Integrin α5 subunit is required for the tumor supportive role of fibroblasts in colorectal adenocarcinoma and serves as a potential stroma prognostic marker

The tumorigenesis of colorectal cancer (CRC) is a complicated process, involving interactions between cancer cells and the microenvironment. The role of α5 integrin subunit in CRC remains controversial, and previous studies mainly focused on cancer cells. Herein, we report an important role of α5 in stroma fibroblasts in the tumorigenesis of CRC. The expression of α5 was found to be located in colorectal tumor stroma rather than in epithelia cancer cells. Immunofluorescence colocalization and gene correlation analysis confirmed that α5 was mainly expressed in cancer‐associated fibroblasts (CAFs). Moreover, experimental evidence showed that α5 expression was required for the tumor‐promoting effect of fibroblast cells. In an in vivo xenograft nude mice model, α5 depletion in fibroblasts dramatically suppressed fibroblast‐induced tumor growth. In an in vitro cell coculture assay, α5 depletion or knockdown reduced the ability of fibroblasts to promote cancer cell migration and invasion compared with wild‐type fibroblasts; moreover, we observed that the expression and assembly of fibronectin were downregulated after α5 depletion or knockdown in fibroblasts. Analysis of the RNA‐Seq data of the Cancer Genome Atlas cohort revealed that high expression of ITGA5 (α5 integrin subunit) was correlated with poor overall survival in colorectal adenocarcinoma, which was further confirmed by immunohistochemistry in an independent cohort of 355 patients. Thus, our study identifies α5 integrin subunit as a novel stroma molecular marker for colorectal adenocarcinoma, offers a fresh insight into colorectal adenocarcinoma progression, and shows that α5 expression in stroma fibroblasts underlies its ability to promote the tumorigenesis of colorectal adenocarcinoma.

The tumorigenesis of colorectal cancer (CRC) is a complicated process, involving interactions between cancer cells and the microenvironment. The role of a5 integrin subunit in CRC remains controversial, and previous studies mainly focused on cancer cells. Herein, we report an important role of a5 in stroma fibroblasts in the tumorigenesis of CRC. The expression of a5 was found to be located in colorectal tumor stroma rather than in epithelia cancer cells. Immunofluorescence colocalization and gene correlation analysis confirmed that a5 was mainly expressed in cancer-associated fibroblasts (CAFs). Moreover, experimental evidence showed that a5 expression was required for the tumor-promoting effect of fibroblast cells. In an in vivo xenograft nude mice model, a5 depletion in fibroblasts dramatically suppressed fibroblast-induced tumor growth. In an in vitro cell coculture assay, a5 depletion or knockdown reduced the ability of fibroblasts to promote cancer cell migration and invasion compared with wild-type fibroblasts; moreover, we observed that the expression and assembly of fibronectin were downregulated after a5 depletion or knockdown in fibroblasts. Analysis of the RNA-Seq data of the Cancer Genome Atlas cohort revealed that high expression of ITGA5 (a5 integrin subunit) was correlated with poor overall survival in colorectal adenocarcinoma, which was further confirmed by immunohistochemistry in an independent cohort of 355 patients. Thus, our study identifies a5 integrin subunit as a novel stroma molecular marker for colorectal adenocarcinoma, offers a fresh insight into colorectal adenocarcinoma progression, and shows that a5 expression in stroma fibroblasts underlies its ability to promote the tumorigenesis of colorectal adenocarcinoma.

Introduction
Colorectal cancer (CRC) is the third most commonly diagnosed cancer globally. Although 5-year relative survival of the patients has reached almost 65% in high-income countries, the disease remains a significant cause of mortality (Brenner et al., 2014;Siegel et al., 2017).
Colorectal cancer is a complex entity composed of multiple cell types and extracellular molecules in which the epithelial cancer cells actively crosstalk with the tumor microenvironment (TME). TME, or tumor stroma, which comprises fibroblasts, immune cells, blood and lymphatic vessels, and extracellular matrix (ECM), plays an important role in modulating the growth of cancer and therapeutic response through cytokines, cell-cell interactions and mechanical sensing (Bahrami et al., 2018;Beauchemin, 2011). CAFs are one of the most abundant cell types in the tumor stroma and has been shown to promote tumor growth and invasion, and confer resistance to chemotherapy (Glentis et al., 2017;Ishimoto et al., 2017;Lotti et al., 2013). Recent studies showed that CAFs could trigger cancer cell migration and invasion through assembling ECM protein-fibronectin or intracellular signaling pathway via integrins Erdogan et al., 2017;Knuchel et al., 2015).
Integrins are a large family of a/b heterodimeric cell adhesion molecules that mediate cell-cell, cell-ECM and cell-pathogen interactions. Integrins play an important role in signal transduction, cell migration and proliferation (Harburger and Calderwood, 2009;Hynes, 2002). The integrin subunit a5 (ITGA5) combines with b1 integrin subunit (ITGB1) to form a5b1 complex and serves as a receptor for fibronectin (Schaffner et al., 2013). Ubiquitination and degradation of a5b1 integrin in lysosome are important for proper fibroblast migration on fibronectin (Lobert et al., 2010), and accumulation of a5 integrin at the plasma membrane by Rab21 is required for CAFs to promote the invasion of squamous carcinoma cells in vitro (Hooper et al., 2010). However, the significance of a5 in tumor stroma and CAFs is not well studied in CRC.
In CRC cells, the role of a5b1 is controversial, as some studies showed that a5b1 is required for colon cancer cell survival, invasion and migration (Lee and Juliano, 2000;Murillo et al., 2004), whereas in another study a5b1 exerted its tumor suppressor-like activity in colon cancer cells by inhibiting HER-2 signaling (Kuwada et al., 2005). Prognostic significance of a5b1 in CRC was identified in a study with a small cohort of patients, and a5b1 expression was shown to increase significantly in the cytoplasmic of cancer cells (Yang et al., 2013). In another study, a5 integrin gene and protein levels were only increased in mucinous carcinomas but not in common colorectal adenocarcinoma (Denadai et al., 2013). These conflicting reports reflect our lack of understanding of the specific role of a5 in CRC.
Herein we proposed a new model assuming that a5 expression in stroma fibroblasts was required in promoting tumor growth in vivo and cancer cell migration and invasion in vitro, accompanied by downregulated expression and assembly of fibronectin after a5 depletion in fibroblasts. Moreover, clinically we observed a positive correlation between a5 and fibronectin expression, and we showed that high expression of a5 was correlated with poor overall survival (OS) of patients with colorectal adenocarcinoma. These data suggest an important role of fibroblast-expressed a5 in tumorigenesis of colorectal adenocarcinoma.

Bioinformatic analysis of TCGA data
We analyzed ITGA5 expression level using the RNA-Seq data in 592 samples of CRC patients that with complete follow-up information, downloaded from The Cancer Genome Atlas cohort (TCGA) database (https://portal.gdc.cancer.gov/). For the clinicopathological parameters, we included patient age, sex, tumor stage, local invasion depth (T), lymph node involvement (N), present of distant metastasis (M), lymphatic invasion and tumor location, all of which were added to the analysis of multivariate Cox regressions.

Hierarchical cluster analyses
Hierarchical cluster analyses were done to obtain a visual representation of gene expression correlation between ITGA5 and CAFs markers. The hierarchical clustering algorithm used is based closely on the average-linkage method as described previously (Eisen et al., 1998) and the object of this algorithm is to compute a dendrogram that assembles all elements into a single tree. Basically, the gene expression clustering was performed using Software GENE CLUSTER 3.0, Stanford University, Califonia, USA and then visualized using Java TREEVIEW, Stanford University, Califonia, USA.

Patients and sample collection
Primary tumor tissues were obtained from patients with colorectal adenocarcinoma (AC) who underwent a surgical resection. A total of 355 formalin-fixed, paraffin-embedded (FFPE) specimens (one from each case) were collected from the Department of Pathology of Shanghai Tenth People's Hospital Affiliated to Tongji University, between 2008 and 2013. Additionally, 24 colorectal AC tissues were collected at the time of surgical resection and were immediately snap-frozen in liquid nitrogen before storage at À80°C. Exclusion criteria were: (1) radiotherapy or chemotherapy before surgery; (2) presence of hereditary or inflammation-associated CRC; (3) the presence of mucinous adenocarcinomas (MAC) or other pathological complications; (4) incomplete follow-up information; (5) non-CRC-related cause of death; (6) poor quality of the tissue. According to the TNM staging system, tumors were classified independently by two pathologists. The clinicopathological parameters included were age, sex, tumor size, tumor location, pathological grade and tumor stage. All tissue samples were collected from patients with appropriate informed written consent and the study was performed in accordance with the ethical standards of the Declaration of Helsinki. The study was approved by the Ethics Committee of the hospital.

Tissue microarray (TMA) construction
Tissue microarray was prepared to examine the expression of a5. Briefly, a tissue core with a 0.4-mm diameter was punched from a representative area that was examined by hematoxylin and eosin (HE) staining from FFPE tissue blocks of the primary tumor and then transferred to a recipient TMA block. Each tissue core was assigned a unique TMA location.

Immunohistochemistry (IHC)
Tissue microarray or whole sections of FFPE specimens were cut into 4-lm sections and placed on polylysinecoated slides. For immunohistochemical staining, the slides were deparaffinized in xylene and rehydrated in graded alcohols and distilled water, followed by antigen retrieval. Antigen retrieval was achieved by microwaving in 10 mM sodium citrate buffer (pH 6.0) for about 20 min. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide in methanol at 37°C for 15 min, and nonspecific protein binding was blocked with 5% goat serum. Specimens were then incubated with primary antibody against a5 (ab150361; Abcam, England or HPA002642, Sigma, St. Louis, MO, USA) overnight in a humidity chamber at 4°C, and secondary antibody was applied for 30 min at room temperature. All sections were visualized with diaminobenzidine and counterstained with hematoxylin.
The percentage of immunostaining and the staining intensity (0, negative; 1+, weak; 2+, moderate; and 3+, strong) of the slides were recorded. A histochemistry score (H-score) was calculated as described previously (Azim et al., 2015) using the following formula: Hscore = Σ(I 9 Pi) = (percentage of cells of weak intensity 9 1) + (percentage of cells of moderate intensity 9 2) + (percentage of cells of strong intensity 9 3), where I = intensity of staining and Pi = percentage of stained tumor cells, producing scores ranging from 0 to 300. The maximum H-score would be 300, corresponding to 100% of cells with strong intensity. All sample slides were scored separately by two pathologists blinded to the clinical information.

Tumorigenicity assay in nude mice
All the animal studies were administered according to the guidelines of Institution Animal Care and Use Committee and all the protocols were approved by Tongji University. Male BALB/c nude mice aged 6 weeks were housed in a specific pathogen-free environment in the Animal Laboratory Unit, Tongji University, China. SW480 cells (or DLD-1 cells, 2 9 10 6 ) with or without CCD-18Co fibroblasts (2 9 10 6 ) were suspended in 0.1 mL of PBS and injected subcutaneously into the flanks of the mice (one injection/mouse). CCD-18Co fibroblasts alone (2 9 10 6 ) were used as control. Tumor growth was monitored two times per week, and tumor volume (V) was monitored by measuring the length (L) and width (W) with a Vernier caliper and calculated using the formula V = (L 9 W 2 ) 9 0.5. After 23 or 26 days, the mice were sacrificed by cervical dislocation by an expert and qualified person, the tumor samples were collected. Tumors were harvested, FFPE and used for IHC staining. The relative expression quantitation of the target gene was determined using 2 ÀDDCt method. For each sample, the assay was performed in triplicate.

Western blotting
The CCD-18Co fibroblast cells were cultured until confluence and then cells starved in serum-free medium for 48 h before myofibroblasts were differentiated by incubation of fibroblast cultures in serum-free medium containing 10 or 20 ngÁmL -1 recombinant transforming growth factor (TGF)-b1 (Peprotech, Princeton, NJ, USA) for 72 h. Subsequently, total cellular RNA was prepared as mentioned above. Cell protein was prepared as described below.

Immunofluorescence microscopy
Immunofluorescence was performed according to standard protocols (Abcam). Paraffin-embedded sections were deparaffinized in xylene and rehydrated in graded alcohols and distilled water, followed by antigen retrieval. Antigen retrieval was achieved by microwaving in 10 mM sodium citrate buffer (pH 6.0) for about 20 min. The slides were blocked at room temperature for 60 min using blocking reagent with 5% goat serum. Specimens were then incubated with primary antibody overnight in a humidity chamber at 4°C. These antibodies included a5 (ab150361; Abcam or MAB1969M; Merck Millipore, Burlington, MA, USA), b1 (ab52971; Abcam), a-SMA (ab124964; Abcam), vimentin (ab92547; Abcam) and fibronectin (ab32419; Abcam). Secondary antibodies were applied for 60 min at 4°C using Alexa Fluor 568 goat anti-rabbit IgG or Alexa Fluor 647 goat antimouse IgG (Invitrogen, Carlsbad, CA, USA). DAPI staining was included in the mounting medium. Images were obtained using a Zeiss 710 confocal microscope, Oberkochen, Germany.
SW480-GFP and CCD-18Co (vector or sgRNA) coculture cells or CCD-18Co (vector or sgRNA) monoculture cells were grown on glass coverslips for 24 h and then cells were fixed with 4% formaldehyde and permeabilized with 0.1% Triton X-100. Followed blocking for 60 min with PBS containing 1% BSA, 5% goat serum, primary antibody was incubated overnight at 4°C with an antibody against a5 (ab150361; Abcam) or fibronectin (ab32419; Abcam). After three rinses, the cells were stained with Alexa Fluor 568 goat anti-rabbit IgG (Invitrogen) for 1 h at room temperature and cells incubated with Phalloidin-iFluor 647 (ab176759; Abcam). DAPI staining was included in the mounting medium. Images were obtained using a Zeiss 710 confocal microscope.

Cell migration and invasion assay
The Transwell migration assay was performed as described (Otomo et al., 2014), SW480-GFP cells (1 9 10 4 ) alone, or together with CCD-18Co (vector or sgRNA) cells (1 9 10 4 ), were suspended in 500 lL of serum-free DMEM, and then plated onto 8.0-lm pore size of 24-well TranswellÒ inserts (Corning, Corning, NY, USA). The bottom well was filled with 750 lL of DMEM containing 5% FBS. After being incubated at 37°C with 5% CO 2 for 16 h, the cells on the upper surface of the Transwell insert were scraped using cotton swabs. The migrated cancer cells on the lower surface were observed under a fluorescence microscope (Leica, Wetzlar, Germany). The same principles were used for the invasion assays; inserts contained a thin MatrigelÒ layer (Corning) and were plated with double-numbered SW480-GFP cells (2 9 10 4 ) alone or together with CCD-18Co (vector or sgRNA) cells (2 9 10 4 ). Invasion was allowed to occur for 22 h.
For testing the effect of a5 knockdown in fibroblasts, DLD-1-GFP cells were plated together with CCD-18Co cells treated with or without siRNA (Mock, siCtr or sia5), and the same principles were used for cell migration and invasion assay.

Statistical analyses
SPSS 22.0 (SPSS Inc., Chicago, IL, USA) and GRAPH-PAD PRISM v.7 (La Jolla, CA, USA) were used for statistical analyses. The Kaplan-Meier method was performed to estimate OS with log-rank test. Cox proportional hazards model was used to compare hazard ratios in both uni-and multivariable analyses. The multivariable analysis included known clinical relevant parameters. Results are presented in the multivariable analysis as hazard ratios including 95% confidence intervals. The correlations between ITGA5 expression and clinicopathological features were analyzed using Chi-square or Fisher's exact test wherever appropriate. Pearson's correlation was used to analyze correlations between the expression of ITGA5 and CAFs markers.
Each experiment was repeated independently a minimum of three times in the same conditions. Acquired data were presented as mean values, and error bars represent the SEM. Student's t-test was used for comparison between two groups, and a one-way ANOVA was used to compare multiple groups with Bonferroni's multiple comparison correction. For all analyses, two-tailed P-values below 0.05 were considered significant: * P < 0.05, ** P < 0.01, *** P < 0.001.

a5 Integrin subunit located in the tumor stroma and mainly expressed in CAFs
To investigate the specific role of a5 integrin subunit in CRC, we first examined the expression level of a5 in freshly isolated tumor and adjacent normal tissues from 24 patients using qPCR. We observed that most tumor tissue samples from patients with colorectal adenocarcinoma (21/24; 87.5%) showed elevated levels of ITGA5 compared with the corresponding adjacent normal tissue samples (P = 0.0003; Fig. 1A). We used IHC to analyze further the expression pattern of a5 integrin in tumor. Surprisingly, we found that a5 was largely expressed in the tumor stroma, whereas only weak to negligible expression levels were observed in tumor epithelial cells. In normal tissues, a5 integrin also mainly located in the lamina propria and muscularis mucosae. The result was confirmed using two completely independent antibodies against a5 in whole-slide tissue sections of colorectal adenocarcinoma and normal adjacent tissue (Fig. 1B). Furthermore, we performed immunofluorescence staining to examine the location of a5 in human normal colorectal tissues. As shown in Fig. 1C, a5 integrin was mainly located in the lamina propria rather than epithelial cells, whereas b1 integrin subunit was largely expressed on epithelial cells as reported (Desgrosellier and Cheresh, 2010).
Carcinomas are heterogeneous tissues composed of cancer cells and stromal cells, and CAFs have been shown to be the most predominant cell type within the stromal components of carcinomas (Orimo and Weinberg, 2006). Thus we analyzed the correlation between the expression levels of ITGA5 and CAF-related marker genes in 24 freshly isolated colorectal adenocarcinoma samples by assessing the mRNA expression using qPCR. The CAF-related markers are comprised mainly of six genes -ACTA2 (a-SMA), VIM (vimentin), PDGFRB (platelet-derived growth factor receptor b, PDGFRb), fibroblast activation protein (FAP), CXCL-12 (stromal cell-derived factor 1a, SDF-1a) and IL-6 (interleukin-6)which were selected based on previous reports (Rasanen and Vaheri, 2010). The heat map showed similar expression patterns between ITGA5 and CAF marker genes ( Fig. 2A), and calculated Pearson's correlation coefficient showed that ITGA5 was significantly correlated with these CAF marker genes (Figs 2B and S1A). The correlation was confirmed by analyzing the RNA-seq database of 517 colorectal adenocarcinomas extracted from TCGA cohort (Fig. S1B). These findings suggested that the expression of ITGA5 was highly correlated with CAF marker genes and ITGA5 might be expressed mainly in CAF cells in tumor stroma. We further confirmed the correlation by costaining FFPE colorectal adenocarcinoma samples with antibodies against a5 (green) and two CAF markers: a-SMA (red) and vimentin (red). We noted that a5 colocalized in the tumor stroma with a-SMA and vimentin, whereas tumor epithelial cells stained less (Fig. 2C). The pathological histology type of the sample was confirmed by HE staining (Fig. S1C). These results strongly suggested that a5 was mainly expressed in CAFs rather than epithelial cancer cells, which implied an important role of a5 in tumor fibroblasts.

ITGA5 knockout in fibroblasts diminished tumor-promoting effect of fibroblasts in nude mice
Cancer-associated fibroblasts are a diverse cell population that can have different characteristics in different tumor types and tissue locales. A prominent component of CAFs is myofibroblasts, an activated form of fibroblasts. Fibroblasts have been widely described to have a profound influence on the development and progression of carcinoma (Noel et al., 1993;Park et al., 2016;Rasanen and Vaheri, 2010), as also confirmed in our in vivo experiments by injecting epithelial colorectal adenocarcinoma cells SW480, alone or with human normal colonic fibroblast cells CCD-18Co, into immunocompromised (nu/nu) mice. The tumorigenic ability of SW480 cells was significantly promoted when co-injected with equal numbers of CCD-18Co cells, with significantly enhanced tumor growth and generated tumors that averaged twice the size of those formed by SW480 cells alone (Fig. S2A,B), whereas CCD-18Co cells alone had no capacity to develop tumors (Fig. S2C). To examine whether a5 expression in fibroblasts could affect tumor progression, we compared the tumor-promoting capacities of CCD-18Co cells with or without a5 expression when co-injected with epithelial colorectal adenocarcinoma cells SW480 or DLD-1 in SCID mice. Our data showed that the tumor-promoting effect of CCD-18Co cells was significantly reduced after a5 was depleted by specific sgRNA (CCD-18Co sgRNA) compared with control cells infected with empty vector (CCD-18Co vector) (Figs 3A,B and S2D,E). Furthermore, Ki67 expression was evaluated by IHC in solid tumor tissues; a significant inhibition of Ki67 signaling was observed in solid tumors from mice co-injected with SW480 and CCD-18Co sgRNA cells compared with SW480 and CCD-18Co vector cells (Fig. 3C). These data indicated that the expression of a5 integrin subunit was required for fibroblasts to promote tumor growth in nude mice.

a5 Depletion in fibroblasts downregulated fibronectin assembly and alleviated cancer cell migration and invasion elicited by fibroblasts
Preexisting normal stromal fibroblasts could potentially convert into myofibroblasts in response to TGFb in vivo, specifically during the course of tumor progression. Myofibroblasts are activated fibroblasts that are commonly characterized by the expression of multiple activation markers, including a-SMA, FAP, PDGFRb and vimentin (Kojima et al., 2010;Midgley et al., 2013). To elucidate why a5 expression was required for fibroblasts to promote tumor growth, we proceeded to examine whether a5 expression could affect fibroblast activation induced by TGF-b. Western blotting was used to assess the protein expression of fibroblast-activated markers in CCD-18Co cells after TGF-b treatment (Fig. S3A). Although CCD-18Co cells expressed a-SMA, FAP, PDGFRb and vimentin in the absence of TGF-b, the presence of gradually elevated concentrations of TGF-b (10 and 20 ngÁmL -1 ) dramatically enhanced the expression levels of a-SMA and FAP (Fig. S3A). Meanwhile, the expression levels of a5 and b1 subunits, especially their ligand-fibronectin, was also progressively upregulated in the presence of elevated concentrations of TGF-b (Fig. S3A). We further examined fibroblast activation induced by TGF-b after a5 was depleted. Western blotting assay confirmed the depletion of a5 expression in CCD-18Co sgRNA cells compared with vector control cells (CCD-18Co vector cells). The depletion of a5 had no effect on the expression of b1 subunit or fibroblast-activated markers, but the expression level of fibronectin was dramatically reduced in CCD-18Co sgRNA cells (Fig. 4A). We confirmed the decreased expression of FN after a5 depletion using qPCR assay (Fig. S3B). In addition, immunofluorescence assay showed that fibronectin expression was significantly decreased in CCD-18Co sgRNA cells, most of the fibronectin existing as spots and almost completely abrogated fibronectin fibrillogenesis, compared with CCD-18Co vector control cells (Fig. 4B). Collectively, these observations suggested that a5 depletion on fibroblasts significantly reduced the expression of fibronectin and affected the ensuing fibronectin assembly.
Fibronectin is a major component in ECM and plays an essential role in orchestrating the assembly of ECM fibers. ECM provides a framework for cell adhesion and movement and is responsible for some morphological changes in tumor cells (Kinsey et al., 2008;Mao and Schwarzbauer, 2005). Thus, we proceeded to investigate the influence of fibroblast-expressed a5 on cancer cell morphology and function. We generated GFP-labeled SW480 (SW480-GFP) cells and monitored the effect of fibroblasts on CRC cell function through contact assay by coculturing SW480-GFP cells with CCD-18Co (vector or sgRNA) cells. We observed that SW480-GFP cells were closely associated with fibroblasts and acquired rather elongated shapes when cultured with CCD-18Co vector cells, whereas in the presence of CCD-18Co sgRNA cells, SW480-GFP cells exhibited a rounded morphology and remained discrete from fibroblasts in fluorescence microscopy (Fig. 4C). We then performed cell migration and cell invasion assay to test the role of fibroblasts on CRC cell function. The migration assay showed that a contact coculture with CCD-18Co vector cells markedly enhanced SW480-GFP cancer cell Transwell migration compared with that of SW480-GFP coculturing with CCD-18Co sgRNA cells (Fig. 4D,F). Similarly, the invasiveness of SW480-GFP cells induced by CCD-18Co vector cells was significantly inhibited by a5 depletion in CCD-18Co cells (Fig. 4E,G). All of the above results implied that a5 expression on fibroblasts played an essential role in enhancing cancer cell migration and invasion, which could be mediated by affecting fibronectin expression and assembly.
Additionally, to confirm the role of a5 expression on fibroblasts in vitro, we tried to knock down the expression of a5 using siRNA in CCD18-Co cells. Cells were treated with no siRNA (Mock) or siRNA targeting integrin a5 (a5 oligo1 + a5 oligo2, sia5); nontargeting siRNA was used as a negative control (siCtr). Western blotting was utilized to determine the ability of siRNA to knockdown a5 expression in CCD-18Co cells. Cells transfected with a5 siRNA showed a reduction in a5 protein expression when compared with the mock and control siRNA-transfected cells (Fig. S4A). Consistently the expression level of fibronectin was reduced in CCD-18Co sia5 cells (Fig. S4A). Immunofluorescence assay showed that fibronectin fibrillogenesis was also reduced in CCD-18Co sia5 cells compared with CCD-18Co Mock and siCtr cells (Fig. S4B). Furthermore, we performed cell migration and cell invasion assay to test the role of fibroblasts on CRC cell function; the result showed that the transmigration and invasiveness of DLD-1-GFP cells were significantly decreased when cocultured with CCD-18Co sia5 cells compared with CCD-18Co wild-type cells (Mock and siCtr) (Fig. S4C,E and D, F). Collectively, the above results confirmed that a5 expression on fibroblasts played an essential role in enhancing cancer cell migration and invasion, possibly mediated by their effect on fibronectin expression and assembly.

Clinical significance of a5 integrin expression in colorectal adenocarcinoma
Colorectal cancer can be classified into several subtypes by histology, in which classical adenocarcinoma (AC) accounts for the large majority of cases, and about 10% of all CRC are MAC, which retained a substantial amount of mucin within the tumor. Several studies have reported on the distinct molecular and genome signatures between these different subtypes (Dunkin et al., 2018;Nitsche et al., 2013;Park et al., 2015). In this study, to investigate the clinical significance of a5 expression in CRC, we initially utilized a bioinformatics approach and analyzed gene expression data from TCGA database. Of the 592 CRC cases that had complete follow-up information, 517 were AC and 75 MAC. We separately stratified the AC and MAC cases into high (above the median level) and low (below the median level) expression of ITGA5 for OS analysis. The result showed that high ITGA5 level was associated with decreased patient survival in the AC group (HR = 1.804; 95% CI 1.208-2.692; log-rank P = 0.0039) (Fig. 5A) but not in the MAC group (HR = 0.8914; 95% CI 0.3390-2.344; log-rank P = 0.05436) (Fig. S5A). In addition, we assessed multivariate Cox regressions that included age, sex, tumor stage, lymphatic invasion, tumor location and ITGA5 expression level in the colorectal AC group. The expression level of ITGA5, age and stage emerged as independent prognostic factors (Fig. 5B). Thus a5 integrin might act as a potential prognostic biomarker for colorectal adenocarcinoma.
Next, we analyzed the correlation between ITGA5 expression level and clinicopathological parameters in AC groups (Table 1). We divided all the cases into two groups according to the expression status of ITGA5. ITGA5 expression level was observed to be significantly associated with local invasion depth (P = 0.002) and lymph node involvement (P = 0.041) ( Table 1). We also observed a significantly correlated expression between ITGA5 and FN (P < 0.0001; r = 0.7548, Pearson's correlation), as shown in Fig. 5C.
To confirm further the clinical significance of a5 expression in Chinese patients with AC, we selected 355 eligible patients diagnosed with colorectal AC in 2008-2013 and assessed integrin a5 expression by IHC staining of the TMA block. The cases were divided into two groups according to the expression status of a5 (high or low) (Fig. 5D). Of the 355 cases, 112 (31.5%) patients died during a median follow up of 68.23 months (from 0.13 to 100.23 months). The association between a5 expression levels and OS of the patients was analyzed using Kaplan-Meier analysis and log-rank test. A statistically significant difference in OS was observed between these two groups of Fig. 4. a5 Depletion downregulated fibronectin assembly in fibroblasts and showed reduced effects on promoting cancer cell migration and invasion. (A) Expression levels of a5 and selected markers were determined by western blotting in CCD-18Co vector cells and CCD-18Co sgRNA cells with or without TGF-b activation. b-Actin served as loading control. A representative result of three independent experiments is shown. (B) Immunostaining of fibronectin (red) in CCD-18Co vector cells and CCD-18Co sgRNA cells. F-actin is stained with phalloidin-647 (red) and DNA was stained with DAPI (blue). Scale bar: 20 lm. (C) a5 Depletion in fibroblasts affected cancer cell morphology in the coculture system. Representative images of SW480-GFP cocultured with control fibroblasts (CCD-18Co vector) or with a5-depleted fibroblasts (CCD-18Co sgRNA). The a5 was shown in red, F-actin was stained with phalloidin-647 (purple) and DNA was stained with DAPI (blue). Scale bar: 20 lm. (D-G) The a5 depletion in fibroblasts showed reduced effects on promoting cancer cell migration and invasion compared with the wild-type fibroblasts. SW480-GFP cells were cultured alone or cocultured with CCD-18Co vector control cells or with a5-  patients; the group with high expression of a5 had a worse prognosis than the other group (HR = 1.639; 95% CI 1.131-2.376; log-rank P = 0.009) (Fig. 5E). The clinicopathological parameters of this independent group were summarized according to the staining status of a5 in Table S1. Furthermore, univariate Cox regression model analysis revealed that older age (P = 0.013), advanced TNM stage (P < 0.001) and high expression of a5 (P = 0.010) were associated with poor prognosis of CRC in terms of OS (Table 2). When these three factors were included into the multivariate Cox regression model, the results showed that older age (HR = 1.753; 95% CI 1.200-2.561; P = 0.004), advanced TNM stage (HR = 2.000; 95% CI 1.372-2.916; P < 0.001) and high expression of a5 (HR = 1.621; 95% CI 1.110-2.366; P = 0.012) were independent prognostic factors for colorectal adenocarcinoma (Table 2). Consistently, ITGA5 was significantly correlated with FN (P < 0.0001; r = 0.7958, Pearson's correlation) in 24 freshly isolated colorectal adenocarcinoma samples by assessing the mRNA expression using qPCR (Fig. 5F). We also examined the colocalization between a5 (green) and fibronectin (red), which showed positive colocalization (Fig. S5B).  Thus, a5 expression was clinically correlated with fibronectin expression and a5 expression in tumor stroma might serve as a potential prognostic marker in colorectal adenocarcinoma.

Discussion
Our efforts uncovered an important role of a5 in a tumor-supporting effect of fibroblasts and suggested that a5 might act as a new stroma molecular signature for colorectal adenocarcinoma. The expression of a5 was shown to be largely expressed in tumor fibroblasts and was required for fibroblasts in promoting tumor growth in vivo and supporting cancer cell migration and invasion in vitro. Clinically, high a5 expression level was associated with poor OS in colorectal adenocarcinoma, and correlation analysis showed that the expression level of a5 was significantly associated with local tumor invasion depth and lymph node metastasis (Table 1), which further suggested the tumor-promoting role of a5 in fibroblasts. Thus stroma expression of a5 might be a therapeutic target for CRC patients with adenocarcinoma. Interestingly, Kuninty et al. (2018) recently reported that a5 integrin could also contribute to tumor-promoting effects of pancreatic stellate cells in pancreatic cancer, and revealed a5 as a therapeutic target in pancreatic cancer, which suggested a ubiquitous role of a5 expressed on fibroblasts to promote tumorigenesis.
To identify the specific cells in the tumor stroma that mainly expressed a5, we performed gene correlation analysis and co-immunostaining assay between ITGA5 (a5 integrin subunit) and CAF markers, as CAFs have been shown to be abundantly present within the stromal components of carcinomas (Orimo and Weinberg, 2006). The results showed an extraordinarily correlated expression between a5 and CAF markers (Figs 2 and S1), suggesting that stroma expression of a5 mainly occurred in CAFs. Myofibroblasts are a prominent component of CAFs and have been suggested to produce pro-invasive signals that induce a more aggressive phenotype in the tumor (Liang et al., 2005;Nakayama et al., 1998). We showed that depletion of a5 greatly reduced the tumor-promoting effect of fibroblasts, implying an important role of a5 in CRC through tumor fibroblasts. Our work was done using a normal colonic fibroblast cell line, which supports our final conclusion, as there are reports showed that preexisting normal stromal fibroblasts could potentially convert into myofibroblasts in response to TGF-b in vivo, specifically during the course of tumor progression (Kojima et al., 2010;Midgley et al., 2013). We understand that primary isolated CAF might support these conclusions better, and therefore more experiments will be performed to further this study using primary CAFs in the future.
In our experiments, the expression level of a5 and fibronectin was significantly increased together with fibroblast activation markers during the process of fibroblast-myofibroblasts conversion induced by TGFb (Figs 4A and S3A). However, a5 depletion in fibroblasts did not affect the expression levels of fibroblast activation markers as shown in the western blotting assay (Fig. 4A), except for significant inhibition of fibronectin expression and assembly (Figs 4A,B and S3B). In a recent report, TGF-b was shown to induce peritumoral fibronectin deposition and tumor infiltration in basal cell carcinoma (Kuonen et al., 2018), which suggested that a5 might affect the expression and assembly of fibronectin through some TGF-b responsive pathways, rather than a general effect on the activation state of fibroblasts. Our following study showed that cancer cell migration and invasion induced by fibroblasts was significantly inhibited by a5 depletion in in vitro coculture assays, which might due to the downregulated assembly of fibronectin. These results were consistent with two recent reports, as fibronectin assembly has been shown to be a new hallmark of CAFs, and the ability of fibroblasts to induce cancer cell invasion was directly correlated with the amount of fibronectin they assembled . Also, alignment of fibronectin fibers by CAFs could promote CAFcancer cell interactions and mediate directional cancer cell migration in coculture assays, whereas knocked down fibronectin expression in CAFs completely disrupted ECM synthesis and organization (Erdogan et al., 2017).
A proangiogenic role of a5b1 has been well demonstrated (Kim et al., 2000;Muether et al., 2007;Schaffner et al., 2013;Taverna and Hynes, 2001) and has been implicated in a phase I clinical study that used dual anti-angiogenic therapy combining a5b1 and vascular endothelial growth factor inhibition in patients with advanced solid tumors (Weekes et al., 2018). Although no overt clinical activity was observed in that study (Weekes et al., 2018) the authors found that the highest expression of fibronectin in hepatocellular carcinoma (HCC) patients showed the best overall response. Evidence has accumulated showing a crucial role of fibronectin assembly in supporting tumor progression (Wang and Hielscher, 2017;Yoneda, 2015); however, the relationship between the expression of fibronectin and the prominent role of integrin a5b1 in tumor stroma has not been well elucidated.
Fibronectin is assembled into fibers through its binding to integrin adhesion receptors. Integrin a5b1 is the major fibronectin receptor and has been shown to be responsible for fibronectin matrix assembly (Kinsey et al., 2008). Consistent with previous studies identifying a crucial role of a5b1 in mediating fibronectin matrix formation in CAFs (Erdogan et al., 2017;Hooper et al., 2010), we found that a5 expression in CCD-18Co fibroblasts indad a great effect on fibronectin assembly (Figs 4B and S4B) by reducing the expression of fibronectin, as confirmed by western blotting assay (Figs 4A and S4A). All of these data reinforced the interactions between a5 and fibronectin. Although we preliminarily found that the downstream PI-3-kinase signaling pathway was significantly influenced after a5 was depleted (data not shown), it remains unclear how a5 depletion affects fibronectin expression and thus more experiments are needed to study the underlying mechanism further.

Conclusions
In summary, we demonstrate in the current study that a5 integrin subunit was mainly located in the colorectal tumor fibroblasts, and we provide evidence that a5 expression is required for fibroblasts to exhibit a tumor-promoting effect, as a5 depletion in fibroblasts dramatically suppresses fibroblast-induced tumor growth in xenograft nude mice and inhibits cancer cell migration and invasion induced by fibroblasts in in vitro coculture assays. We also observed that the expression and assembly of fibronectin were downregulated after a5 was depleted in fibroblasts. Our study offered fresh insight into colorectal adenocarcinoma progression and shows that a5 expression on tumor stroma might serve as a prognostic marker and therapeutic target in colorectal adenocarcinoma.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. (A) Pearson's correlation between the expressions of ITGA5 and CAF marker genes (PDGFRB, FAP, IL-6 and CXCL-12) in 24 fresh colorectal adenocarcinoma samples. Expression was determined by qPCR in triplicate. (B) Heat map of mRNA transcript levels of CAF marker genes in 517 colorectal adenocarcinomas extracted from TCGA database. Each row represents a gene and each column represents a case. Cases were organized from left to right by the ascending order of ITGA5 expression level. The dendrogram was drawn with the hierarchical clustering method. The color scheme representing gene expression levels (row Z-score) is illustrated above the heatmap. (C) Representative HE staining of the samples used for double immunofluorescent staining in Fig. 3C. Scale bar: 50 lm. The growth curves of xenograft tumors in nude mice (n = 7 in each group) that were injected with adenocarcinoma cancer cells (DLD-1) alone, DLD-1 and transfected fibroblasts with a5 depletion (SW480+ CCD-18Co sgRNA) or DLD-1 and transfected fibroblasts with vector control (SW480+ CCD-18Co vector). Tumor size was measured at the indicated time points (days). Tumors were excised for histology at the last time point. Data are presented as mean AE SEM. *** P < 0.001 vs control (two-way ANOVA with Bonferroni post-tests). (E) Photographs of dissected tumor samples. Fig. S3. (A) Expression levels of a5 and selected markers were determined by western blotting in CCD-18Co cells with or without TGF-b activation. A representative result of three independent experiments is shown on the left. Quantification of three independent blots for a5 expression as a ratio relative to the expression level of control cells is shown on the right. Data are presented as mean AE SEM. *** P < 0.001 vs control (one-way ANOVA). (B) Transcript levels of ITGA5 and FN determined by qPCR. Experiments were performed in triplicate and Student's t-test was used for statistical analysis. Fig. S4. (A) Expression levels of a5 and fibronectin were determined by western blotting in CCD-18Co cells treated with no siRNA (Mock) or siRNAs (a5 oligo1+ a5oligo2) targeting integrin a5 (sia5); non-targeting siRNA was used as a negative control (siCtr). b-Actin served as loading control. A representative result of three independent experiments is shown. (B) Immunostaining of fibronectin (red) in CCD-18Co cells treated with or without siRNA. F-actin is stained with phalloidin-647 (red) and DNA is stained with DAPI (blue). Scale bar: 20 lm. (C-F) The a5 knockdown in fibroblasts shows reduced effects on promoting cancer cell migration and invasion compared with the wild-type fibroblasts. DLD-1-GFP cells were cocultured with wild-type CCD-18Co cells (Mock and siCtr) or with a5 knockdown CCD-18Co cells (sia5) in Transwell inserts with or without Matrigel. (C,E) Serum-induced Transwell migration was calculated after 16 h. DLD-1-GFP cells migrating to the lower chamber were observed under a fluorescence microscope and were counted. Scale bar: 200 lm. (D,F) Serum-induced cell invasion were calculated after 22 h. DLD-1-GFP cells that invaded through the Matrigel to the lower chamber were observed under a fluorescence microscope and were counted. Scale bar: 200 lm. Error bars, SEM (n = 3). * P < 0.05 (one-way ANOVA).  Table S1. Characteristics of 355 CRC cases according to a5 staining status in the independent cohort.