Abstract
TGFβ plays a crucial role in the tumor microenvironment by regulating cell–cell and cell–stroma interactions. We previously demonstrated that TGFβ signaling on myeloid cells regulates expression of CD73, a key enzyme for production of adenosine, a protumorigenic metabolite implicated in regulation of tumor cell behaviors, immune response, and angiogenesis. Here, using an MMTV-PyMT mouse mammary tumor model, we discovered that deletion of TGFβ signaling on myeloid cells (PyMT/TGFβRIILysM) affects extracellular matrix (ECM) formation in tumor tissue, specifically increasing collagen and decreasing fibronectin deposition. These changes were associated with mitigated tumor growth and reduced metastases. Reduced TGFβ signaling on fibroblasts was associated with their proximity to CD73+ myeloid cells in tumor tissue. Consistent with these findings, adenosine significantly downregulated TGFβ signaling on fibroblasts, an effect regulated by A2A and A2B adenosine receptors. METABRIC dataset analysis revealed that patients with triple-negative breast cancer and basal type harbored a similar signature of adenosine and ECM profiles; high expression of A2B adenosine receptors correlated with decreased expression of Col1 and was associated with poor outcome. Taken together, our studies reveal a new role for TGFβ signaling on myeloid cells in tumorigenesis. This discovered cross-talk between TGFβ/CD73 on myeloid cells and TGFβ signaling on fibroblasts can contribute to ECM remodeling and protumorigenic actions of cancer-associated fibroblasts.
TGFβ signaling on fibroblasts is decreased in breast cancer, correlates with poor prognosis, and appears to be driven by adenosine that accelerates tumor progression and metastasis via ECM remodeling.
Introduction
Tumor microenvironment consists of multiple cellular and noncellular components, creating a niche for tumor development. Complex interactions of its components, including cells, extracellular matrix (ECM), dissolved or/and immobilized on acellular surfaces active molecules, etc., are critical for tumor outcome and require thorough investigation (1).
TGFβ contributes to normal organogenesis and tissue homeostasis and, importantly, plays a crucial role in cancer progression by regulation of different aspects of tumor behavior (1–3). Pleiotropic functions of TGFβ both as a tumor suppressor and a tumor promoter are well known and continue to be examined extensively by many researchers. In tumor microenvironment, TGFβ regulates functioning of both parenchyma and stroma (1, 2, 4). The effects of TGFβ include regulation of numerous ECM-related genes—collagen type I (Col1), collagen type III (Col3), fibronectin, as well as matrix-modifying enzymes such as matrix metalloproteinases type 2 (MMP2), MMP9, and lysyl oxidase homologue 4 (LOXL4; ref. 2), suggesting an essential role of TGFβ in behavior of cancer-associated fibroblasts (CAF). CAFs directly and indirectly regulate tumor cells phenotype, their proliferation, invasiveness, and metastatic abilities. Generally, CAFs regulate tumor progression by directing production and remodeling of ECM structures, angiogenesis, and secretion of wide specter of cytokines (4–6). TGFβ impact on tumor-infiltrating immune cells has become recognized in recent years. TGFβ regulates the recruitment, motility, activation, and function of immune cells. It drives polarization of M2 macrophages and N2 neutrophils, inhibits NK-cell maturation, and promotes Treg differentiation that can alter immune system net effect from anti- to protumorigenic (7, 8).
On the other hand, inadequate TGFβ signaling in different cells of tumor microenvironment can also be detrimental. For instance, at the late phase of tumor development, the loss of TGFβ signaling in carcinoma cells changes their behavior to more aggressive (9). Fang and colleagues have shown that lack of TGFβ signaling in fibroblasts leads to increased tumor weight and number of metastasis (10). In contrast to carcinoma cells and fibroblasts, the loss of TGFβ signaling in myeloid cells correlates with better outcome (11). Recently, our group demonstrated that TGFβ signaling in myeloid cells regulates expression of CD73 ectoenzyme, which is in pair with CD39 is responsible for extracellular ATP breakdown to adenosine. Adenosine, known promoting agent in tumor microenvironment, directly affects behavior of neoplastic cells and participates in formation of the immunosuppressive niche and angiogenesis (3). Adenosine works through G-protein–coupled receptors A1, A2A, A2B, and A3. These receptors have different affinity to adenosine, conveying signal via different downstream pathways, and their expression and effects are varied and depend on cell type and environment (12, 13). Building on the seminal discoveries by the Sitkovsky group, who was first to provide the genetic and pharmacologic in vivo evidence for nonredundant functioning of extracellular adenosine-A2 receptors-cAMP axis in regulation of inflammation and tumor protection (14, 15), a conceptually new field of anticancer therapy by targeting the hypoxia-extracellular adenosine-A2AR/A2BR signaling pathways has been developed, leading to current promising clinical trials (16).
Mechanisms of changes in TGFβ signaling during tumor progression on tumor or stromal cells may be diverse and include mutations that lead to the absence of key proteins of TGFβ pathway, insufficient downstream signaling, or downregulation by another signaling pathway. Here, we demonstrate how TGFβ signaling on a single-cell type can shape net response of tumor microenvironment and contribute to cancer outcome. Specifically, we show that TGFβ signaling in myeloid cells affects tumorigenic properties of CAFs that participate in tissue architectural rearrangements associated with adverse outcome. Using single-cell image analysis, we tested our hypothesis that CD73+ myeloid cells via adenosine production and activation of adenosine receptors regulate CAF's response to TGFβ. Following analysis of human METABRIC datasets and outcome of patients with breast cancer shows that A2B adenosine receptors can play an important role in this process.
Materials and Methods
Transgenic mice
Experiments were performed on MMTV-PyMT/TGFβRIIfloxed and MMTV-PyMT/TGFβRIILysM-KO mice (FVB background) established and maintained as described previously (3, 17). To generate MMTV-PyMT/TGFβRIILysM-KO mice, we first crossed LysM-Cre mice (FVB background, kindly provided by Timothy Blackwell, Vanderbilt University Medical Center, Nashville, TN) with MMTV-PyMT mice and then MMTV-PyMT/TGFβRIIfloxed mice with MMTV-PyMT/LysM-Cre mice. The studies were approved by IACUC at Vanderbilt University Medical Center (Nashville, TN).
Cell line
Immortalized mouse tumor mammary fibroblasts were generated in Dr. Harold Moses laboratory (Vanderbilt University Nashville, TN), used in previous publications (18–20), and gifted to us. Fibroblasts were grown in T-75 flasks (Thermo Fisher Scientific) in DMEM medium (Gibco) supplemented with 10% FBS and 1% Antibiotic–Antimycotic (Gibco). Cell lines were not authenticated. Cells were Mycoplasma-free, tested using LookOut Mycoplasma PCR Detection Kit (Sigma), and used within 20 passages of thawing. For Western blotting assay, fibroblasts were transferred to 10-cm cell culture dishes (Corning). When cells achieved 90%–95% confluency, they were treated with TGFβ, 1 ng/mL (R&D Systems) and NECA (5′-N-Ethylcarboxamido adenosine, Sigma-Aldrich).
Histology
Tumor tissue samples were fixed in 10% neutral buffered formalin for 12–24 hours at room temperature. Fixed samples were embedded in paraffin blocks, and 5-μm sections were cut. Hematoxylin and eosin (H&E) and all IHC staining procedures were performed by TPSR core at Vanderbilt University Medical Center (Nashville, TN). H&E staining was used for routine morphologic analysis. To study ECM elements, we performed picrosirius red staining (PRS) and IHC staining for laminin (DAKO), fibronectin (Abcam), and elastin (Abcam). Macrophages, neutrophils, endothelial cells, and proliferating cells were detected using anti-F4/80 (Novus Biologicals LLC), anti-Ly6G (Abcam), anti-CD31 (Dianova), and anti-Ki67 (Cell Signaling Technology) antibodies, respectively. Fluorescent staining was performed with anti-αSMA (Abcam) with secondary anti-mouse Alexa Flour 488, pSMAD3 (Abcam) with secondary anti-rabbit-Biotin and Streptavidin Alexa Flour 647, F4/80 (Novus Biologicals LLC) with anti-rat Alexa Flour 750, CD73 (R&D Systems) with anti-sheep Alexa Flour 568. Pictures were taken on Keyence BZ-X710. Whole-image scanning was performed on Aperio Versa 200 (Leica).
Western blotting
Tissue protein extracts were analyzed by Western blotting for total SMAD2/3, pSMAD3, and collagen type I. Running conditions: NuPAGE 4%–12% Bis-Tris gel (Invitrogen), 100 V, 2 hours for total SMAD2/3 and pSMAD3, or handmade 6% acrylamide gel (Bio-Rad recommendations), 100 V, 5 hours for collagen I. Transferring conditions: iBlot nitrocellulose gel transfer stacks (Invitrogen) for total SMAD2/3 and pSMAD3, or 0.45 μm nitrocellulose membrane, 20 V, overnight for collagen. Blocking conditions: Tris-buffered saline (Corning), 2.5% milk (Bio-Rad), 0.05% Tween 20 (Sigma-Aldrich), 1 hour. Primary antibodies: anti-total SMAD2/3 rabbit antibodies (Abcam), anti-pSMAD3 rabbit antibodies (Abcam), and anti-collagen I rabbit antibodies (Abcam), overnight, 4°C. GAPDH was used as a housekeeping control. Secondary antibodies: anti-rabbit and anti-mouse antibodies HRP-conjugated, 2 hours (Promega).
Protein extraction
For collagen detection by Western blotting, samples of tumor tissue were homogenized by pestle in RIPA protein lysis buffer and then treated by Sonicator W-225 (Heat Systems-Ultrasonic, Inc.), at pulse regime, 40% of pulse duty for 10 seconds and 40% from maximum power. After sonication, samples were incubated with DNAse for 5 minutes, 25°C.
ELISA analysis
ELISA was performed using DuoSet Kit (R&D Systems) following the manufacturer's protocol.
RNA extraction and gene expression analysis
RNA extraction was performed from previously frozen tissues following the manufacturer's instructions (Qiagen). Analysis of gene expression was performed by NanoString technique with PanCancer Pathways Panel (NanoString Technologies). This panel shows the functional state of basic cancer pathways such as cell cycle and apoptosis, RAS, PI3K, STAT, MAPK, TGFβ, Hedgehog, Wnt, Notch, chromatin modification, transcriptional regulation, and DNA damage control. A total of 12 samples (6 control and 6 experimental) were used for NanoString analysis, and gene expression quantification was made using the NanoString nCounter Platform (NanoString Technologies).
Gel contraction assay
Immortalized mouse mammary fibroblasts and CD11b+ cells, extracted from mouse bone marrow, were embedded in collagen gel. The collagen gel was prepared from rat-tail collagen type I (Corning), 10 × Dulbecco's Modified Eagle Medium (DMEM, Gibco), 10 × NaHCO3, and sterile ddH2O. Concentration of rat tail collagen type I in the gel mixture was 3 mg/mL. Cells/gel mixture was placed in 24-well plates for 48 hours. TGFβ was added at concentration of 1 ng/mL after 48-hour incubation, and collagen gel plaques were detached from well's bottom for successful contraction.
Image analysis
ECM analysis
ECM reach regions were imaged using PRS in IF field in Texas red filter, with 20× objective. Analysis of ECM fibers was performed using CurveAlign (http://loci.wisc.edu/software/curvealign) and CtFIRE (http://loci.wisc.edu/software/ctfire) software packages.
Area-based analysis
For image quantification, ImageJ v.1.51k was used (ImageJ). To measure area that was occupied by different stages of cancer progression, free hand selection tool was used. To quantify number of IHC and picrosirius red–positive pixels, 10 regions of interest (ROI) were taken. Threshold options demonstrated maximum pixels of interest and simultaneously minimum of background pixels. Results are presented as a percentage of positive area.
Single-cell analysis
To perform single-cell analysis of multiplexed fluorescent-stained images, image analysis pipeline was built on KNIME (Knime.com) platform (Knime 3.6 with integrated image processing and analysis extension v. 1.7.0.201906270525). DAPI-stained images were used to generate pixel classification model in Ilastik (https://www.ilastik.org) followed by integrated from ImageJ watershed algorithm for nuclear segmentation. Cell ROIs were generated by circular outgrow of nuclear masks. Single-cell features were extracted by aligning nuclear or cell ROI's masks to specific fluorescent stain images. Geometrical, statistical, and texture features were extracted for each segmented cell. For classifications, single cells on subset of images were annotated as “positive” or “negative” based on the fluorescent signal. This training set with “Ground Truth” and appended features was used as input for XGBoost Tree Ensemble Learner within KNIME to generate a prediction model that was applied to a whole dataset. The output of this prediction algorithm is a probability value, calculated for each cell on input features, to be close to “positive” or “negative” cell class for each trained stain. Investigator-defined probability cutoff determined a binary cell classification (“positive”/“negative”) for “fibroblast” (“αSMA positive” probability value > 0.6), “macrophage” (“F4/80 positive” probability value >0.6). To quantify number of myeloid cells with CD73 expression, signal intensities normalized to DAPI intensity (sums fluorescent signals) were graphed, and gates of cells with high F4-80/high CD73 were made in a manner similar to FACS analysis and considered double-positive cells. Total cell number and specific class cell number per image were quantified, and percent calculations were made. For pSmad3 signal quantification fibroblast (αSMA+) data were filtered and grouped per image (per mouse) followed by extraction of median fluorescent intensity of pSmad3.
Statistical analysis
Results were presented as mean ± SEM. Multiple comparisons between groups were performed using one-way ANOVA followed by Dunnett procedure for multiplicity adjustment. Two-group comparison was performed using two-sample t tests or Wilcoxon rank-sum test as appropriate. The gene expression was normalized across all arrays. Gene symbols were assigned using the manufacturer-provided annotation, but the analysis was performed at probe level. Correlation analyses for the gene expressions between the signatures of interest were performed using data representing 562 patients (white, 35–70 years old) from METABRIC and up to 151 (female, white, 35–70 years old, not Hispanic or Latino; varies by cancer type; Supplementary Fig. S1) patients from TCGA. The association between RFS and A2b was analyzed using the univariable Cox proportional hazard regression model. When a subtype–gene expression interaction was present, stratified Kaplan–Meier survival curves were created to compare “high versus low” of the A2b and A2a gene expression for each subtype. All tests were statistically significant at two-sided 5% level. All analyses were conducted using GraphPad Prism 8.0 Software (GraphPad Prism) or R version 3.6.0.
Results
TGFβ signaling in myeloid cells regulates tumor development
Previously, we have generated mice with spontaneous tumor formation of mammary gland (MMTV-PyMT) without TGFβ receptor II on myeloid cells (LysM-Cre) PyMT/TGFβRIILysM. We have reported these mice have decreased tumor growth and, notably, the number of lung metastases (3). Because changes in ECM are essential in metastasis (21, 22), we decided to evaluate a morphology of tumor tissue, isolated from PyMT/TGFβRIILysM (experimental) and PyMT/TGFβRIIWT (control) animals to determine a mechanism by which TGFβ in myeloid cells contribute to metastasis.
Tumor tissues were isolated on 4th week after tumor had been first palpated (6–7 weeks of age), as described earlier (3). In the beginning of tumor formation, the MMTV-PyMT model shows well-differentiated luminal adenoma that usually presents in the main collective duct, which then progresses to the poor-differentiated adenocarcinoma (Fig. 1A). After formation of primary tumor around main collective duct, additional foci of cancer formation occur in peripheral ducts (23), thus allowing us to observe several stages of tumor development. Therefore, we separated the areas correspondent to different stages of tumor progression and performed histologic and IHC analysis of these areas. As seen in Fig. 1A, two different areas can be distinguished in tumor tissue: early carcinoma (EC) and late carcinoma (LC). The EC was represented by ducts covered by 3–4 layers of tumor cells and ducts completely filled by tumor cells. Neoplastic cells in those regions have high level of differentiation and demonstrate secretory function. The structure of LC regions was represented by high amounts of tumor cells with high grade of nuclear pleomorphism, high levels of mitosis in the absence of duct structures. Connective tissue was presented by solitary fibers and formed disorganized web. We found that tumors isolated from PyMT/TGFβRIIWT mice have increased LC areas in comparison with tumors from PyMT/TGFβRIILysM mice (Fig. 1A).
To detect proliferation rate, we performed Ki67 IHC staining. In EC areas, Ki67+ cells are mostly located in peripheral regions of ducts and usually form groups of different sizes. In LC regions, Ki67+ cells are located mostly separately and rarely form large groups. We found that number of Ki67+ cells in EC areas from PyMT/TGFβRIILySM tumors is decreased compared with corresponding areas from PyMT/TGFβRIIWT tumors (Fig. 1B).
Blood vessels in tumors are generally localized in connective tissue strands. Using CD31 staining, we observed an increased number of blood vessels in EC regions, with no difference between PyMT/TGFβRIIWT and PyMT/TGFβRIILysM animals. In LC areas, a higher density of blood vessels was observed in tumors of PyMT/TGFβRIIWT animals compared with corresponding areas from PyMT/TGFβRIILySM tumors (Fig. 1C).
Myeloid cells were localized primarily in connective tissue strands of EC and LC regions, especially near blood vessels. Also, a small number of macrophages and neutrophils was localized between neoplastic cells and in the necrosis regions. We detected that PyMT/TGFβRIILysM tumors have lower numbers of neutrophils (Gr1+), especially in EC regions, compared with PyMT/TGFβRIIWT tumors (Fig. 1D). Similar to neutrophils, F4/80+ were localized in connective tissue strands. We also found a decrease in number of macrophages in both EC and LC areas of tumors isolated from PyMT/TGFβRIILysM mice compared with corresponding areas from PyMT/TGFβRIIWT tumors (Fig. 1E).
To elucidate how the lack of TGFβ signaling in myeloid cells changes behavior of tumor microenvironment, we used NanoString Technologies (CancerPath panel). Hierarchical clustering analysis showed two clusters that consisted of WT and KO samples mixed together (Fig. 2A). This suggests that the lack of TGFβ signaling in myeloid cells does not significantly change gene expression profile of tumors. Some of the genes that have demonstrated change in expression are related to PI3K and Wnt signaling pathways (Fig. 2B–D). This observation of the lack of significant changes in the expression of cancer-related genes discordant to differences in tumor growth rate directed us to test the hypothesis that changes in structure and functions of ECM associated with TGFβ signaling in myeloid cells could be a key contributor to observed tumor phenotype.
TGFβ signaling in myeloid cells regulates ECM deposition in tumor tissue
To make detailed analysis of ECM, we used PRS. Recently, it has been shown to be more informative in analysis of collagen deposition in tissue than the second harmonic generation (SHG; ref. 24). We found that in LC area of tumor tissue, collagen formed fibers of variable length and width, and these fibers had different directions forming a web-like structure. In EC areas, collagen fibers surround ducts and often form wide strands. We detected that tumors from PyMT/TGFβRIILysM mice contained higher amount of collagen deposition by PRS especially in EC area compared with corresponding areas of PyMT/TGFβRIIWT tumors (Fig. 3A). Using polarized light on PRS, we were able to distinguish mature and immature collagen in the tissue. Collagen fibers in all regions of tumors were represented mainly by mature collagen in both types of mice. However, the amount of immature collagen in EC regions was higher in PyMT/TGFβRIILysM tumors compared with control group (Fig. 3A). This is indicative of a more intensive collagen synthesis, secretion, and assembly in that area during tumor development. In addition to PRS, we confirmed our finding by immunoblotting analysis, which also revealed an increased amount of collagen in MMTV-PyMT tumors without TGFβRII signaling in myeloid cells compared with control tumors (Fig. 3B).
In addition, we analyzed other basic ECM components—laminin, fibronectin, and elastin. Laminin is one of the major components of basal membrane of gland ducts (22). Laminin was represented by threads of variable length, which during tumor progression became shorter and did not form uninterrupted structure of basal membrane. In some areas, laminin was nearly absent. We found that PyMT/TGFβRIILysM tumors contain more laminin in LC area compared with PyMT/TGFβRIIWT tumors, but found no difference in laminin content in their EC areas (Fig. 3C). Furthermore, we have also observed a reduced amount of fibronectin only in LC regions of PyMT/TGFβRIILysM tumors compared with PyMT/TGFβRIIWT, whereas the presence of fibronectin in EC areas was almost similar in both tumors (Fig. 3D). In contrast, we did not find any elastin-positive fibers in ECM, but some of neoplastic cells contained elastin-positive structures in cytoplasm. However, there was no difference visually between PyMT/TGFβRIIWT and PyMT/TGFβRIILysM tumors.
CD73+ myeloid cells regulate TGFβ signaling in fibroblasts
TGFβ is a major regulator of collagen secretion by fibroblasts (25). Because we have detected an increased collagen amount in tumor tissue of PyMT/TGFβRIILysM mice, we next determined whether TGFβ signaling is affected in tumor tissue of these mice. Whole-tumor tissue homogenates demonstrated higher phosphorylation of SMAD2/3 in PyMT/TGFβRIILysM versus control animals, despite the fact that these mice are lacking TGFβ signaling in myeloid cells (Fig. 4A). Next, we measured concentration of total and active TGFβ protein by ELISA and found no difference between experimental mice and control group (Fig. 4B). Furthermore, we analyzed TGFβ signaling in isolated cell populations focusing specifically on fibroblasts, the cells that primarily secrete collagen. We used multiplex immunofluorescent staining for pSMAD3, αSMA, F4/80, and CD73 to detect activity of TGFβ signaling (pSMAD3) and to distinguish activated fibroblasts (αSMA+) and myeloid cells (F4/80+). Importantly, we included CD73 stain in antibody panel to confirm previously published flow cytometry data that CD73 is decreased in myeloid cells with deletion of TGFβ signaling (3). Visually, we observed that tumor and stromal cells located in close proximity to F4/80+ cells had reduced pSMAD3 signal (Fig. 4C). To extract quantitative data to support this observation, we performed single-cell analysis of multicolor-stained slides. For this, we built an image analysis pipeline utilizing KNIME analytical platform. First, using DAPI-stained nuclei (Fig. 4D, left panel), we generated nuclear segmentation (Fig. 4D, middle panel). Next, generated cell masks were applied to single-channel images (stained for F4/80, αSMA, CD73, and pSmad3) to extract cell features. In parallel, supervised machine learning was conducted on single-channel images. For this, training set was annotated with positive and negative cell examples. These annotations were combined with extracted cell features to generate model for quantification of cell class probabilities. Cell phenotyping was achieved using binary probabilities (positive/negative) for αSMA (activated fibroblast) and F4/80 (macrophage; Fig. 4D, right panel). We have confirmed our previous finding (3) that cells with deleted TGFβRII have downregulated CD73 by showing that numbers of CD73+F4/80+myeloid cells are decreased in tumor tissue of PyMT/TGFβRIILysM mice compared with control (Fig. 4E). In addition, we found that tumor tissue of PyMT/TGFβRIILysM mice contain more fibroblasts (αSMA+F4/80−) with higher median fluorescence for pSMAD3 compared with control (Fig. 4F). These data are consistent with a hypothesis that CD73+ myeloid cells can negatively regulate TGFβ effects on tumor-associated fibroblasts. (Fig. 4G).
To test this hypothesis, we performed a gel contraction assay by mixing mouse mammary tumor fibroblasts with myeloid cells into gel, and after 24-hour set time, stimulated them with TGFβ (Fig. 4H). We found that fibroblasts' response to TGFβ was decreased when they were mixed with WT myeloid cells versus myeloid cells with deleted TGFβRII. CD73+ myeloid cells provide a vast supply of adenosine in tumor tissue (3). We have also demonstrated that CD73, which is expressed on myeloid cells, is upregulated by TGFβ signaling. To determine whether adenosine can be associated with observed decrease of TGFβ signaling on fibroblasts, we treated mouse mammary tumor fibroblasts with TGFβ and NECA (a stable analogue of adenosine). We found a significant downregulation of TGFβ-stimulated phosphorylation of SMAD2/3 and collagen accumulation (Fig. 4I). Incubation cells with adenylate cyclase activator Forskolin recapitulated this effect of adenosine, demonstrating a link between an adenosine A2 receptor–mediated signaling and a TGFβ signaling pathways. These data suggest that adenosine, generated by CD73+ myeloid cells–infiltrating tumor tissue may downregulate TGFβ effects on tumor-associated fibroblasts via activation of adenylate cyclase–driven adenosine receptors (A2 type). These myeloid cell interactions with fibroblasts resulting in downregulation of TGFβ on the latter could be associated with protumorigenic properties of CAF and represent a promising pharmacologic target.
Association of adenosine receptors and ECM components with outcome of patients with cancer
To gain insight into adenosine signaling in human breast cancers, we have used open source METABRIC datasets to assess the association between the expressions of adenosinergic-related genes (CD73, A2aR, and A2bR) and ECM-related genes. Analysis of METABRIC data showed that high expression of ADORA2B (low affinity A2b adenosine receptor) was associated with high expression of NT5E (Fig. 5A). We did not observe an association between ADORA2A (high affinity A2a adenosine receptor) and NT5E (CD73). In addition, we found that high expression of ADORA2B, but not ADORA2A, was associated with the lower expression of main ECM genes—COL1A1, COL1A2, COL11A1 (Fig. 5B). This finding demonstrates similarity to animal model pattern—increased adenosine signaling in conjunction with downregulated TGFβ signaling. Using TCGA data, we found that only patients with head and neck squamous adenocarcinoma have changes close to patients with breast cancer (Supplementary Fig. S1). Stratifying patients with breast cancer by stage and number of positive lymph nodes, we found that most significant changes appear in patients with stage I cancer or with 1–3 positive lymph nodes (Supplementary Fig. S2).
Analysis of survival of patients with breast cancer (http://kmplot.com) showed that the high expression of ADORA2B was associated with shorter survival for basal type and triple-negative type of breast cancer only. Importantly, ADORA2A correlative data is contrasting ADORA2B data in these cancer types, demonstrating adverse impact of A2B adenosine receptor in breast cancer that could be mediated through its ability to downregulate TGFβ response (Fig. 5C).
Discussion
In this study, we demonstrated that TGFβ signaling in myeloid cells impacts tumor architecture and malignant potential. In a mouse model of myeloid cell–specific TGFβ receptor II knockout, deletion of TGFβ signaling resulted in significant changes in ECM composition in tumor tissue, that is, increased amounts of collagen and laminin and decreased deposition of fibronectin. Given a profound role of TGFβ in regulation of collagen production, we sought to assess its amount and signaling status in tumor tissue. Although no difference in total and active TGFβ protein was detected in tumor tissue homogenates from control and PyMT/TGFβRIILySM mice, TGFβ signaling was increased in the latter group. Previously, we have demonstrated that TGFβ signaling in myeloid cells regulates CD73-mediated production of protumorigenic adenosine (3). In this study, we discovered the next branch of TGFβ tumorigenic mechanism mediated by CD73+ myeloid cells that decrease TGFβ signaling in fibroblasts resulting in formation of a less prominent tumor stroma. We demonstrated, for the first time, that stimulation of adenosine receptors directly disturbs TGFβ effects on cancer-associated fibroblasts resulting in altered fibroblasts functions that may affect tumor progression, invasion, and metastasis. In addition, we performed METABRIC human data analysis that revealed inverse correlation between expression of Col1 and A2B adenosine receptor, whereas A2A receptor expression was not associated with any of ECM genes. Moreover, using survival data of patients with breast cancer (http://kmplot.com), we found that patients with basal and triple-negative breast cancer, which also had a high level of A2B receptor expression had poor outcome; on the contrary, the A2A receptor expression was found to be associated with better outcome in these patients.
TGFβ is a well-known regulator of growth, differentiation, and migration, which are essential for tumor progression. Recently, multidimensionality of its effects came to be recognized with the realization that TGFβ exhibits actions on all cells within the tumor microenvironment, resulting in a complex net outcome. This outcome is dependent on a type and functional status of target cells and availability of partners, such as cytokines and active metabolites of signaling pathways (2). Numerous reports showed that the loss of TGFβ signaling in cancer-associated fibroblasts correlates with poor prognosis (10, 26), the same holds truth for the loss of its signaling in tumor cells (27–29). Our recent studies showed that the loss of TGFβ signaling on myeloid cells, on the contrary, reduces tumor growth and metastasis (3). This study provides insight into mechanism by which TGFβ, acting on myeloid cells, can contribute to cancer progression. Our results, which demonstrate a relative decrease in area occupied by late carcinoma in PyMT/TGFβRIILySM mice, are indicative of reduced rate of tumor progression in these mice, and provide in vivo evidence that TGFβ signaling in myeloid cells promotes tumor progression. Finding of enhanced TGFβ signaling in tumor tissue lysates and decreased expression of cancer-related genes like PI3K, JAK-STAT, Ras, and MAPK in PyMT/TGFβRIILySM mice provides evidence for overall low-grade tumors in these mice.
Matrix composition, amount, physical properties, and tumor cell–ECM interactions are crucial for cancer progression (21, 22, 30–33). Fibrillar collagen is the most prevalent ECM protein in the body (34), contributing to mechanical properties of normal tissues and tumors (35). While changes occurring in organization and mechanical properties of ECM in breast cancer are reported to correlate with poor outcome (21, 36–38), the mechanisms of such changes remain poorly understood. Overall collagen accumulation is considered to be a hallmark of breast cancer progression (39). However, our observation of increased deposition of Collagen 1 in PyMT/TGFβRIILySM mice is challenging this dogma. In contrast, another ECM protein, fibronectin, was significantly decreased in late carcinoma regions of PyMT/TGFβRIILySM tumors, compared with corresponding regions of PyMT/TGFβRIIWT tumors. Fibronectin has been shown to be produced by both cancer-associated fibroblasts and cancer cells themselves. Generally, fibronectin expression in breast cancer is associated with poor clinical outcome (3, 22). Thus, the data presented here suggest that TGFβ signaling in myeloid cells regulates a balance of deposition of main ECM proteins—collagen and fibronectin, both of which are significantly involved in tumor progression.
Fibroblasts are principal cells maintaining ECM integrity in normal and diseased tissues (40). Differences in tumor stroma deposition in a mouse model with deleted TGFβ signaling on myeloid cells, led us to hypothesize that TGFβ signaling in myeloid cells can affect functions of fibroblasts in tumor microenvironment contributing to observed changes.
We have previously demonstrated that TGFβ signaling in myeloid cells regulates the expression of CD73 ectoenzyme on mature myeloid cells (3). Tumor tissue, in comparison with normal tissue, is enriched with extracellular ATP (eATP), released from cells as a part of damage-associated molecular pattern (DAMP; ref. 41). The eATP, in general, exerts antitumorigenic and proinflammatory effects (42). However, eATP is degraded further by CD39/CD73 enzymatic pair to adenosine (43), a known protumorigenic metabolite (16). As a result, the concentration of adenosine in tumor microenvironment is elevated (44–46). There is strong evidence that adenosine modulates immune response during cancer progression, as well as behavior of neoplastic cells (16). In this study, we demonstrated that adenosine downregulates TGFβ-induced phosphorylation of SMAD2/3 and collagen synthases on fibroblasts, indicating a new interplay of adenosinergic and TGFβ signaling on another cell type, cancer-associated fibroblasts. Relatively high concentration of adenosine in tumor tissue can be maintained only with sustained activity of CD39/CD73 enzymatic pair (3). Using gel contraction assay, we detected an increased fibroblast response to TGFβ mixed with cells having downregulated expression of CD73. Therefore, we suggested that myeloid cells having ability to generate adenosine downregulate TGFβ-dependent functions of fibroblasts.
It has been demonstrated that adenosine forms immunosuppressive niche, enhances angiogenesis, and changes neoplastic cells behavior to more aggressive (16). We assume that under healthy condition, general biological role of observed phenomenon is to reduce destructing effect related with inflammation. Indeed, Ohta and coauthors observed similar adenosine-dependent effects on T cells in melanoma cancer. They suggested a mechanism that explains insufficient effects of T cells in tumor microenvironment where hypoxic condition and excessive cell death determine high concentration of adenosine in cancer tissue, which through A2A and A2B receptors drive T cells to acquire immunosuppressive phenotype (14, 15, 47). In this study, we demonstrated that stimulation of adenosine receptors directly affects TGFβ actions on cancer-associated fibroblasts, resulting in altered fibroblasts functions that may affect tumor progression, invasion, and metastasis. Adenosine can exert its actions by interacting with A1, A2A, A2B, or A3 subtypes of adenosine receptors of P1 purinergic family. Specific P1 receptor repertoire expressed on variety of cells, including fibroblasts, endothelial cells, immune cells, and tumor cells can be different, creating unique dynamic adenosinergic milieu. Both A2 subtypes of adenosine receptors increase intracellular cAMP by stimulation of adenylate cyclase, whereas A1 and A3 inhibit this enzyme (12, 13). Our observation that the adenylate cyclase activator forskolin, similar to adenosine, inhibits TGFβ-induced SMAD phosphorylation in fibroblasts suggests the involvement of A2 subtypes of adenosine receptors in adenosine/TGFβ pathway's cross-talk. To find association between A2 subtypes of adenosine receptors and cancer, we performed METABRIC human data analysis, which revealed inverse correlation between expression of Col1 and A2B adenosine receptors, whereas the expression ofA2A receptors was not associated with any of ECM genes. Moreover, using survival data of patients with breast cancer (http://kmplot.com), we found that patients with basal and triple-negative breast cancer, which also had a high level of A2B receptor expression, had poor outcome; on the contrary, the A2A receptor expression was found to be associated with better outcome in these patients.
It has been previously demonstrated that an increased expression of CD73 in human breast cancer correlates with poor outcome (44). Our findings that A2A and A2B expression is associated with ECM protein production and regulation of TGFβ signaling revealed a new mechanism by which adenosine may play a central role in tumor tissue matrix deposition, thus affecting tumor progression. Our data show that the high levels of A2B receptor expression in basal and triple-negative types of human breast tumors are associated with lower survival. This may indicate a potential role of A2B adenosine receptors in promoting tumor development via its ability to downregulate TGFβ response in fibroblasts demonstrated in our study.
There is an increased interest in development of pharmacologic approach to inhibit adenosine metabolism during tumor progression. In present times, anticancer clinical trials investigators mostly concentrate their attention on CD39, CD73, and A2A adenosine receptors (48–50). Our data show that A2B receptors can also be a valuable target in anticancer therapy. The primary strategy of adenosine receptors inhibition in anticancer treatment is blocking of adenosine signaling predominantly in immune cells because it works as immune checkpoint (12, 13, 44–46, 48, 51, 52). Our new data indicate that adenosine signaling in fibroblasts can play an important role in tumor progression. Our study showed that stimulation of adenosine receptors on fibroblasts can downregulate their TGFβ signaling and polarize them to cells with procancerogenic properties.
In summary, the interaction between myeloid cells and CAFs, demonstrated in our study, provides new insights into processes, which determine pleiotropic functions of TGFβ in tumor development. Better understanding of mechanisms that are critical for tumor microenvironment formation and evolution provides a basis for development of novel therapeutic approaches based on more precise targeting and manipulation of adenosinergic and TGFβ signaling.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: G. Vasiukov, T. Blackwell, I. Feoktistov, S.V. Novitskiy
Development of methodology: G. Vasiukov, A. Zijlstra, S.V. Novitskiy
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Vasiukov, T. Novitskaya, A. Zijlstra, S.V. Novitskiy
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Vasiukov, P. Owens, F. Ye, Z. Zhao
Writing, review, and/or revision of the manuscript: F. Ye, Z. Zhao, H.L. Moses, I. Feoktistov, S.V. Novitskiy
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.V. Novitskiy
Study supervision: S.V. Novitskiy
Acknowledgments
This work was supported by NIH grant R01CA200681 (to S.V. Novitskiy).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.