Cancer-associated fibroblast-derived Gremlin 1 promotes breast cancer progression

Background Bone morphogenetic proteins (BMPs) have been reported to maintain epithelial integrity and to antagonize the transforming growth factor β (TGFβ)-induced epithelial to mesenchymal transition. The expression of soluble BMP antagonists is dysregulated in cancers and interrupts proper BMP signaling in breast cancer. Methods In this study, we mined the prognostic role of BMP antagonists GREMLIN 1 (GREM1) in primary breast cancer tissues using in-house and publicly available datasets. We determined which cells express GREM1 RNA using in situ hybridization (ISH) on a breast cancer tissue microarray. The effects of Grem1 on the properties of breast cancer cells were assessed by measuring the mesenchymal/stem cell marker expression and functional cell-based assays for stemness and invasion. The role of Grem1 in breast cancer-associated fibroblast (CAF) activation was measured by analyzing the expression of fibroblast markers, phalloidin staining, and collagen contraction assays. The role of Grem1 in CAF-induced breast cancer cell intravasation and extravasation was studied by utilizing xenograft zebrafish breast cancer (co-) injection models. Results Expression analysis of clinical breast cancer datasets revealed that high expression of GREM1 in breast cancer stroma is correlated with a poor prognosis regardless of the molecular subtype. The large majority of human breast cancer cell lines did not express GREM1 in vitro, but breast CAFs did express GREM1 both in vitro and in vivo. Transforming growth factor β (TGFβ) secreted by breast cancer cells, and also inflammatory cytokines, stimulated GREM1 expression in CAFs. Grem1 abrogated bone morphogenetic protein (BMP)/SMAD signaling in breast cancer cells and promoted their mesenchymal phenotype, stemness, and invasion. Moreover, Grem1 production by CAFs strongly promoted the fibrogenic activation of CAFs and promoted breast cancer cell intravasation and extravasation in co-injection xenograft zebrafish models. Conclusions Our results demonstrated that Grem1 is a pivotal factor in the reciprocal interplay between breast cancer cells and CAFs, which promotes cancer cell invasion. Targeting Grem1 could be beneficial in the treatment of breast cancer patients with high Grem1 expression. Electronic supplementary material The online version of this article (10.1186/s13058-019-1194-0) contains supplementary material, which is available to authorized users.


Background
Although carcinomas, which account for approximately 90% of human cancers, are derived from the epithelia, the tumor stroma exerts a powerful influence on cancer behavior, such as tumor cell growth, invasion, metastasis, and evading immune responses. The tumor stroma consists of cancer-associated fibroblasts (CAFs); vascular, inflammatory, and immune cells; and extracellular matrix (ECM) residing within or in the vicinity of a tumor [1]. CAFs are differentiated from quiescent fibroblasts and are associated with increased expression of myofibroblastic markers, such as vimentin, αsmooth muscle actin (αSMA), fibroblast activation protein (FAP), and fibroblast-specific protein 1 (FSP1, also known as S100A4) [2]. Tumors, including those from the breast, often display desmoplasia (a fibrillar network) that is mainly caused by CAFs, in that they produce and remodel ECM components, including collagen, fibronectin, and laminin [3]. The increased stiffness and abnormal physical structure of the ECM can promote tumor cell growth and metastatic dissemination and are also critical for the generation and maintenance of the CAF phenotype [3].
Bone morphogenetic proteins (BMPs) are secreted growth factors that belong to the transforming growth factor β (TGFβ) family [4]. Signaling by BMPs is initiated by binding their cognate transmembrane serine/ threonine kinase receptors, which triggers the phosphorylation of intracellular SMAD1/5/8 (R-SMADs). Activated R-SMADs can form heteromeric complexes with SMAD4 that accumulate in the nucleus, where they can regulate transcriptional responses in concert with other DNA-binding transcription factors [4]. BMP signaling can elicit diverse and complex biological processes in development and disease, including cancer [5]. Many secreted BMP antagonists, which sequester BMP ligands and prevent their binding to receptors, have been identified [6]. Accumulating evidence indicates that several cancer types show dysregulated BMP signaling caused by a disequilibrium of BMPs and their antagonists. For example, BMP antagonists such as Noggin, Follistatin, and Chordin like (Chrdl)1 have been linked to induce osteoclast differentiation and promote osteolytic bone metastases [7,8]. The BMP antagonist Coco permits a few dormant breast cancer cells to escape the quiescent state imposed by BMP signaling and thereby establish metastases [9].
In several cancers, Grem1 reduces the negative effect of BMPs on stemness, proliferation, migration, and invasion of cancer cells [22][23][24]. In hereditary mixed polyposis syndrome, GREM1 is predominantly expressed in the epithelium of the large bowel, where it disrupts homeostatic intestinal morphogen gradients and initiates colonic tumorigenesis [25,26]. GREM1 was also detected at the colorectal cancer desmoplastic invasion front, highlighting a potential role in cancer metastasis [27]. High levels of GREM1 gene expression were observed in the stromal fibroblasts of many types of cancer [23,28,29], suggesting that CAFs are a potential source of Grem1. However, the effects of Grem1 on CAFs' function and on the interaction between (breast) cancer cells and fibroblasts are unclear.
The results presented here support the idea that Grem1 is a clinical predictor of a poor prognosis in breast cancer. Mechanistically, Grem1 produced by CAFs promoted fibroblast activation in an autocrine manner and stimulated breast cancer cell stemness and invasion in a paracrine manner. Grem1 could be an attractive therapeutic target to interfere with breast cancer progression.

Data mining of gene expression in clinical patient samples and 52 breast cancer cell lines
In-house and publicly available gene expression datasets GSE2034 [30], GSE5327 [31], GSE2990 [32], GSE7390 [33], and GSE11121 [34] were used for GREM1 (and transforming growth factor β1/2/3 (TGFB1/2/3), interleukin 1β (IL1B), and tumor necrosis factor α (TNFA)) expression in lymph node-negative, non-(neo-)adjuvant-treated primary breast cancer patients with available metastasis-free survival data, leading to a cohort of 867 patients. Using the GSE41313 dataset [35], GREM1, BMP, and BMP receptor expression was assessed in silico in 52 breast cancer cell lines. Breast cancer dataset GSE14548 [28] was investigated to explore the GREM1 expression in breast epithelium and stroma; this dataset was obtained using tissues from normal breast; grade I, II, and III ductal carcinoma in situ (DCIS), and invasive breast cancer tissues that were micro-dissected using a laser capture technique. In addition, the colorectal cancer dataset GSE39396 [36] was analyzed for GREM1 expression; epithelial cells, leukocytes, fibroblasts, and endothelial cells were isolated by flow cytometry. Data were gathered from Gene Expression Omnibus (http:// www.ncbi.nlm.nih.gov/geo/). Raw.cel files were processed using frozen robust multiarray analysis (fRMA) parameters (median polish) [37], after which the batch effects were corrected using ComBat [38].

GREM1 RNA in situ hybridization (ISH)
A matched breast cancer, adjacent (adenosis or hyperplasia, and cancer free) and adjacent normal tissue microarray (TMA) was purchased from US Biomax (BR724). GREM1 RNA in situ hybridization was conducted with an RNAscope GREM1 Probe (312831-C2, Advanced Cell Diagnostics) and a 2.5 HD Detection Kit -BROWN (322300, Advanced Cell Diagnostics). All procedures were performed by strictly following the manufacturer's instructions. The ISH results were scanned by a digital slide scanner (Pannoramic 250 Flash III, 3DHISTECH). The presence of intracellular brown punctate dots was considered as positive staining. The signal intensity was scored utilizing a 5-point system: 0, no signals visible; 1, weak signals barely visible; 2, visible signals but not intensive; 3, moderate intensive signals; and 4, intensive signals. Scoring was evaluated independently by two observers with similar outcomes.

Plasmids, lentiviral transduction, and generation of stable cell lines
The human GREM1 complementary DNA (cDNA) was cloned from cDNA by PCR and inserted into the pCDH lentiviral vector. pLV-mCherry has been described by our laboratory before [41]. pUltra-Smurf (blue fluorescent protein AmCyan) was obtained from Addgene (48974, Addgene). Human GREM1 lentiviral shRNAs were obtained from the Sigma MISSION shRNA library. Five shRNAs were tested, and the two most effective shRNAs TRCN0000063833 (sh#1) and TRCN0000063837 (sh#2) were used.
Lentiviruses were produced by co-transfecting cDNA expression plasmids or shRNAs with helper plasmids pCMV-VSVG, pMDLg-RRE (gag/pol), and pRSV-REV into HEK293T cells using polyethyleneimine (PEI). Cell supernatants were harvested 48 h after transfection and stored at − 80°C. MDA-MB-231 and MCF7 cells were labeled with mCherry by infecting for 24 h with mCherryexpressing lentiviral supernatants diluted 1:1 with normal culture medium in the presence of 5 ng/ml of polybrene (107689, Sigma-Aldrich). Forty-eight hours after infection, cells were placed under neomycin (A1720, Sigma-Aldrich) selection. 19TT and W21 cells were labeled with AmCyan and subjected to positive fluorescence-activated cell sorting (FACs). To obtain GREM1 stable expressing cell lines, M1, M2, MDA-MB-231, and W21 cells were infected and selected with puromycin (P9620, Sigma-Aldrich). Puromycin was used at 1 μg/ml to maintain selection pressure. After infection with GREM1 targeting shRNAs, 19TT cells were used within short term as 19TT cells are puromycin resistant already.

Stimulation with conditioned medium (CM) or cytokines
MDA-MB-231 and MCF7 cells were grown to 70-80% confluency, washed two times with PBS, and incubated in serum-free DMEM for 24 h. Conditioned medium (CM) was then collected and passed through a 0.45-mm syringe filter (SLHP033RB, Merck Millipore).

Quantitative real-time PCR (qRT-PCR)
Total RNAs were isolated using the NucleoSpin RNA II kit (740955, BIOKE´). A total of 1 μg of RNA was reverse transcribed using the RevertAid First Strand cDNA Synthesis Kit (K1621, Thermo Fisher Scientific). Real-time PCR was conducted with GoTaq qPCR Master Mix (A6001, Promega) using CFX Connect Detection System (1855201, Bio-Rad). All target gene expression levels were normalized to GAPDH. The sequences of primers used to detect target human genes in qRT-PCR were listed in Additional file 1 Table S1.

CAGA-luciferase reporter assay
HEK293T cells were seeded on a 24-well plate at approximately 5 × 10 4 cells per well. The next day, cells in each well were co-transfected with 0.1 μg TGFβ/ SMAD-inducible (CAGA) 12 luciferase transcriptional reporter construct [42] and 0.08 μg β-galactosidase expression construct using PEI. After overnight incubation, cells were starved with serum-free medium. Eight hours later, serum-free media were removed and replaced by CM from breast cancer cell lines. One nanogram per milliliter TGFβ3 treatment was performed as a standard. After another overnight incubation, luciferase and β-galactosidase activities were measured. The luciferase activity was normalized based on the β-galactosidase activity.

Collagen gel contraction assays
The contraction assay [44] was performed to evaluate the contractility of 19TT cells with GREM1 knockdown or GREM1-overexpressing W21 cells. Collagen gels were prepared by mixing fibroblast cell suspensions in serumfree medium and type I collagen (Corning, 354249) solution. The final cell density was 2.0 × 10 5 cells/ml with 1 mg/ml collagen. A 0.5-ml mixture was casted into each well of a 24-well plate and allowed to polymerize for 30 min at 37°C. Following gelatinization, another 0.5 ml of serum-free DMEM was added to the gel. Changes of gels were recorded by using a ChemiDoc Imaging System (17001402, Bio-Rad) at a fixed distance above the gels at 24, 48, and 72 h. The surface area of the gels was quantified by ImageJ software. The percentage of contraction was calculated using the formula 100% × (well surface area − gel surface area)/well surface area.
Three-dimensional spheroid invasion assay mCherry-labeled MDA-MB-231 or MCF7 cells and coculture (1:1 mixture) with W21 or 19TT groups were prepared at 1000 cells/ml in complete DMEM. Drops of the single-cell suspension (30 μl) were placed onto the lids of 10-cm dishes, which were inverted over dishes containing 10 ml PBS. Hanging drop cultures were incubated 7 days allowing sufficient sedimentation and formation of one spheroid per drop. Images were taken by an inverted fluorescent microscope (DMi8, Leica Microsystems).
The 3D spheroid invasion assay was performed according to our previous study [45] with slight modifications. Single spheroids were embedded in the center of each well of a flat-bottom 96-well plate pre-coated with 50 μl of collagen mixture. Type I collagen (354249, Corning) was neutralized according to the manufacturer's protocol. The collagen mixture was prepared by diluting neutralized collagen with serum-free medium to a final concentration of 2 mg/ml. Eight spheroids generated by each experimental setting were randomly chosen for embedding. After spheroid embedding, another 50 μl of collagen mixture was overlaid onto the collagen matrix in each well. The plate was incubated for 30 min at 37°C to solidify the gels. Thereafter, 50 μl of serum-free medium was added to each well to prevent the surface from dehydrating. Plates were placed under standard cell culture conditions. Images were taken at days 0, 2, and 4 after embedding by using inverted fluorescence microscopy (DMi8, Leica Microsystems). The invasion was quantified by measuring the area occupied by cells using ImageJ software.

Embryonic zebrafish intravasation and extravasation assay
Zebrafish xenograft breast cancer cell experiments were performed by injecting fluorescently labeled breast cancer cells into embryos at 48 h post-fertilization as described before [41]. Briefly, approximately 400 mCherry-labeled MDA-MB-231 cells were injected into the perivitelline space or the duct of Curvier (DoC) of transgenic zebrafish embryos (fli: enhanced green fluorescent protein (EGFP)), whose vasculature is marked in green. For co-injection, mCherry-labeled MDA-MB-231 cells and AmCyan-labeled W21 or 19TT cells were mixed at a ratio of 1:1. Then, approximately 400 mixed cells were injected into the zebrafish perivitelline space. Zebrafish embryos were maintained at 34°C after injection, a compromise for both the fish and the human cell lines. Three days post-injection (dpi) into the perivitelline space, the MDA-MB-231 cells that intravasated from the cell mass toward the embryonic fish body within the head and tail regions were imaged and counted under a confocal microscope (SP5 STED, Leica Microsystems). At 5 dpi into the DoC, the number of MDA-MB-231 cells that extravasated individually from the circulation into the collagen fibers of the tail fin or the number of clusters formed by M2 cells collectively was analyzed. At least 200 zebrafish embryos were injected for each condition. After verification by microscopy, only correctly injected and viable zebrafish were used for experimental analysis. All experiments were repeated at least two times independently, and representative experiments are shown.

Statistical analysis
Statistical analyses were performed using GraphPad Prism version 7.0. Numerical data from triplicates are presented as the mean ± standard deviation (s.d.), except for the analysis of zebrafish experiments, where a representative result is expressed as the mean ± standard error (s.e.m). Experiments were analyzed with an unpaired Student's t test. P < 0.05 was considered statistically significant.

Results
High GREM1 expression in breast tumors is associated with a poor prognosis BMPs have been reported to maintain epithelial integrity and to antagonize TGFβ-induced epithelial to mesenchymal transition (EMT), an important process for cancer cell invasion and metastasis [5]. Many soluble BMP antagonists have been described to be misexpressed and to interrupt proper BMP signaling in breast cancer [7][8][9]. We examined the prognostic role of soluble BMP antagonists in primary breast cancer using an in-house and publicly available cohort of 867 untreated lymph nodenegative breast cancer patients (see the "Methods" section for datasets that were used). The median follow-up time of metastasis-free survival (MFS) was 94.1 months (range from 1 to 299.4 months). High expression of GREM1 was found to be associated with a poor prognosis among all BMP antagonists that were examined. As shown in Fig. 1a, according to the GREM1 mRNA expression level, the subjects were divided evenly into three quantiles: low, middle, and high. GREM1 expression was inversely associated with MFS in this cohort, i.e., higher expression, and poorer outcome (low vs high: HR (hazard ratio) = 1.35, 95% confidence interval (CI) 1.15-1.57, log rank P = 0.00018; low vs middle: HR = 1.41, CI 1.02-1.96, P = 0.036; middle vs high: HR = 1.31, CI 0.98-1.74, P = 0.065). A similar association was observed when dividing Pearson's coefficient tests were performed to assess the statistical significance subjects into two quantiles (Additional file 2: Figure S1a). Furthermore, high expression of GREM1 correlated with a poor prognosis in all the breast cancer molecular subtypes examined: human EGF receptor (HER)2 + , triple − , estrogen receptor (ER) + , and ER − (Additional file 2: Figure S1b-e). Therefore, GREM1 is a poor prognostic marker of metastasis-free survival in breast cancer regardless of the subtype.

GREM1 is expressed in cancer-associated fibroblasts
When we examined the GREM1 expression in 52 human breast cancer cell lines by mining previously published datasets (see the "Methods" section), we found that only 3 breast cancer cell lines express low (MDA-MB-436 and HCC38) or intermediate (SUM149PT) levels of GREM1; all other 49 cell lines had no detectable expression (Fig. 1b). To explore the source of GREM1 expression, we stained GREM1 RNA by using in situ hybridization (ISH) in a breast cancer TMA, which comprised 24 matched cases of invasive ductal carcinoma, adjacent tissue, and adjacent normal tissue. As shown in Table 1 and Fig. 1c, we identified variable amounts of GREM1 expressed in fibroblast-like cells, i.e., CAFs, whereas there were no detectable levels of GREM1 in cancer-adjacent normal tissues or adjacent cancer-free breast tissues. None of the epithelial cells of breast cancers included in this study showed GREM1-positive expression. The GREM1 expression in breast cancer tissue samples is thus mainly caused by the presence of tumor stroma. Moreover, using the inhouse and publicly available primary breast cancer datasets, we observed a significant positive correlation between GREM1 and markers for CAFs and tumor matrix stiffness/desmoplasia, such as FAP, fibronectin (FN)1, fibrillin (FBN)1, collagen (COL)1A1, thrombospondin (THBS)2, and a-actin (ACTA)2 (Fig. 1d, Additional file 2: Figure S1f). Taken together, these results suggest that CAF-derived Grem1 might play a pivotal role in promoting breast tumor progression.

TGFβ secreted by cancer cells and inflammatory cytokines induces GREM1 expression
Analysis of GREM1 in tissue sections revealed that only the CAFs in close proximity to the cancer cells (tumor-stromal interface) showed high GREM1 RNA expression (Fig. 1c, bottom panel). We therefore explored the possibility that fibroblasts express GREM1 in response to factors secreted by cancer or inflammatory cells. We first collected CM from M1 immortalized normal breast cells and breast cancer cell lines MCF7 and MDA-MB-231. Treatment of 19TT CAFs (Fig. 2a) or W21 MSCs (Additional file 3: Figure S2a) with MDA-MB-231 or MCF7 cell CM resulted in a significant increase in GREM1 mRNA levels. There was no effect of M1 CM on GREM1 expression. To explore the factors that are responsible for inducing GREM1 expression in CAFs, we analyzed by data mining the expression of TGFB1/2/3 and inflammatory cytokines in breast cancer cell lines as well as in breast cancer tissues. We found that TGFB1/2/3 are highly expressed in both breast cancer cell lines and tissues. Inflammatory cytokines, including IL1B and TNFA, were expressed in breast cancer tissues but only at very low levels in breast cancer cell lines (Fig. 2b, c). IL1B and TNFA expression in breast cancer tissues are thus likely caused by the stromal cells present in breast cancer tissue samples. Challenging 19TT CAFs (Fig. 2d) or W21 MSCs (Additional file 3: Figure S2b) with TGFβ3, TNFα, and IL1β promoted GREM1 mRNA expression. Next, we analyzed whether TGFβ is secreted by cancer cells. MDA-MB-231 and MCF7, but not M1, were found to express active TGFβ (Fig. 2e). Importantly, the GREM1 expression-inducing activity of MDA-MB-231 or MCF7 cells could be blocked by a TGFβ-neutralizing antibody (Fig. 2f). Taken together, TGFβ secreted by cancer cells is the main determinant for inducing GREM1 expression by CAFs. Within the tumor-stroma niche, inflammatory cells secreting cytokines may also contribute to GREM1 expression by CAFs.

Grem1 increases mammosphere formation
BMPs are reported to be inhibitors of cell stemness, suggesting that secreted Grem1 might oppositely affect the stem traits [9,22,46]. First, we confirmed that BMPs and BMP receptors are indeed expressed in breast cancer cell lines (Fig. 1b, Additional file 4: Figure S3a and b). Then, mammosphere formation assay was performed to assess the effect of Grem1 on mammary stem cell activity in vitro. GREM1overexpressing M1 cells exhibited twofold more sphere formation compared to control cells in each of two subsequent passages (Fig. 3a). The administration of exogenous rhGrem1 showed a similar effect on mammosphere formation of M1 cells, whereas the administration of exogenous BMP2 mitigated sphere formation ability. The latter could be reversed by the concurrent administration of rhGrem1 (Fig. 3b). The surface expression of CD44 +/high CD24 −/low cells has been considered a stem population marker of breast cancers or cell lines [46]. Flow cytometry analysis demonstrated a significant increase in the  CD44 +/high CD24 −/low cell subpopulation in GREM1overexpressing M1 cells compared to the control (Fig. 3c). qRT-PCR revealed that GREM1 OE (Fig. 3d) or rhGrem1 (Additional file 4: Figure S3c) increased the expression of transcriptional regulators YAP, TAZ, SOX2, and OCT4, which have been implicated in maintaining breast cancer stemness. Moreover, M1 cells treated with rhGrem1 or the BMP type I receptor inhibitor LDN193189 [47] also displayed more CD44 +/high CD24 −/low cells than nontreated control cells (Fig. 3e). These results suggest that Grem1 enhances the mammosphere formation of M1 cells by repressing BMP signaling.

Grem1 promotes breast cancer cell invasion
To further characterize the role of Grem1 in breast cancer, we stably expressed Grem1 in the breast cancer cell lines M2 and MDA-MB-231 with a lentiviral vector. In a way, these transfected cell lines are somewhat reminiscent to the few breast cancer cell lines that express GREM1. In these GREM1-overexpressing cell lines, BMP-induced SMAD1/5/8 phosphorylation (Fig. 4a) and expression of BMP target genes ID1 and 3 (Fig. 4b) were clearly inhibited. Notably, the mRNA levels of the mesenchymal markers SLUG, SNAI1, VIM, and NCAD were increased by ectopic GREM1 expression (Fig. 4c, d) or exogenous rhGrem1 treatment (Additional file 5: Figure S4b), suggesting that Grem1 induces a slightly more mesenchymal phenotype in these breast cancer cells. To test whether exposure to Grem1 also results in more invasive behavior, we introduced these cells into the blood circulation of embryonic zebrafish via DoC injection and examined extravasation 5 days post-injection (dpi) in the avascular tail fin area. Compared to the vector control, the GREM1 overexpression group showed a higher number of extravasated M2 cell clusters (Fig. 4e) or MDA-MB-231 single cells (Fig. 4f). The BMP/SMAD signaling could be inhibited by exogenous administration of rhGrem1 (Additional file 5: Figure S4b). Next, we injected MDA-MB-231 cells suspended in PBS supplemented with or without rhGrem1 into the perivitelline space of embryonic zebrafish and examined the level of cells in circulation at 3 dpi. Exogeneous rhGrem1 increased cellular intravasation significantly, as more cells were found in the head and tail regions of zebrafish embryos (Fig. 4g).

Grem1 promotes fibroblast activation
Grem1 is associated with fibrosis [17][18][19]. To explore the role of Grem1 in fibroblast activation, we first compared GREM1 mRNA expression levels in foreskin fibroblasts, 19TT breast cancer CAFs, and HM, W18, and W21 human mesenchymal stem cells (MSCs). M2, MCF7, and MDA-MB-231 cancer cells served as a negative control. MSCs, which are considered fibroblast precursors and can differentiate into fibroblasts [2], showed the lowest expression of GREM1; GREM1 expression in the foreskin fibroblasts, which represent normal fibroblasts, was significantly higher than that in MSCs, and 19TT CAFs showed the highest levels (Fig. 5a), indicating that GREM1 expression increases during different stages of fibroblast activation. We next knocked down GREM1 in 19TT CAFs. As shown in Fig. 5b, two shRNAs-mediated GREM1 knockdown increased the mRNA expression of both ID1 and ID3 and decreased the expression of TGFβ signaling components and their target genes (plasminogen activator inhibitor (PAI-1), fibroblast activation markers, and matrix metalloproteinases (MMPs). GREM1 knockdown in 19TT CAFs also led to decreased protein levels of FN1, S100A4, collagen I, FAP, and αSMA (Fig. 5c). This result suggests that Grem1 is a pivotal factor in fibroblast activation.
To examine whether Grem1 affects the cytoskeletal changes, we stained the cells with fluorescein-conjugated phalloidin to visualize filamentous (F)-actin. Indeed, GREM1 knockdown in 19TT CAFs resulted in less prominent stress fibers and less organized bundles in the cytoplasm (Fig. 5d). More significantly, the ability of 19TT CAFs to contract collagen gels (a 3D model widely used for evaluating fibroblast-mediated matrix remodeling capacity) decreased significantly with GREM1 knockdown (Fig. 5e). Moreover, GREM1 overexpression in (See figure on previous page.) Fig. 2 TGFβ secreted by breast cancer cells and inflammatory cytokines induce GREM1 expression in CAFs. a GREM1 expression in 19TT CAFs after treatment with conditioned medium (CM) from breast cell lines (M1, MDA-MB-21, or MCF7). The expression was normalized to the parallel time control of normal medium treatment. The results are expressed as the mean ± s.d, n = 3. Student's t test, *P < 0.05, ***P ≤ 0.001. b TGFB1, TGFB2, TGFB3, TNFA, and IL1B mRNA levels in 52 breast cancer cell lines. The expression levels were categorized into four groups: background, low, intermediated, and high. c TGFB1/2/3, TNFA, and IL1B expression in primary breast cancer samples. The expression level was categorized into four groups: background, low, intermediated, and high. d TGFβ3 (5 ng/ml), TNFα (10 ng/ml), or IL1β (10 ng/ml) induce GREM1 expression in 19TT cancer-associated fibroblasts (CAFs). Expression was normalized to the parallel time control of buffer treatment. The results are expressed as the mean ± s.d., n = 3. Student's t test, *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001. e Measurements of TGFβ activity in CM from breast cancer cell lines using a CAGA luciferase (LUC) reporter assay in HEK293T cells as a readout. TGFβ-neutralizing antibody (10 ng/ml) was added to demonstrate that luciferase activity in CM is due to the TGFβ activation and not activins or nodal. Recombinant TGFβ was added to control for the functionality of the assay. The value is normalized to β-galactosidase (βGal) activity. The results are expressed as the mean ± s.d, n = 3. Student's t test, ***P ≤ 0.001. f The induction of GREM1 expression in 19TT CAFs by CM from MCF7 and MDA-MB-21 is blocked by TGFβ-neutralizing antibody. The results are expressed as the mean ± s.d, n = 3. Student's t test, **P ≤ 0.01 Fig. 3 Grem1 maintains stemness in M1 cells. a GREM1 overexpression (OE) induces more mammosphere formation in M1 cells. Left, representative images of mammospheres at 7 days. Right, the number of spheres formed per 1000 cells plated. The primary spheres were disintegrated and replated further. Secondary spheres formed were counted. The results are expressed as the mean ± s.d., n = 3. Student's t test, **P ≤ 0.01. b Pro-mammosphere formation ability of recombinant human Grem1 (rhGrem1) protein (500 ng/ml) can be neutralized by BMP2 (50 ng/ml). Left, representative images of spheres at 7 days; right, the number of spheres formed per 1000 cells plated. The results are expressed as the mean ± s.d., n = 3. Student's t test, **P ≤ 0.01. c Flow cytometry analysis shows that GREM1 OE in M1 cells increases the stem population (CD44 +/high CD24 −/low ). d GREM1 OE in M1 cells upregulates stem cell transcription factors. GAPDH was used as an internal control. The results are expressed as the mean ± s.d., n = 3. Student's t test, *P < 0.05, ***P ≤ 0.001. e Flow cytometry analysis showing that 2 days of treatment with rhGrem1 (500 ng/ml) or the BMP type I receptor inhibitor LDN193198 (120 nM) also leads to an increase in the CD44 +/high CD24 −/low population   Figure S5a) induced intensive myofibroblast-like characteristics (Additional file 5: Figure S5b-e). Consistent with this result, W21 MSCs treated with rhGrem1 or the selective BMP receptor kinase inhibitor LDN193189 exhibited an upregulation of genes which were inhibited by GREM1 knockdown in 19TT CAFs (Additional file 6: Figure S5f). Overall, these observations imply that Grem1 is closely associated with the fibrogenic phenotype of breast CAFs.

Fibroblast-derived Grem1 promotes breast cancer cell invasion in a 3D spheroid model
Previous studies have indicated that CAFs are propellants of cancer cell invasion [1,2]. Prompted by the profibrotic role of Grem1, we further explored the roles of Grem1 in fibroblast-mediated cancer cell invasion using a 3D spheroid model. As illustrated in Additional file 7: Figure S6a, spheroids were produced from hanging drop co-cultures of mCherry-labeled breast cancer cells, MCF7 or MDA-MB-231 cells, and AmCyan-labeled 19TT CAFs. CAFs with or without GREM1 knockdown were mixed with these breast cancer cells at a 1:1 ratio. The various resulting spheroids were embedded in collagen gel. As shown in Additional file 7: Figure S6b, the monocultured MCF7 spheroid showed a collective cell invasion phenotype in collagen, and in the presence of 19TT cells, the increased invasion of CAFs was measured at day 4. However, upon GREM1 depletion in the 19TT CAFs, the co-culture spheroids showed strongly reduced invasion (Fig. 6a). Likewise, GREM1 knockdown in the CAFs reduced the invasion of MDA-MB-231 cells in MDA-MB-231 and 19TT co-culture spheroids at days 2 and 4 (Fig. 6b).

Fibroblast-derived Grem1 promotes breast cancer cell intravasation
Next, we examined the role of fibroblast-expressed Grem1 in breast cancer cell invasion in vivo. We injected mCherry-labeled MDA-MB-231 cells into the perivitelline space of zebrafish in the absence or presence of either AmCyan-labeled W21 MSCs, foreskin fibroblasts, or 19TT CAFs. As depicted in Fig. 7a, intravasation of the MDA-MB-231 cells was significantly increased when co-implanted with W21, validating a previous study in which MSCs promoted cancer metastasis [48]. Importantly, this intravasation was much more enhanced by the foreskin fibroblasts and even more so by the 19TT CAFs, suggesting a correlation with their GREM1 expression level. Indeed, the ectopic expression of GREM1 in W21 cells resulted in enhanced MDA-MB-231 cell intravasation upon co-injection (Additional file 8: Figure S7). Consistent with this result, GREM1 knockdown mitigated the promotion role of 19TT CAFs on MDA-MB-231 cell intravasation (Fig. 7b).

Discussion
Our work has uncovered a strong association between high GREM1 expression in breast tumor biopsies and a poor prognosis. We provide mechanistic insights into GREM1's key role in facilitating breast cancer progression using in vitro and in vivo studies. Grem1 is highly expressed by CAFs at the invasion front; its expression can be promoted by factors, such as TGFβ released by breast cancer cells and inflammatory cytokines. Grem1 mediates the fibrogenic activation of CAFs in an autocrine manner. Grem1 has a direct effect on cancer cell invasion and stemness, evidenced by the fact that it promoted a slightly more mesenchymal/stemness phenotype in breast cancer cells. It could also contribute indirectly to this process via its potent effects on fibroblast activation. In this way, Grem1 promotes the formation of a microenvironment conducive to breast cancer cell invasion. Thus, Grem1 is a key determinant of the mutual interplay between breast cancer cells and CAFs (Fig. 7c).
Although we found an association between Grem1 and poor breast cancer prognosis, the prognostic significance of Grem1 in different cancer types is not consistent. For example, Grem1 expression correlates with progression-free survival in pancreatic neuroendocrine tumors [49] and colorectal cancer [50], but it is an indicator of poor progression-free survival in cervical cancer [51]. Grem1 may have different roles in different tumor types, but this may be dependent on the experimental setup, the analysis of expression in complete tumors vs stromal expression specifically, and/or the determining the levels of RNA vs protein. For instance, when testing (See figure on previous page.) Fig. 5 GREM1 knockdown in 19TT breast cancer-associated fibroblasts (CAFs) attenuates fibrotic characteristics. a qRT-PCR comparison of relative GREM1 expression in M2, MCF7, MDA-MB-231, HM, W18, and W21 fetal mesenchymal stem cells (MSCs), foreskin fibroblasts, and 19TT CAFs. GAPDH was used as an internal control. The results are expressed as the mean ± s.d., n = 3. Student's t test, **P ≤ 0.01, ***P ≤ 0.001. b qRT-PCR analysis of the selected genes, BMP targets, TGFβ pathway constituents/targets, fibroblast activation markers, and matrix metalloproteinases in 19TT CAFs with/ without shRNA-mediated GREM1 knockdown. GAPDH was used as an internal control. The results are expressed as the mean ± s.d., n = 3. Student's t test, *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001. c Western blot analysis to detect the changes in indicated proteins after GREM1 knockdown in 19TT CAFs. d 19TT CAFs with/without GREM1 knockdown were stained with fluorescein-phalloidin (green) to visualize F-actin. DAPI was used for nuclear staining (blue). e Collagen gel contraction assay. 19TT CAFs with/without GREM1 knockdown were embedded in collagen gels. After 24, 48, and 72 h, the area of each gel (white dash circle) was imaged and quantified. Left, representative images of contracted gels. Right, the percentage of gel contraction. Quantification is shown in the "Methods" section. The results are expressed as the mean ± s.d., n = 3. Student's t test, **P ≤ 0.01 commercial antibodies on tissue sections, including sections of GREM1-deficient animals, we found that the detected signals may not have been specific for Grem1 (data not shown). To avoid these putatively non-specific measurements, we determined GREM1 mRNA levels by in situ hybridization.
The mRNA detection method revealed that GREM1 was exclusively expressed by CAFs. Our findings are supported by our data mining of publicly available datasets. We analyzed breast cancer dataset GSE14548 generated by Ma and colleagues [28], which separated epithelial and stromal tissues, and in this dataset, GREM1 was found mainly expressed in the (invasive) breast cancer stroma, and there was no GREM1 expression observed in normal epithelium and stroma (Additional file 9: Figure. S8a). In addition, we mined a colon cancer dataset GSE39396 in which epithelial cells, leukocytes, fibroblasts, and endothelial cells were separately isolated by FACs and thereafter profiled. Consistent with our results, only fibroblasts were found to express GREM1 (Additional file 9: Figure S8b).
We found that GREM1 expression in CAFs is particularly high in close vicinity of cancer cells. This is consistent with previous reports in which Grem1 was found to be highly expressed in CAFs in the microenvironment of basal cell carcinoma (and other tumors) compared to normal tissue counterparts [23] and a study of colorectal Right, relative invasion area was quantified as the area difference at days 2 and 4 relative to that at day 0. The results are expressed as the mean ± s.d., n = 8. Student's t test, *P < 0.05, **P ≤ 0.01 cancer, in which Grem1 was found to be expressed at the invasion fronts in CAFs and to mediate the loss of cancer cell differentiation [3]. We identified TGFβ secreted by cancer cells as a strong driver of GREM1 expression by CAFs. Moreover, clinical breast cancer samples were also found to highly express TGFB1/2/3 suggesting that these findings are of clinical relevance. Such invasion fronts are rich in inflammatory cells [52]. Consistent with this result, we found that inflammatory cytokines IL1β and TNFβ induced GREM1 expression in CAFs. Moreover, GREM1 expression correlated with mesenchymal marker expression in tumor samples. The latter observation indicates that Grem1 at the invasion front may contribute to the desmoplastic phenotype (Fig. 7). We observed a striking activation of fibrogenesis in fibroblasts and in CAFs by Grem1. Depletion or ectopic expression of Grem1 in CAFs demonstrated that Grem1 expression is positively linked to the expression of TGFβ ligands and target genes, mesenchymal markers, extracellular matrix proteins, and matrix metalloproteinase (MMP) remodeling factors at the mRNA level, as well as with fibronectin, S100A4, collagen I, FAP, and αSMA at the protein level. In addition, Grem1 promoted actin stress fiber formation and collagen gel contraction. These expression patterns are a characteristic of a fibrogenic response and fibroblast activation. The Grem1-induced responses may be mediated by TGFβ pathway activation; TGFβ is a strong inducer of fibrogenesis and an activator of fibroblasts [53]. With TGFβ being a strong inducer of Grem1 and vice versa, it may act in a feed-forward loop.
Multiple studies have shown that CAFs create a microenvironment suitable for cancer cell invasion [1,2], which we further demonstrated in this study in vivo by co-injection of breast cancer cells with fibroblasts/CAFs into the zebrafish perivitelline space. Thus, the profibrogenic ability of Grem1 could contribute to its role in promoting cancer cell invasion mediated by activated fibroblasts and CAFs. 3D co-culture of breast cancer cells with CAFs in collagen demonstrated that Grem1 is critical for invasion. In accordance with these results, Grem1 strongly promoted intravasation in a zebrafish coinjection xenograft model. Moreover, by injecting ectopic Grem1-producing M2 and MDA-MB-231 cells into the DoC of zebrafish embryos, we found that Grem1 strongly promoted the extravasation of cancer cells. These results may explain the clinical association between Grem1 expression in tumors and a poor prognosis of MFS.
Mechanistically, Grem1 exerts its effects by antagonizing selective BMPs [11]. Consistent with this notion, we found that BMP-induced SMAD1/5 phosphorylation is inhibited in breast cancer cells and in CAFs. In addition, depletion of endogenous Grem1 in CAFs upregulates BMP/SMAD-dependent ID1/ID3 expression while the addition of rhGrem1 has the opposite effect. Moreover, treatment with a selective BMP receptor kinase inhibitor mimicked the effect of exogenous Grem1 protein by promoting mammosphere formation and fibrogenic marker expression. However, our results do not exclude the possibility that Grem1 also can act via BMP-independent pathways [24]. For example, induction of TGFβ expression by Grem1 may occur independently of BMP antagonism. Grem1 was found to promote cell viability, migration, and invasion in glioma [54] and the invasive phenotype of mesothelioma [55] by activating TGFβ/SMAD signaling. Moreover, Grem1 may promote breast tumorigenesis by acting on signaling pathways distinct from TGFβ family signaling; in renal tubular cells, Grem1 has been reported to signal via the vascular endothelial growth factor receptor 2 (VEGFR2) pathway [21] to promote angiogenesis [56] and to mediate inflammation and the infiltration of immune-inflammatory cells [57]. Furthermore, Grem1 may act directly or indirectly by sequestering BMP on endothelial cells and immune cells and thereby promote tumorigenesis. Irrespective of the precise mechanisms, our results demonstrate potent protumorigenic effects of Grem1 on cancer cells and CAFs in vitro in mono-and in co-culture, as well as a key in vivo role for Grem1 in stimulating extravasation and for Grem1producing CAFs in mediating the intravasation of breast cancer cells. These two processes, extravasation and intravasation, are the key steps in the dissemination and distant colonization of primary cancer cells. The results are expressed as the mean ± s.e.m., n = 2. Student's t test, **P ≤ 0.01. c Schematic of the working model of Grem1 function in breast cancer progression. Grem1 expression in fibroblasts is induced by factors (such as TGFβ from breast cancer cells or maybe other stromal cells (that produce inflammatory cytokines). Grem1 could activate fibroblasts into CAFs. CAFs might present a desmoplastic microenvironment, thereby promote cancer cell invasion. Grem1 itself could promote the stemness and invasion of breast cancer cells