CD109 acts as a gatekeeper of the epithelial trait by suppressing epithelial to mesenchymal transition in squamous cell carcinoma cells in vitro

There is increasing evidence that the expression of CD109, a GPI-anchored cell surface protein is dysregulated in squamous cell carcinoma (SCC). However, the functional role of CD109 in SCC progression is poorly understood. In current study, we demonstrate that CD109 is a critical regulator of epithelial phenotype in SSC cells. CD109 levels inversely correlate with TGF-β signaling, EMT, migration, and invasion in cultured SCC cells. CRISPR/Cas9-mediated knockout CD109 (CD109 KO) in SCC cells represses epithelial traits and promotes the mesenchymal phenotype, as evidenced by elevated expression of mesenchymal proteins and markers of epithelial to mesenchymal transition. Treatment with recombinant CD109 protein causes CD109 KO cells to regain their epithelial traits. CD109 loss results in pronounced alterations of gene expression as detected by microarray analysis and in dysregulation of 15 important signalling pathways as shown by KEGG pathway cluster analysis. Validation using 52 human oral SCC tumor samples show that CD109 levels inversely correlate with tumor grade and the activation state of one such pathway, the TGF-β signaling pathway. Taken together, our findings highlight a novel role for CD109 as a gatekeeper of the epithelial phenotype by regulating TGF-β pathway in SCC cells.

molecular changes indicative of enhanced EMT traits. Further validations using 52 human oral SCC samples show that CD109 levels inversely correlate with TGF-β signalling and tumor grade. In summary, CD109 functions as a "gatekeeper" of the epithelial phenotype and represses mesenchymal traits through modulation of the canonical TGF-β pathway to inhibit EMT.
As shown by IHC staining (Fig. 1A), CD109 expression was barely detectable in normal epithelia (Fig. 1A top) but was strongly expressed in the well-differentiated OSCC concomitantly with weak expression of TGF-β and P-Smad2 (Fig. 1A middle). In contrast, poorly differentiated tumors exhibited weakly staining for CD109, but markedly intense staining for both TGF-β and P-Smad2 (Fig. 1A bottom). These observations suggest an inverse relationship between CD109 expression and TGF-β/P-Smad2 levels in OSCC tissues.
To confirm this observation, we assessed the correlation between CD109, TGF-β and P-Smad2 in relation to tumor clinical stages and pathological grades. As summarized in Table 1, about 41% of early clinical stage (T1 + T2) patients' tumors exhibited strong CD109 expression, whereas only 7.8% of late stage (T3 + T4) patients' tumors had strong expression of CD109 (χ 2 test: P = 0.01), indicating that CD109 expression was markedly higher in early clinical stage OSCC tumors. Consistent with this result, CD109 expression was significantly higher in low grade OSCC (Grade I + II) than in high grade tumors (Grade III). About 48% of lower grade tumors (Grade I + II) exhibited strong CD109 expression while only 16.3% of Grade III tumors stained strong for CD109 (χ 2 test: P < 0.01). Altogether, the above results suggest that strong CD109 staining is associated with early clinical stage and low-grade tumor in OSCC patients.
In contrast, as shown in the Table 1/ Fig. 1, TGF-β staining is significantly higher in poorly differentiated tumors when compared to well-differentiated tumors. More than 62.8% of high-grade tumors (Grade III) expressed high levels of TGF-β while only 14% low grade (Grade I + II) tumors showed similar expression levels of TGF-β. (χ 2 test: P = 0.01). When correlated with the clinical stage, we observed that strong TGF-β stains preferentially in advanced stages of OSCC. Thus, 44.2% of late clinical stage tumors (T3 + T4) expressed high levels of TGF-β compared to 21.2% of early clinical stage tumors (T1 + T2). Although we observed a clear association of strong TGF-β expression with advanced tumor, the size of our cohort was too small to reach statistical significance (χ 2 test: P = 0.2). Similarly, the expression of P-Smad2 was found to be significantly higher in late clinical stage of tumors. About 43% of advanced clinical stage tumors (T3 + T4) exhibited strong P-Smad2 staining, whereas only 4.2% of early clinical stage tumors (T1 + T2) exhibited similar expression levels (χ 2 test: P = 0.001). A similar association was observed between tumor pathological grade and P-Smad2 expression: more that 48% of high grade (Grade III) tumors expressed high levels of P-Smad2 while only 7.3% of low grade (Grade I + II) tumors showed high levels of P-Smad2 expression (χ 2 test: P = 0.05) ( Table 1/ Fig. 1B). These results suggest that TGF-β/P-Smad2 pathway activation is associated with high tumor grade and advanced clinical stage in OSCC patients.
Finally, we also calculated odds ratios of the association between CD109, TGF-β/P-Smad2, clinical stage and tumor grade in OSCC (Table 1). Patients with tumors staining strongly for CD109 were less likely to develop to advanced clinical stage (OR = 0.119; 95% CI = 0.031 to 0.45). Similarly, these patients were also less likely to develop poorly differentiated tumors (OR = 0.1667; 95% CI = 0.0422 to 0.6579). Conversely, patients with tumors strongly expressing TGF-β or P-Smad2 are more likely to be classified as advanced tumor stage (TGF-β: OR = 2.09; 95% CI = 0.6485 to 6.742) (P-Smad2: OR = 18.67; 95% CI 3.532 to 98.66). Similarly, these patients were more likely to develop poorly differentiated tumors (TGF-β: OR = 10.5; 95% CI 2.085 to 52.87; P-Smad2: OR = 4.24; 95% CI 2.9095 to 19.79). These observations are consistent with previous findings of others and us [10][11][12][13][14][15][16]19 and support the notion that CD109 negatively regulates TGF-β/Smad signaling, as our group previously demonstrated 7-9 . CD109 is heterogeneously expressed in cultured SCC cells and its levels inversely correlate with TGF-β signaling, EMT marker expression, cellular migration, and invasion in vitro. We next examined whether CD109 is heterogeneously expressed in 2 different SCC cell lines (A431 and FaDu). At 80% confluence, A431cells were sorted for CD109 expression into: CD109 high (Top 10%), CD109 low (Lower 10%) and CD109 Med (10-90%) ( Fig. 2A). These sorted cells were maintained in culture for 3 weeks, then sorted again as above (Fig. 2B). A431 cells sorted in the three CD109 expression groups faithfully maintained this heterogeneity during subsequent culture (Fig. 2B). Interestingly, the levels of Alk5 (TGF-β receptor I), and P-Smad2 (downstream effector of TGF-β signalling) were markedly lower in CD109 high A431 cells compared to those in CD109 low cells (Fig. 2C,D). To evaluate whether cellular levels of CD109 modulate EMT responses in SCCs, we measured expression of EMT markers in A431 cells and found that the expression of the mesenchymal proteins Fibronectin, α-SMA and the EMT transcription factors Snail, Slug, and Zeb2 were significantly attenuated in CD109 high A431cells (Fig. 2E,F). This was further confirmed by immunofluorescence microscopy showing that CD109 high cells exhibit markedly reduced mesenchymal proteins and EMT transcription factors expressions (Fig. 2G,H). We then evaluated the effect of CD109 levels on SCC cell migration by an in vitro wound healing assay and found that cell migration rates were significantly reduced in CD109 high versus CD109 low A431 cells, (2019) 9:16317 | https://doi.org/10.1038/s41598-019-50694-z www.nature.com/scientificreports www.nature.com/scientificreports/ regardless of the presence or absence of exogeneous TGF-β (Fig. 2I,J). To determine the effect of CD109 on invasiveness, we carried out matrigel invasion assays. Our results demonstrate that CD109 high A431 cells exhibit a 2-fold reduction in cell invasion compared to CD109 low counterparts (Fig. 2K,L). To rule out the possibility that these results were specific to A431 cells, we repeated these experiments on FADU cells, a model cell line of oral squamous carcinoma and obtained comparable results as with the A431 cells (Fig. 3). Taken together, these observations demonstrate that SCC cells heterogeneously express CD109 and that CD109 low SCC cells exhibit enhanced TGF-β signaling, EMT marker expression as well as elevated cellular migration and invasion compared to CD109 high cells.
Generation and verification of CD109-knockout A431 cell lines. To further investigate the role of CD109 in SCC cells, we used the CRISPR-Case9 gene editing system to generate stable CD109-knockout SCC cell lines 20 . gRNAs that target exon1 or exon2 of the CD109 gene were designed (Fig. 4A) and CD109 negative cells were subsequently generated by single cell sorting to produce clonal CD109 KO cell lines (Fig. 4B). Although we worked on both A431and FaDu cells, we were able to successfully generate only A431 CD109 KO cell lines (CD109-KO) because FaDu cells were unable to survive the single cell sorting process. The efficacy of CD109 knockout was confirmed at the mRNA level (Fig. 4C) and protein level (Fig. 4D,E). We then characterized the mutations by PCR on the genomic DNA isolated from the selected single clones. The sizes of PCR products were different between parental and KO cells (Fig. 4F-H) which suggested that both exon 1 and exon 2 were mutated due to cas9-mediated DNA breaks at the CD109 locus. We then performed PCR to detect exon 3, 4 and 33, intron An inverse relationship between CD109 expression and TGF-β/P-Smad2 in OSCC tissues was observed. (B) Distribution of the clinical stages relative to CD109, TGF-β and P-Smad2 protein expression in OSCC. Patient with high staining of CD109 (Score 2-3) were associated with early clinical stage (T1 + T2) whereas patient with high staining of TGF-β and P-Smad2 were associated with advanced clinical stage (T3 + T4). (C) Distribution of the tumor grade relative to CD109, TGF-β and P-Smad2 protein expression in OSCC. High staining of CD109 (Score 2-3) were associated with well-differentiated status (Grade I + II) whereas high level of TGF-β and P-Smad2 were associated with poorly-differentiated status (Grade III). Chi-square has been used to find the significance of study parameters on categorical scale between two or more groups. A value of P < 0.05 was considered statistically significant. All analyses were performed with Prism GraphPad.
www.nature.com/scientificreports www.nature.com/scientificreports/ 2, as well as the promoter region of CD109, and confirmed that these regions were not affected (Fig. 4G,H), suggesting the specificity of exon 1 or exon2 disruption. We also designed primers to detect SLC17A5 the closest gene upstream to CD109 and found that it is present in the CD109 KO cells (Fig. 4F), suggesting that only the region encompassing exons 1 and 2 of CD109 was deleted without affecting neighboring genes. Sequencing the PCR amplicons corresponding to exons 4 and 33 confirmed that these two exons were parts of the CD109 gene ( Supplementary Fig. S1). Altogether, these results demonstrate that CD109-KO SCC A431cell lines were successfully generated by CRISPR/Cas9 and the mutation is in a frame deletion of exon1 and exon2 resulting in a truncated mutant without affecting the neighboring genomic regions.

Loss of CD109 enhances cell proliferation and supresses apoptosis in SCC cells. Interestingly,
we observed that CD109-KO cells consistently demonstrate dramatically enhanced proliferation and supressed apoptosis even at high cell density, as evidenced by the cumulative growth curve (Fig. 5A,B). In contrast, parental A431 cells dramatically underwent apoptosis at full confluence, resulting in massive cell death, as indicated by decreased number of viable cells (Fig. 5A,B) as well as decreased ki67 (Fig. 5C,D) and increased Caspase 8 expressions when compared to CD109 KO cells (Fig. 5E,F). Importantly, CD109-KO cells showed no evidence of cell death/apoptosis and continually proliferated without cell death and apoptosis, resulting in multiple-layered cell growth (Fig. 5A). Furthermore, both mRNA and protein levels of tumour suppressors TP53 and TP63, which are known to promote apoptosis, and Caspase 3, were significantly downregulated in CD109 KO cells ( Fig. 5G-I). These results suggest that loss of CD109 enhances proliferation and suppresses apoptosis by decreasing the expression of TP53, TP63 and Caspase family genes in SCC in vitro.

CD109 KO SCC cells acquire a mesenchymal phenotype, and exhibit enhanced TGF-β signaling, EMT, motility and invasiveness. Strikingly, CD109-KO cells lose their epithelial tight cell-cell contacts
and acquired a mesenchymal phenotype characterized by a spindle-like fibroblastic shape (Figs 4E and 5A). Consistent with these morphological changes, RT-PCR revealed that mRNA levels of fibronectin (FN), vimentin (VIM), a-SMA and collagen were significantly increased in CD109-KO cells (Fig. 6A). In line with these results, as detected by western blots and immunofluorescence microscopy, the mesenchymal proteins such as vimentin, fibronectin, N-cadherin and EMT master transcription factors Zeb2, Snail were markedly increased whereas epithelial protein marker E-cadherin was markedly down-regulated in CD109 KO cells ( TGF-β is a potential EMT inducer and our group has identified that CD109 is a co-receptor of TGF-β and counteracts TGF-β function 7,8 . We therefore hypothesized that the enhanced EMT response in CD109-KO cells might be due to enhanced TGF-β signaling because of the absence of CD109. Indeed, we observed that the levels of P-Smad2, a downstream effector of TGF-β signaling, were markedly elevated in CD109 KO cells as shown by Western blot (Fig. 6D) and immunofluorescence (Fig. 6E,F), indicating an enhanced TGF-β signaling activity in CD109-KO cells. Collectively, these data suggest that CD109 is required to maintain an epithelial phenotype and supresses the EMT process by inhibiting the TGF-β pathway.
Since enhanced TGF-β signalling and EMT phenotype are usually accompanied by enhanced migration and invasive capacity, we then evaluated the effect CD109 loss on these endpoints by a wound healing assay (Fig. 6G,H) and matrigel invasion assay (Fig. 6I,J). The loss of CD109 increased cellular migration in the wound healing assay by 50% (Fig. 6G,H) and also had a substantial effect on cellular invasiveness, leading to a 3-fold increase of invasion in matrigel trans-well assays (Fig. 6I,J), suggesting an important role of CD109 in the regulation of the invasive behavior of SCC cells. Together, our results show that CD109-KO SCC cells exhibit dramatically increased migratory and invasive behavior.  truly depends on CD109 loss and to examine if these effects could be rescued by recombinant human CD109 protein (rhCD109), we treated CD109-KO cells with 10 uM rhCD109 protein. We observed that the KO-cells regained the epithelial phenotype and forfeited their spindle-shaped mesenchymal morphology (Fig. 7A). This was further confirmed by the upregulation of E-cadherin (CDH1) and downregulation of mesenchymal protein FN, a-SMA, and the EMT markers Snail and Twist in the rhCD109 treated SCC cells (Fig. 7B,C). Consistent with this result, immunofluorescence microscopy revealed markedly increased E-cadherin (CDH1) expression (Fig. 7D,G), and decreased expression of a-SMA (Fig. 7E,G) and Snail (Fig. 7F,G), in the rhCD109 treated KO cells. More importantly, the levels of E-cadherin, a-SMA and Snail in rhCD109-treated (rescued) CD109 KO cells reached similar levels as in parental A431 cells, as evidenced by Western blot (Fig. 6B,C) and immunofluorescence microscopy ( Fig. 7D-G). The observation that rhCD109 treatment can rescue the epithelial phenotype of CD109 KO-cells demonstrates that the loss of CD109 plays a causative role in mediating the above observed effects and that the phenotype of CD109-KO cells is unlikely to result from potential off-target mutations produced by the CRISPR/Cas9. Overall, our data support the notion that CD109 functions a gatekeeper of epithelial traits of the SCC cells by suppressing the EMT process via inhibition of the TGF-β pathway. . To obtain further insight into how the loss of CD109 affects cellular signalling pathways in SCC cells, we subjected the dysregulated genes to a KEGG pathway cluster analysis, which revealed that 15 important pathways are significantly deregulated as a consequence of CD109 loss (Fig. 8A), including signalling pathways associated with: miRNA expression (Fig. 8A); cancer and apoptosis ( Fig. 8A and Supplementary Excel File-1); cellular adhesion, collagen and ECM deposition (Fig. 6b and); TGF-β and EMT ( Fig. 8A and Supplementary Excel File-1); and matrix metalloproteinases (MMPs) (Supplementary Excel File-1). Together, the above results suggest that loss of CD109 led to dysregulation of genes involved in TGF-β signaling, EMT process, ECM, cell migration/adhesion, apoptosis, remodeling and composition of the cytoskeleton.

Discussion
CD109 is a GPI-anchored cell surface glycoprotein that is expressed in a wide variety of cell types including skin cells, mesenchymal stem cells and tumor cells [13][14][15][16][17]21,22 . Our group identified CD109 as a component of the TGF-β receptor system and has shown that it negatively regulates TGF-β signaling 7,8 . We have also reported that membrane-anchored CD109 promotes localization of the TGF-β receptors into the caveolar compartment and enhances TGF-β receptor degradation 7,9 . We also demonstrated that CD109 inhibits TGF-β-induced EMT through differential regulation of Smad2/3 and MAP kinase pathways in keratinocyte and SCC cell lines (Bizet et al., in preparation). In addition, CD109 can be released from the cell surface by enzymes such as phosphatidylinositol phospholipase C 23 . Furthermore, CD109 can be processed by furinase and cleaved into a soluble CD109 that can be secreted into culture medium 19 . Interestingly, the soluble form of CD109 can also bind TGF-β1 with high affinity and sequester it away from the TGF-β signaling receptors 23 .
Although TGF-β pathway has been implicated in EMT and metastasis, and CD109 counteracts TGF-β function, so far few studies have reported on the relationship among CD109, TGF-β, EMT, migration and invasiveness in SCC cells 24,25 . Here, we initially demonstrate that SCC cells express heterogenous levels of CD109 and that CD109 expression is inversely correlated with TGF-β signaling, EMT process, motility and invasion of SCC cells. We also show that CD109 KO SCC cells acquire a mesenchymal phenotype, and loss of CD109 promotes EMT process, cell motility and invasiveness likely by enhancing the TGF-β/Smad-Sail/Zeb2 axis. Moreover, the effect of the deletion of CD109 in SCC cells can be rescued by rhCD109 as evidenced by the finding that the rhCD109 subpopulations. 10,000 cells were seeded on a BioCoat ™ Matrigel ® Invasion Chamber for 24 hours. Cells that invaded through the matrigel-coated membrane were stained with 1% crystal violet, photographed, and counted. The levels of CD109 are inversely corelated with cell invasion. All the results are expressed as the mean ± S.D. of three independent experiments. Significance is calculated using a One-Way ANOVA *P < 0.05. **P < 0.01 and ***P < 0.001. The graphs display the raw data. Scale bars: 30 μm, 100 μm and 300 μm, as indicated. (2019) 9:16317 | https://doi.org/10.1038/s41598-019-50694-z www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ It is known that the expression of TGF-β is increased in various tumor types and in general TGF-β expression correlates with poor patient prognosis [26][27][28] . It is also well documented that TGF-β signaling promotes induction of EMT, invasiveness, and metastasis [29][30][31] . TGF-β has been shown to have a dual role in cancer progression: it is a tumor suppressor in early cancer development and it switches to become a tumor promoter in late stages of tumor by promoting EMT and metastasis 32,33 . As cancer progresses to advanced stages, cancer cells lose www.nature.com/scientificreports www.nature.com/scientificreports/ their response to TGF-β-induced growth suppression and become sensitive to the tumor-promoting effects of TGF-β 31 . We hypothesized that when the tumor-promoter function of TGF-β predominates, CD109 can repress TGF-β induced EMT, leading to a decrease in motility and invasion. Indeed, we found that the loss of CD109 enhances TGF-β-induced EMT, up-regulates mesenchymal proteins while downregulating E-cadherin and enhancing migration and invasion in A431 cells. Together, these findings suggest that CD109 plays an important role in the modulation of migration and invasion of SCC cells by blocking TGF-β-mediated EMT. Consistent with this notion, CD109-positive basal-like breast carcinoma shows reduced fat invasion as compared to the CD109-negative counterparts, suggesting that CD109 expression may affect biological properties of cancer cells 15 . Moreover, reduced expression of CD109 on hepatocellular carcinoma tumor vessels is associated with large tumor size, microvascular invasion, and advanced tumor stage and CD109 is an independent risk factor for disease-free survival (P = 0.001) 34 .
Loss of E-cadherin not only causes deregulation of cytoskeleton and loss of cell polarity but also leads to increased cell motility and invasiveness 35 . E-cadherin mediates cell-cell adhesion, promotes cell clusters and restricts cell motility toward the extracellular matrix 36 . In the present study, we found that loss of CD109 diminished the E-Cadherin expression while rhCD109 protein rescued the E-cadherin expression in CD109-KO cells, suggesting that CD109 can maintain the E-cadherin expression.
Smads play a crucial role in mediating the intracellular signalling of TGF-β and the TGF-β-Smads2/3 pathway is often aberrantly regulated in cancer 37 . Interestingly, we show that the deletion of CD109 results in an enhanced phosphorylation of Smad2, in an induction of slug, snail and Zeb2 as well as the suppression of E-cadherin with concomitant increase in mesenchymal proteins, such as fibronectin, vimentin and a-SMA. Conversely, the effect of the deletion of CD109 in SCC cells can be rescued by rhCD109, resulting in CD109-KO cells regaining the epithelial phenotype and losing the mesenchymal morphology. This is associated with upregulation of E-cadherin and downregulation of mesenchymal protein fibronectin, vimentin, a-SMA, and the EMT markers.
Our findings highlight a new role of CD109 as a gatekeeper of epithelial cancer cell phenotype and indicate that the underlying mechanism involves inhibition of TGF-β signalling leading to the attenuation of migration and invasion of SCC cells. A schematic model depicting the possible mechanism underlying CD109 function as a gatekeeper of the epithelial phenotype by inhibiting the EMT process is shown in Fig. 8C. In the absence of CD109, TGF-β-induced EMT is unopposed resulting in the loss of epithelial traits and in the acquisition of a mesenchymal phenotype of SCC cells (Fig. 8C). Together, the findings from current study suggest that CD109 is a master regulator of epithelial traits and gatekeeper of epithelial phenotype. CD109 has the potential to play an essential role in SCC progression and is a potential target for therapeutic intervention in SCC in vitro.
Although previous work from our laboratory has provided the first evidence to indicate that CD109 is a TGF-β co-receptor and later emphasized its role as a negative regulator of TGF-β signaling [7][8][9] , there is now increasing evidence to suggest that CD109 is more than a TGF-β co-receptor that inhibits TGF-β signaling. A recent report provides evidence to indicate CD109 enhances EGF signaling while attenuating TGF-β signaling, in SK-MG1 glioblastoma cells 38 . Furthermore, CD109 has been proposed to act as a novel pro-metastatic factor that can regulate metastatic ability in lung adenocarcinoma by modulating Jak-Stat3 activity in vivo 39 . It is therefore possible that the consequences of loosing CD109 in cancer progression and metastasis are more complex. While CD109 has the ability to inhibit TGF-β-induced EMT and to enhance migration of cancer cells, it may promote tumor initiation and metastasis by enhancing the EGF or the JAK-STAT3 pathway. A precise definition of the molecular pathways regulated by CD109 and the consequences of their dysregulation in cancer progression and metastasis remain to be determined.

Methods
Study population. SCC samples were surgically removed at the Department of Otorhinolaryngology Head and Neck-Surgery (McGill University). This study was approved by the Research Ethics Committee at the Jewish General Hospital (Protocol #11-093) and carried out in accordance with the approved guidelines and regulations. All experimental protocols were approved by Research Review Office (RRO, Canada). Patients were advised of the procedures and provided written informed consent. The eligibility criteria included previously untreated patients, without a second primary tumor and submitted for treatment in the same institution.
Tissue microarray (TMA) platform. Tissue cores with a dimension of 1.0 mm from each specimen were punched and arrayed in duplicate on a recipient paraffin block using a Tissue Microarrayer (Beecher Instruments ® , Silver Springs, MD, USA). After cutting sections from the recipient block and transferring these with adhesive tape to coated slides for subsequent UV cross-linkage (Instrumedics Inc ® , Hackensack, NJ, USA), the slides were dipped in a layer of paraffin to prevent oxidation and stored in a freezer at −20 °C. Immunohistochemistry (IHC) analysis and scoring. IHC was carried out on TMA as described earlier 40 . Incubations with the primary antibodies were conducted overnight at 4 °C for: mouse monoclonal anti-CD109 (Cat. No. 556039, BD Biosciences). TGFBR1/ALK5 Antibody (aa281-295, LS-C312882), anti-P-Smad2 (Cat. No. ab188334, Abcam) and anti-TGFβ1 (Cat. No. ab92486, Abcam). The sections were washed and incubated with secondary antibodies (Advanced ™ HRP Link, DakoCytomation, K0690, Denmark) followed by the polymer detection system (Advanced ™ HRP Link, DakoCytomation). Reactions were developed with a solution containing 0.6 mg/mL of 3,3′-diaminobenzidine tetrahydrochloride (DAB, Sigma, St Louis, MO, USA) and 0.01% H 2 O 2 and then counter-stained with Mayer's hematoxylin. Positive controls (a tissue known to contain the antigen under study) were included in all reactions in accordance with manufacturer's protocols. The negative control consisted in omitting the primary antibody and incubating slides with PBS and replacing the primary antibody with normal serum. The IHC reactions were performed in duplicate on different TMA levels, representing four-fold redundancy for each case. The second slides were 25 sections deeper than the first, resulting in at least 250 μm of distance between the two sections with different cell samples for each tumor.
IHC analysis was performed blindly to the clinical aspects and conducted by two independent certified pathologists. Each core was scanned in low power field to choose the most stained area predominant in at least 10% of tumor cells. The presence of a clearly visible dark brown precipitation was considered positive for immunostaining. Slides were analyzed as previously described 41 in accordance to the staining intensity: 0, no visible reaction; 1, weak expression, 2: strong positivity. For statistical analysis, the samples were categorized into groups: negative and positive cases. Wound healing assay. A431 CD109 high , CD109 med and CD109 low cells were seeded at a density of 6 × 10 5 cells/well into Costar ® 6 Well Clear TC-Treated Multiple Well Plates (Product #3516, Corning Inc, USA) and cultured for ~48 h or until the cells had reached ~90% confluency. Cells were then pre-incubated with serum-free medium (SFM) for 24 h to inhibit cell proliferation 42 . The monolayer of A431 cells were scratched across the centre with a sterile 200 μl pipette tip to create a cell-free line. The culture medium was aspirated and washed three times to remove cellular debris. The culture plates were replenished with serum free DMEM in the absence or presence of 100 pm TGF-β1 (Cat. No. 7754-BH-005, R&D). Photographs were taken immediately (0 h) and 24 h after the scratch and Image-J software was used to measure the width of the wound area. The experiments were repeated 3 times. Cell migration was expressed as percentage of the scratch area filled by migrating cells at 24 h post scratch: migration rate = (T0 hr scratch width − T24 hr scratch width)/T0 hr scratch width) × 100%.

Matrigel invasion assay. The Matrigel invasion assay was done by using the BD Biocoat Matrigel Invasion
Chamber (Cat. No. 354480; pore size: 8 mm, 24-well; BD Biosciences) according to the manufacturer's protocol. Briefly, the control inserts or matrigel-coated inserts were rehydrated with plain DMEM for 2 h before use. Cells (5 × 104 cells) in 500ul Dulbecco's modified Eagle's medium without serum were seeded on the upper chamber; the lower chamber was filled with DMEM supplemented with 10% FBS as chemoattractant. After 24 hours, cells on the upper side of the membrane were wiped off; cells on the lower side of the membrane were fixed for 10 min by cold methanol and stained with 0.1% crystal violet solution for 20 min and washed with PBS 3 times. Cell numbers were counted using an inverted microscope at ×200 magnification with 10 fields of view, and the mean values were taken as the invasive cell number. All assays were done in triplicate for at least three independent experiments RNA isolation and quantitative real-time PCR. Total RNA was isolated from cultured cells using RNeasy mini Kit (Cat. No. 74104, QIAGEN). Reverse transcription (RT) was done on 1 μg of cellular total RNA using the MML-V reverse transcriptase (Invitrogen). Quantitative real-time PCR (qPCR) was performed using the Platinum SYBR Green SuperMix (Invitrogen) and an ABI Prism 7500 Real-Time PCR apparatus (Applied Biosystems). Primer sets used were as follows: