BMP4 promotes the metastasis of gastric cancer by inducing epithelial–mesenchymal transition via ID1

Gastric cancer (GC) is the fifth leading cause of global cancer mortality, accounting for 723,100 deaths annually (Jemal et al., 2011). Surgical resection is an effective treatment for GC at an early stage. However, the majority of advanced-stage GC patients with metastasis miss the opportunity for surgical resection, resulting in a grim five-year survival rate of less than 30% (Kamangar et al., 2006). Recurrence and metastasis after surgery are the major reasons for poor prognosis and are a tremendous obstacle for the successful treatment of GC patients. Despite great efforts in the battle against GC metastasis, further research into new treatment approaches and the underlying molecular mechanism are needed.

Epithelial cancer cells gain motility and invasiveness in a dynamic and reversible process known as epithelial–mesenchymal transition (EMT) to execute multiple steps of the invasion–metastasis cascade (Le et al., 2014; Vanharanta and Massagué, 2013). Cells undergoing EMT are characterized by the downregulation of junction proteins, such as E-cadherin (also known as CDH1), and the loss of cell–cell connections, cell–matrix contacts and epithelial polarity; these cells concomitantly gain more mesenchymal traits and undergo cytoskeletal changes to increase motility and invasiveness (Thiery, 2002). The E-cadherin and β-catenin (also known as CTNNB1) complex stabilizes epithelial cells by forming cell–cell junctions (Buckley et al., 2014). EMT-related transcription factors (TFs), such as Twist (also known as TWIST1 in humans) and Snail (also known as SNAI1 in humans), have been reported to induce EMT by inhibiting E-cadherin transcription directly or indirectly (Peinado et al., 2007; Yang et al., 2004, 2018). Reorganization of the extracellular matrix (ECM) facilitates tumor cell migration and invasion during tumor progression. The activation of matrix metalloproteases (MMPs) can trigger matrix remodeling and has been considered to favor the EMT process (Tao et al., 2013).

EMT can be induced by some alternative signaling pathways, including the Wnt, Notch and transforming growth factor β (TGF-β) pathways (Gonzalez and Medici, 2014). Bone morphogenesis proteins (BMPs), a large subgroup of the TGF-β superfamily, have been shown to mediate a variety of biological functions in cell proliferation, cell differentiation and cell fate during embryonic development (Murali et al., 2005; van Wijk et al., 2007). In recent years, a role of BMPs in the pathogenesis and development of cancer has emerged. BMPs transduce signals by binding to the BMP receptor (BMPR) complex. Once ligands bind, BMPR1 heterodimerizes with BMPR2 to trigger the phosphorylation of the receptor-regulated SMAD1/5/8 pathway. Activated SMAD1/5/8 subsequently translocates into the nucleus to regulate target gene expression (Morikawa et al., 2013). Bone morphogenetic protein 4 (BMP4), an extracellular signaling molecule that belongs to the BMP family, has been implicated as a pivotal regulator in tumor progression. The functional effect of BMP4 is complex and tissue specific (Deng et al., 2007; Ma et al., 2017a; Rothhammer et al., 2007; Zeng et al., 2017). BMP4 has been suggested to modulate cisplatin sensitivity in GC (Ivanova et al., 2013). However, the effects of BMP4 on GC metastasis and the underlying molecular mechanisms remain to be elucidated.

We hypothesized that BMP4 might affect GC metastasis by regulating EMT; thus, we carried out this study to determine the effects of BMP4 on EMT-regulated GC metastasis both in vitro and in vivo and to determine the molecular signaling mechanisms involved.

To determine whether the expression level of BMP4 was altered in human GC specimens, we first investigated BMP4 mRNA expression in 53 pairs of cancer tissues and adjacent normal gastric tissues from gastric cancer patients by reverse transcription qPCR (RT-qPCR). As displayed in Fig. 1A, the BMP4 mRNA level was substantially higher (more than 2-fold) in cancer tissues than that in adjacent normal gastric tissues, with a median fold change of 3.36. The protein expression of BMP4 and EMT markers (E-cadherin and vimentin) was then examined by immunohistochemistry (IHC). Based on the IHC scoring standard (Tao et al., 2013), 28.85% (45/156) and 71.15% (111/156) of GC tissues were defined as low and high BMP4 expression, respectively, among 156 GC patients in a retrospective study cohort. Typical IHC images of BMP4 and EMT markers in normal and GC tissues are shown in Fig. 1B. The results of IHC showed that positive staining of BMP4 and vimentin was localized in the cytoplasm of GC cells, whereas positive staining of E-cadherin was mainly localized on the membrane of GC cells. Moreover, Spearman correlation analysis revealed that the expression of BMP4 was significantly negatively correlated with the expression of epithelial marker E-cadherin (Spearman's ρ=−0.213, P=0.007) but positively associated with the expression of the mesenchymal marker vimentin (Spearman's ρ=0.160, P=0.037).

We further investigated the correlation between BMP4 expression and the clinicopathological characteristics of 156 GC patients in the retrospective study cohort. High BMP4 expression was found to be statistically associated with tumor invasion, lymph node metastasis and tumor, lymph node and metastasis (TNM) stage (P<0.05; Table 1). Kaplan–Meier survival analysis further revealed that high BMP4 expression was associated with shorter overall survival (OS) and recurrence-free survival (RFS) in GC patients in the retrospective study cohort (P<0.01 for both measures of survival; Fig. 1C). The median OS was 28.0 months in patients with low BMP4 expression and 19.0 months in patients with high BMP4 expression (P=0.0086; Fig. 1C). The median RFS in patients with low BMP4 expression was significantly longer than that in patients with high BMP4 expression (low BMP4, 20.0 months; high BMP4, 12.0 months; P=0.0078). Multivariate Cox regression analyses showed that high BMP4 expression served as an independent prognostic factor for RFS and OS in GC (P<0.05; Table 2). Taken together, elevated BMP4 expression was independently correlated with the poor prognosis of GC patients.

The expression of BMP4 in five GC cell lines and the human gastric mucosal epithelial cell line GES-1 was analyzed. The RT-qPCR results showed that BMP4 was upregulated in GC cells compared to expression levels in GES-1 cells (Fig. 2A). Western blotting analysis also showed that the protein level of BMP4 was higher in GC cells than in GES-1 cells (Fig. 2B). GC cell lines SGC7901 and BGC803 were chosen for further experiments. To determine whether BMP4 contributed to GC migration and invasion, we treated GC cells (SGC7901 and BGC803) with human recombinant BMP4 (100 ng/ml) or its receptor antagonist noggin (200 ng/ml) for 48 h. Wound healing assays, as shown in Fig. 2C, demonstrated that BMP4 treatment significantly increased the migration ability of GC cells (Fig. 2C; P<0.01) whereas noggin treatment decreased migration of GC cells. As shown in Fig. 2D, transwell invasion assays showed that BMP4 treatment distinctly increased the numbers of invaded cells in both SGC7901 and BGC803 cell lines, whereas noggin treatment decreased the numbers of invaded cells in both cell lines (Fig. 2D; P<0.01). Taken together, these results suggested that BMP4 was upregulated in GC cell lines and promoted GC cell migration and invasion in vitro.

Based on the important role of EMT in tumor metastasis, and the association of BMP4 with EMT in GC tissues, a series of assays were applied to analyze the role of BMP4 in EMT in GC cells. After being treated with BMP4, GC cells displayed an EMT-like conversion, and treatment with noggin was observed to block this conversion. As shown in Fig. 3A, the morphology of GC cells in the BMP4-treated groups was scattered and spindle-like with stretched pseudopodia, whereas cells in the control groups displayed a condensed and pebble-like shape. RT-qPCR, western blotting and immunofluorescence staining were subsequently applied to detect the expression levels and colocalization of EMT markers in BMP4- and/or noggin-treated GC cells. Compared to the blank control, BMP4 treatment resulted in downregulated mRNA and protein expression of the epithelial marker E-cadherin, whereas the expression of mesenchymal markers (vimentin and fibronectin) was upregulated in GC cells (Fig. 3B,C). Moreover, mRNA and protein expression of EMT-related TFs (Twist and Snail) were observably increased in BMP4-treated GC cells (Fig. 3B,C). However, noggin treatment played an opposite role to that of BMP4 in EMT progression. Cellular immunofluorescence colabeling with E-cadherin and vimentin antibodies in GC cells showed that, compared with the blank control, BMP4 treatment downregulated expression of E-cadherin, shifted cell morphology to a mesenchymal phenotype and upregulated expression of the cytoskeleton protein vimentin (Fig. 3D). Noggin treatment blocked these changes in GC cells (Fig. 3D). Our data suggested that BMP4 induced EMT, while noggin inhibited EMT in GC cells.

To further confirm the role of BMP4 in promoting metastasis and regulating EMT in GC, we established a xenograft metastasis model by stably infecting GC cells with a BMP4 siRNA lentivirus (si-BMP4) or with a negative control siRNA lentivirus (si-Control). RT-qPCR and western blotting confirmed the infection efficiency (Fig. 4A; P<0.001). The infected cells were injected into mice via the tail vein, and the mice were sacrificed 1 month later. As shown in Fig. 4B, the weight of the mice was not obviously different between the si-BMP4 and si-Control groups at the early stage after tumor cell implantation. From the 15th day, the weight of the mice in the si-Control group began to decrease gradually. The mice in the si-BMP4 group had significantly greater weight loss than the mice in the si-Control group on the 30th day (Fig. 4B; P<0.05). Because of the reported close association between MMP activity and EMT-regulated tumor metastasis (Deryugina and Quigley, 2006), we applied a fluorescence imaging system to detect MMP activity in lung metastatic foci in the xenograft metastasis model on the 30th day of GC cell implantation, before the mice were sacrificed. Fig. 4C shows that, compared to the si-Control group, the mean MMP activity, represented by fluorescence intensity, was markedly inhibited or barely detected when BMP4 was knocked down in GC cells (mean±s.d. MMP activity for si-BMP4 versus si-Control: SGC7901, 5.93±0.78 versus 17.90±0.77, P<0.001; BGC803, 0.93±0.34 versus 14.20±1.57, P<0.01) (Fig. 4C). Hematoxylin and eosin staining confirmed the inhibition of lung metastasis by BMP4 siRNA in GC cells in vivo. As shown in Fig. 4D, lung metastatic foci were observed more frequently in the si-Control group than in the si-BMP4 group [si-Control vs si-BMP4: SGC7901, 80% (4/5) versus 20% (1/5); BGC803, 60% (3/5) versus 0% (0/5)], and the si-Control group harbored more metastatic nodules than the si-BMP4 group (Fig. 4D; P<0.001). To further determine the role of BMP4 in EMT regulation in vivo, we evaluated the expression levels of EMT markers in another BMP4-knockdown subcutaneous xenograft model. IHC pictures of tumor tissue sections from the flanks of the nude mice demonstrated that BMP4 knockdown upregulated E-cadherin expression but downregulated vimentin expression substantially compared to that in the si-Control group (Fig. 4E), revealing that BMP4 knockdown inhibited EMT progression and metastasis of GC cells in vivo.

To clarify the potential molecular mechanisms underlying BMP4-promoted EMT and metastasis in GC, we compared 84 TGF-β-related gene expression profiles between BMP4-treated and noggin-treated cells by using a human TGF-β/BMP signaling pathway PCR array. BMP4 treatment of BGC803 cells dramatically increased the mRNA expression level of the inhibitor of DNA binding 1 (ID1), with a 9.5-fold change in mRNA expression compared to noggin-treated cells (Table S1). SMAD1/5/8, a downstream target of BMP-receptor binding, was also investigated. Results of western blotting showed that phosphorylation of SMAD1/5/8 occurred 30 min after BMP4 treatment (Fig. 5A). Subsequent RT-qPCR and western blotting analyses corroborated the upregulation of ID1 in BMP4-treated GC cells (Fig. 5B; P<0.01). In addition, ID1 expression was assessed in GC tissues from 156 patients in the retrospective study cohort and found to be positively correlated with BMP4 expression (Fig. 5C; Spearman's ρ=0.402, P=0.026). Expression of ID1 and phospho-SMAD1/5/8 in the xenograft subcutaneous model was also inhibited in the BMP4 knockdown groups (Fig. 4E), indicating that SMAD and ID1 signaling play an important role in BMP4-regulated GC cells.

To further clarify the underlying mechanism of BMP4-mediated enhancement of invasiveness and EMT progression in GC cells, we utilized three candidate pairs of siRNAs to manipulate ID1 expression, and ID1 siRNA sequence #3, which had the highest ID1 inhibition efficiency, was selected for subsequent transient transfection of GC cells (Fig. 5D). Compared to treatment with BMP4 alone, the co-administration of ID1-siRNA and BMP4 attenuated the EMT-enhancing effects of BMP4 by upregulating the epithelial marker E-cadherin and by downregulating the mesenchymal marker vimentin and EMT-related TFs (Twist and Snail) in GC cells (Fig. 5E,F). Moreover, knockdown of ID1 inhibited the mRNA and protein expression levels of MMP2 and MMP9 in BMP4-treated GC cells (Fig. 5E,F). Functionally, knockdown of ID1 suppressed BMP4-enhanced migratory and invasive abilities in both SGC7901 and BGC803 cells (Fig. 5G,H; P<0.05).

Because knockdown of ID1 attenuated the promotion of EMT and invasion by BMP4 treatment, we further designed a rescue experiment to determine whether overexpression of ID1 reversed the inhibition of EMT and invasion in BMP4 knockdown GC cells (SGC7901/si-BMP4 and BGC803/si-BMP4 cells). We transfected the ID1 overexpression plasmid or control vector into GC cells with stable BMP4 knockdown (si-BMP4+ID1 or si-BMP4+control). The overexpression of ID1 was confirmed by RT-qPCR and western blotting (Fig. 6A,B). As shown in Fig. 6A,B, overexpression of ID1 in si-BMP4 GC cells reversed the effects of EMT inhibition caused by BMP4 knockdown, resulting in increased expression of mesenchymal markers and Snail, and decreased expression of the epithelial marker E-cadherin. Changes in MMP2 and MMP9 expression were consistent with changes in the mesenchymal markers (Fig. 6A,B). Further functional assays revealed that ID1 abrogated the migratory and invasive phenotypes inhibited by knockdown of BMP4 in vitro (Fig. 6C,D; P<0.05). Therefore, ID1 was responsible, at least in part, for the observed role of BMP4 in the regulation of EMT and invasion.

Taken together, we identify a role for BMP4 in the promotion of invasion and metastasis of GC cells and propose a mechanism responsible for this process, whereby abnormal BMP4 expression activates SMAD1/5/8 signaling by binding BMPR, which leads to the phosphorylation of SMAD1/5/8 and its translocation into the nucleus to regulate the expression of the target gene ID1. By upregulating ID1, expression of MMPs and EMT are induced to promote tumor metastasis (Fig. 6E).

Despite the great number of efforts made in anticancer therapy over several decades, metastasis remains a major clinical obstacle in the successful treatment of gastric cancer. TGF-β signaling has been revealed to exert different effects depending on the type of tumor. BMP4 has been reported to be a novel prognostic predictor in glioma and hepatocellular carcinoma (HCC) (Bao et al., 2013; Ma et al., 2017b; Wu and Yao, 2013). BMP4 was not detected in the normal antrum but was found to be expressed in H. pylori-infected stomach cells and able to influence the fate of gastric epithelial cells, resulting in inflammatory cell influx (Bleuming et al., 2006). Moreover, BMP4 is a key element in the intestinal differentiation of gastric cells that is active in intestinal metaplasia (Barros et al., 2008), implying a role for BMP4 in the development of gastric cancer. However, the role of BMP4 in gastric cancer is still controversial. Wu et al. (2010) reported that the activation of BMP4 signaling inhibited the proliferation of gastric cancer cells. Other studies have indicated that BMP4 is an oncogenic factor. Shen et al. (2013) found that tumor-associated macrophages enhanced BMP signaling and promoted gastric cancer cell invasion. Ivanova et al. (2013) also found that BMP4 is highly expressed in GC tissues, and that targeting BMP4 using shRNA sensitized GC cells to cisplatin. Consistently, we demonstrated that BMP4 was overexpressed in GC tissues, and that this correlated with the prognosis of patients. We also identified BMP4 as an independent prognostic factor for GC patients. Herein, we provide the first evidence that BMP4 acts as a central mediator of the EMT process in GC cells and, by acting in this fashion, plays a key role in tumor invasion and metastasis both in vitro and in vivo. Moreover, we found that the downstream target ID1 was responsible for the effects of BMP4 on EMT-regulated tumor metastasis, as profiled by PCR array.

Activation of the BMP4 signaling pathway has been determined to play an important role in tumor metastasis regulation, and its exact effect is cell type-specific. Cao et al. (2014) demonstrated that BMP4 inhibited breast cancer metastasis by blocking myeloid-derived suppressor cell activity, and Hu et al. (2017) determined that BMP4 overexpression inhibited metastasis of esophageal squamous cancer. However, other reports have demonstrated that BMP4 promotes metastasis in HCC (Zeng et al., 2017), bladder cancer (Martínez et al., 2017), colorectal cancer (Yokoyama et al., 2017) and lung adenocarcinoma (Chen et al., 2016). When assessing the metastasis of GC, we first showed that BMP4 potentiated GC cell motility and invasiveness in vitro, whereas knockdown of BMP4 inhibited the metastasis of GC cells both in vitro and in vivo, identifying BMP4 as a tumor-promoting factor in GC.

Metastasis is a multistage process involving complex mechanisms and functional alterations of invasive tumor cells, which starts when cells lose intercellular connections, and individual cells are released from the primary tumor (Joyce and Pollard, 2009). EMT, through which cancer cells acquire invasive and migratory abilities, has been found to induce cancer cell phenotypes that allow the execution of multiple steps of the invasion–metastasis cascade in malignant progression (Huber et al., 2005; Thiery, 2002). Reduction of the expression of tight junction proteins (for example, E-cadherin), which play an essential role in the EMT process, contributes to ECM remodeling and tumor metastasis (Thiery, 2009; Zeisberg and Neilson, 2009). Twist and Snail, as transcription factors critical to EMT, are considered to induce EMT and thus promote tumor metastasis (Xu et al., 2018; Yang et al., 2004; Zeisberg and Neilson, 2009). Notably, Twist was found to promote tumor metastasis by inducing the formation of invadopodia, which are cellular structures that directly invade other cells (Eckert et al., 2011). In this study, we emphasize the effects of BMP4 on EMT regulation in GC cells. BMP4 treatment decreased the expression of E-cadherin and upregulated the expression of mesenchymal markers and transcriptional factors (Twist and Snail), suggesting that BMP4 induces EMT in GC cells. Another key functional trait of BMP4 is its regulatory effect on MMPs, which indispensably contribute to tumor invasion and metastasis. The functional mechanisms by which MMP2 and MMP9 facilitate metastasis are suggested to be degradation of the ECM (Jabłońska-Trypuć et al., 2016) and promotion of pseudopod formation (Brooks et al., 1996; Yu and Stamenkovic, 1999). Moreover, another mechanism that accounts for the promotion of metastasis by MMPs is their regulatory effect on EMT progression via interactions with cadherins. MMPs have been shown to regulate E-cadherin expression, resulting not only in the loss of cell adhesion but also in the promotion of tumor cell invasion and metastasis (Christofori and Semb, 1999; Sancéau et al., 2003). We observed morphological changes typical of EMT and formation of membrane protrusions (pseudopodia) after BMP4 treatment in GC cells. Live fluorescence imaging showed decreased MMP activity in lung metastatic foci when BMP4 expression was knocked down in GC cells in vivo, supporting a role for BMP4 in regulation of MMP activity.

We further investigated the signaling responsible for BMP4-regulated EMT and metastasis in GC cells. As suggested by the results of our PCR array, ID1 was identified as an important downstream target of BMP4 in GC cells. Members of the ID protein family, which inhibit the ability of basic helix-loop-helix (bHLH) transcription factors to bind to DNA, thus interrupting their regulatory functions in many developmental and differentiation processes, have been found to regulate tumor progression (Benezra et al., 1990; Han et al., 2004). ID1, one of the most important members of the ID family, has been shown to participate in cell dedifferentiation and metastasis and is linked to poor prognosis in human cancers (Lasorella et al., 2014; Mody et al., 2017). It has been suggested that ID1 is involved in the regulation of tumor invasiveness and metastasis in human malignant diseases (Ciarrocchi et al., 2011; Cubillo et al., 2013; Pillai et al., 2011). We found that ID1 expression noticeably decreased when BMP4 was knocked down. BMP4-induced EMT, motility and invasiveness were attenuated by knocking down ID1. The rescue experiment, conducted by co-transfection of BMP4 siRNA and an ID1 overexpression plasmid, confirmed the role of ID1 in BMP4-induced biological effects. ID1 is also likely to contribute to BMP4-promoted tumor invasiveness via the regulation of MMPs. Previous studies have shown that ID1 overexpression significantly induced the secretion of MMP2 and/or MMP9 in some human cancer cells (Coppe et al., 2004; Nieborowska-Skorska et al., 2006). Accordingly, we demonstrated that knockdown of ID1 attenuated the upregulation of BMP4-induced MMP2 and MMP9 expression, whereas overexpression of ID1 rescued the inhibition of MMP2 and MMP9 expression caused by BMP4 knockdown. Considering the observed mediator effects of ID1, we have identified ID1 as an indispensable target of BMP4, and found that inhibition of ID1 was capable of blocking GC invasion enhanced by BMP4.

Interestingly, knockdown of ID1 was found to greatly, but not completely, inhibit BMP4-induced EMT in GC cells, suggesting that some transcription factors or signaling molecules other than ID1 might be involved in mediating BMP4 function in GC cells. More detailed investigation of the underlying molecular mechanisms by which ID1 is involved in the functions of BMP4 in GC is required.

It is worth mentioning that the EMT process is reversible. EMT occurs in the initial step of metastasis, which facilitates tumor cell dissemination from the primary site and subsequent invasion into the blood circulation. When transferred to distant metastatic sites, EMT-phenotype tumor cells might undergo MET (mesenchymal–epithelial transition), switching to re-gain epithelial cell characteristics and transplanting to form metastatic foci (Nieto, 2013; Pei et al., 2019). The switch between the two states is dynamic, and the state in different stages of cancer progression needs to be detected and monitored.

In summary, our study found that expression of BMP4 was significantly upregulated in GC tissues and correlated with poor prognosis in GC patients. BMP4 was upregulated in GC cells, and BMP treatment induced EMT and consequential tumor migration and invasion, whereas noggin, which antagonizes BMP4, suppressed these functions in vitro. Furthermore, knockdown of BMP4 inhibited GC cell metastasis and EMT in vivo. ID1 acted as an important mediator in BMP4-regulated EMT and invasion by activating SMAD1/5/8. Our data suggest a novel role for BMP4 in GC progression. Further studies are needed to understand the detailed mechanism of BMP4-regulated GC progression.

A total of 209 specimens were randomly collected from gastric cancer patients who were admitted to Xiangya Hospital, 3rd Xiangya Hospital and Hunan Province Tumor Hospital, Central South University, Changsha, Hunan, China. Fifty-three paired specimens of tumor tissues and adjacent normal gastric tissues were randomly obtained from July 2011 to June 2015. In the retrospective study cohort, 156 tissue specimens were randomly collected between January 2010 and June 2013. All specimens were routinely processed for a pathological diagnosis according to the WHO classification. Patient staging data were defined according to the NCCN (National Comprehensive Cancer Network) Clinical Practice Guidelines in Oncology (2016, Version 3). No patients received radiotherapy, chemotherapy or immunotherapy before surgery. The samples were snap-frozen in liquid nitrogen and stored at −80°C for subsequent RNA extraction or formalin-fixed and paraffin-embedded immunohistochemistry (IHC). The study was approved by the Xiangya Hospital Research Ethics Committees, and this study complied with the ethical guidelines of the Helsinki Declaration. Informed consent was obtained from all patients. Overall survival (OS) was determined as the time interval between GC resection or diagnosis by gastroscopic biopsy and death or the last observation. Patients alive at the end of the follow-up and those who died with no sign of GC recurrence were censored for anonymity. Recurrence-free survival (RFS) was calculated as the time from the date of tumor resection or GC diagnosis to the date of first conclusive evidence of recurrence.

Total RNA was extracted using TRIzol Reagent (Invitrogen, Carlsbad, CA), and cDNA was synthetized using the PrimeScript Kit (TaKaRa Bio, Dalian, China) according to the manufacturer's protocols. RT-qPCR was performed in triplicate using a SYBR Green fluorescence-based assay (TaKaRa Bio) on an ABI ViiA 7 real-time PCR system (Applied Biosystems, Carlsbad, CA). The primers for real-time PCR are listed in Table S2. Relative mRNA expression levels were calculated using the 2^−ΔΔCt method [ΔCt=Ct (targeting gene)−Ct (GAPDH)] and were normalized to the internal control of GAPDH.

Tissues were fixed in 10% formalin, dehydrated, and embedded in paraffin. Then 4 μm-thick sections were processed for analyses. After dewaxing, hydrating, antigen retrieval, and endogenous peroxidase activity blocking, the sections were incubated with different primary antibodies at 4°C overnight, followed by incubation with secondary antibodies for 30 min at room temperature. DAB (Beyotime Biotechnology, Shanghai, China) and hematoxylin (Beyotime Biotechnology) staining was then performed. The primary antibodies for IHC are listed in Table S3. Immunoreactivity for each tested protein was scored according to the percentage of positive-staining cells and staining intensity, as previously described (Tao et al., 2013).

Human normal gastric musical epithelial cells (GES-1) and five human gastric cancer cell lines (SGC7901, BGC803, AGS, BGC823 and MKN45) were obtained from the Cell Bank of Typical Culture Preservation Committee of Chinese Academy of Science (Shanghai, China). The cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) (Gibco, Grand Island, NY), 100 U/ml penicillin and 100 μg/ml streptomycin sulfate (Gibco) and were incubated at 37°C under an atmosphere of 95% air and 5% CO2. For in vitro experiments, cells were stimulated with recombinant human BMP4 (PeproTech, Rocky Hill, NJ) at a concentration of 100 ng/ml or with the BMP4 receptor antagonist, noggin (PeproTech), at a concentration of 200 ng/ml. BMP4 and noggin were reconstituted according to the manufacturer's description.

Total protein was extracted in RIPA lysis buffer, separated by SDS–PAGE and then transferred onto PVDF membranes (Millipore, Bedford, MA). The membranes were blocked in TBS-T containing 5% non-fat milk at 37°C for one hour and incubated with primary antibodies at 4°C overnight followed by incubation with the HRP-conjugated secondary antibody (1:5000; ab6721, ab6728; Abcam, Cambridge, UK) for one hour at 37°C. The primary antibodies for western blotting are listed in Table S3. The signals were detected by an enhanced chemiluminescence kit (Millipore), and the bands were automatically visualized using a ChemiDoc XRS+ system (Bio-Rad, Hercules, CA) and quantitatively analyzed with Image Lab software (Bio-Rad). GAPDH expression was used as the internal loading control.

Wound healing assay: GC cells were seeded in 6-well plates for 24 h. A straight line was scratched across the confluent cell monolayers with a 200 μl pipette tip. The cells were washed with PBS three times and subsequently cultured in fresh medium with 1% FBS for 48 h. The wound healing of the scratched cells was photographed under a DMIL LED AE2000 inverted microscope (Leica, Wetzlar, Germany).

The invasion ability of GC cells was assessed using Matrigel-coated upper inserts containing polycarbonate filters with an 8 μm pore size (BD Biosciences, New Jersey, US). The culture medium containing 10% FBS was placed in the lower chambers to act as a chemoattractant. A total of 3×10⁴ cells suspended in 200 μl serum-free RPMI 1640 were seeded in the upper chambers and incubated at 37°C for 48 h. The cells that penetrated the Matrigel-coated filter were fixed with methanol and stained with 0.1% Crystal Violet hydrate solution, and the stained cells were examined under a microscope.

Cells were seeded in 24-well plates covered with slides overnight and then treated with BMP4 and/or noggin for 48 h. After washing, fixing and blocking, the slides were incubated with primary antibodies against E-cadherin (ab40772, Abcam; 1:200 dilution) and vimentin (ab8069, Abcam; 1:400 dilution) at 4°C overnight followed by the corresponding Alexa Fluor 546- and 488-conjugated secondary antibodies (Invitrogen; 1:500 dilution) for 1 h at room temperature. The nuclei were stained with DAPI for 3 min. Immunofluorescence images were captured using a TCS SP5 laser confocal microscope (Leica).

For the BMP4 expression knockdown in vivo experiments, a BMP4-RNA interference (RNAi) lentiviral vector (GV118-si-BMP4) was constructed. Four candidate BMP4 siRNAs and negative controls were designed and synthesized by GeneChem (GeneChem Co. Ltd., Shanghai, China). The knockdown efficiency of BMP4 siRNA was confirmed by RT-qPCR and western blotting, and BMP4 siRNA1 had a maximum efficiency of 81.6%, so it was selected for further experiments (Fig. S1A,B). The sequences of the BMP4 siRNAs are listed in Table S4. SGC7901 and BGC803 cells were infected with the lentivirus with an optimal multiplicity of infection (MOI) of 20 TU/ml and 40 TU/ml, respectively, following the manufacturer's protocol (GeneChem Co. Ltd.). Cells were treated with puromycin (InvivoGen, San Diego, CA) to produce stably transfected cell lines for further animal experiments.

The RT2 Profiler PCR Array [Human TGF-β/BMP Signaling Pathway (Qiagen, Valencia, CA)] was used to profile downstream targets of BMP4 in GC cells. An RNeasy Mini Kit (Qiagen) was used to extract total RNA from cells. cDNA was synthesized using an RT² First Strand Kit (Qiagen) according to the manufacturer's protocol. The PCR array was performed using a ViiA 7 real-time PCR system (Applied Biosystems) to determine which genes changed more than 2-fold between the BMP4-treated and noggin-treated groups.

Based on the sequence of ID1 mRNA (GenBank Accession: X77956.1), three chemically synthesized siRNAs were purchased from GenePharma (Shanghai, China). siRNA1: sense strand, 5′-GGUCACGUUUGGUGCUUCUTT-3′; antisense strand, 5′-AGAAGCACCAAACGUGACCTT-3′. siRNA2: sense strand, 5′-AUCACCGACUGAAAAUAUUTT-3′; antisense strand, 5′-AAUAUUUUCAGUCGGUGAUCA-3′. siRNA3: sense strand, 5′-CGCCGGAUCUGAGGGAGAATT-3′; antisense strand, 5′-UUCUCCCUCAGAUCCGGCGAG-3′. The siRNA negative control used was: sense strand, 5′-UUCUCCGAACGUGUCACGUTT-3′; antisense strand, 5′-ACGUGACACGUUCGGAGAATT-3′. SGC7901 and BGC803 cells were seeded in 24-well plates overnight and then transfected with siRNA separately using Lipofectamine RNAiMAX Transfection Reagent (Invitrogen) according to the manufacturer's instructions. siRNA3-mediated knockdown of ID1 (75% in SGC7901 cells and 85% in BGC803 cells) was confirmed by RT-qPCR and western blotting.

To enhance the expression of ID1, the human wild-type cDNA of ID1 was cloned into the MV/MCS/IRES/EGFP/Neo vector. The control vector used in this study was an empty vector. Plasmids of ID1 and the control vector were purchased from GenePharma (Shanghai, China). For rescue experiments, GC cells with stable knockdown of BMP4 with lentivirus (si-BMP4 cells) were seeded into 6-well plates overnight and then transfected with ID1 plasmid or control vector using Lipofectamine 3000 (Invitrogen) according to the manufacturer's protocol. Cells were collected after 48 h of transfection for further analyses.

To investigate the effects of BMP4 on gastric cancer metastasis in vivo, a xenograft animal model was constructed in male BALB/c mice (5 weeks old). These mice were bred and maintained under defined conditions at the Animal Experiment Center of Central South University (SPF grade). Briefly, five BALB/c mice in each experimental group were injected with GC cells stably transfected with si-BMP4 or si-Control (2×10⁶ cells in 0.1 ml RPMI 1640 per mouse) via tail veins. Weight changes of nude mice in each group were monitored every 5 days. After 1 month following implantation, the metastatic foci in the lungs were visualized using an FMT-4000 3D Fluorescence Molecular Tomography Imaging System (PerkinElmer, Boston, MA) by injecting MMPSense 750 FAST Fluorescent Imaging Agent (PerkinElmer) via the tail vein (2 nmol agent formulated in 1× PBS). The mean tumor fluorescence intensity was equal to the total MMP fluorescence signal/tumor size, which was used to reflect the mean MMP activity in vivo. All animal experiments were conducted at the Animal Institute of Central South University and were approved by the Medical Experimental Animal Care Commission of Central South University.

Statistical analysis was performed using SPSS 21.0 software (SPSS Inc., Chicago, IL). Quantitative values are presented as the mean±s.d. or median (and range). Paired t-tests and Student's t-tests were used for paired and unpaired continuous data, respectively. The χ² test was applied for categorical data. Cumulative OS and RFS were evaluated using the Kaplan–Meier method and the log-rank test. A Cox proportional hazards regression model was used for multivariate analysis of the parameters. P<0.05 was considered to be statistically significant.