Long Non-Coding RNA MIR31HG Promotes the Transforming Growth Factor β-Induced Epithelial-Mesenchymal Transition in Pancreatic Ductal Adenocarcinoma Cells

The epithelial-to-mesenchymal transition (EMT) describes a biological process in which polarized epithelial cells are converted into highly motile mesenchymal cells. It promotes cancer cell dissemination, allowing them to form distal metastases, and also involves drug resistance in metastatic cancers. Transforming growth factor β (TGFβ) is a multifunctional cytokine that plays essential roles in development and carcinogenesis. It is a major inducer of the EMT. The MIR31 host gene (MIR31HG) is a newly identified long non-coding (lnc)RNA that exhibits ambiguous roles in cancer. In this study, a cancer genomics analysis predicted that MIR31HG overexpression was positively correlated with poorer disease-free survival of pancreatic ductal adenocarcinoma (PDAC) patients, which was associated with upregulation of genes related to TGFβ signaling and the EMT. In vitro evidence demonstrated that TGFβ induced MIR31HG expression in PDAC cells, and knockdown of MIR31HG expression reversed TGFβ-induced EMT phenotypes and cancer cell migration. Therefore, MIR31HG has an oncogenic role in PDAC by promoting the EMT.


Introduction
In 2020, pancreatic cancer became the seventh leading cause of cancer mortality worldwide, and there are nearly as many pancreatic cancer patient deaths (466,000) as there are new cases (496,000) due to its poor prognosis [1]. Pancreatic ductal adenocarcinoma (PDAC) accounts for more than 85% of patients, and it is the most prevalent and aggressive pancreatic cancer type [2]. PDAC has a very low 5-year survival rate (of <8%) because of late diagnoses, high rates of metastasis, and drug resistance [3,4]. The recurrence rate remains significant even in patients with early-stage disease [5]. The majority of PDAC patients (80%~85%) are not suitable for surgical resection, which is the only curative therapeutic option. Furthermore, approximately half of PDAC patients have metastases when first diagnosed, also negating curative resection [4]. Gemcitabine-based chemotherapy is indispensable for treating unresectable pancreatic cancer [5]. However, current regimens

Upregulation of MIR31HG Is Associated with Disease-Free Survival in PDAC Patients
Previously, only one study investigated the role of MIR31HG in pancreatic cancer [10], which found that MIR31HG is overexpressed in PDAC tissues and cell lines. Downregulation of MIR31HG inhibits in vitro and in vivo PDAC cell growth by repressing cell cycle progression and inducing apoptosis. In addition, they identified that a tumor-suppressive miR-193b directly targets MIR31HG, and MIR31HG also competes for miR-193b binding to its messenger (m)RNA targets [10]. To obtain a greater understanding of the role of MIR31HG in PDAC, MIR31HG expressions in normal and PDAC tissues were obtained from the Gene Expression Profiling Interactive Analysis (GEPIA) database [32]. As shown in Figure 1A, MIR31HG was overexpressed in PDAC tumor tissues according to The Cancer Genome Atlas (TCGA) PanCancer Atlas dataset [33]. Because only four normal pancreatic tissues existed in this dataset and an additional 167 normal pancreatic tissues were from Genotype-Tissue Expression (GTEx) data [34], two other pancreatic cancer cohorts (GSE16515 [35] and GSE28735 [36]) were obtained from the Gene Expression Omnibus (GEO) database [37]. Consistently, MIR31HG was found to be overexpressed in tumor tissues in these two cohorts ( Figure 1B). To investigate the prognostic role of MIR31HG in PDAC patients, Kaplan-Meier overall and disease-free survival plots correlated with MIR31HG expression were obtained from the GEPIA and PROGgeneV2 databases [38]. As shown in Figure 1C and 1D, MIR31HG overexpression was not correlated with overall survival in either TCGA or GSE28735 datasets. However, MIR31HG expression was correlated with disease-free survival in the TCGA dataset, suggesting that MIR31HG overexpression may be correlated with recurrence and metastasis in PDAC patients. Because the disease-free survival status in the GSE28735 cohort was not available, whether MIR31HG overexpression was also correlated with patient's disease-free survival was unknown. cohorts (GSE16515 [35] and GSE28735 [36]) were obtained from the Gene Expression Omnibus (GEO) database [37]. Consistently, MIR31HG was found to be overexpressed in tumor tissues in these two cohorts ( Figure 1B). To investigate the prognostic role of MIR31HG in PDAC patients, Kaplan-Meier overall and disease-free survival plots correlated with MIR31HG expression were obtained from the GEPIA and PROGgeneV2 databases [38]. As shown in Figure 1C and 1D, MIR31HG overexpression was not correlated with overall survival in either TCGA or GSE28735 datasets. However, MIR31HG expression was correlated with disease-free survival in the TCGA dataset, suggesting that MIR31HG overexpression may be correlated with recurrence and metastasis in PDAC patients. Because the disease-free survival status in the GSE28735 cohort was not available, whether MIR31HG overexpression was also correlated with patient's disease-free survival was unknown. MIR31HG expressions in pancreatic tumor and normal tissues were obtained from the GEPIA database, while 151 pancreatic tumors and four normal tissues were from the TCGA PanCancer Atlas dataset. Another 167 normal pancreatic tissues were from the GTEx database; (B) MIR31HG expressions in normal and pancreatic cancer tissues were obtained from two pancreatic cancer cohorts (GSE16515 and GSE28735) in the GEO database; (C) Kaplan-Meier overall and disease-free survival plots for cancer patients (TCGA PanCancer Atlas dataset) with high and low MIR31HG expression were generated using the GEPIA database; (D) Kaplan-Meier overall survival plots for cancer patients (GSE28735) with high and low MIR31HG expression were generated using PROGgeneV2 database.

Upregulation of MIR31HG is associated with the EMT gene signature in PDAC patients
To investigate molecular alterations correlated with MIR31HG overexpression in PDAC, microarray gene expression profiles in GSE16515 and GSE28735 were analyzed using Gene Set Enrichment Analysis (GSEA) software [39,40] for enrichment of cancer hallmarks [41]. We found that seven cancer hallmarks, including TGFβ signaling, were commonly enriched in the two PDAC cohorts (Figure 2A). TGFβ signaling has both protumorigenic and tumor-suppressive roles in PDAC, which depends on the tumor stage [42,43]. TGFβ suppresses tumors in the early stages of carcinogenesis, because it serves as an antimitogen that stops cell-cycle progression during the G1 phase [44]. During pancreatic carcinogenesis, changes in TGFβ signaling components are prevalent. For example, mutations in the SMAD4 and TGFβ type II receptor (TGFBRII) genes may make cancer (A) MIR31HG expressions in pancreatic tumor and normal tissues were obtained from the GEPIA database, while 151 pancreatic tumors and four normal tissues were from the TCGA PanCancer Atlas dataset. Another 167 normal pancreatic tissues were from the GTEx database; (B) MIR31HG expressions in normal and pancreatic cancer tissues were obtained from two pancreatic cancer cohorts (GSE16515 and GSE28735) in the GEO database; (C) Kaplan-Meier overall and disease-free survival plots for cancer patients (TCGA PanCancer Atlas dataset) with high and low MIR31HG expression were generated using the GEPIA database; (D) Kaplan-Meier overall survival plots for cancer patients (GSE28735) with high and low MIR31HG expression were generated using PROGgeneV2 database.

Upregulation of MIR31HG Is Associated with the EMT Gene Signature in PDAC Patients
To investigate molecular alterations correlated with MIR31HG overexpression in PDAC, microarray gene expression profiles in GSE16515 and GSE28735 were analyzed using Gene Set Enrichment Analysis (GSEA) software [39,40] for enrichment of cancer hallmarks [41]. We found that seven cancer hallmarks, including TGFβ signaling, were commonly enriched in the two PDAC cohorts (Figure 2A). TGFβ signaling has both protumorigenic and tumor-suppressive roles in PDAC, which depends on the tumor stage [42,43]. TGFβ suppresses tumors in the early stages of carcinogenesis, because it serves as an antimitogen that stops cell-cycle progression during the G 1 phase [44]. During pancreatic carcinogenesis, changes in TGFβ signaling components are prevalent. For example, mutations in the SMAD4 and TGFβ type II receptor (TGFBRII) genes may make cancer cells resistant to TGFβ's antimitogenic activity [42,43]. Thus, TGFβ promotes cancer invasion, angiogenesis, and metastasis by inducing the EMT in late-stage carcinogenesis [42,43]. Although the EMT cancer hallmark was not commonly enriched in these two pancreatic cancer cohorts, we found that the EMT tended to be enriched in MIR31HG-high expressing pancreatic cancer patients ( Figure 2B). Therefore, MIR31HG overexpression was correlated with the EMT gene signature in PDAC. Consistent with our study, colorectal cancer patients with higher MIR31HG expression were characterized by elevated TGFβ and EMT gene expressions [15]. cells resistant to TGFβ's antimitogenic activity [42,43]. Thus, TGFβ promotes cancer invasion, angiogenesis, and metastasis by inducing the EMT in late-stage carcinogenesis [42,43]. Although the EMT cancer hallmark was not commonly enriched in these two pancreatic cancer cohorts, we found that the EMT tended to be enriched in MIR31HG-high expressing pancreatic cancer patients ( Figure 2B). Therefore, MIR31HG overexpression was correlated with the EMT gene signature in PDAC. Consistent with our study, colorectal cancer patients with higher MIR31HG expression were characterized by elevated TGFβ and EMT gene expressions [15].

Figure 2.
Gene set enrichment analysis (GSEA) revealed the potential role of MIR31HG in transforming growth factor β (TGFβ) signaling and the epithelial-to-mesenchymal transition (EMT): (A) A GSEA was performed to enrich 50 cancer hallmarks in pancreatic tumor tissues; (B) A GSEA was performed to enrich the EMT gene signature in two pancreatic cancer cohorts (GSE16515 and GSE28735). FDR, false-discovery rate.

MIR31HG enhances the TGFβ-induced EMT in PANC-1 cells
To investigate whether MIR31HG participates in the TGFβ-induced EMT, PANC-1 cells were transfected with MIR31HG small interfering (si)RNA and then challenged with TGFβ. Cell morphological observations showed that cells had converted to a spindleshaped, fibroblast-like morphology by 48 h TGFβ treatment, which was prevented by MIR31HG knockdown ( Figure 4A). Real-time qPCR and Western blot analyses showed that MIR31HG knockdown suppressed TGFβ-induced downregulation of CDH1 and CLDN4 and upregulation of COL1A1 at both the mRNA and protein levels ( Figure 4B and 4C). On the other hand, MIR31HG overexpression slightly enhanced 24 h TGFβ treatment-induced cell morphological changes and upregulation of COL1A1 and VIM at the mRNA or protein level ( Figure 4D-4F). However, MIR31HG knockdown or overexpression alone was insufficient to induce significant changes in cell morphology and EMT marker expressions (Figure 4), suggesting that MIR31HG participates in the EMT when TGFβ signaling is activated. Acquisition of a motile capacity is an important feature of the EMT [48]. Thus, the migrating activity of PANC-1 cells was measured by a wound-healing

MIR31HG Enhances the TGFβ-Induced EMT in PANC-1 Cells
To investigate whether MIR31HG participates in the TGFβ-induced EMT, PANC-1 cells were transfected with MIR31HG small interfering (si)RNA and then challenged with TGFβ. Cell morphological observations showed that cells had converted to a spindleshaped, fibroblast-like morphology by 48 h TGFβ treatment, which was prevented by MIR31HG knockdown (Figure 4A). Real-time qPCR and Western blot analyses showed that MIR31HG knockdown suppressed TGFβ-induced downregulation of CDH1 and CLDN4 and upregulation of COL1A1 at both the mRNA and protein levels ( Figure 4B,C). On the other hand, MIR31HG overexpression slightly enhanced 24 h TGFβ treatment-induced cell morphological changes and upregulation of COL1A1 and VIM at the mRNA or protein level ( Figure 4D-F). However, MIR31HG knockdown or overexpression alone was insufficient to induce significant changes in cell morphology and EMT marker expressions (Figure 4), suggesting that MIR31HG participates in the EMT when TGFβ signaling is activated. Acquisition of a motile capacity is an important feature of the EMT [48]. Thus, the migrating activity of PANC-1 cells was measured by a wound-healing assay. As shown in Figure 5, TGFβ increased the cell-migrating activity, which was rescued by silencing the MIR31HG expression. Consistently, MIR31HG knockdown alone did not affect cell-migratory activity. assay. As shown in Figure 5, TGFβ increased the cell-migrating activity, which was rescued by silencing the MIR31HG expression. Consistently, MIR31HG knockdown alone did not affect cell-migratory activity. MIR31HG or control siRNA for 24 h and then exposed to 5 ng/mL TGFβ for another 24 h. Total RNAs were examined by a real-time qPCR for MIR31HG and EMT marker expressions. Results are shown as the mean ± standard deviation (SD). * p<0.05 and *** p<0.001 or # p<0.05, ## p<0.01, and ### p<0.001 indicate statistical significance between groups. ns, no statistical significance between indicated groups; (C) PANC-1 cells were transfected with MIR31HG or control siRNA for 24 h and then exposed to 5 ng/mL TGFβ for another 48 h. Total protein lysates were examined by Western h and then exposed to 5 ng/mL TGFβ for another 24 h. Total RNAs were examined by a realtime qPCR for MIR31HG and EMT marker expressions. Results are shown as the mean ± standard deviation (SD). * p < 0.05 and *** p < 0.001 or # p < 0.05, ## p < 0.01, and ### p < 0.001 indicate statistical significance between groups. ns, no statistical significance between indicated groups; (C) PANC-1 cells were transfected with MIR31HG or control siRNA for 24 h and then exposed to 5 ng/mL TGFβ for another 48 h. Total protein lysates were examined by Western blotting for EMT marker expressions. Representative images of each protein were obtained from the same or different blots. The band intensity was quantified and the related ratio to untreated si-NC cells was shown; (D) PANC-1 cells were transfected with a MIR31HG-encoding or control plasmid for 24 h and then exposed to 5 ng/mL TGFβ for another 24 h. Cell morphology was observed under bright-field microscopy. The arrows indicate cells with spindle-shaped, fibroblast-like morphology. Scale bar, 25 µm; (E) PANC-1 cells were transfected with a MIR31HG-encoding or control plasmid for 24 h and then exposed to 5 ng/mL TGFβ for another 24 h. Total RNAs were examined by a real-time qPCR for MIR31HG and EMT marker expressions. Results are shown as the mean ± standard deviation (SD). * p < 0.05, ** p < 0.01, and *** p < 0.001 or # p < 0.05, ## p < 0.01, and ### p < 0.001 indicate statistical significance between groups. ns, no statistical significance between indicated groups; (F) PANC-1 cells were transfected with a MIR31HG-encoding or control plasmid for 24 h and then exposed to 5 ng/mL TGFβ for another 48 h. Total protein lysates were examined by Western blotting for EMT marker expressions. Representative images of each protein were obtained from the same or different blots. The band intensity was quantified and the related ratio to untreated pMS2 cells was shown. blotting for EMT marker expressions. Representative images of each protein were obtained from the same or different blots. The band intensity was quantified and the related ratio to untreated si-NC cells was shown; (D) PANC-1 cells were transfected with a MIR31HG-encoding or control plasmid for 24 h and then exposed to 5 ng/mL TGFβ for another 24 h. Cell morphology was observed under bright-field microscopy. The arrows indicate cells with spindle-shaped, fibroblastlike morphology. Scale bar, 25 μm; (E) PANC-1 cells were transfected with a MIR31HG-encoding or control plasmid for 24 h and then exposed to 5 ng/mL TGFβ for another 24 h. Total RNAs were examined by a real-time qPCR for MIR31HG and EMT marker expressions. Results are shown as the mean ± standard deviation (SD). * p<0.05, ** p<0.01, and *** p<0.001 or # p<0.05, ## p<0.01, and ### p<0.001 indicate statistical significance between groups. ns, no statistical significance between indicated groups; (F) PANC-1 cells were transfected with a MIR31HG-encoding or control plasmid for 24 h and then exposed to 5 ng/mL TGFβ for another 48 h. Total protein lysates were examined by Western blotting for EMT marker expressions. Representative images of each protein were obtained from the same or different blots. The band intensity was quantified and the related ratio to untreated pMS2 cells was shown.
Cancer cells employ the EMT to gain invasive ability [48]. It was shown that MIR31HG knockdown inhibited the invasion of AsPC-1 and PANC-1 pancreatic cancer cells [10]. Whether MIR31HG knockdown also inhibits TGFβ-induced cell invasion warrants further investigation. The above study together with ours also implies that the effects of MIR31HG on cancer cell migration and invasion are uncoupled as reported earlier in other cancer cell types [49,50]. In addition, previous studies in non-small cell lung cancer and osteosarcoma cells also identified a promoting role of MIR31HG in the EMT [51,52]. However, they found that MIR31HG knockdown was sufficient to upregulate mesenchymal markers and downregulate epithelial markers [52], suggesting that a cancer type-specific role of MIR31HG may exist.
Our results indicated that MIR31HG alone did not seem to impact the EMT pheno-
Cancer cells employ the EMT to gain invasive ability [48]. It was shown that MIR31HG knockdown inhibited the invasion of AsPC-1 and PANC-1 pancreatic cancer cells [10]. Whether MIR31HG knockdown also inhibits TGFβ-induced cell invasion warrants further investigation. The above study together with ours also implies that the effects of MIR31HG on cancer cell migration and invasion are uncoupled as reported earlier in other cancer cell types [49,50]. In addition, previous studies in non-small cell lung cancer and osteosarcoma cells also identified a promoting role of MIR31HG in the EMT [51,52]. However, they found that MIR31HG knockdown was sufficient to upregulate mesenchymal markers and downregulate epithelial markers [52], suggesting that a cancer type-specific role of MIR31HG may exist.
Our results indicated that MIR31HG alone did not seem to impact the EMT phenotype. However, loss-of-function and gain-of-function experiments indicated that COL1A1 gene expression was regulated by MIR31HG among the EMT markers tested in this study, suggesting that COL1A1 may be a potential target of MIR31HG. Interestingly, COL1A1 expression can be suppressed by miR-193 family members [53,54]. Because MIR31HG acts as an endogenous sponge of miR-193b in PDAC [10], we hypothesized that MIR31HG may also compete for miR-193 binding to COL1A1 mRNA.
In addition to its role in cancer, MIR31HG also plays a role in adipocyte differentiation (adipogenesis) [55]. Overexpression of MIR31HG reduces adipocyte differentiation in vitro and in vivo. In contrast, knockdown of MIR31HG inhibits the expression of an adipogenicrelated gene, fatty acid binding protein 4 (FABP4), via histone modification of its gene promoter, leading to suppression of adipocyte differentiation [55]. Interestingly, it has been reported that EMT-derived breast cancer cells, but not epithelial cancer cells, can be trans-differentiated into post-mitotic and functional adipocytes, leading to the repression of primary tumor metastasis [56]. It would be interesting to explore whether such phenomena also exist in PDAC and the involving role of MIR31HG in cancer cell-adipocyte transdifferentiation in the future.
In conclusion, we found that MIR31HG overexpression in PDAC was positively associated with patients' disease-free survival, but not overall survival. In addition, MIR31HG overexpression was correlated with upregulation of TGFβ signaling and the EMT. In vitro experiments confirmed the induction of MIR31HG by TGFβ treatment in PDAC cells. Knockdown of MIR31HG expression reversed the TGFβ-induced EMT. However, this study has several potential biases and limitations. First, the expression of MIR31HG and its clinical impact in PDAC patients were only analyzed using TCGA, GSE16515, and GSE28735 datasets. Further validation using clinical samples is still needed. Second, only one cell line was used in the in vitro experiments. Cell type variations may exist. In addition, animal experiments are needed to validate the in vitro observations. Third, the mechanisms of how TGFβ upregulates MIR31HG expression and how MIR31HG promotes the TGFβ-induced EMT have not yet been elucidated. Therefore, more investigations are required for a full understanding of the oncogenic role of MIR31HG in PDAC and a better validation of our conclusions.

Cell Culture and Treatment
A human pancreatic cancer cell line (PANC-1) from the Bioresource Collection and Research Center (BCRC; Hsinchu, Taiwan) was kindly provided by Prof. Hsin-Yi Chen (Taipei Medical University, Taipei, Taiwan). Cells were cultured in Dulbecco's modified Eagle medium (DMEM) containing 10% fetal bovine serum (FBS), 2 mM L-glutamine, 1 mM sodium pyruvate, and 1% antibiotic-antimycotic, and were grown at 37 • C in a humidified CO 2 (5%) incubator. For TGFβ treatment, cells were first serum-starved for 24 h and then treated with TGFβ in serum-free medium. The recombinant human TGFβ1 (#100-21) was purchased from PeproTech (Rocky Hill, NJ, USA). Cellular morphological changes were photographed at 20× magnification under an inverted microscope (IX71, Olympus, Tokyo, Japan).

Wound-Healing Assay
Cells (10 6 ) were grown in six-well plates for 24 h to reach a monolayer at more than 90% confluence. A wound in each well was created with a pipette tip. Floating cells were washed away with prewarmed phosphate-buffered saline (PBS), and then cells were treated with TGFβ in serum-free medium. Wounds were photographed in the same position at 0 and 48 h. Wound sizes were analyzed by an ImageJ plugin (Wound_healing_size_tool) [57]. The wound closure rate was calculated by the following formula: Wound closure (%) = [(A 0h − A 48h )/A 0h ] × 100%, where A 0h is the initial wound size at time zero and A 48h is the wound size after 48 h.

Statistical Analysis
Statistical analyses were performed using GraphPad Prism 9 (GraphPad Software, San Diego, CA, USA). Data were initially tested for normality using the Kolmogorov-Smirnov test. Parametric (normally distributed) data were analyzed using an unpaired two-tailed Student s t-test. Nonparametric (non-normally distributed) data were analyzed with the two-tailed Mann-Whitney test. A p value of <0.05 was considered statistically significant.

Data Availability Statement:
The data supporting this study can be obtained from the public databases or are available on request from the corresponding author.