Gallbladder cancer-associated broblasts promote vasculogenic mimicry formation and tumor growth of gallbladder cancers via upregulating the expression of NOX4, a poor prognosis factor, through activating IL-6-JAK-STAT3 signal pathway

Background Cancer-associated broblasts (CAFs) and vasculogenic mimicry (VM) play important roles in the occurrence and development of tumors. However, the relationship between CAFs and VM formation, especially in gallbladder cancer (GBC) has not been claried. In this study, we investigated whether gallbladder CAFs (GCAFs) can promote VM formation and tumor growth and explored the underlying molecular mechanism. Methods A co-culture system of human GBC cells and broblasts or HUVECs was established. VM formation, proliferation, invasion, migration, tube formation assays, CD 31 -PAS double staining, optic/electron microscopy and tumor xenograft assay were used to detect VM formation and malignant phenotypes of 3-D co-culture matrices in vitro, as well as the VM formation and tumor growth of xenografts in vivo, respectively. Microarray analysis was used to analyze gene expression prole in GCAFs/NFs and VM (+)/VM (-) in vitro. QRT-PCR, western blotting, IHC and CIF were used to detected NOX4 expression in GCAFs/NFs, 3-D culture/co-culture matrices in vitro, the xenografts in vivo and human gallbladder tissue/stroma samples. The correlation between NOX4 expression and clinicopathological and prognostic factors of GBC patients was analyzed. And, the underlying molecular mechanism of GCAFs promoting VM formation and tumor growth in GBC was explored. Results GCAFs promote VM formation and tumor growth in GBC; and the nding was conrmed by facts that GCAFs induced proliferation, invasion, migration and tube formation of GBC cells in vitro, and promoted VM formation and tumor growth of xenografts in vivo. NOX4 is highly expressed in GBC and its stroma, which is the key gene for VM formation, and is correlated with tumor aggression and survival of GBC patients. The GBC patients with high NOX4 expression in tumor cells and stroma have a poor prognosis. The underlying molecular mechanism may be related to the up-regulation of NOX4 expression through paracrine IL-6 mediated IL-6/JAK/STAT3 signaling pathway. Conclusions GCAFs promote VM formation and tumor growth in GBC via upregulating NOX4 expression through the activation of IL-6-JAK-STAT3 signal pathway. NOX4, as a VM-related gene in GBC, is overexpressed in GBC cells and GCAFs, which is related to aggression and unfavorable prognosis of GBC patients. phosphate oxidase 4, NADPH oxidase 4; reactive oxygen species; DMEM, Dulbecco’s modied Eadles medium; HUVEC, human umbilical vein endothelial cell; α-SMA, α-smooth muscle actin; FAP, broblast activation protein; IHC, immunohistochemistry; CIF, co-immunouorescence; FBS, fetal bovine serum; PBS, phosphate buffer CO 2 , carbon dioxide; 3-D, three-dimensional; HE, hematoxylin and eosin; PAS, periodic acid-Schiff; OD, optical density; SPF, specic pathogen-free; TEM, transmission electron OS, overall survival; IVT, in vitro transcription; FC, fold change; GO, Gene Ontology; SI, staining index; enzyme-linked immunosorbent assay; IL-6, interleukin-6; DAPI, diamidine phenylindole; qRT-PCR, quantitative reverse transcription-polymerase chain reaction; JAK, Janus kinase; STAT, and activator of transcription; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; RIPA, assay; ECL, enhanced chemiluminescent; HR, Hazard ratio; CI, condence


Introduction
Gallbladder cancer (GBC) is a highly malignant primary tumor of the biliary tract. Due to the lack of speci c clinical manifestations and effective monitoring indicators, the early diagnosis rate of GBC is low. At the same time, the disease also has the characteristics of high invasion and metastasis, insensitivity to chemo-radiotherapy, resulting in disappointing surgical and drug treatment results and poor prognosis; according to relevant reports, its 5-year survival rate is only about 5% [1][2][3]. Therefore, it is of great signi cance to understand the pathogenesis of GBCs and explore the effective prognosis indicators and drug targets, so as to improve the diagnosis, treatment and prognosis of GBC patients.
It is generally believed that the occurrence and development of tumor is not only related to tumor itself, but also depends on the complex interaction between malignant cells and their microenvironment. Tumor microenvironment (TME), also known as tumor stroma, is composed of extracellular matrix (ECM) and a variety of stromal cells [4][5]. Cancer-associated broblasts (CAFs) are the most important stromal cells in TME, which play an important role in tumor growth and development, and participate in many biological processes such as tumor proliferation, invasion and metastasis, angiogenesis [6][7][8][9][10], and are related to poor prognosis [11]. In these biological processes, angiogenesis and effective blood supply are the basic conditions for tumor growth and metastasis [12]. In recent years, studies have found a novel tumor blood supply pattern, called vasculogenic mimicry (VM), which occurs in certain highly aggressive malignancies, and is closely related to poor clinical results and poor prognosis [13][14][15][16]. It has become one of the potential targets for anticancer therapy [17][18][19][20]. We previously reported that VM exists in GBC and is associated with the patient's poor prognosis. The mechanism of VM formation in GBC may be closely related to the activation of PI3K/MMPs/Ln-5γ2 and/or EphA2/FAK/Paxillin signaling pathways [14][15][16][17][18][19]. Recently, it was reported that CAFs can promote VM formation in hepatocellular cancer (HCC) [21]. However, so far, the mechanism of VM formation is not clear, and the relationship between gallbladder cancer-associated broblasts (GCAFs) and VM formation in GBC has not been reported.
NOX4 (nicotinamide adenine dinucleotide phosphate oxidase 4, NADPH oxidase 4) is a number of NADPH oxidase (NOX) family [22,23]. NOX family plays vital roles in signaling transduction, cell growth, apoptosis, differentiation and tumor development. Long-term oxidation/reduction imbalance is considered to be one of the important factors leading to carcinogenesis. Oxidative stress induces the development of cancer due to excessive production of reactive oxygen species (ROS) by members of the NOX family [24][25][26][27]. It was reported that NOX4 is overexpressed in many malignancies and participates in malignant processes such as proliferation and metastasis [22,28]; NOX4 can also promote tumor angiogenesis by regulating the production of vascular endothelial growth factor, which further results in poor prognosis [29].
In this study, we rstly found that GCAFs can promote VM formation and tumor growth of GBC; NOX4, as a key gene of VM formation in GBC, is over-expressed in GBC and its stroma, which is related to poor prognosis. At the same time. We also proved that GCAFs promote VM formation and tumor growth of  [30]) were used in this study. These cells were propagated in Dulbecco's modi ed Eadles medium (DMEM, Gibco, USA) supplemented with 10% fetal bovine serum (FBS, Ausbian, Australia) and 0.1% gentamicin sulfate (Gemini Bioproducts, Calabasas, Calif). The established human umbilical vein endothelial cell line (HUVEC, gift from Department of Pathophysiology, Shanghai Medical College, Fudan University) was cultured in endothelial cell medium (ECM, Sciencell, USA) with 10% FBS. Human GCAFs and normal broblasts (NFs) were isolated from the clinical specimens of human GBC tissues and adjacent normal tissues, and identi ed by the detection of the stromal markers α-smooth muscle actin (α-SMA) and broblast activation protein (FAP) using immunohistochemistry (IHC), co-immuno uorescence (CIF) and western blotting. The established GCAFs and NFs (the cells used in this experiment were between the 4th and 9th generation) were incubated in DMEM/F-12 medium (Gibco; USA) supplemented with 10% FBS. Co-cultures of GBC cells and GCAFs or NFs were performed as previously described [11]. All of the cells were maintained in a carbon dioxide (CO 2 ) incubator (SANYO MCO-175, Japan) at 37˚C with a 5% CO 2 atmosphere.

VM formation assay in vitro
The Transwell chamber (aperture 0.4 µm, diameter 6.5 mm) was used to establish a co-culture system of human GBC cells and broblasts. Human GBC cell lines (GBC-SD, SGC-996 and OCUG-1) and broblasts (GCAFs or NFs) were used to prepare cell suspensions (GBC cells, 4 × 10 4 ·ml − 1 ; broblasts, 2 × 10 4 ·ml − 1 ). Matrigel and rat-tail type I collagen three-dimensional (3-D) matrices (ABI, USA) were prepared as described previously [15], coated on the bottom of the lower chamber respectively for comparative test. Cells were divided into GBC cell (GBC-SD), GBC cell (GBC-SD) + NFs and GBC cell (GBC-SD) + GCAFs groups. In the analysis of VM related gene expression pro le, cells were divided into GBC cell (GBC-SD, SGC-996 or OCUG-1) group and GBC cell (GBC-SD, SGC-996 or OCUG-1) + GCAFs co-culture group. 50 µl GBC cell suspensions (5 × 104·ml-1) were injected into the lower chamber containing different gels separately and incubated for 2 h, then DMEM/F12 medium containing 100 U·ml − 1 penicillin and streptomycin (containing 0.2% FBS) was added. The same amount of (200 µl) broblast suspensions (2 × 10 4 /ml) or serum-free medium were added to the upper chamber. The culture medium was changed every 2-3 days. The formation of VM in 3-D matrices from GBC cells or co-cultures was performed by hematoxylin and eosin (H&E) staining and immunohistochemical periodic acid-Schiff (PAS) staining (without hematoxylin counterstain) as described previously [15,18,19], and was observed under an inverted optical microscope (Caikang XDS-100) every day, and the number of VM formation was recorded in 5 visual elds at 200 magni cations.

Assays of malignant phenotypes of GBCs triggered GCAFs in vitro
To verify that GCAFs promoted VM formation of GBCs, in following experiments we detected malignant phenotypes including proliferation, invasion, migration and tube formation of GBC cells triggered by GCAFs in vitro.
Cell invasion was assessed using the Transwell chambers with aperture of 8 µm and diameter 6.5 mm (Corning, USA). Matrigel (200 µl/well, BD, USA) was coated on the bottom of the upper chamber to simulate the basement membrane and extracellular matrix. Cells were grouped as above. GBC-SD, SGC-996 and TJ-GBC2 cell suspensions (1 × 10 5 ·ml − 1 ; 200 µl) were respectively inoculated on the gel in the upper chambers. Lower chambers were inoculated with 700 µl NFs, GCAFs (5 × 10 4 ·ml − 1 ) or serum-free medium. After 24 h of culture, the cells invaded through the basement membrane were stained with Giemsa (Beyotime, China) and counted under an inverted optical microscope (Caikang XDS-100, Shanghai, China). All experiments were performed in triplicate.
Cell migration was assessed with a wound healing assay. Cells were grouped as above. 100 µl GBC cell suspensions (5 × 10 4 cells/well) or equal volume co-culture cells (50 µl GBC cell suspensions, 50 µl NFs or GCAFs; 5 × 10 4 ·ml − 1 ) containing NFs, GCAFs or serum-free medium were inoculated into 96-well wounding plate (Coster, USA) with culture medium and cultured in a single layer for 24 h until 90% cells fused. Then, a scratch tester was used to scratch a wound at the central bottom of 96-well plate. Cell Tumor Xenograft assay in vivo The xenograft experiments were performed in accordance with the o cial recommendations of Chinese Community Guidelines, and were approval from Research Ethical Review Broad in Tongji University (Shanghai, China). BALB/C nu/nu mice (equal numbers of male and female mice, 4-week old, about 20 g) were purchased from Shanghai Laboratory Animal Center, Chinese Academy of Sciences, and housed under speci c pathogen-free (SPF) conditions. The mice were randomly divided into GBC-SD group and GBC-SD + GCAFs group, 10 mice in each group. 0.2 ml serum-free medium containing GBC-SD or GBC-SD + GCAFs co-culture cell suspensions (5.0 × 10 6 ·ml − 1 ) were respectively injected subcutaneously into the right axilback of the nude mice. Tumor xenograft size i.e. the maximum diameter (a) and minimum diameter (b) was measured with calipers twice a week. The tumor volume was calculated by the following formula: V (cm 3 ) = Πab2/6. After 7 weeks, mice were sacri ced and xenograft specimens were used for western blotting, or were para n-embedded, depara nized, hydrated and were then used for immunohistochemistry (IHC) staining and Co-immuno uorescence (CIF) staining, respectively.
VM formation assay of tumor xenografts in vivo VM formation assay of xenograft sections in vivo was conducted by H&E staining, CD31-PAS double staining and transmission electron microscopy (TEM) as described previously [15,18,19]. Histomorphologic appearance and VM characteristic of the tumor xenografts in vivo were observed under an inverted optical microscope (Caikang XDS-100) and a JEOL-1230 TEM (Japanese Electronics, Japan).

Patients and clinical specimens
From 2007 to 2011, 85 patients with GBC, 10 patients with gallbladder precancerous lesion or benign lesion were recruited from Tongji Hospital, Tongji University (Shanghai, China). This study was conducted in accordance with the o cial recommendations of ethical standards, the Declaration of Helsinki and the Chinese Community Guidelines, and was approved by the Ethics Committee and the Institutional Review Board of Tongji Hospital. A written informed consent was obtained from each patient. A total of 115 gallbladder tissue specimens including 105 para n-embedded specimens (85 GBC, 10 gallbladder precancerous or benign lesion specimens) and 10 fresh GBC specimens con rmed by operation and histopathology were used in this study. All GBC patients had not received chemotherapy or radiotherapy before surgery. Curative resection (R0 resection) was de ned as no residual tumor status, whereas microscopic (R1 resection) and macroscopic residual tumor (R2 resection) was de ned as non-curative resection. To reduce effects directly related to surgery, patients who died within one month after surgical resection were not included. Two independent pathologists who blinded to the patients' clinical status veri ed diagnoses of these GBC samples. According to WHO criteria and the Nevin stage system, detailed clinicopathological and follow-up data were collected from the patient's medical records and completed by a telephone survey, routine visit record and address. Clinical outcome was followed from the date of surgery to the date of death or until the end of September 30, 2011. Cases lost during follow-up were regarded as censored data for the survival analysis. The median follow-up period for all patients was 18.6 (range, 1-60) months. The 5-year overall survival (OS) rate was 11.8% (10/85). Demographic and clinicopathological data are summarized in Table 1.  Then the newly generated cRNA was synthesized, puri ed and labeled. Finally, after hybridizing and cleaning with a GeneChip Hybridization Wash and Stain kit (Affymetrix, USA), the Genechip Array scanner 3000 (Affymetrix, USA) was used to scan the assays to nd out the differentially expressed genes between GCAFs and NFs. Array data were normalized by log scale robust multi-array analysis and analyzed by R-Project software. The gene expression was considered signi cant if the fold change (FC) value was > 1.5 or < 0.67, and P < 0.05. Gene Ontology (GO) analysis was used for functional enrichment analysis, and gene set enrichment analysis and Fisher exact analysis were used to perform statistical analysis of GO. In order to explore the differences of gene expression pro le between GCAFs and NFs, potentially relevant up-or down-regulated genes involved in biological processes were selected for veri cation.
Affymetrix Human lncRNA array (Affymetrix, USA) was used to analyze the expression pro le of VM related genes in VM (+) and VM (-) groups in vitro. Transcriptome library construction, transcriptome assembly and annotation protocols were provided by Shanghai Oe Biotech Co., Ltd., China. The Pearson correlation between its expression value and each mRNAs expression value was calculated for each lncRNA. For function prediction of lncRNAs, the co-expressed mRNAs for each differentiated lncRNA were calculated and then a functional enrichment analysis of this set of co-expressed mRNAs was carried out. The enriched functional terms were used as the predicted functional term of given lncRNA. The coexpressed mRNAs of lncRNAs were identi ed by calculating Pearson correlation with correlation P-value < 0.05. Then the hypergeometric cumulative distribution function was used to calculate the enrichment of functional terms in annotation of co-expressed mRNAs. The cis-regulation regions were identi ed by the following procedures. For each lncRNAs, we identi ed the mRNAs as "cis-regulated mRNAs" when: (1) the mRNAs loci are within 300 k windows up-and down-stream of the given lncRNA, (2) the Pearson correlation of lncRNA-mRNA expression is signi cant (P-value of correlation ≦ 0.05).

Immunohistochemistry (IHC) and enzyme-linked immunosorbent assay (ELISA) in vitro and in vivo
IHC was used to detect the expression of NOX4 protein in sections from the 3-D co-culture samples in vitro, nude mice xenografts in vivo and human gallbladder tissues or stroma. After depara nizating and inactivating endogenous peroxides, the sections (4 µm thick) were pretreated with bovine serum albumin V working solution (Beijing Solarbio Science &Technology Co., China), then incubated with primary antirabbit NOX4 (Sigma, USA), secondary anti-rabbit IgG (Maixin, China) and 3, 3-diaminobenzidine (DAB) solution, and stained with hematoxylin according to the manufacturer's instructions. Phosphate buffer saline (PBS; Thermo Fisher Scienti c, USA) was used to replace the primary antibody for negative control. The expression of NOX4 was observed under an optical microscope (Olympus, Japan). In order to score the stains, ve random elds of each section were observed or more than 500 cells counted per slide. In addition, the expression of NOX4 in different human gallbladder tissue/stroma was evaluated by a semiquantitative system with the staining index (SI). The SI scoring criteria are as follows: (positive cell percentage score) (staining intensity score). The positive cell percentage was scored from 0 to 4 as follows: 0 (no positive cells), 1 (1%-25%), 2 (26%-50%), 3 (51%-75%) and 4 (76%-100%). The staining intensity was scored from 0 to 3 as follows: 0 (negative), 1 (weak), 2 (moderate), and 3 (strong). 3 of SI score were used to distinguish between low (IS ≤ 3) and high (IS > 3) protein expression.
The expression of interleukin-6 (IL-6) protein product in supernatant from the 3-D co-culture samples in vitro or nude mice xenografts in vivo was determined by ELISA using the Human IL-6 ELISA Kit (Abcam, UK) according to the manufacturer's protocol. All samples were analyzed in triplicate.
Co-immuno uorescence (CIF) staining in vivo CIF staining with stromal markers such as α-SMA and FSP-1 was used to con rm the expression of NOX4  The gray value and gray coe cient ratio of every protein were analyzed and calculated with Image J analysis software (National Institutes of Health).

Statistical analysis
All data were expressed as mean ± SD (standard deviation), and statistically analyzed by SPSS 22.0 software (IBM, USA). The SI was analyzed using Kruskall-Wallis test with Dunn's post-hoc comparison.
Student t-test was used in independent sample analysis. The χ 2 test was used to analyze the relationship between NOX4 expression and GBC patients' clinicopathologic parameters. Bivariate correlations between two independent variables were analyzed by calculating the Spearman's correlation coe cients. The survival analysis was calculated by Kaplan Meier method and compared by log rank test. The Cox's regression model was used for univariate and multivariate analysis of prognostic factors. P < 0.05 was considered to be statistically signi cant.

GCAFs promote VM formation of GBC cells in vitro
A 3-D co-culture system of GBC-SD and GCAFs/NFs was established to investigate the effect of GCAFs on the VM formation of GBC. As showed in Fig. 1, GBC-SD cells and GBC-SD + NFs groups formed vasculogenic-like networks when cultured on Matrigel and rat-tail collagen-matrix for 2 days, and had VM formation for 14 days; but GBC-SD + GCAFs group formed patterned vasculogenic-like networks for only 19 hours, had VM formation for only one week; and the tube number of vasculogenic-like networks in GBC-SD + GCAFs was signi cantly more than that of alone GBC-SD cells and GBC-SD + NFs (P < 0.05). Furthermore, PAS positive, cherry-red materials found in granules and patches in the cytoplasm of GBC-SD cells appeared around the signal cell or cell clusters. These results showed that GCAFs can promote VM formation of GBC-SD cells in vitro.

GCAFs promote the proliferation, invasion, migration and tube formation of GBC cells in vitro
To verify that GCAFs promote VM formation of GBC cells, we detected malignant phenotypes including proliferation, invasion, migration and tube formation of GBC cells triggered by GCAFs in vitro using a coculture model of GBC cells (GBC-SD, SGC-996, TJ-GBC2) and broblasts (GCAFs, NFs). As shown in Fig. 2, the proliferation ability of GBC cell + GCAFs groups were signi cantly enhanced compared to alone GBC cell or GBC cell + NFs groups (all P < 0.05; Fig. 2A); the number (relative invasive ability) of cells that invaded the basement membrane in GBC cell + GCAFs groups was signi cantly more than that of alone GBC cell or GBC cell + NFs groups (all P < 0.05; Fig. 2B); the cell migration rate in GBC cell + GCAFs coculture groups was signi cantly higher than that of alone GBC cell groups or GBC cell + NFs co-culture groups (all P < 0.05; Fig. 2C). Furthermore, tube formation assay showed that tube structure was observed in HUVEC + GBC-SD + GCAFs groups at 12 h; no obvious tube structure was seen in HUVECs or HUVEC + GBC-SD + NFs cultured in Matrigel for 12 h, and the formation of tube structure cannot be observed until about 18 h. At 48 h, obvious tube formation was observed in HUVEC, HUVEC + GBC-SD, HUVEC + GBC-SD + NFs and HUVEC + GBC-SD + GCAFs groups; but among them, the number of the lumen formed in HUVEC + GBC-SD + GCAFs group was signi cantly more than that of the other three groups (all P < 0.01; Fig. 2D). Taken together, GCAFs can promote the proliferation, invasion, migration and tube formation of GBC cells/HUVECs, and then promote VM formation.

GCAFs promote VM formation and tumor growth of GBC xenografts in vivo
A nude mouse xenograft model was established to further verify whether GCAFs can promote VM formation and tumor growth of GBC in vivo. As shown in Fig. 3, size, volume and growth curves of tumor xenografts in GBC-SD + GCAFs groups were signi cantly larger than that of GBC-SD group (P < 0.05; Fig. 3A); H&E staining ( Fig. 3Bb1) and CD 31 -PAS double staining (Fig. 3Bb2) of xenografts section in GBC-SD and GBC-SD + GCAFs groups showed that tumor cells were lined with vessel-like structure, with single or multiple red blood cell inside, without any evidence of tumor necrosis, and PAS positive substances were arranged in a channel-like arrangement; furthermore, TEM (Fig. 3Bb3) clearly visualized single or multiple red blood cells in the central of tumor nests in the xenografts; but the number of blood vessel-like structures, erythrocytes and PAS positive substances in GBC-SD + GCAFs group were signi cantly more than those of GBC-SD group (Fig. 3Bb1-2). These results further con rmed that GCAFs have the ability to promote tumor growth and VM formation in GBC.

NOX4, as a GCAFs-derived VM-related gene, is highly expressed in the process of VM formation in GBC triggered by GCAFs
In order to understand which genes are involved in the VM formation triggered by GCAFs in GBC, we analyzed the gene difference expression pro le of GCAFs/NFs and VM(+)/VM(-) in vitro using Affymetrix GeneChip Human 1.0ST array and Affymetrix Human lncRNA array. As shown in Fig. 4, a total of 466 upregulated genes (FC > 1.5) and 596 downregulated genes (FC < 0.67) were identi ed in GCAFs/NFs according to the inclusion criteria. As one of 16 angiogenesis-related genes, NOX4 was signi cantly upregulated (FC = 2.58) in GCAFs compared with NFs. Furthermore, Affymetrix Human lncRNA array was used to analyszed VM-related gene expression pro le in GBC. Venn modal analysis (upregulation, > 2 times; or downregulation, < 0.5 times) showed that the expression of NOX4, JAK and STAT3 was signi cantly upregulated in 154 VM-related genes (Fig. 5). These data preliminary showed that NOX4, as a GCAFs-derived VM-related gene, is highly expressed in the progress of VM formation triggered by GCAFs in GBC.
To verify that NOX4 expression is upregulated during the formation of VM induced by GCAFs in GBC, we further detected the expression of NOX4 in vitro and in vivo using qRT-PCR, western blot and IHC. As shown in Fig. 6, NOX4 mRNA expression in GCAFs (Fig. 6A.a1) or in 3-D matrices of GBC cell (GBS-SD, SGC-996, OCUG-1) + GCAFs co-culture (Fig. 6a2) was signi cantly upregulated compared with NFs or alone GBC cells in vitro, which was consistent with the results of microarray analysis. And, the expression of NOX4 protein in GBC cell + GCAFs co-culture groups was signi cantly higher than that of GBC-SD + NFs co-culture or alone GBC-SD in vitro (all P < 0.05; Fig. 6A.a3-4), but no statistical difference was observed between GBC-SD + NFs and GBC-SD groups. Furthermore, the expression of NOX4 in GBC xenografts in vivo was detected by IHC (Fig. 6B.b1) and western blotting (Fig. 6B.b2). The results showed that NOX4 was also high expressed in GBC-SD + GCAFs group compared with GBC-SD group (all P < 0.05); and the expression of NOX4 mRNA in GBC-SD + GCAFs group was signi cantly upregulated compared with GBC-SD group (P < 0.05; Fig. 6B.b3). All together, these data veri ed that NOX4, as a GCAFs-derived VM-related gene, was highly expressed in VM formation triggered by GCAFs; in other words, GCAFs upregulated NOX4 expression in the process of VM formation in GBC.
To further verify whether NOX4 was over-expressed in VM formation of GBC triggered by GCAFs, we detected the expression of NOX4 protein in human gallbladder tissue or stroma samples using IHC and CIF staining. As shown in Fig. 7 (Fig. 7A). In addition, CIF staining with stromal markers α-SMA and FSP-1 showed that NOX4 (green) overlapped or co-localized with α-SMA and FSP1 (red or brown) positive stroma in GBC (Fig. 7B). These results further con rmed that NOX4, as a key gene for VM formation, was over-expressed in GBCs-induced VM formation of GBC.

NOX4 is related to aggression and unfavorable prognosis of GBC patients
Here, We further discussed the relationship between NOX4 expression and clinical, pathological and prognostic factors in GBC patients. SI score was used to distinguish NOX4 expression: IS ≤ 3, NOX4 low expression; >3, high expression. In the tumor cells of GBC, NOX4 was highly expressed in 50 cases (58.8%) and low in 35 cases (41.2%); in stromal cells, the expression of NOX4 was high in 55 cases (64.7%) and low in 30 cases (35.3%). As shown in Table 1-2, chi-square analysis and spearman rank correlation analysis indicated that the expression of NOX4 in tumor/stromal cells was not related to age, gender, tumor location, tumor size, histological type, lymph node metastasis and resection method (all P > 0.05), but signi cantly associated with tumor differentiation, liver metastasis, vascular invasion, Nevin staging and VM (all P < 0.05) in GBC patients.   Fig. 7C, Kaplan-Meier analysis showed that the survival time of GBC patients with high NOX4 expression in GBC tumor (Fig. 7C1, P = 0.026) and stroma (Fig. 7C2, P = 0.020) was signi cantly shorter than that of GBC with low NOX4 expression.
All together, these results con rmed that the expression of NOX4, especially in the tumor stroma, is an independent prognostic factor of GBC. The high expression of NOX4 in GBC tumor/stroma is closely related to VM formation, and indicates poor prognosis.
In order to explore the underlying molecular mechanism of GCAFs promoting VM formation of GBC, according to the above experimental results, we constructed a network regulation model of GCAFs promoting VM formation in GBC. As shown in Fig. 8, in this network regulation model, NOX4, STAT3 and JAK were regulated in the VM-related key genes.
Considering that IL-6, which is autocrine-secreted by tumor cells, especially, paracrine-secreted by stromal cells such as CAFs in TME, participates in tumor development by activating of some signal pathways such as STAT3 [33][34][35][36][37], and the IL-6/JAK/STAT3 signaling axis is involved in occurrence and development of cancer [38], we assume that GCAFs promote VM formation in GBC via upregulating NOX4 expression through paracrine IL-6 mediating IL-6/JAK/STAT3 signaling pathway. In order to verify the hypothesis, we rst detected the expression of IL-6 in GCAFs/NFs in vitro using western blotting and ELISA. As shown in Figure 9, IL-6 and its product were highly secreted and upexpressed in the culture supernatant of GCAFs compared with NFs in vitro (all P<0.05). Furthermore, we detected the expression of related signaling pathway genes in 3-D culture/co-culture matrices using ELISA, western blotting and qRT-PCR. The results showed that the expression of IL-6 product in the supernatant of 3-D co-culture matrices, and the expression of JAK1, JAK2 and STAT3 at protein and mRNA levels in the 3-D co-culture matrices of GBC-SD+GCAFs was signi cantly higher than that of GBC-SD culture or GBC-SD+NFs coculture (all P<0.05); but no statistical difference was observed in the expression of these genes between GBC-SD and GBC-SD+NFs groups ( Figure 10A). Finally, we detected the expression of related signaling pathway genes in the nude mouse xenografts in vivo using ELISA, IHC and western blotting ( Figure 10B). The results showed that the expression of IL-6 in the supernatant of GBC-SD+GCAFs group was signi cantly higher than that of GBC-SD group (P<0.05), and the protein expression of JAK1, JAK2 and STAT3 in GBC-SD+GCAFs group was signi cantly higher than that in GBC-SD group (all P<0.05). All together, these results suggested that GCAFs promote VM formation of GBC via upregulating NOX4 expression through paracrine IL-6 mediating IL-6/JAK/STAT3 signaling pathway.

Discussion
The interaction between TME and tumor cells has been a hotspot in the study of carcinogenesis [4,5].
CAFs, as the most critical stromal cells in TME, participate in the process of tumor growth and development [6][7][8][9][10][11]39]. As an important supplement to tumor microcirculation, VM occurs in certain highly aggressive malignancies, and is closely related to tumor progression and poor prognosis [13][14][15][16]40]. However, the relationship between CAFs and VM formation has rarely been reported. Recently, some studies have con rmed that CAFs can promote VM formation in some tumors, such as HCC [21] and gastric cancer [41,42]. In this study, we observed that GCAFs can promote VM formation and tumor growth of GBCs in vitro and in vivo. Therefore, we rst believed that GCAFs has the ability to promote tumor growth and VM formation in GBC.
Recently, it was reported that CAFs promote VM formation in HCC cells by secreting TGF-β and SDF1 [21]; CAFs-derived HGF promote vascularization in gastric cancer [41]. So whether there are speci c genes regulation in the process of GCAFs-triggered VM formation in GBC? Here, we performed microarray analysis for GCAFs/NFs using Affymetrix GeneChip Human 1.0ST array, and for VM (+)/VM (-) using Affymetrix Human lncRNA array. In the process of analyzing the differentially expressed gene pro les of GCAFs and NFs, we found that 16 genes in GCAFs were related to tumor angiogenesis. Among them, the expression of NOX4 was signi cantly increased.At the same time, the result of lncRNAs microarray showed that the expression of NOX4, JAK and STAT3 was signi cantly increased in 154 VM-related genes of GBC. As mentioned above, NOX4, as a member of the NADPH oxidase family, plays vital roles in proliferation, metastasis and angiogenesis of tumors through the production of ROS [29,43,44]. Then, we detected NOX4 expression in vitro and in vivo using qRT-PCR, western blot and IHC. The results showed the expression of NOX4 in GCAFs was signi cantly up-regulated at both mRNA and protein levels. Finally, we further detected the expression of NOX4 protein in human gallbladder tissue or stroma samples using IHC and CIF. The results showed NOX4 expression in epithelium or stroma was signi cantly higher than that in gallbladder precancerous lesions and benign lesions. All together, these data veri ed the high expression of NOX4 not only in GBC cells/stroma, but also in the process of VM formation induced by GCAFs. In other words, GCAFs upregulate the expression of NOX4 during VM formation in GBC. It has been con rmed that NOX4 is highly expressed in many types of tumor cells such as pancreatic cancer [45], renal cell carcinoma [46] and gastric cancer [47]. However, current researches on NOX4 are mainly focused on tumor cells, and the role of NOX4 in VM formation is rarely mentioned.
Thus, our nding rstly con rmed that NOX4 is highly expressed in GBC and its stroma, which is a GCAFs-derived key gene for VM formation.
In order to investigate the clinical value of the overexpression of NOX4, we further studied the relationship between the expression of NOX4 and clinicopathological characteristics/prognostic factors in patients with GBC. It was recently reported that the expression of NOX4 is closely related to the tumor size and prognosis in gastric cancer [47]; the prognosis of colorectal cancer patients with high expression of NOX4 was poor [48]. In this study, the results showed that the high expression of NOX4 in GBC cells, especially in GBC stroma (i.e. GCAFs) was signi cantly correlated with tumor differentiation, liver metastasis, vascular invasion and Nevin staging, specially VM formation. The multivariate analysis con rmed that tumor differentiation degree, liver metastasis, vascular invasion and NOX4 stroma expression were the independent prognostic factors for the OS of GBC patients. The survival time of GBC patients with high NOX4 expression was signi cantly shorter than that of GBC patients with low expression. These results rstly con rmed that NOX4 is related with aggression and unfavorable prognosis in patients with GBC; the high expression of NOX4 in GBC and its stroma predicts a poor prognosis.
At present, the mechanism of CAFs promoting tumor VM formation is not clear. It was reported that CAFs promoted VM formation in HCC by paracrine TGF-β and SDF1 [21]; CAFs-derived HGF promoted vascularization in gastric cancer via PI3K/AKT and ERK1/2 signaling [41]; CAFs induced vasculogenic mimicry in gastric cancer cells through the role of EphA2-PI3K signaling [42]. In order to explore the underlying molecular mechanism of GCAFs promoting VM formation in GBC, in view of aforementioned microarray analysis results, we found that among 154 VM-related genes, NOX4, JAK and STAT3 are involved in VM formation of GBC. So, we constructed a gene network regulation model in which GCAFs promote VM formation according to lncRNAs chip analysis. The results showed that JAK1, JAK2, STAT3 and NOX4 were regulated in the VM-related key genes.
Paracrine is one of the important ways in which CAFs act on tumor cells, CAFs can promote tumor cell growth and angiogenesis by secreting a variety of cytokines [31,32]. It was reported that the JAK/STAT3 signaling pathway plays an important role in mediating the multiple effects of interleukin-6 (IL-6) on tumor proliferation, invasion and metastasis [38]. IL-6 is an important in ammatory cytokine secreted by tumor cells or stromal cells [33], which is highly expressed in some tumors, and participate in a variety of malignant biological processes such as apoptosis, proliferation, metastasis and angiogenesis by activating a variety of signal pathways, such as STAT3, ERK and MAPK [34][35][36][37]. So, IL-6/JAK/STAT3 signaling pathway was believed as a key signaling pathway for tumor progression [38], and has been con rmed in many types of solid tumors such as HCC [49], breast cancer [50] and lung adenocarcinomas [51]. Considering that IL-6, which is paracrine-secreted by stromal cells such as CAFs, participates in tumor development by activating the IL-6/JAK/STAT3 signaling axis [38], we assume that GCAFs promote VM formation of GBC via upregulating NOX4 expression through paracrine IL-6 mediating IL-6/JAK/STAT3 signaling pathway. In order to verify the hypothesis, we detected the mRNA and protein expression of IL-6, JAK1, JAK2 and STAT3 in vitro and in vivo using ELISA, IHC, western blotting and qRT-PCR. The results showed that IL-6 was highly secreted and highly expressed in GCAFs, and these pathway genes were highly expressed during the VM formation in GBC. All together, these results rstly suggested that GCAFs promote VM formation in GBC via upregulating NOX4 expression through paracrine IL-6 mediating IL-6/JAK/STAT3 signaling pathway.

Conclusions
Collectively, our study rstly demonstrate that GCAFs promote VM formation and tumor growth in GBC via upregulating NOX4 expression through paracrine IL-6 mediated IL-6-JAK-STAT3 signal pathway. NOX4, as a key gene of VM formation in GBC, is highly expressed in tumor and stroma, which is related to the progression and poor prognosis of GBC patients. The present ndings may be of importance to explore a promising novel strategy for diagnosis, prognostic judge and anti-VM target treatment in human GBC.

Declarations
Ethics approval and consent to participate  GCAFs promote the malignant phenotypes of GBC cells/HUVECs. (A), Proliferation assay, the proliferation ability of GBC cell+GCAFs group was signi cantly enhanced compared with alone GBC cell group and GBC cell+NFs group (all *P<0.05). (B), Transwell invasion assay (Giemsa stain, ×200), the number (relative invasion ability) of cells that invaded through the basement membrane in GBC cell+GCAFs group was signi cantly more than that of GBC cell group and GBC cell+NFs group (all *P<0.05). (C), Wound healing assay, the cell migration rate of GBC cell+GCAFs group was signi cantly stronger than that of GBC cell group and GBC cell+NFs group (8h: all *P<0.05; 24h: all #P<0.05). (D), Tube formation assay, at 12h: HUVEC group (-); HUVEC+GBC-SD (+), HUVEC+GBC-SD+NFs (+), *P<0.05; HUVEC+GBC-SD+GCAFs (+), #P<0.05. At 48h: obvious tubular formation was observed in all the four groups, the number of tube formed in HUVEC+GBC-SD+GCAFs group was signi cantly more than that in the other three groups (#P<0.01), but no statistical difference was observed between HUVEC+GBC-SD group and HUVEC+GBC-SD+NFs group.    NOX4 is highly expressed in human GBC tissue and stroma, and is associated with poor prognosis. A (IHC, ×200; E=epithelium, S=Stroma). The expression (cytoplasmic and/or nuclear brown staining) of NOX4 in GBC cells and stroma (n=85) was signi cantly upregulated compared with gallbladder precancerous (adenomas, n=10; P=0.007, P=0.003) and benign lesions (cholecystitis, n=10; both P=0.001), but no statistical difference between precancerous lesions and benign lesions (both P>0.05).