Introduction
Transforming growth factor-beta (TGFβ) is a homodimeric polypeptide, which includes three isoforms: TGFβ1, TGFβ2 and TGFβ3. Secreted TGFβ binds to TGFβ receptor II (TβRII) and forms a heterodimeric complex with TGFβ receptor I (TβRI). The activated TβRI phosphorylates intracellular Smad2 and Smad3 (canonical TGFβ pathway). Simultaneously, phosphorylation of TβRII activates PI3K, MAP3k1, PP2A, RHOA and others (non-canonical pathway) [
1]. TGFβ plays a major role in the regulation of tumor initiation, progression, and metastasis, which requires TβRII for signaling [
1].
It has been published that decreased expression or loss of TβRII correlates with an increased risk of developing invasive breast cancer [
2]. Contrary to this fact, in mouse models of cancer, the inhibition of TGFβ signaling with the expression of dominant-negative TβRII (DNRII) or deletion of TβRII increases cellular proliferation without initiating tumor development [
3],[
4]. Therefore, the assumption is that attenuated TGFβ signaling alone is insufficient for transformation. In our previous research article it was indicated that deletion of TβRII in mammary epithelial of mouse mammary tumor virus (MMTV)-polyoma middle T antigen (PyMT) mice results in shortened tumor latency and a five-fold increase in lung metastases compared to MMTV-PyMT tumors with intact TGFβ signaling [
5],[
6]. The mechanisms behind this phenotypic difference are correlated with the increased expression of CXCL1, CXCL5 and CCL20 [
7],[
8]. Abrogated TGFβ signaling in carcinoma cells can indirectly promote progression of MMTV-PyMT tumor and metastasis by polarization T cells to Th17 cells via accumulation of CD11b
+Gr1
+ cells [
9]. Additionally, epithelial TGFβ signaling regulates fibroblast recruitment and activation. Our recent article confirmed the fact that fibroblast-stimulated carcinoma cells utilize TGFβ signaling to drive single-cell migration, but migrate collectively in the absence of TGFβ signaling, which promotes mammary tumor invasion [
10].
Mammary tumorigenesis has been examined through the use of numerous transgenic mouse models with wide utilization of the MMTV promoter/enhancer to drive expression in mammary epithelium. Overexpression of ErbB2 (Neu, human epidermal growth factor 2 (HER2)) or a constitutively active version of this receptor in the mammary epithelium leads to the development of metastatic mammary tumors [
11]-[
13]. Concurrently, overactivation of the ErbB2 pathway correlates with poor clinical prognosis in breast cancer patients [
14]. Using Neu-induced mammary tumor models with increased activity of TGFβ signaling (MMTV/ALK5 and MMTV/TGFβ1), it was possible to induce that active TGFβ signaling accelerates metastasis and the number of circulating tumor cells [
15]-[
17]. The loss-of-function experiments through the expression of soluble betaglycan or a DNIIR has been reported to suppress metastasis in Neu-induced mammary tumors [
16],[
18].
Based on these data we decided to examine the connection between TGFβ and Neu signaling in mammary tumor progression using MMTV-Neu and MMTV-Neu
activated induced tumorigenesis [
13]. The transgenic strains in conjunction with mice expressing DNIIR were used in the mammary epithelium to investigate the effect of attenuated TGFβ signaling on tumorigenesis and metastasis. We found that attenuation of TGFβ signaling with DNIIR prolonged tumor latency and dramatically enhanced pulmonary metastasis. The mechanism was different from that reported for the MMTV-PyMT model with conditional deletion of TβRII. Increased chemokine secretion through the knockout of carcinoma cells with resultant influx of CD11b
+Gr1
+ myeloid cells increased metastasis [
9]. In the MMTV-Neu model with DNIIR, there was no difference in chemokine secretion increase by the carcinoma cells and no increase in immature myeloid cell infiltration. Instead, there was reported increased secretion of vascular endothelial growth factor (VEGF), diminished pericyte coverage of vessels, and increased vessel leakiness and vasculogenesis. These symptoms likely act as the mechanism for the increased number of metastases. Lastly, analysis of human breast cancer transcriptome databases demonstrated a significant correlation between decreased
TGFBR2 and increased
VEGFA gene expression similar to what was observed in the mouse models. Higher
VEGFA gene expression was correlated with poor survival only in HER2-positive (HER2+) patients.
Methods
Mice and cell lines
All studies were performed on 202Mul and NK1Mul mice. To generate the mice with DNIIR-dominant-negative TβRII (202Mul/DN and NK1Mul/DN) 202Mul or NK1Mul mice ordered from Jackson Laboratory (Bar Harbor, ME, USA) and mice with expressed dominant-negative TβRII were crossed [
19]. The mice are proven to be on pure FVB background. The studies were approved by IACUC at Vanderbilt University Medical Center, Nashville, TN, USA.
The 202Mul and 202Mul/DN carcinoma cell lines were derived from primary tumors of 202Mul and 202Mul/DN mice, established and cultured in DMEM/F12 with 5% adult bovine serum. These carcinoma cells were implanted into the mammary fat pad of the #4 mammary gland via collagen plugs (CP) (5 × 105 cells/plug). CP were prepared by suspending carcinoma cells (5 × 105/plug) in collagen solution (50 mkl/plug), followed by pipette of this mixture (50 mkl) to 12-well dishes and incubation at 37°C for 45 minutes to solidify the gel. Then CP were overlaid with medium and incubated for an additional 4 to 18 hours at 37°C. The collagen mixture contained rat-tail collagen type I (BR Biosciences, San Jose, CA, USA), 10 × Earle's Balanced Salt Solution (EBSS, Gibco), NaHCO3, 1 M NaOH, and sterile ddH2O. The size of tumors was determined by direct measurement of tumor dimensions at 2 to 3 day intervals using calipers.
Flow cytometry analysis
Single-cell suspensions were made from the spleens of tumor-bearing mice [
20],[
21], and tumor tissues [
22]. Excised tumors were chopped into small pieces, incubated in DMEM (Gibco, Life Technologies, Grand Island, NY, USA) with no serum, 1 mg/mL collagenase I (Sigma, St. Louis, MO, USA), and 1 mg/mL Dispase II (Roche) for 2 hours at 37°C, and then passed through a cell strainer. Total cell numbers were counted, and CD45
+ cell populations that represented tumor-infiltrating host immune cells were analyzed by flow cytometry. After treatment with FcR Blocking Reagent (Miltenyi Biotec Inc., Auburn, CA, USA), tumor single-cell suspensions (10
6 cells/mL) were labeled using fluorescein-conjugated antibodies (Abs) (Biolegend, eBiosciense, BD, all from San Diego, CA, USA) for 20 minutes on ice. Data acquisition was performed on a LSRII flow cytometer (BD Immunocytometry Systems, Franklin Lakes, NJ, USA), and the data were then analyzed with FlowJo software. Nonviable cells were excluded using 4',6-diamidino-2-phenylindole (DAPI). Antigen negativity was defined as having the same fluorescent intensity as the isotype control.
ELISA
Cytokine levels in conditional media and tissue lysates were measured using the mouse CXCL1, CXCL5, MCP-1, VEGF, and IL-6 ELISA Duo kits (R&D Systems, Minneapolis, MN, USA) following the manufacturer's protocol.
Histology, IHC, and IF staining
Tissues were embedded directly in an optimal cutting temperature compound without fixation or placement in 10% formalin overnight, and then embedded in paraffin and sectioned at 5 μm. Sections were de-waxed in xylene and rehydrated in successive ethanol baths. For immunohistochemistry (IHC), the MOM kit was used (Vector). H&E and CD34 staining were performed in Translational Pathology Shared Resources (Vanderbilt University, Nashville, TN, USA). For immunofluorescence (IF) staining, primary and secondary antibodies were diluted in 12% BSA, and then mounted in DAPI that contained a SlowFade medium (Invitrogen). Antibodies used for staining were NG2 (1:200; Abcam), CD31 (1:200; BD Biosciences), 5-bromo-2’-deoxyuridine (BrdU) (BD Biosciences). Quantification of staining was performed using ImageJ software (National Institutes of Health, Bethesda, MD, USA) in accordance with the recommended guidelines. H&E and IHC sections were photographed using the OLYMPUS BX41 microscope and OLYMPUS DP2-BSW software. Slides for H&E and CD34 staining of lungs were scanned using the Leica SCN400 slide scanner with 20 × objective. Slides were photographed using a ZEISS Axioplan 2 microscope, and then numbered using MetaMorph software.
Whole-lung mounting
Mice were sacrificed by anesthetic overdose. Lungs were processed as described in the previously published article [
23]. The tumor nodules in lungs were then counted.
Cytokine antibody array
Cells (106) were plated on a 6-wells plate in 3 mL of DMEM/F12 with 5% of adult bovine serum. Conditional medium was collected after 18 hours, and secreted proteins were screened using the RayBio Mouse Cytokine Antibody Array C Series 1000 (RayBiotech Inc., Norcross, GA, USA) according to the manufacturer's instructions.
Western blot analysis
Cells or tissue were lysed in radioimmunoprecipitation assay buffer containing protease inhibitors cocktail (Roche Diagnostics, Indianapolis, IN, USA). Total protein concentrations were quantified with the Pierce BCA Protein Assay Kit (Pierce Biotechnology, Rockford, IL, USA). Equal amounts of protein (30 to 60 μg/well) were resolved in NuPAGE Novex 4 to 12% Bis-Tris polyacrylamide gel in the presence of 1 × MES buffer (2-(
N-morpholino)ethanesulfonic acid; Invitrogen) and transferred to a polyvinylidene fluoride membrane Immobilon-FL (Millipore Bioscience Research Reagents, Temecula, CA, USA). Anti-Akt (Cell Signaling, 9272), ph-Akt (Cell signaling, 4060), actin (Sigma, A2066) and secondary anti-Rabbit (Thermo Scientificm 31462) were used at 1:1,000, 1:1,000, 1:2,000 and 1:5,000 dilutions, respectively. After treatment with appropriate peroxidase-conjugated secondary antibody, the bands were visualized with an enhanced chemiluminescence method [
24]. The intensity of protein bands was quantified by a densitometer using ImageJ 1.45's software (National Institutes of Health).
Proliferation assays
For in vivo experiments BrdU incorporation was used by injecting 100 μL of 1 mg/mL BrdU 2 hours prior to performing euthanasia of animals. For in vitro experiments 3H-thymidine incorporation was performed for 2 hours prior to conducting measurement with a scintillation counter, whereby mean cpm were normalized to untreated cells. Cells were plated in 24-well culture dishes at 4 × 104 per well. TGFβ1 (R&D Systems, Minneapolis, MN, USA) treatment was performed in normal serum-containing media for 24 hours.
Statistical analysis
Data were presented as mean ± standard error of the mean (SEM). Multiple comparisons between the treatment groups and the control untreated group were performed using one-way analysis of variance (ANOVA) followed by Dunnett's procedure for multiplicity adjustment. Two-group comparisons were performed using the two-sample
t-test. Among the 1,056 human breast tumor tissue samples from The Cancer Genome Atlas Breast Cancer (TCGA BRCA) depository, 531 samples have both gene expression and clinical data available and were therefore used for the following analysis. Samples with low versus high TβRII expression were compared to their CXCL1, CXCL5, MCP-1, IL-6, and VEGF expression levels usingthe Wilcoxon rank-sum test for all patients as well as stratified by estrogen receptor (ER), progesterone receptor (PR), and HER2 status separately. The findings were validated using six independent GEO datasets (4992, 6532, 2990, 12093, 3494, and 10886). However, HER2 status was not available in these GEO datasets. Breast cancer subtype classifiers were available in the literature. In this study patient subtype was predicted using the PAM50 classifier (
R package, genefu 1.0.9 [
25]-[
27]). The association between distant metastasis-free survival (DMFS) and VEGF was analyzed using publicly available databases (GEO, EGA, and TCGA) for breast, ovarian, and human lung tumors from [
28]. All tests are two-sided and significant at the 5% level. All statistical analysis for human breast cancer data was performed in R 3.0.2 [
29].
Discussion
Important roles of TGFβ and HER2 signaling in tumor initiation and progression have been established in a large number of studies. To examine the role of TGFβ signaling in HER2+ breast cancer, we used MMTV-Neu mice with DNIIR. In our studies 202Mul mice with overexpression of wild-type ErbB2 and NK1Mul mice with mutant activated ErbB2 were utilized. In our experiment, we did not incorporate any specific pathways as was executed in the Siegel
at al. publication [
16]. In our analysis of the TCGA database (Additional file
1: Figure S7) we observed that in fewer than 5% of patients ErbB2 mutated and in about 15% of patients the ErbB2 receptor was amplified. This result was supported by recently published work by Bose
et al. [
34], in which researchers observed ErbB2 mutation in only a small percentage of HER2+ patients. Based on this fact we believe that our GEM tumor model is highly appropriate for investigation of the HER2+ type of breast cancer with attenuated TGFβ signaling.
In our GEM model the major differences between mice with intact and disrupted TGFβ signaling were increased tumor latency in parallel with the increased number of lung metastases (Figure
1). Changes in metastasis were different when compared with mouse models where ErbB2 was mutated by activated Grb2 or Shc signaling pathways [
16]. The increased tumor latency was also opposite to MMTV-PyMT/TGFβRII-KO mice with deletion of
Tgfbr2 in the mammary epithelium [
5], as previously studied in our laboratory. However, the increase in lung metastases in the MMTV-Neu/DNIIR mouse model was similar to MMTV-PyMT/TGFβRII-KO mice. This indicates that any manipulation to diminish TGFβ signaling in GEM models will lead to increased metastasis regardless of tumor-driving oncogenic transformations.
Increased tumor latency in MMTV-Neu/DNIIR mice (202Mul/DN, NK1Mul/DN) versus control mice (202Mul, NK1Mul) could be first due to cell-cycle dysregulation, where TGFβ signaling plays an important role, and second, due to dysregulation of chemokine expression. Feng
et al. demonstrated that the immune cells provide a source of tropic support to transformed epithelial cells, just as they do to normal epithelium during wound healing, and play a primary role in tumor initiation [
35]. We found that mice with DNIIR expression had downregulated levels of CXCL1 and CXCL5 (Figure
3) and as a result, fewer CD11b
+Gr1
+ cells in tumor tissue (Additional file
1: Figure S3).
Based on our current study and Siegel
et al. [
19] we can make a basic conclusion that attenuated TGFβ signaling in HER2+ tumor models, with active Shc and Grb2 pathways, decreases the probability of lung metastasis development. With intact ErbB2, attenuated TGFβ signaling has the opposite effect in spite of the fact that tumor latency increases.
Tumor tissues in mice with DNIIR had increased vasculogenesis, increased vessel size, leakiness, and decreased number of pericytes (Figure
2). Clinical data showed that a low number of pericytes correlated with poor patient prognosis [
36],[
37]. Simultaneously, disruption of pericytes also enhanced metastasis [
38]. In parallel with a decreased number of pericytes, we discovered that the size of vessels in tumor tissue was larger in mice with DNIIR. We hypothesized that there were probably two different mechanisms involved in increased vasculogenesis in DNIIR mice; and increased angiogenesis in parallel with the disruption of vessel support by pericytes. It is likely that an increase in vessel leakage leads to an increased number of metastases [
38].
In our previously published articles, we indicated that deletion of
Tgfbr2 leads to an increase in chemokine expression in mammary and pancreatic epithelium [
7],[
39]. Researchers have also linked deletion of
Tgfbr2 to an increase in mammary fibroblasts [
40]. In the mammary tumorigenesis studies, the major differences were found in CXCL1 and CXCL5 expression, which play an important role in the migration of neutrophils and myeloid-derived suppressor cells (CD11b
+Gr1
+).
An increased number of myeloid cells in MMTV-PyMT mice could be a basic mechanism driving decreased tumor latency and increased number of lung metastases. In our current study with attenuated TGFβ signaling in MMTV-c-Neu mice, we discovered an opposite effect in which levels of CXCL1/CXCL5 as well as CCL2 (MCP-1) decreased (Figure
3) in parallel with an increased level of VEGF. Comparison analysis of chemokine and VEGF secretions in MMTV-PyMT cells with DNIIR (Additional file
1: Figure S3) showed the same effect. There was an opposite result when TGFβRII was deleted. Our laboratory published a study [
41], which showed that DNIIR system could not completely inhibit the non-canonical TGFβ pathway versus the canonical (SMAD-dependent). This information correlates with our data on increased pAKT (Figure
4, Additional file
1: Figure S5) in MMTV-Neu/DNIIR cells, which is downstream of the non-canonical TGFβ pathway. We propose that it is a significant mechanism in the differential regulation of chemokines and VEGF secretion. We can conclude that the MMTV-Neu/DNIIR GEM tumor model is a model of spontaneous mammary carcinogenesis with a diminished canonical TGFβ pathway (SMAD-dependent) with the still-active non-canonical TGFβ pathway. There is also a significant amount of data to support a dose-response mediating TGFβ-induced phenotypic changes [
42],[
43]. Thus, it is likely that the observed gene expression changes are the result of altered TGFβ response due to the specific amount of TGFβ induced in cells, which express the DNIIR. Such observations could not be made with conditional knockout of
Tgfbr2 as TGFβ signaling is completely abrogated. Other authors also found increased levels of pAKT when combining activated TGFβ signaling (ALK5
T204D, TGFβ1
S223/225) with Neu-induced tumorigenesis [
15],[
17]. It can be explained by overactivation of both the canonical and non-canonical TGFβ pathways. An increased level of VEGF can explain increased vasculogenesis and vessel leakage in MMV/c-Neu DNIIR mice and could likely be involved in the observed increase in lung metastases.
Previously, we found that deletion of
Tgfbr2 leads to an increase in the number of CD11b
+Gr1
+ cells in MMTV-PyMT mice. In Neu-induced carcinogenesis, we did not observe many differences in immune components in mice with DNIIR versus control mice. The number of CD11b
+Gr1
+ cells decreased in tumor tissue from MMTV-c-Neu DNIIR mice that could result from downregulation of chemokines. Simultaneously the number of T cells decreased in parallel with the number of CD11b
+Gr1
+ cells in the MMTV-c-neu DNIIR tumors. Usually, increased numbers of CD11b
+Gr1
+ cells correlate with suppression of T cell proliferation [
44], but in our model we did not observe these changes to take place. Therefore, we suggest that decreased number of T cells could be due to the increased VEGF secretion in DNIIR mice. It has been shown that VEGF strongly inhibits T-cell development via VEGFR2 [
32]. Also, VEGF receptors are capable of inhibiting dendritic cell function and, thus, we hypothesized that anti-tumor immune response in mice with DNIIR diminished as a result of higher levels of VEGF.
Based on these findings, we conclude that human HER2+ breast cancer associated with decreased TGFβ signaling would also correlate with deceased expression of CXCL1/5 chemokines and increased VEGF. We observed that decreased TGFBR2 expression in human breast cancer patients correlated with decreased CXCL1, but not with CXCL5 in HER2+, PR + and ER + tumors. MCP-1 and IL6 decreased and VEGF level increased in all types of breast tumors with low TGFBR2 expression. To our surprise, increased VEGF expression in human breast cancer patients correlated with reduced DMFS only in HER2+ patients. The same outcomes were observed in our mouse studies.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
SVN and EF were involved in study conception, all experiments and data analyses, and drafting of the manuscript. AEG and AC performed mice genotyping, isolation and establishment of mammary tumor cell lines. MA assisted in in vitro cell maintenance and experiment coordination. DP performed immunoblotting. PO, DRY and MWP provided critical insight about the study design and experimental interpretation. ZZ, FY and YS provided statistical data analysis of human cancer datasets. HLM was a primary contributor to study conception, design and experimental implementation. All authors read and approved the final manuscript.