TM4SF1 promotes EMT and cancer stemness via the Wnt/β-catenin/SOX2 pathway in colorectal cancer

The cancer tissues (T) and paired adjacent tissues (N) were obtained from the Gastrointestinal Surgery Department of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China, from July 2012 to April 2017. None of the patients received any radiochemotherapy before the operation. Immunohistochemistry staining was performed as described elsewhere [23], and specific antibodies were used as follows: TM4SF1, β-catenin (1:100 dilution, Abcam, USA) and SOX2, CD133 (1:80 dilution; Abcam, USA). The expression of TM4SF1 was evaluated according to the intensity of the staining (0, 1+, 2+ and 3+) and the percentage of positive cells, which was scored as 0 (0%), 1 (1–25%), 2 (26–50%), 3 (51–75%) or 4 (76–100%). The staining index (SI) was calculated as follows: SI = (intensity score in 1) × (positive staining score in 2). SI < 3 was classified as low expression, while SI ≥4 was classified as high expression. Furthermore, the study was approved by the Human Subjects Protection Committee of the Tongji Medical College, Huazhong University of Science and Technology, China.

The CRC cell lines LoVo and NCM460 were cultured in DMEM (Boster, China). The HCT116, SW480, RKO, FHC, and DLD1 cell lines were cultured in RPMI 1640 medium (HyClone, USA). All of the cell lines were cultured in medium supplemented with 10% foetal bovine serum (FBS, ScienceCell). Primary antibodies against the proteins TM4SF1, N-cadherin, vimentin, SOX2, MMP9, CD133, Par3 and β-catenin were purchased from Abcam (USA), while those against E-cadherin, c-Myc, and ZO1 were purchased from Santa Cruz (USA), and those against Smad2 and CD44 were purchased from Sigma (USA). The second antibody, affinity-purified, biotinylated rabbit anti-rat IgG, was purchased from Sigma (USA). Accession numbers are available in Table S1.

To transiently silence the expression of TM4SF1, siRNAs (Ribo-Bio, Guangzhou, China) were transfected into SW480 and LoVo cells by using Lipofectamine 3000. TM4SF1 shRNA (Shanghai Gene Chem Co, Ltd., China) was used to establish stable cell lines to knock down the expression of TM4SF1. The SOX2, TM4SF1 cDNA, c-Myc cDNA, and β-catenin cDNA plasmids (Shanghai GeneChem Co, Ltd., China) were used to upregulate the expression of SOX2, c-Myc and β-catenin. All transfection procedures were performed according to the manufacturer’s instructions.

Real-time PCR was performed as described elsewhere [9]. The primers for RT-PCR were designed in our laboratory and synthesized by Sangon (Shanghai, China, Table S2). mRNA expression was quantitated using the 2-(△Ct sample - △Ct control) method. WB and band density analyses were conducted as described previously [23]. Whole cells were lysed in RIPA buffer containing phosphatase inhibitor and protease inhibitor cocktail. The bicinchoninic acid (BCA) protein assay kit (Pierce, Rockford, USA) was used to determine protein concentrations. Antibodies against proteins are listed in Supplementary Table S1. The bands of interest in the Western blots were normalized to GAPDH or tubulin.

Migration and invasion were examined by Boyden chamber assay (8-μm pore). CRC cells were resuspended in 200 μL FBS-free medium (1 × 10⁵ cells) and added to the top chamber (BD, USA). Medium supplemented with 10% FBS was added to the lower chamber. After 24 h, the cells were fixed and stained, and the number of cells in six randomly selected fields was counted under the microscope. The invasion assay was similarly conducted with a modified Boyden chamber whose upper chamber was coated with Matrigel (BD Bioscience, USA), and the rest of the protocol was performed in a similar manner as the migration assay.

Cells (1 × 10⁵) were cultured in 6-well plates. After 16 h, the complete medium was replaced with a low concentration of serum fresh medium (2%). Consistently shaped wounds were scratched with a 10-μL pipette tip across each well after the cells reached 90% confluence, based on a technique described elsewhere [24]. The cells were gently washed with PBS twice to remove loose cells, and serum-free medium was added. To ensure that the wounds with the same wound area were comparable, multiple positioning marks were made at the center of the denuded surface. The scratch zones were photographed by inverted microscopy at 0 and 24 h. Axio Vision Rel. 4.8 software was used for the measurements and to determine the migrating ability of cancer cells. The data presented were repeated three times.

Cell viability was examined using a CCK-8 assay (Sevenbio, China). Cells were seeded in 96-well plates at a density of 4 × 10³ cells per well in 200 μL medium for 24 h, 48 h and 72 h. The absorbance was detected at 450 nm after the cells were treated with 10% CCK-8 at 37 °C for 2 h. Cell viability was calculated as a ratio of optical density (OD) values of drug-treated samples to those of controls.

Tumour sphere formation was performed as described elsewhere [25, 26]. Cells (4 × 10⁴) were seeded in six-well ultra-low attachment plates per well (Corning, NY, USA) in sphere formation medium: serum-free DMEM/F-12 (Invitrogen) supplemented with β-FGF (10 ng/mL, Invitrogen), B27 (20 ng/mL, Invitrogen), human EGF (20 ng/mL, Invitrogen) and IGF (20 ng/ml, Cell Signaling). Cells were subsequently cultured at 37 °C in an atmosphere containing 5% CO2 to form tumourspheres. After 10–14 days, the images of cells were captured by inverted microscopy at a magnification of × 100 or × 200. The number of tumourspheres was counted and plotted, and the percentage of tumourspheres with diameters of 50–100 μm, 100–150 μm or > 150 μm was calculated and plotted using ImageJ software.

The IF assay was carried out as described previously [27]. Cells were cultured on coverslips in 24-well plates for 24 h, fixed with 4% formaldehyde, blocked with 5% bovine serum albumin, and permeabilized with 0.5% Triton X-100. After that, the washed cells were incubated with polyclonal antibodies against TM4SF1 (Abcam, 1:50), SOX2 (Abcam, 1:80), Zo1 (Abcam, 1:100), and vimentin (Abcam, 1:50), followed by incubation with mouse/goat anti-rabbit antibodies labelled with Cy3 (Beyotime Institute of Biotechnology, 1:800). After incubation with DAPI (Biosharp Biotech, Hefei, China, 1:1000) for 5 min, the cells were observed under a fluorescence microscope within 4 h.

We performed ChIP assays following the instructions of the EZ-ChIP™ kit (Millipore, Billerica, MA). The following antibodies were used for immunoprecipitation: rabbit anti-c-Myc (Abcam, USA) and rabbit anti-IgG (Abcam, USA). The ChIP DNA sample or 1% total input (5 μL) was precipitated, washed, dried, and resuspended in water. The enrichment of the specific amplified region was analysed by qRT-PCR.

Male 4-week-old nude mice were purchased from Beijing HFK Bioscience Co., Ltd. (Beijing, China). Stably transfected cells were injected into the flank of the mice (20 mice in each group). We measured the size of xenografts using callipers (calculated volume = shortest diameter² *longest diameter/2) weekly. The subcutaneous tumours were harvested at 4–5 weeks and then measured and weighed. Then, tumour tissue sections were stained by immunohistochemistry. The metastasis models were induced by tail vein injection as described in previous studies [23]. The mice were anesthetized by chloral hydrate (4%, 0.2 ml/20 g) and sacrificed by cervical dislocation and then their lung tissues were collected. The number and size of lung metastatic tumors were scored and analyzed. The experiment was completed in the Surgery Experimental Animal Center of Tongji Medical College and was approved by the Institutional Animal Care and Use Committee of Tongji Medical College, Huazhong University of Science and Technology.

We investigated the expression of TM4SF1 (cancer vs. normal) in the Oncomine (https://​www.​oncomine.​org), The Cancer Genome Atlas (TCGA, https://​tcgadata.​nci.​nih.​gov/​tcga/​) and the Gene Expression Omnibus (GEO) (https://​www.​ncbi. nlm.​nih.​gov/​gds) databases, and the results suggested that TM4SF1 was significantly upregulated in CRC compared with in adjacent normal tissues (P < 0.01, Fig. 1a, b). The IHC examination revealed an obvious abundance of TM4SF1 protein in CRC tissues (Fig. 1c). In addition, high TM4SF1 expression was significantly correlated with tumour T stage and lymph node metastasis (Fig. 1d, e, P < 0.01) but not with age, sex, tumour size, or tumour differentiation (P > 0.05, Table 1). Moreover, Kaplan-Meier analysis indicated that CRC patients with elevated expression of TM4SF1 suffered from poor survival (Fig. 1f, P < 0.01), which was consistent with the studies conducted by Sveen, Smith and Marisa in the R2 Genomic Analysis Platform (Fig. 1g-i, P < 0.01, https://​hgserver1.​amc.​nl/​cgi-bin/​r2/​main.​cgi). As shown in Fig. 1j, k TM4SF1 was significantly upregulated in CRC tissues compared with normal tissues (fold change = 1.59, P < 0.01). Then, we found that the expression of TM4SF1 in CRC cell lines (HCT116, LoVo, RKO, SW480, DLD1) was higher than that in normal colon epithelial cells (NCM460 cells and FHC cells, Fig. 1l).

Specific shRNAs (sh-Control, sh-TM4SF1#1/2) and TM4SF1 plasmids were transfected into SW480 and LoVo cells, and the expression of TM4SF1 was confirmed by qPCR and WB (Fig. 2a and Fig. S1a). Then, a wound healing assay indicated that depletion of TM4SF1 significantly suppressed scratch wound healing and TM4SF1 overexpression enhanced the migration of CRC cells (Fig. 2b and Fig. S1b). Consistent with these results, the Transwell assay confirmed that TM4SF1 silencing inhibited the migration and invasion of SW480 and LoVo cells (Fig. 2c). In contrast, cells with TM4SF1 overexpression exhibited more aggressive migratory and invasive potential (Fig. S1c). qPCR and WB analysis showed that TM4SF1 knockdown increased the expression of E-cadherin and ZO1 and decreased the expression of vimentin, N-cadherin and MMP9 (Fig. 2d), while TM4SF1 overexpression increased the expression of vimentin, N-cadherin, and β-catenin and decreased the expression of E-cadherin (Fig. S1d). As shown in Fig. 2e, sh-TM4SF1-transfected SW480 cells presented an increased distribution of ZO-1 on the cell membrane and a decreased expression of vimentin in the cytoplasm or nucleus. Interestingly, TM4SF1-silenced SW480 cells showed a conversion from a spindle-like mesenchymal phenotype with obvious characteristics of the interstitial cells to a cobblestone-like shape, as observed under a phase contrast microscope (Fig. 2f). In contrast, TM4SF1 overexpression resulted in decreased expression of ZO-1 and increased expression of vimentin (Fig. S1e). And TM4SF1-overexpressing cells exhibited a more mesenchymal phenotype than control cells, as observed under a phase contrast microscope (Fig. S1f).

To investigate whether TM4SF1 is involved in EMT induced by TGF-β1, SW480 and LoVo cell lines were treated with recombinant human TGF-β1 protein at different concentrations (0, 10, 20 ng/mL) for 48 h. The results showed that TGF-β1 significantly promoted the migration and invasion of CRC cells (Fig. 3a, b and Fig. S1g, h) and increased the expression of Smad2, vimentin, N-cadherin, and MMP9 while decreasing the expression of E-cadherin (Fig. 3c). Furthermore, we found that TM4SF1 deficiency markedly decreased the cell migration and invasion induced by TGF-β1 (Fig. 3a, b), enhanced the expression of epithelial markers (E-cadherin), and suppressed the expression of mesenchyme markers (N-cadherin, vimentin, Smad2, MMP9) at both the protein level and the mRNA level compared to that in the control group (Fig. 3d). These results suggested that TM4SF1 is involved in the process of EMT induced by TGF-β1 in CRC cells.

Cancer stemness has been one of the most important potential mechanisms leading to CRC tumourigenesis and progression [21]. To elucidate whether TM4SF1 is associated with the stem cell-like properties of CRC cells, we performed a sphere formation assay and found that knockdown of TM4SF1 expression significantly decreased the sphere formation and TM4SF1 overexpression enhanced sphere formation capability of both cell lines (Fig. 3e and Fig. S2a). Further, we detected the effect of TM4SF1 on cell proliferation. The CCK-8 assay revealed that the depletion of TM4SF1 strongly diminished CRC cell growth. Conversely, TM4SF1 overexpression significantly promoted the proliferation of SW480 and LoVo cells (Fig. S2b). To further examine the function of TM4SF1 in cell resistance to chemotherapy, we treated the infected CRC cells with fluorouracil. The CCK-8 assay revealed that the viability of sh-TM4SF1-transfected cells was markedly lower than that of the control cells (Fig. 3f). In addition, we found that the IC50 value of TM4SF1 knockdown cells was 50 nmol/L, which is lower than that of the matched group. Consistently, silencing TM4SF1 significantly decreased the expression of stemness markers such as CD133, CD44, SOX2, and ALDHA1 (Fig. 3g), while TM4SF1 overexpression increased the expression of CD133 and SOX2 (Fig. S1d). These data indicated that TM4SF1 can regulate stemness and fluorouracil resistance of CRC cells.

To investigate the molecular mechanism of TM4SF1 in EMT and cancer stemness, we performed RNA-Seq to identify the differentially expressed genes (DEGs) between TM4SF1-silenced CRC cells and control cells. Gene set enrichment analysis (GSEA) indicated that the Wnt signalling pathway was one of the most impaired pathways in TM4SF1-deficient CRC cells (Fig. 4a). The depleted cells showed decreased expression of Wnt/β-catenin target genes (c-Myc, Axin2, TCF7, MMP7) (Fig. 4b). In addition, we found that TM4SF1 knockdown reduced the levels of total β-catenin and the nuclear translocation of β-catenin (Fig. 4c). Based on these observations, we evaluated whether activation of Wnt signalling was able to reverse the suppressive effects of TM4SF1 knockdown on EMT and stemness. Therefore, TM4SF1-deficient cells were treated with or without LiCl (a GSK-3β inhibitor that activates the β-catenin-mediated Wnt signalling pathway). We found that LiCl activated Wnt/β-catenin signalling and upregulated the expression of Wnt/β-catenin target genes (c-Myc, Axin2, TCF7, MMP7) in TM4SF1-deficient cells (Fig. 4d and Fig. S3a). In addition, LiCl markedly suppressed the expression of E-cadherin and enhanced the expression of MMP9, vimentin, CD133, and CD144 at both the protein and mRNA levels (Fig. 4e and Fig. S3b). Further, we found that activation of β-catenin rescued the EMT phenotype and diminished the TM4SF1 deficiency-mediated inhibitory effects on cell migration and invasion (Fig. 4f, g) and cell stemness (Fig. 4h). Then we further detected whether depletion of β-catenin or inhibition of Wnt/β-catenin pathway could reverse the promoting effect of TM4SF1 overexpression in CRC cells. The results suggested that the depletion of β-catenin and Wnt/β-catenin pathway inhibitor, XAV-393, could attenuated the regulatory effect of TM4SF1 overexpression on the expression of Wnt/β-catenin target genes (c-Myc, Axin2, TCF7, MMP7) and the EMT and stemness related markers in CRC cells (Fig. S4a, b). And sh-β-catenin or XAV-939 could markedly reduce TM4SF1-overexoression mediated migration and tumoursphere formation in SW80 and LoVo cells (Fig. S4c, d, e). These results suggested that TM4SF1 may maintain stemness and the mesenchymal phenotype in a Wnt/β-catenin signalling-dependent manner.

It has been revealed that SOX2 is overexpressed in many cancer stem progenitor cells and is involved in chemotherapy resistance and the metastasis and recurrence of cancer [28‐30]. Hence, we asked whether SOX2 is necessary for the TM4SF1-induced EMT and stemness of CRC cells. We investigated the expression of SOX2 (normal vs. cancer) in the GEO database (https://​www.​ncbi.​nlm.​nih.​gov/​gds, GSE70880), which revealed that SOX2 was overexpressed in CRC compared to normal controls (P < 0.05 Fig. 5a). Spearman’s rank correlation analysis revealed that there was a strong positive correlation between the expression levels of TM4SF1 and SOX2 in CRC tissues (r = 0.53, P < 0.05, Fig. 5b). Further, we found that SOX2 was significantly overexpressed in CRC tissues compared to the paired normal tumour adjacent tissues (Fig. 5c, d). As shown in Fig. 5e, the expression of SOX2, as well as Axin2, MMP7, and c-Myc, was significantly downregulated in sh-TM4SF1 SW480/LoVo cells. Then, we constructed a stable SOX2 overexpression vector in TM4SF1 deficient cells, which showed that overexpression of SOX2 significantly potentiated the migration and invasion of SW480 and LoVo cells (Fig. 5f, g). SOX2 overexpression significantly promoted sphere formation in TM4SF1-deficient cells (Fig. 5h). As shown in Fig. 5i, compared to control cells, SW480-SOX2 cells presented decreased distribution of ZO-1 on the cell membrane and increased expression of vimentin and SOX2 in the cytoplasm. Furthermore, we found that upregulation of SOX2 rescued the TM4SF1 deficiency-mediated downregulation of N-cadherin, vimentin, MMP9, CD133, and CD44 and upregulation of the expression of E-cadherin (Fig. 5j and Fig. S5a). All these results indicated that SOX2 upregulation abolished the suppressive effect of sh-TM4SF1 on EMT and stemness. On this basis, we further investigated the effect of Wnt/β-catenin activation on the TM4SF1-mediated regulation of SOX2 expression. We treated SW480/LoVo-sh-TM4SF1 cells with LiCl as described previously. The activation of Wnt signalling restored the expression of c-Myc and SOX2, which was reduced by sh-TM4SF1 (Fig. 6a). Further, we found that SOX2 deficiency significantly downregulated the expression of SOX2 but not the expression of c-Myc in the LiCl-treated cells (Fig. 6b).

In the canonical Wnt/β-catenin signalling pathway, T-cell factor/Lymphoid enhancer factor (TCF/LEF) transcription factors activate the target genes by recruiting the β-catenin transcriptional coactivator and binding to the Wnt responsive DNA elements [31]. However, we did not identify a consensus TCF/LEF binding site in the promoter region of SOX2, which suggests that Wnt/β-catenin signalling may not modulate SOX2 directly and that the target genes may mediate Wnt/β-catenin/SOX2 signalling. It has been reported that c-Myc could activate SOX2 gene transcription by binding to the SOX2 promoter in lung cancer cells [32]. To further investigate the function of c-Myc in the regulation of SOX2 expression in CRC cells, SW480/LoVo cells were transiently transfected with si-c-Myc or si-Control. The results revealed that c-Myc knockdown significantly decreased the expression of SOX2 (Fig. 6c). c-Myc overexpression increased endogenous SOX2 expression at the mRNA and protein levels, suggesting that c-Myc may exert transcriptional regulation of SOX2 (Fig. 6d). Then, we identified an essential binding site for c-Myc (5′-CACATG-3′) in the SOX2 promoter (Fig. 6e), and conducted a ChIP assay with c-Myc-specific antibodies in SW480 cells to investigate whether c-Myc could bind to the sequence of the SOX2 promoter region. Cross-linked chromatin was prepared from SW480 cells, immunoprecipitation was performed using either the anti-c-Myc antibody or IgG, and the sequence containing the putative c-Myc binding site was amplified. As the figure shows (Fig. 6f), c-Myc could bind to the SOX2 promoter. In addition, we found that SOX2 knockdown restored migration and sphere formation in c-Myc-overexpressing cells (Fig. 6g, h). We found that SOX2 deficiency decreased the expression of N-cadherin, CD133, and CD44 and upregulated the expression of E-cadherin. The above results clearly demonstrated that Wnt/β-catenin/c-Myc regulates the expression of SOX2 in CRC cells (Fig. 6i).

To validate the function of endogenous TM4SF1 in vivo, we constructed stably transfected SW480 cells, subcutaneously injected them into BALB/c-nu mice and then measured tumour growth in the xenograft mouse model weekly (Fig. 7a). As shown in Fig. 7b, TM4SF1 knockdown markedly suppressed tumour growth in vivo compared to that of the controls. Then we examined the expression of TM4SF1, SOX2, and β-catenin in the xenografts by IHC, WB and, qPCR; the results revealed that the xenograft derived from TM4SF1 knockdown cells showed significantly diminished expression of TM4SF1, SOX2, β-catenin, and CD133 (Fig. 7c, d and Fig. S6a, b). Furthermore, the effect of TM4SF1 on tumour metastasis was examined by using mouse tail-vein injection model. We determined the metastatic nodules in the lungs 6 weeks after inoculation by haematoxylin and eosin (H&E) staining, and found that compared with the control groups, both the size and number of pulmonary metastatic nodules were suppressed in the TM4SF1-knockdown groups (Fig. 7e and Fig. S6c). These results indicate that TM4SF1 is involved in tumourigenesis and tumour metastasis in vivo.