval, and blocking conditions are shown in Additional file
3. Secondary antibody (1:250) and signal amplification (avidin-biotin complex (ABC)) were performed at room temperature prior to chromogen-detection with ImmPACT 3,3-diaminobenzidine (DAB) according to the manufacturer’s instructions. Tissues were counterstained with hematoxylin, dehydrated, and mounted for microscopic analysis. For immunofluorescence experiments, tissues were processed as described without performing peroxidase quenching. Alexa-488 conjugated streptavidin (1:100) was incubated at room temperature for 1 h prior to mounting with anti-fade media and subsequent epifluorescence imaging.
All animals were allowed to progress to end stage (tumor diameter 1.5 cm). At this time, lungs were collected and fixed in 10% neutral buffered formalin. After 24 h, surface nodules on all lung lobes were counted under a dissection microscope. The left lobe was then sectioned in four step-sections of 50 μm with two serial sections at each step. GFP immunofluorescence was performed as described to assess micrometastatic load. Total numbers of GFP+ lesions were counted per × 10 field of view (FOV) over seven FOVs per lung. The area of each lesion was calculated and the total sum of lesion area determined per FOV. Three independent lungs were analyzed for each cell line/genotype combination over two 50-μm steps, using ImageJ [
39].
Immunoblotting
Snap-frozen tumor pieces were homogenized in ice cold 40 mM Tris, 276 mM NaCl, 20% glycerol, 2% NP-40, 4 mM EDTA, 20 mM NaF, 2 mM sodium orthovanadate, 40 μg/ml phenylmethane sulfonylfluoride (PMSF), and 50 μg/ml aprotinin. Briefly, 30 μg of tumor homogenate was fractionated by SDS-PAGE, transferred to polyvinylidene fluoride (PVDF) membranes, and then probed with appropriate antibodies (ERK1/2, 1:2000; pERK1/2, 1:5000; STAT5, 1:50000; pSTAT5, 1:1000; ERα, 1:1000; AKT, 1:2000; pAKT S473, 1:1000; PRLR, 1:1000). Signals were visualized by enhanced chemiluminescence and quantified by densitometry (UVP Visionworks). All end-stage tumors were analyzed.
Collagen imaging and analysis
Paraffin-embedded, 5-μm sections of early lesions were stained with picrosirius red as previously described [
40]. Polarized light microscopy with a × 10 objective was utilized to assess birefringence that is specific to fibrillar collagens. Multiphoton microscopy and second harmonic generation (SHG) imaging were performed as previously described [
41]. Briefly, collagen was visualized with a laser excitation of 890 nm with a 445 ± 20 nm emission filter to detect backwards SHG focused on the sample with a × 20/0.75NA objective. The overall fiber alignment analysis was performed with the CurveAlign software following fiber detection utilizing CT-FIRE [
42,
43]. The alignment coefficient was based on the orientation of the collagen fibers in the same image and ranges from 0 to 1, where the larger the alignment coefficient, the more aligned the fibers.
Statistical analysis
Statistical analyses were performed in GraphPad Prism v4.0 or SAS/STAT v9.4. Data were tested for normality using the Shaprio-Wilk test. Data were analyzed using the unpaired t tests (normal data) or Mann-Whitney U non-parametric test (non-normal data) as indicated. Tumor growth and CTC data were analyzed using two-way repeated measures analysis of variance (ANOVA), taking into account values across days in the same animals. Collagen alignment was quantified using repeated measure group comparison with the SAS Proc Mixed model to take into account image–image and tumor − tumor variation for each cell line/genotype combination. Significance of the data are presented as *p < 0.05, **p < 0.01, ***p < 0.001, or ****p < 0.0001.
Discussion
Metastasis of ERα + breast cancer is poorly understood. ECM characteristics such as alignment of the collagen fibers in and around the tumor correlate with poor survival [
13], demonstrating the importance of this component of the tumor microenvironment. However, the role of hormones in these events, and the consequences of changes in the ECM for hormone actions have not been clear. While some evidence supports a role for elevated PRL exposure in tumor progression [
18], conflicting data on PRL-induced signal effectors suggest that other factors modulate PRL responses. We previously demonstrated in vitro that a dense/stiff collagen-I matrix shifts the balance of PRL signals from physiological to tumor progressive, and permits PRL and estrogen to reorient collagen-I fibers that mimic aggressive ERα + breast cancers in women [
28,
29]. Here, we report that the predictions from these studies are confirmed in an immunocompetent mouse model in vivo: collagen-I accumulation promoted ERK1/2 and AKT activation and decreased STAT5 phosphorylation in PRL-induced ERα + mammary carcinomas, driving local invasion of the primary tumor, realigning collagen fibers, mobilizing tumor epithelia, and enhancing pulmonary metastases. These data indicate that the ECM can alter hormonal signals to drive aggressive behavior of ERα + tumors, providing mechanistic insight into their metastasis.
Extensive in vitro studies of human breast cancer cells have shown that the PRLR, like other cytokine receptors, can activate multiple signaling pathways [
17]. However, the determinants of the spectrum of signaling pathways, and their respective roles in disease in vivo have been unclear. In normal mammary development, PRL signals primarily through JAK2/STAT5 to direct expansion and differentiation of the mammary epithelia in concert with ovarian steroids [
49,
50]. Consistent with clinical data demonstrating that high STAT5 activation in breast cancers predicts favorable outcomes [
19‐
21], constitutive activation of STAT5 in mouse models leads to well-differentiated tumors [
51]. Interestingly, signals through this pathway are required for initiation, but not progression, of PRL-induced cancers in NRL-PRL females [
52]. However, many ERα + cancers that develop in this model are very aggressive, and display highly activated ERK1/2 and AKT [
33,
38], indicating activation of non-JAK2/STAT5 pathways. The data presented herein substantiate that the ECM is a potent determinant of the signaling cascades and outcomes of PRL actions in vivo, illuminating the apparent disparity between PRL exposure and activated STAT5 in the progression of clinical breast cancer. Moreover, our findings suggest that targeting non-canonical PRL signals may be of therapeutic benefit.
Mammographic density is a strong predictor of breast cancer risk [
53‐
55]. However, epidemiologic data linking ECM density and breast cancer aggression are inconsistent [
56,
57]. Mammographically dense tissue partly comprises increased fibrillar collagen [
58]. Although this can stiffen the ECM [
7,
59], the properties of density and stiffness are not always linked. For example, at involution following weaning, the mammary ECM contains higher levels of collagen I, yet the ECM is less stiff than that in the nulliparous gland [
60]. This suggests a fundamental difference in ECM architecture between physiologic states and cancer which deserves further study. Interestingly, we have observed that the ECM features - stiffness and collagen-I ligand density - exert distinct effects on PRL-initiated signals, using polyacrylamide hydrogels in vitro: elevated density reduces PRL phosphorylation of STAT5, whereas stiffness augments PRL signals to ERK1/2 [
61]. The alterations in both signaling cascades elicited by reduced collagen I degradation and the increase in collagen fibers aligned with invasive projections of the tumor mass observed in the current in vivo study suggest that PRL-expressing ERα + cancer can directly or indirectly remodel dense collagen matrices to increase stiffness and alignment. Further, PRL itself may contribute to collagen density. PRL and mammographic density are epidemiologically linked [
62,
63] and PRL enhances the expression of mammary ECM components such as
Col1a1 [
64], and
Tnc [
65], which promotes cancer cell invasion [
66]. Taken together with the current data, these observations begin to outline a model where the characteristics of PRL and ECM work together to promote the invasion and metastasis of hormone-responsive breast cancer.
Mouse models of ERα + breast cancer are limited, and few develop distant metastases [
31,
32]. Although patient-derived xenografts are proving useful in elucidating the behavior of some breast cancer subtypes in vivo, they require highly immunocompromised hosts [
67,
68], a substantial limitation in light of the accumulating evidence on the importance of the immune response in tumor progression and metastasis [
69]. Moreover, ERα + breast tumors have been difficult to grow in mice, preventing the widespread use of patient-derived xenografts for this subtype. Interestingly, mice genetically engineered to express human PRL appear to increase successful transplantations of human primary ERα + breast cancer [
70]. Our syngeneic model allows examination of the behavior of ERα + mammary cancers, including metastasis, permitting us to demonstrate potent interactions between the features of the ECM and PRL that drive aggression in vivo.
Acknowledgements
We would like to thank Dr Ruth Sullivan for helpful discussions on the collagen environment, Kyle Wegner for technical assistance with picrosirius red staining and image processing, Adib Keikhosravi for useful input on imaging and analysis, Dr Chad Vezina for helpful discussions on collagen imaging and providing immunofluorescent microscopy capabilities, Dr Joan Jorgensen for providing the eGFP antibody and Melanie Iverson for assistance with the immunohistochemical analysis.