NADPH Oxidase 4 Promotes Hypoxia-induced Epithelial-to-mesenchymal Transition via Histone Modication in Pancreatic cancer

Hypoxia is a characteristic of the tumor microenvironments within Pancreatic cancer (PC) which has been linked to its malignancy. Oxidative stress, characterized by NADPH oxidase (NOX) activation, and epithelial-to-mesenchymal transition (EMT) could be induced by hypoxia which involved in tumor progression and metastasis. However, the relationship between hypoxia-induced oxidative stress and EMT has not been claried, and the regulatory mechanism of NADPH oxidase is still unknown. was performed using GraphPad Prism v6.0 (GraphPad Inc., La Jolla, CA, USA) software. All data were reported as the mean ± SD. The differences between two groups were analyzed using T-test and the differences among multiple groups were analyzed using one-way ANOVA or two-way ANOVA followed by Tukey’s test. Correlations between two groups were analyzed by the Person's Rank-Order method. Kaplan-Meier curve (Log-rank tests) was used to determine any signicant associations of patient outcome and hypoxia score or NOX4. P<0.05 was considered statistically signicant.


Introduction
Hypoxia is a vital feature of the tumor microenvironment, including pancreatic cancer (PC) (1,2).
Increased desmoplasia and resulting insu cient perfusion are important causes of hypoxia in PC (3,4).
Triggered by hypoxia, tumor cells activate a variety of molecular pathways in order to adapt to environmental change (5). Epithelial-to-mesenchymal transition (EMT) is a process of transforming tumor cells from epithelial to mesenchymal cell types which could be induced by hypoxia, contributing to tumor invasion and metastasis (6,7). In this regard, most studies proposed that hypoxia-inducible factor 1 subunit alpha (HIF1α) is held accountable for the induction of EMT (8,9). However, this view has been updated by the latest studies showing that the change of histone modi cations under hypoxia occurred more rapidly than HIF1α induction. This rapid response in histone modi cation is able to induce EMT process in Hela cells (10,11). However, the mechanism of hypoxia-induced histone methylation and its association with EMT in PC is not yet clear.
Oxidative stress is another crucial process in tumor cells after exposure to hypoxia. And NADPH oxidase (NOX), a main reactive oxygen species (ROS) producer, can be activated in response to hypoxia (12,13). Different from the inevitable physiological "leakage" of the mitochondrial respiratory chain during normal function, the enzymes NADPH oxidase are the primary sources of non-mitochondrial ROS production, and more importantly, they generate ROS in a regulated manner (14). Hence, the NADPH oxidase has been suggested as a possible cellular oxygen sensor (15). More importantly, NADPH oxidase has also been reported to induce EMT in breast cancer cells (16). Therefore, NADPH oxidase may be involved in hypoxiainduced EMT in PC.
In this study, we took advantage of GEO and TCGA database to establish a list of hypoxia-associated genes. Further analysis uncovered a prognostic value of hypoxia-related gene sets, which is also associated with EMT-related pathways. Then, we identi ed NOX4, the enzyme activated by hypoxia in PC, activated EMT pathway by regulating intracellular ROS levels and by modifying chromatin methylation modi cation.

Materials And Methods
Analysis of pancreatic cancer gene expression data from GEO and TCGA database The Hypoxia-related gene expression signature consisted of 200 genes and EMT-related gene expression signature consisted of 200 genes were obtained from the gene set of HALLMARK_HYPOXIA and HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION in The Molecular Signatures Database (MSigDB, https://www.gsea-msigdb.org/). Microarray gene expression data from GSE15471 and GSE16515 were used to screen for genes that were highly expressed in PC tissues and then merged with 200 hypoxiarelated genes. Level3 TCGA RSEM Gene expression data and clinical information for pancreatic cancer were downloaded from UCSC (https://xena.ucsc.edu/public). We got 30 highly expressed hypoxia-related genes in pancreatic cancer tissues. Then we calculated the hypoxia score of TCGA samples using the 30 genes as previously described (17). Brie y, for each gene, samples with the top 50% of expression value were given a score of +1, and samples with the bottom 50% of expression value were given a score of -1. The hypoxia score in each sample was the sum of the scores of 30 genes (Table S1). Similarly, EMT score of TCGA samples was calculated using 200 EMT-related genes using the same method (Table S2). TCGA subtype de ned by Bailey et al.(18), Collisson et al.(19), and Mo tt et al. (20) were derived from the most recent TCGA pancreatic adenocarcinoma subclassi cation (21).

Functional Analysis and GSEA
Spearman correlation analysis was used to nd genes related to hypoxia scores (Table S2) and NOX4 expression. The R package "clusterPro ler" (22) was applied for the Gene Ontology (GO) analysis and KEGG analysis of genes positively related to hypoxia score. Hallmark gene sets were enriched using Metascape (23). Gene-set enrichment analysis (GSEA) was applied to enrich hallmark gene sets related to NOX4 in TCGA samples. NOX4 knockdown and overexpress NOX4 short-hairpin RNAs (shRNAs) or scrambled control shRNA were designed, synthesized and packaged into lentivirus particles (Corues Biotechnology, Nanjing, China). Cells were plated into six-well plates (3×10^5 cells per well). Before transfection, the culture medium was replaced with DMEM with 10%FBS and 1 µg/mL Polybrene (GeneChem, Shanghai, China). Infectious lentivirus particles were harvested for 72 hours after transfection.

RNA extraction and RT-PCR
Total RNA was isolated from cells using RNAiso Plus Reagent (Takara, Kusatsu, Japan). RT reactions were performed using the PrimeScript™ RT Master Mix (Takara). Then quanti cation of mRNA expression was performed using SYBR® Advantage® qPCR Premix (Takara) in a total reaction volume of 20 µl according to the manufacturer's instructions. The reaction was performed using the LightCycler® 96 system (Roche Diagnostics, Basel, Switzerland). ACTB was used as an internal control. The primer sequences used were shown in the supplementary material.

Immunoblot analysis
Cell or tissue homogenate was used for immunoblot analysis. Speci c information was described in supplementary materials.

Immunohistochemistry
Para n sections of tissues from 50 PC patients were used for immunohistochemical detection of NOX4 expression. The detailed protocol was presented in the supplementary material.
Cell Counting Kit-8 (CCK8) assay CCK8 assay (Dojindo, Kumamoto, Japan) was used to detect the cell viability of PC cells according to the manufacturer's instructions. HPAC cells were planted in 96-well plates (5000 cells each) and incubated with 10 μL CCK8 for 2 h at 37 °C. The absorbance was recorded at 450nm.

Migration and invasion assays
Migration and invasion assays were proceeded in transwell chamber (Corning, NY, USA) with (invasion) or without (migration) Matrigel matrix (Corning) in a 24-well plate. 5x10 6 cells resuspended in serum-free DMEM were added in the up chamber, and DMEM medium with 20% FBS were added in the bottom chamber. Then cells were xed with 4% paraformaldehyde and stained with 1% crystal violet. Cells were imaged and counted using a 20× microscope.

Experimental mice
Four-week-old male BALB/c nude mice (weighing 16-18 g) were purchased from Changzhou Cavens Experimental Animal Co. Ltd (Changzhou, China). To establish the subcutaneous transplanted model, HPAC cells were injected subcutaneously to the ank region of nude mice of 6 weeks old. Tumors were measured using vernier callipers twice a week, and the volume of tumors was calculated using the formula (length x width2/2). After one month, all mice were sacri ced, and tumors were collected. For lung-metastasis xenografts, 1×10 6 HPAC cells suspended in 100 µL cold PBS were injected into the lateral tail vein (4 for each group). After one month, all mice were sacri ced, and lung tissues were xed with 4% paraformaldehyde. H&E staining was used to evaluate the proportion of metastatic lesions. All animals were approved by the Ethics Committee of Nanjing Drum Tower Hospital. All animals used in this study were treated humanely and followed guidelines set by the Animal Care Committee. The study was approved by the Ethics Review Committee for Animal Experimentation at Nanjing Drum Tower Hospital (Nanjing, China).

CHIP-PCR
Chromatin Immunoprecipitation(CHIP)-PCR was performed to analyze the effect of NOX4 on the binding of H3K4ME3 to the SNAIL1 promoter sequence. Detailed information was presented in the supplementary material.

Statistical analysis
All bioinformatics analyses were performed using R (https://www.r-project.org/) and Rstudio software (B Corps™, DE, USA). All statistical analysis was performed using GraphPad Prism v6.0 (GraphPad Inc., La Jolla, CA, USA) software. All data were reported as the mean ± SD. The differences between two groups were analyzed using T-test and the differences among multiple groups were analyzed using one-way ANOVA or two-way ANOVA followed by Tukey's test. Correlations between two groups were analyzed by the Person's Rank-Order method. Kaplan-Meier curve (Log-rank tests) was used to determine any signi cant associations of patient outcome and hypoxia score or NOX4. P<0.05 was considered statistically signi cant.

Result
Evaluation of hypoxia-related gene expression in PC First, we quanti ed hypoxia-related gene expression in gse15471 and ges16515 using 200 genes from hallmark HYPOXIA gene set (Molecular Signatures Database [MsigDB]). This analysis identi ed 30 shared hypoxia-related genes that were overexpressed in PCs, compared to normal controls in both databases (Figure 1a, b). Figure 1c showed the expression of these 30 genes across PAAD tumor samples in TCGA dataset. Then, we used these 30 genes as hypoxia-related gene signature, and a hypoxia score was calculated for each individual sample according to the expression levels of these genes (described in the method part, Table S1). Next, we explored the prognostic signi cance of hypoxiarelated gene expression in TCGA PAAD dataset. Tumors with the top 50% of hypoxia score values had a worse overall survival (OS) and progression-free survival (PFS) than those with the bottom 50% of hypoxia score values. And also, hypoxic tumors were more likely to have a higher grade and pathologic stage ( Figure 1f). To verify whether hypoxia scores are related to speci c molecular subtypes, we grouped TCGA PAAD samples by the four-groups classi cation of Bailey et al. (18), the three-group classi cation of Collisson et al. (19), and the two-group classi cation of Mo tt et al (20). This analysis revealed that the subtypes with unfavorable prognosis (squamous in Bailey clusters, quasimesenchymal (QM) in Collisson clusters and basal-like in Mo tt clusters) were more likely to have a higher hypoxia score (Figure 1g).

Epithelial-mesenchymal transition was induced in hypoxia samples
Then, we calculated the pearson correlation coe cient between each hypoxia score and all the other genes and ranked them according to the correlation coe cient in TCGA PAAD samples, and a total of 385 genes showed a signi cant positive correlation with hypoxia score (R≥0.5) (Table S2). Their biological processes and pathways were analyzed using the GO terms of biological processes and KEGG pathways.
The GO analysis showed that the biological processes such as extracellular structure organization and extracellular matrix organization were signi cantly enriched (Figure 2a), and the KEGG pathway analysis identi ed focal adhesion and PI3K-AKT signalling pathway as the enriched pathways ( Figure 2b). Next, we used the Hallmark gene sets from MSIGDB (https://www.gsea-msigdb.org/) to analyze the enriched sets of same 385 genes, which showed EMT pathway is the most enriched one (Figure 2c). Figure 2d showed the expression of EMT markers consisting of 200 genes increases along with hypoxia score. As above mentioned, we then calculated an EMT score for each individual sample and tested if it was correlated with the hypoxia score. This analysis revealed that EMT and hypoxia score was strictly correlated with each other (Figure 2e). To experimentally validate this, we tested the expression of EMTrelated genes in two PC cell lines treated with/without hypoxia (1% O 2 ). This analysis revealed that the mRNA and protein expression of vimentin (VIM), cadherin 2 (CDH2) and SNAIL1 increased in HPAC and Panc1 cell lines upon hypoxia exposure for 24 hours. Accordingly, the mRNA and protein expression of cadherin 1 (CDH1) decreased in both cell lines (Figure 2f, 2g).
Since hypoxia was described to stimulate the production of ROS, we rst evaluated the correlation between hypoxia-related expression and NOX family members which are responsible for generating nonmitochondrial ROS (24). This analysis revealed a signi cant correlation between NOX4 expression and hypoxia score (Figure 3a). We used the median value to distinguish groups with high and low expression of NOX4 in TCGA samples and carried out a GSEA analysis. Importantly, this analysis identi ed EMT and hypoxia pathways were signi cantly (with a cutoff for FDR 5% and p value < 0.01) enriched in high expression group of NOX4 (Figure 3b).
Further, we detected the location of NOX4 in the bulk tissues of two PC patients. NOX4 was expressed in both broblasts and epithelial cells (Supplementary Figure 1a). To further verify the expression of NOX4 after hypoxia, we examined the mRNA and protein expression of NOX4 after hypoxia exposure for 0-24 hours. We found that NOX4 expression increases at a very early stage (Figure 3c, 3d). We also found that the level of ROS increased steadily after hypoxia exposure in HPAC cells (Supplementary Figure 1b). Since superoxide is generated from NOX4 and quickly converts to H2O2 through dismutation (25), we hypothesized that H2O2 was partially responsible for NOX4 and hypoxia-induced EMT phenotype. Indeed, H2O2 activated an EMT program in HAPC and Panc1 cells in a dose-dependent manner (Figure 3e). In line, the level of ROS increased signi cantly, when NOX4 was overexpressed in HAPC and Panc1 cells (Figure 3f). Correspondingly, EMT process is also activated in these NOX4-overexpressing cells (Figure 3g, Supplementary Figure 1c Hypoxia-induced NOX4 activation was independent of HIF1α, but dependent on TGFβ1 HIF1α is a vital transcription factor induced by hypoxia. We then analyzed whether NOX4 is a downstream gene of HIF1α, however, we found the upregulation of NOX4 preceded to HIF1α induction upon hypoxia exposure (Figure 3c). And also, the hypoxia-induced NOX4 expression was not affected after the knockdown of HIF1α using siRNA (Figure 3i). Given that transforming growth factor beta 1 (TGFβ1) is a NOX4 inducer, we found TGFβ1 secreted by HPAC cells increased signi cantly after hypoxia ( Figure 3j). Then, we treated HPAC cells with TGFβ1 in time and concentration gradient, and similarly to previous studies, NOX4 is signi cantly activated by TGFβ1 in PC cells (Figure k). TGFβ1 neutralizing antibody treatment attenuated hypoxia-induced NOX4 activation in HPAC cells (Figure 3l). Therefore, TGFβ1 may be the upstream signal of NOX4 during hypoxia. NOX4 is highly expressed by PC and positively correlated with the degree of hypoxia in tumor tissue To investigate the relevance between NOX4 expression and hypoxia in PC tissues, we compared the IHC scores of NOX4 in HI1Fα-positive and negative PC tissues within a cohort of 56 PC patients from our centre and found that NOX4 was highly expressed in hypoxic PC tissues (Figure 4a). We also found that PC patients with higher NOX4 expression had worse overall survival and higher histologic grade than those with lower NOX4 expression (Figure 4b,c). We also analyzed NOX4 expression in GEO databases (GSE15471 and GSE16515) and found that NOX4 expression is up-regulated in bulk PC tissues ( Figure  4d). Similar results were found in 7 PC tissues compared with 6 benign pancreatic tissues collected in our centre (Figure 4e). We also observed that PC cell lines had a higher level of NOX4, as compared to human ductal pancreatic epithelial cells (HPDE cell line, Figure 4f). By analyzing TCGA data, the expression of NOX4 was signi cantly correlated with tumor stage and grade (Figure 4g,4h).

NOX4 promotes the proliferation and metastasis of PC cells
To investigate the functional signi cance of NOX4, we rst performed cell proliferation assay after the knockdown and overexpression of NOX4 in HPAC cells. After the knockdown of NOX4, the cell growth was inhibited, while the overexpression of NOX4 promoted the growth of HPAC cells (Figure 5a,b). Then we established subcutaneous transplanted model to investigate the role of NOX4 on the proliferation of HPAC cells in vivo. The volume of xenograft tumors was signi cantly increase after NOX4 overexpression ( Figure 5c) and decrease after the knockdown of NOX4 (Figure 5d). And also, we used GLX351322 (a NOX4 inhibitor) to treat HPAC cells, the growth of cells was also inhibited (Figure 5e).
In migration and invasion assays, we found that NOX4 overexpression enhanced the migration and invasion of HPAC cells (Figure 5f). Likewise, a similar outcome was observed in HPAC cells after NOX4 knockdown ( Figure 5g). Then, a lung-metastasis xenograft mouse model was established to study the potential role of NOX4 in PC metastasis in vivo. The percentage of metastatic area were signi cantly reduced after NOX4 knockdown (Figure 5h).

NOX4 up-regulates SNAIL1 expression by increasing histone methylation
The previous study has shown that hypoxia induced a robust increase of histone methylation markers. In particular, chromatin immunoprecipitation followed by deep sequencing (ChIP-sequencing) of H3K4me3, a methylation marker associated with active gene transcription, identi ed that EMT was up-regulated after hypoxia exposure. To investigate whether NOX4 signi cantly affected histone methylation modi cations, we tested various histone methylation markers after overexpressing NOX4. We found that NOX4 induced an increase in histone methylation (Figure 6a). Then we focused on the H3K4me3, as we veri ed the previous ndings that the expression of H3K4me3 was indeed signi cantly induced by hypoxia in HPAC cells (Figure 6b). Compellingly, the knockdown of NOX4 reversed the upregulation of H3K4me3 caused by hypoxia (Figure 6c). Then, we analyzed the ChIP-sequencing results of PC cell line for H3K4ME3 in the GEO database (GSE945856), and found a peak in the promoter region of SNAIL1.
Thus, we performed ChIP-PCR of H3K4me3 at ChIP-sequencing peak for SNAIL1 in HPAC cells after altering the expression of NOX4. The results revealed a marked increase in H3K4me3 at SNAIL1 after NOX4 overexpression or 24 hours of hypoxia exposure (Figure 6d,e). When NOX4 was knocked down with shRNAs, it was partially compromised (Figure 6e).

Discussion
Hypoxia is an adverse living condition for tumor cells; however, it leads to aggressive phenotypes in a variety of tumors (4,26,27). As we presented, hypoxia-related gene signature was prognostic and linked with up-regulated EMT pathway. NOX4-induced oxidative stress and rapid changes in chromatin modi cation status was a necessary processes facilitating this hypoxia-induced EMT process.
We quanti ed hypoxia score in the TCGA using 30 hypoxia-related genes that are highly expressed in PC and described the correlation between hypoxia score and clinical parameters in these samples. Similar to previous studies in other tumors (28,29), samples with high hypoxic score exhibited a worse prognosis.
Molecular subtypes of PC have been de ned by several studies and linked with prognosis and response to treatment (18)(19)(20). Here, we showed that samples with high expression of hypoxia-related genes were more concentrated in the subtype with the worst prognosis (squamous in Bailey clusters, QM in Collisson clusters and basal-like in Mo tt clusters). These results suggested that hypoxia is a driving factor that promoted the transformation of tumors to a more aggressive phenotype.
Hypoxia triggers varied molecular responses in tumor cells. To better understand the potential regulatory mechanism of PC cells under hypoxia, we used hypoxia-related gene signature to predict possible downstream pathways. Here, we found a strong positive correlation between hypoxia-related gene expression and EMT process in PC samples. EMT was induced by hypoxia in a variety of tumors including non-small cell lung cancer (NSCLC), ovarian carcinoma (30,31) and the mechanism involves the regulation of SNAIL, twist family bHLH transcription factor (TWIST) and snail family transcriptional repressor 2 (SLUG) (32,33). Stabilization of HIF-1a is a crucial transcription factor caused by intratumoral hypoxia which can induce EMT binding directly to the promoter of TWIST and SNAIL (33,34). However, we found a rapid induction of NOX4 after hypoxia in PC cells which signi cantly promotes the EMT process even before the activation of HIF1α.
In vivo and in vitro experiments demonstrated NOX4 overexpress or inhibition in pancreatic cancer cells caused changes of proliferation and invasion ability. And these were consistent with the analysis of clinical data from TCGA and our database. NOX4 has been reported to be activated to promote the antiapoptotic ability of PC (35,36). It has also been shown that NOX4 can promote tumor metastasis in some tumors such as human colorectal cancer and non-small cell lung cancer (37,38). More importantly, GLX351322, a NOX4 selective inhibitor, could attenuate proliferation ability of pancreatic cancer cells, which means that PC may bene t from NOX4-targeting therapy.
Changes in chromatin modi cation after hypoxia, especially histone methylation, are important newly discovered mechanisms in recent years (10). This rapid oxygen sensing mechanism caused a series of critical changes of important pathways to make tumor cells respond to hypoxia. We rstly demonstrate that NOX4 can induce stable histone methylation after hypoxia which leads to the regulation of important pathways including EMT. And this process of oxygen sensing seems to be earlier than the activation of HIF1α.
In summary, we demonstrate for the rst time that up-regulation of NOX4 after hypoxia can induce histone methylation through altering intracellular ROS levels. Up-regulation of histone modi cations, especially activation of H3K4ME3, associated with active gene transcription of SNAIL1 which cause rapid and robust EMT process. However, the mechanism of hypoxia-induced NOX4 activation and how NOX4 affects histone methylation still need further investigation.

Conclusions
In summary, our study established a list of 30 hypoxia-related genes in PC which was prognostic and linked with up-regulated EMT pathway. Further analysis found NOX4 was induced by hypoxia and crucial Medical School, and the informed consent forms were obtained from patients enrolled in this study.
All animals used in this study were treated humanely and the study was approved by the Ethics Review Committee for Animal Experimentation at Nanjing Drum Tower Hospital (Nanjing, China).

Consent for publication
The content of this manuscript has not been previously published and is not under consideration for publication elsewhere.  after 1%O2 treatment for 24 hours were assessed by qRT-PCR. The e ciency of hypoxia was validated using HIF1α. ***p < 0.001, **p < 0.01, *p < 0.05 vs 20% O2. The data were presented as the mean ± SD.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.