Ephrin Receptor A4 Expression Enhances Migration, Invasion and Neurotropism in Pancreatic Ductal Adenocarcinoma Cells

Background/Aim: We sought to identify the mechanisms of perineural invasion in pancreatic ductal adenocarcinoma (PDAC). Materials and Methods: We utilized in vitro cancer cell-nerve co-culture models comprising human PDAC cell lines (MIA Paca2 and PANC-1) and a dorsal root ganglion (DRG) isolated from neonatal mice. We compared gene expression profiles between cell lines with/without DRG conditioned medium (DRG-CM) using RNA-sequencing (RNA-seq). Results: Migration, invasion, and neurotropism were significantly enhanced in MIA Paca2 but not in PANC-1 cells co-cultured with DRGs. Among 285 genes which showed significant differences in expression levels between cell lines in RNA-seq, we focused on Ephrin receptor A4 (EPHA4), which was upregulated in MIA Paca2 cells treated with DRG-CM. The abilities of migration, invasion, and neurotropism enhanced by DRG co-culture were abolished when EPHA4 was knocked down by siRNA in MIA Paca2 cells. Conclusion: EPHA4 can be a potential target gene to regulate perineural invasion in PDAC cells.

Pancreatic ductal adenocarcinoma (PDAC) is a devastating malignancy with dismal prognosis and ranks as the fourth highest cause of cancer-related death in the United States (1). Most patients with PDAC present with locoregional spread or metastatic disease at the time of diagnosis (2). Most cases recur even after local disease control has been achieved by surgical intervention and are strongly refractory to systemic chemotherapy (3). Thus, improved recognition of the biological behavior of PDAC, particularly the aggressive nature of its invasion, is urgently needed.
Perineural invasion (PNI) is a common pathological feature of PDAC, as evidenced by its high incidence observed in surgical specimens of PDAC (4,5). Previous reports have shown that PNI contributes to locally advanced and/or metastatic disease progression in patients with PDAC (6,7). In the past, the anatomical proximity between the pancreas and the periarterial plexus has been implicated in the development of PNI in PDAC (8). More recently, it has been suggested on the basis of neurotropic theory, that the nerves and invading tumor cells interact with each other using neurotrophins to establish PNI (9,10). However, the detailed molecular mechanisms of PNI in PDAC remain to be fully elucidated.
There are few ideal in vitro experimental models to study PNI in malignant disease. Dai et al. have established an in vitro co-culture model using a human PDAC cell line and mouse dorsal root ganglion (DRG), and demonstrated neurite outgrowth from the DRG and enhanced colony formation by PDAC cells, suggesting a mutual growth support (11). Our previous study has utilized this in vitro cancer cells-nerve cells co-culture model with modifications and investigated the role of Tenascin C, an extracellular matrix protein, by evaluating the interactions between PDAC and nerve cells (12).
In this study, we sought to identify the molecules in PDAC cells that are responsible for PNI. First, we evaluated the differences in the migration, invasion, and neurotropic abilities of PDAC cells using an in vitro cancer cells-nerve cells co-culture model. Second, using RNA sequencing (RNA-seq), we compared the gene expression profiles of two PDAC cell lines (MIA Paca2 and PANC-1) treated with or without DRG-conditioned medium (DRG-CM). Of these genes, Ephrin receptor A4 (EPHA4) was suggested to be a potential target gene that regulates PNI in PDAC cells.

Materials and Methods
Cell culture. The human pancreatic cancer cell lines, MIA Paca2 and PANC-1, were purchased from RIKEN Bioresources Cell Bank (BRC Cell Bank, Ibaraki, Japan) and provided by the Cell Resource Center for Biomedical Research, Tohoku University (Sendai, Japan). MIA Paca2 was routinely grown in complete Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Carlsbad, CA, USA) supplemented with 10% Fetal Bovine Serum (FBS), at 37˚C in a humidified atmosphere saturated with 5% CO 2 . PANC-1 was grown in RPMI1640 (Wako, Osaka, Japan) with 10% FBS at 37˚C in a humidified atmosphere saturated with 5% CO 2 .
Isolation of DRG from neonatal mice. The following animal procedures were performed according to the guidelines of the Committee on Experimental Animals of Hamamatsu University School of Medicine . The procedures of mouse DRGs' isolation were described by Ayala et al. (13). Briefly, neonatal (1day-old) ICR mice (Japan SLC, Shizuoka, Japan) were anesthetized by isoflurane and euthanized by cervical dislocation. Each DRG was isolated by performing an anterior laminectomy and microscopic dissection from the lumbar spinal region.
Migration and invasion assays using cancer cell-neuron vertical coculture model. In vitro cell migration and invasion were evaluated using a cell culture insert and a Matrigel invasion chamber (Becton Dickinson, Bedford, MA, USA) separated by an 8 μm-pore filter membrane in 24 well-plates. For the cancer cell-nerve vertical coculture model, four DRGs collected from neonatal mice were seeded in 20 μl of Matrigel (#356231, Matrigel ® Growth Factor Reduced Basement Membrane Matrix, Corning, NY, USA) in the lower chamber, and incubated at 37˚C saturated with 5% CO 2 in a humidified atmosphere for 20 min to allow Matrigel polymerization ( Figure 1A). Then, the lower chamber was loaded with 0.75 ml of medium (DMEM or RPMI1640) containing 1% FBS. In the migration assay, 3×10 4 MIA Paca2 cells or 1×10 4 PANC-1 cells in 0.5 ml of serum-free medium were seeded in the upper chamber. In the invasion assay, 5×10 4 of MIA Paca2 cells or 3×10 4 PANC-1 cells were seeded. Wells with the same amount of Matrigel in the lower chamber without DRG seeding were used as controls (Matrigel-co). After the confirmation of DRG outgrowth 48 h after cultivation ( Figure 1A), non-migrated or non-invaded cells in the upper chamber were gently removed using cotton swabs. The migrated or invaded cells that passed through the membrane were stained with Diff Quick solution (International Reagents, Japan), and five randomly selected 100× magnification fields per membrane were counted under an optical microscope. Three wells for each condition were used in one experiment.
Neurotropism assay using a cancer cell-neuron horizontal coculture model. The protocol of the cancer cell-DRG horizontal coculture model was modified as described previously (12,14). Briefly, 5×10 4 of MIA Paca2 or PANC-1 cells were suspended in 5 μl of Matrigel drop and placed approximately 1 mm away from a 5 μl "DRG" Matrigel drop (Figure 2A). To exclude the possibility of unspecific migration of cancer cells, an additional 5 μl "Blank" Matrigel drop was positioned at the opposite site. The dishes were then placed in an incubator set at 37˚C saturated with 5% CO 2 in a humidified atmosphere for 20 min to allow for Matrigel polymerization. Each cell-suspended or blank Matrigel was connected with a 1 mm-long Matrigel plug, "Spacer" (Figure 2A). After incubation for additional 20 min for "Spacer" polymerization, the Matrigel drops were carefully submersed in 2 ml of DMEM or RPMI1640 supplemented with 1% FBS. The co-cultures were incubated at 37˚C with 5% CO 2 in a humidified atmosphere for 8 days. Representative photographic images of the adjacent and opposite areas of the cancer cell suspension were captured using a microscope (Eclipse TE2000-U, Nikon, Tokyo, Japan) and imaging system (AQUACOSMOS, Hamamatsu Photonics K. K, Shizuoka, Japan). For quantitative analysis of neurotropism in cancer cells, we defined the migration distance of cancer cells towards the DRG as an α1 parameter, and that away from the DRG as α2 parameter, and calculated the cancer neurotropic index as α1/α2 ratio ( Figure 2B). Images of migrating cancer cells were captured and fused using a microscope (Biozero, KEYENCE, Osaka, Japan), and the distances were measured using ImageJ software (15). This horizontal coculture model was performed in at least five biological replicates.

Preparation of DRG-Conditioned Medium (CM).
Twenty DRG cells were isolated from a single mouse and seeded each in a 5 μl drop of Matrigel on a 60 mm dish. After DRG suspensions, the dish was incubated at 37˚C saturated with 5% CO 2 in a humidified atmosphere for 20 min to allow Matrigel polymerization. Then, 4 ml of medium (DMEM or RPMI1640) containing 1% FBS was carefully submersed in the dish and incubated for additional 72 h. After confirming the axonal growth of DRG cells, the supernatant was filtered through a 0.22 μm filter (Millex-GV, Merck KGaA, Dermstadt, Germany) and stored at -30˚C. The medium collected from these steps was used as DRG-conditioned medium (DRG-CM). CM collected from the dish coated with the same amount of Matrigel without DRGs, was used as a control (Matrigel-CM). RNA-seq data processing. Gene expression was normalized by the fragments per kilobase of exon per million mapped fragments mapped (FPKM) method and filtered as previously described (16). Principal component analysis (PCA) mapping was performed using the log2 FPKM value in the R software (17). Differentially expressed genes (DEGs) with statistical significance were identified through the following filtering steps: the default threshold was Log 2 (FoldChange) ≥1 and q value <0.1 (false discovery rate was adjusted with the Benjamini-Hochberg method). The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed with the DAVID bioinformatics tool (version 6.8) (18). p-Values were obtained using Fisher's exact test. Raw data analysis was performed by TaKaRa, Inc.
Quantitative RT-PCR. Total RNA was extracted from cultured cells as described above, and reverse transcription was performed using the Primer script RT Reagent kit (Takara Bio, Otsu, Japan) according to the manufacturer's protocol. The cDNA was amplified by quantitative RT-PCR (qRT-PCR) on a Thermal Cycler Dice Real Time System II (Takara Bio) using the Thunderbird 1PCR Mix (Toyobo Life Science, Osaka, Japan). All PCR reactions were run in triplicates, and the relative levels of gene expression normalized to the control was calculated using 2nd Derivative Maximum methods. Sequences of primers used for amplification were as follows: EphA4 Western blotting. Cells were lysed in chilled lysis buffer supplemented with complete protease and phosphatase inhibitor cocktail (Roche, Basel, Switzerland). Protein concentrations were determined using a bicinchoninic acid protein assay kit (Takara Bio). Protein extracts were subjected to 9% polyacrylamide gel electrophoresis followed by electroblotting onto an Immobilon-Polyvinylidene fluoride membrane (Millipore, Billerica, MA, USA). After blocking for 30 min with 5% skim milk, the membranes were incubated at 4˚C overnight with primary antibodies including anti-EPHA4 (1:1,000, Invitrogen, 4C8H5) and anti-β-actin (1:1,000, No 5125, Cell Signaling Technology, Danvers, MA, USA). The next day, the membranes were incubated for 1 h at room temperature with horseradish peroxidaseconjugated secondary antibodies. Immunoreactive bands were visualized using enhanced chemiluminescence plus western blotting detection reagent ((GE Healthcare, Little Chalfont, UK)), and fusion software (Vilber-Lourmat, Collégien, France).

cells (PDAC) and mouse Dorsal Root Ganglion (DRG). A) Schematic images of the in vitro horizontal co-culture model with a PDAC cell line and DRG. B) An illustration showing the distance covered by migrated tumor cells toward DRG (α1) and that away from the DRG toward neurite outgrowth (α2). C, D) Representative photos of in vitro horizontal co-culture models using PDAC cell lines (C: MIA Paca2 cells, D: PANC-1 cells) and DRG, which were co-incubated for 5 and 8 days, respectively. Magnified micrographs of adjacent and opposite areas are shown. E, F) The distance migrated by the PDAC cells (E: MIA Paca2 cells, F: PANC-1 cells) toward DRG (α1) and away from DRG (α2) were calculated (biological replicate, n=6), respectively. [Black bar in fusion: 200 μm, white bar in magnifying view: 300 μm, yellow dotted line: the migrated distance or PDAC cells toward DRG (α1), orange dotted line: the migrated distance or PDAC cells away from DRG]. d: Days.
parametric variables, respectively. p-Values <0.05 were considered statistically significant. All calculations were performed using SPSS 24.0 software (SPSS Inc., IL, USA).

Migration and invasion abilities of MIA Paca2 but not PANC-1 cells were increased in in vitro vertical co-culture models.
Initially, we investigated the migration and invasion abilities of cancer cells toward neurons using in vitro vertical coculture models ( Figure 1A). The number of migrated and invaded MIA Paca2 but not PANC-1 cells were significantly increased in the DRG co-culture model (DRG-co) compared to those in the control model (Matrigel-co) ( Figure 1B and C).

MIA Paca2 but PANC-1 cells migrated toward DRG
neurites in the in vitro horizontal model. Next, we investigated the neurotropism of PDAC cells using an in vitro horizontal cancer-neuron co-culture model (Figure 2A and B). In MIA Paca2 cells, cancer cell colonies extended toward the DRG, and longer extensions of cancer cell colonies were observed in the adjacent area of the DRG compared to the opposite side at 5 and 8 days after cocultivation ( Figure 2C). Quantitative analysis showed that the extended distance of MIA Paca2 cells toward the DRG was significantly longer than that toward the blank Matrigel. Conversely, such neurotropic behavior was not observed in PANC-1 cells ( Figure 2D).  Transcriptome signature of PDAC cell lines treated with DRG-CM or Matrigel-CM. Next, comprehensive gene expression in both PDAC cell lines (MIA Paca2 and PANC-1 cells) treated with/without DRG-conditioned medium was profiled using RNA-seq analysis. In total, 27,340 genes were identified, and principal component analysis (PCA) mapping showed that the two cell lines had remarkably distinct gene expression profiles ( Figure 3A). This PCA map also revealed a separation of samples according to the culture treated with Matrigel-CM or DRG-CM in MIA Paca2 cells; however, these differences were not observed in PANC-1 cells ( Figure 3A).  Figure 3D) that belonged to the axon guidance pathway, we focused on EPHA4 as a predicted target gene responsible for PNI in PDAC cells. qRT-PCR and western blot analyses revealed that the expression of EPHA4 mRNA and EPHA4 protein was elevated in MIA Paca2 cells treated with DRG-CM compared to that in cells treated with Matrigel-M, whereas EPHA4 was faintly expressed in PANC-1 cells ( Figure 3E and F).

EPHA4 silencing abrogates neurotropism in a PDAC cell line.
To examine the role of EPHA4 in the in vitro co-culture model, we silenced EPHA4 gene expression in the MIA Paca2 cell line using siRNAs ( Figure 4A and B). EPHA4 knockdown abolished the migration and invasion abilities of MIA Paca2 cells that were enhanced in the presence of DRG ( Figure 4C and D). Furthermore, in the horizontal co-culture model, the migration of MIA Paca2 cells toward the DRG was reduced by knockdown of EPHA4 ( Figure 4E and F).

Discussion
Various mechanisms have been proposed for the development of PNI in cancer cells. Recent studies have proposed the concept of neuron-cancer cell crosstalk; nerve cells have been shown to infiltrate the tumor microenvironment and actively stimulate cancer cell growth by releasing neurotransmitters. In turn, the secretion of neurotrophic growth factors by cancer cells drives the outgrowth of neurons in solid tumors (10). This reciprocal interaction between neurons and cancer cells provides new insights into identifying the novel molecular mechanisms involved, and finally leads to the development of anti-neurogenic therapies. In this regard, the in vitro cancer cell-neuron co-culture models utilized in the present study can provide a proper simulation of the cancer-neuron crosstalk, which evaluates the neurotropism of cancer cells by placing each cell type separately and observing the abilities of cancer cells to migrate and invade in a timedependent manner. This co-culture model has been widely accepted to mimic the conditions of PNI in vivo as described in various studies (14,19,20).
The present study showed that the migratory and invading abilities of MIA Paca2 cells were enhanced in the presence of DRG cells in a vertical co-culture model, as shown in Figure 1. These results indicated that the aggressive nature of MIA Paca2 cells was enhanced by certain soluble factors released by the cancer cell-neuron co-culture. Next, using the horizontal co-culture model shown in Figure 2, we found a significant migration of MIA Paca2 cells toward neurons. This implies that nerve cells have the potential to determine the polarity of cancer cells by releasing soluble factors in the tumor-neuron microenvironment. Interestingly, this neurotropism of cancer cells was only seen in MIA Paca2, but not in PANC-1 cells.
We then performed a comprehensive transcriptome analysis using MIA Paca2 and PANC-1 cells with/without DRG-CM treatment. This analysis was conducted as follows: 1. to clarify the basic differences in the molecular profiles of MIA Paca2 and PANC-1 cells, especially the potential to receive the signals from the soluble factors released by neurons; and 2. to investigate the molecular changes of MIA Paca2 cells induced by neurons.
PCA mapping using a total of 27,340 genes showed that each cell line had a distinct gene expression profile ( Figure  3A), which reflected the differences in their origins. In the KEGG pathway analysis of genes that were significantly differentially expressed between MIA-M and PAN-M cells, axon guidance genes were the most enriched. Axon guidance signaling pathways play important roles in normal neuronal migration and positioning during embryonic development (21). Recent studies have suggested that they have also been implicated in cancer cell growth, survival, invasion, and angiogenesis (22). Furthermore, Jurcak et al. reported that the genes involved in axon guidance can be involved in the regulation of PNI in PDAC (23). Their previous reports are in line with our result that genes in the axon guidance pathway are responsible for the neurotropism of cancer cells that lead to PNI. Among genes in the axon guidance pathway, we focused on EPHA4 as a target gene related to PNI, which was a significantly upregulated receptor gene in MIA Paca2 cells treated with DRG-CM compared with the control treatment ( Figure 3C and D).
EPHA4 is a member of the erythropoietin-producing hepatocellular (Eph) family of receptor tyrosine kinases. Eph receptors are classified into two subfamilies, type A and B, according to their binding affinities for Ephrin ligands that are categorized into two subclasses, glycosylphosphatidylinositol anchor (A type) and transmembrane domain (B type) (24). These Eph receptors and Ephrin ligands have been implicated in a variety of biological functions, including axon guidance and migration of neural crest cells in the nervous system and establishment of segmental boundaries (25). EphA receptors mainly bind Ephrin A ligands and induce bi-directional cellto-cell contact signaling pathways, such as adhesion, migration, and invasion by modifying the organization of the actin cytoskeleton and influencing the activities of integrins and intercellular adhesion molecules (26). In cancer cells, Eph-Ephrin signals can function as both tumor promoters and suppressors (26). In certain cellular contexts, Eph receptors activated by ephrins have lost the ability to suppress tumorigenicity, and have acquired oncogenic ability (26). In addition, in terms of ligands of the Eph receptor, there exists a cleaved, soluble form of Ephrin molecule that can provide longer-range functions (27). Wycosky et al. have demonstrated that Ephrins function as soluble, monomeric factors during cancer maintenance and/or progression as well as in normal developmental/physiological processes (28). In this study, EPHA4 knockdown by siRNA in MIA Paca2 cells prevented their migration, invasion, and neurotropic abilities, which were enhanced in the presence of neurons (Figure 4). These results and those of previous studies supported our speculation that the presence of certain soluble forms of ligands, such as Ephrin A, in the tumor-nerve microenvironment, enhanced EPHA4 expression in MIA Paca2 cells, resulting in the migration of cancer cells toward neurons, which finally leads to the establishment of PNI.
This study has several limitations. First, we did not investigate the upstream factors that enhanced EPHA4 expression. As upstream molecules, soluble forms of Ephrin or some other proteins bound to EPHA are relevant candidates. Comprehensive analysis of the conditioned medium extracted from a cancer cell-neuron co-culture model is warranted. Second, the clinical impact of EPHA4 expression in PDAC is lacking in this study. Previous studies have suggested that EPHA4 overexpression was related to poor prognosis and metastasis in PDAC cells (29,30). However, no reports have shown the clinicopathological relationships between PNI and EPHA4 expression. Immunohistochemical analysis or in situ hybridization studies using clinical samples are needed to confirm the clinical importance of our study.
In conclusion, we investigated the mechanisms of PNI using a cancer cells-nerve cells co-culture model. EPHA4 can be a target molecule that regulates PNI in PDAC.

Conflicts of Interest
All Authors have no conflicts of interest to disclose in relation to this study.