Pancreatic cancer tissues were obtained from patients who underwent pancreatic resection at the Kyushu University Hospital between January 2010 and February 2012. The mean age of the patients was 66 years (range 36-85 years), and 43% were females. The study was approved by the Ethics Committee of Kyushu University (reference no. 28-189) and conducted in accordance with the Ethical Guidelines for Human Genome/Gene Research enacted by the Japanese Government and the Helsinki Declaration. Written informed consent was obtained from all patients who agreed to the use of their samples in the present research.

Primary pancreatic cancer tissues from patients were evaluated by hematoxylin and eosin (H&E) staining. Expression levels of cytokeratin-19 (CK-19), E-cadherin, vimentin, and the lymphatic endothelial cell marker D2-40 were examined by immunohistochemistry. The paraffin-embedded patient tissues were sliced into 4-μm-thick section, deparaffinized in xylene, and rehydrated through a graded series of ethanol concentrations. Endogenous peroxidase activity was blocked by incubation with 3% hydrogen peroxide in methanol for 30 min. Antigen retrieval was performed by boiling in a microwave oven (citrate buffer, pH 6.0). Sections were incubated with anti-CK19 antibody (1:500; cat. no. sc376126), anti-E-cadherin (1:500; cat. no. sc8426) (both from Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA), anti-vimentin (1:500; cat. no. ab92547; Abcam, Cambridge, MA, USA), anti-S100P (1:500; cat. no. sc374547; Santa Cruz Biotechnology, Inc.), and anti-D2-40 (cat. no. 413451; Nichirei Biosciences Inc., Tokyo, Japan) overnight at 4°C. Sections were stained with EnVision System-HRP Labeled Polymer Anti-Mouse (cat. no. K4001; Dako, Carpinteria, CA, USA) at room temperature for 40 min. The labeled antigens were visualized using 3,3-diaminobezidine tetrahydrochloride as the chromogen at room temperature for 60 sec. Staining was performed on serial sections. Images were acquired using a confocal laser-scanning microscope (BZ-X700; Keyence Corporation, Osaka, Japan).

Pancreatic cancer cell lines AsPC-1, BxPC-3, Capan-1, CFPAC-1, Hs766T and SW1990 (American Type Culture Collection, Manassas, VA, USA), Panc1 (RIKEN BioResource Center, Tsukuba, Japan), MIAPaCa-2 and SUIT-2 (Japanese Collection of Research Bioresources Cell Bank) were maintained in DMEM (cat. no. D5523-10L; Sigma-Aldrich; Merck KGaG, Darmstadt, Germany) supplemented with 10% FBS (cat. no. 10270; Thermo Fisher Scientific, Inc., Waltham, MA, USA), streptomycin (100 μg/ml), and penicillin (100 U/ml) at 37°C with humidified 90% air and 10% CO₂. LECs (Japanese Collection of Research Bioresources Cell Bank) were cultured in EGM2MV medium (Lonza Group, Ltd., Basel, Switzerland) and were immortalized by previously reported methods (20). In all experiments, immortalized LECs were used. Human umbilical vein endothelial cells (HUVECs) were obtained from Lonza Group, Ltd. (cat. no. C2517A) and maintained in EBM-2 medium (Lonza Group, Ltd.).

LECs were seeded in EGM2MV medium on 24-well plates and allowed to grow to confluence. Spheroids were prepared using cancer cells stained with CellTracker Green CMFDA (Thermo Fisher Scientific, Inc.). Twelve spheroids were transferred into each well. During the incubation period, frames were captured at 15-min intervals with a confocal laser-scanning microscope (BZ-X700; Keyence Corporation) and used to create a time-lapse video. The CCID area was measured at 4 h following spheroid transfer using ImageJ (v.1.48u; National Institutes of Health, Bethesda, MD, USA).

KPCL (LSL-KrasG12D/+; LSL-Trp53R172H/+; Pdx-1-Cre; ROSA26LSL-Luc/+) transgenic mice were constructed as previously described (21). At the onset of distress and/or abdominal distension, animals were sacrificed, and primary tumors and metastatic tumors were excised. Pancreatic cancer cells from KPCL mice were established in our laboratory from excised primary tumors and metastatic tumors using the outgrowth method (22). Cancer cell lines derived from KPCL mice were named as follows: i) number of the serial mouse that the cell line was derived from; and ii) origin of the cancer cells (AC, ascites-derived; PC, primary tumor-derived; LyC, lymph node metastasis-derived; LC, liver metastasis-derived). Experimental protocols involving animals were approved by the Animal Experiment Committee, Graduate School of Medical Sciences, Kyushu University (permit nos. A26-131-0 and A30-309-0).

Cancer cells were cultured in DMEM (Sigma-Aldrich; Merck KGaA) with 10% FBS (Thermo Fisher Scientific, Inc.), rinsed in PBS, trypsinized, and then prepared for spheroid formation and image acquisition. The cells were distributed into ultra-low-attachment-round bottomed 96-well plates (Corning, Inc., Corning, NY, USA) at a density of 1,000 cells/well. The plates were centrifuged at 190 × g for 6 min. Following centrifugation, cells were cultured as usual. Images were captured at 1 and 24 h later, and the area of maximum horizontal section was measured using ImageJ. Spheroid cohesion was compared at 24/1 h.

LECs (4×10⁴/well) were cultured in monolayers in 96-well collagen I-coated plates overnight. Collagen I was used as the principal extracellular matrix molecule. Pancreatic cancer cells were labeled with CellTracker Green CMFDA (Thermo Fisher Scientific, Inc.). Pancreatic cancer cells (4×10⁴/well) were added to 96-well collagen I-coated plates containing confluent LECs, and cells were incubated for 3 h at 37°C. The plates were then washed three times with 200 μl of PBS to remove the non-adherent tumor cells. The number of adhered pancreatic cancer cells was determined in five random fields at ×200 magnification using a confocal laser-scanning microscope (BZ-X700).

Total RNA was isolated from cultured cells using a High Pure RNA Isolation kit with DNase digestion (Roche Diagnostics GmbH, Mannheim, Germany). RNA quality was evaluated using the Agilent 2200 TapeStation system (Agilent Technologies, Inc., Santa Clara, CA, USA) for microarray analysis. RNA was labeled and hybridized to the Agilent SurePrint G3 Human Gene Expression Microarray 8 × 60K Ver.3.0 (Agilent Technologies, Inc.). Data analysis was performed using Feature Extraction software (Agilent Technologies, Inc.).

The paraffin-embedded patient tissues were sliced to a thickness of 4 μm. These were the same tissues that were used for immunohistochemistry. Endogenous peroxidase activity was blocked with methanol containing 0.3% hydrogen peroxidase. Subsequently, 3% hydrogen peroxide was applied as blocking reagent at room temperature for 30 min. Antigen retrieval was performed by boiling in a microwave oven (citrate buffer, pH 6.0). Sections were incubated with rabbit anti-LYVE-1 (1:1,000; cat. no. ab14917; Sigma-Aldrich; Merck KGaA) and mouse anti-S100P (1:500; cat. no. sc374547; Santa Cruz Biotechnology, Inc.) antibodies overnight at 4°C. Sections were then incubated for 1 h with Alexa 546-conjugated anti-mouse immunoglobulin G (IgG) and Alexa 488-conjugated anti-rabbit IgG (both from Thermo Fisher Scientific, Inc.). Nuclear DNA was counterstained with DAPI (0.05 mg/ml). Staining was performed on serial sections. Images were acquired using a fluorescence microscope (BZ-X700; Keyence Corporation).

Total RNA was extracted from cultured cells using a High Pure RNA Isolation kit (cat. no. 11828665001; Roche Diagnostics GmbH). The extracted RNA was quantified by the absorbance at 260 nm, and its purity was evaluated by the 260/280 ratio of absorbance with an ND-1000 spectrophotometer (NanoDrop Technologies; Thermo Fisher Scientific, Inc.). RT-qPCR was performed using the iTaq Universal SYBR-Green One-Step kit and CFX96 Touch Real-Time PCR Detection systems (Bio-Rad Laboratories, Inc., Hercules, CA, USA). Reactions were incubated at 50°C for 10 min; 95°C for 1 min; 95°C for 10 sec; 60°C for 30 sec for 39 cycles; 60°C for 5 sec; 95°C for 5 sec. The results were analyzed using the 2^-ΔΔCq method (23). Primers were purchased from Takara Bio, Inc. The human ACTB gene and GAPDH gene were used as endogenous controls. The following primers were used: S100P, forward, 5′-GCACCATGACGGAACTAGAGACA-3′ and reverse, 5′-CAGGTCCTTGAGCAATTTATCCAC-3′; ACTB, forward, 5′-TGGCACCCAGCACAATGAA-3′ and reverse, 5′-CTAAGTCATAGTCCGCCTAGAAGCA-3′; GAPDH, forward, 5′-GCACCGTCAAGGCTGAGAAC-3′ and reverse, 5′-TGGTGAAGACGCCAGTGGA-3′.

Whole-cell lysates were prepared with PRO-PREP solution (Intron Biotechnology, Inc.) from subconfluent cells. For SDS-PAGE, 20 μg protein was separated by gel electrophoresis on 4-15% Mini-PROTEAN TGX Precast gels (cat. no. 456-1084; Bio-Rad Laboratories, Inc.) and transferred to Trans-Blot Turbo Mini PVDF Transfer Packs (cat. no. 170-4156; Bio-Rad Laboratories, Inc.) using a Trans-Blot Turbo Transfer Starter System (Bio-Rad Laboratories, Inc.). The membranes were incubated with 5% milk as blocking reagent at room temperature for 1 h. The membranes were incubated overnight at 4°C with anti-S100P (1:200; cat. no. sc374547; Santa Cruz Biotechnology, Inc.) and anti-β-actin (1:5000; cat. no. ab8227; Abcam) antibodies, and then probed with horseradish-peroxidase-conjugated-secondary antibodies (1:2,000; cat. no. 7076; Cell Signaling Technology, Inc., Danvers, MA, USA) for 1 h. Immunoblot detection was performed using chemiluminescence with a ChemiDoc XRS System (Bio-Rad Laboratories, Inc.). Quantity One software (version 4.6.6; Bio-Rad Laboratories, Inc.) was used for densitometry.

Cell migration was assessed after incubation for 24 h using uncoated Transwell chambers. To assess the effect of S100P on the migration of LECs, recombinant human S100P protein (Abcam), the receptor for advanced glycation end-products (RAGE) antagonist peptide (Tocris Bioscience, Bristol, UK), or recombinant human IL-6 protein (PeproTech, Inc., Rocky Hill, NJ, USA) was added to the lower chambers in 750 μl of EGM2MV medium. LECs (2×10⁴/well) in 250 μl of EGM2MV medium were seeded in each upper well and incubated for 24 h. Migrated cells at the bottom of the chamber were fixed with 70% ethanol and stained with H&E, and five random fields at ×200 magnification were counted under a confocal microscope (BZ-X700).

LECs at 90% confluence were transfected with siRNA (Sigma-Aldrich; Merck KGaA) by electroporation using a Nucleofector System (Lonza Group, Ltd.) according to the manufacturer's recommendations. The following siRNAs directed against human S100P were used in the study: siRNA-1, sense, 5′-AGGCUUCCUGCAGAGUGGA-3′ and antisense, 5′-UCCACUCUGCAGGAAGCCU-3′; siRNA-2, sense, 5′-GGAUGCCGUGGAUAAAUUG-3′ and antisense, 5′-CAAUUUAUCCACGGCAUCC-3′; and siRNA-3, sense, 5′-CAAGGAUGCCGUGGAUAAA-3′ and antisense, 5′-UUUAUCCACGGCAUCCUUG-3′. To confirm knockdown specificity, a negative control siRNA (cat. no. SIC001; Sigma-Aldrich; Merck KGaA) was used. Transfected cells were used in subsequent experiments at 24-72 h post-transfection. S100P expression was assessed 72 h post-transfection by RT-qPCR.

Results are presented as mean ± SD. Comparisons between groups were performed using one-way ANOVA followed by the Tukey-Kramer multiple comparisons test. Survival analyses were conducted using the Kaplan-Meier method, and the curves were compared using the log-rank test. All statistical analyses were performed using JMP 12 software (SAS Institute, Inc., Cary, NC, USA). P<0.05 was considered to indicate a statistically significant difference.

To investigate whether collective invasion occurs in human pancreatic cancer tissue, human pancreatic cancer tissue samples were stained with H&E, anti-D2-40 (a marker for lymphatic endothelial cells) and anti-CK19 (a marker for cancer cells). In addition, the tissues were stained for the epithelial marker E-cadherin and the mesenchymal marker Vimentin. Immunohistochemical staining of human pancreatic cancer tissues revealed that clusters of tumor cells were present in the lymphatic ducts around the primary tumor (Fig. 1A), and in the subcapsular sinuses (Fig. 1B) and the lymphatic cortex of the lymph nodes (Fig. 1C). Thus, embolic tumor cell clusters were observed in sites that are known to be routes of lymphatic metastasis. These clusters of tumor cells expressed epithelial markers but not mesenchymal markers (Fig. 1A-C). Clusters of cancer cells that penetrated the wall of lymphatic ducts around the primary tumor were frequently observed (Fig. 1D). These findings suggested that cancer cells invaded in the form of clusters, and without having undergone EMT, to form lymph node metastases.

Spheroids of pancreatic cancer cells were placed on top of lymphatic endothelial monolayers. LECs beneath these spheroids migrated and formed large cell-free areas in the lymphatic endothelial monolayer, called CCIDs (Fig. 2A). These in vitro CCIDs were similar to the defects observed in the lymphovascular walls at the sites of tumor cell invasion in human tissue. Observation of CCID formation using a confocal microscope revealed that defects were formed in the LEC layer immediately underneath spheroids (Fig. 2B). When spheroids of human fibroblasts were used, CCIDs were smaller compared with those formed with cancer cells (Fig. 2A and C). CCID formation in LECs was more pronounced compared with HUVECs (Fig. 2D).

First, the spheroid cohesion ability of pancreatic cancer cells and their ability to adhere to LECs were considered as candidate factors related to CCID formation. CCID area (Fig. 3A), spheroid cohesion (Fig. 3B), and adhesion to LECs (Fig. 3C) were investigated using each cell line (human pancreatic cancer cells and KPCL mouse-derived pancreatic cancer cells). The CCID area differed for each cell line (Fig. 3A). Similarly, spheroid cohesion (Fig. 3B) and adhesion to LECs (Fig. 3C) differed for each cell line. The CCID area was not significantly different between human and KPCL mouse-derived cell lines. Spheroid cohesion was significantly higher in KPCL mouse-derived cell lines compared with human cell lines (P<0.01). Adhesion to LECs was significantly higher in human cell lines compared with KPCL mouse-derived cell lines (P=0.02). However, they did not exhibit common features. The correlation between each result and CCID was analyzed by Pearson product-moment correlation coefficient. Neither spheroid cohesion (P=0.80) nor adhesion to LECs (P=0.49) displayed a significant correlation with CCID formation (Fig. 3D and E).

To evaluate whether secreted cancer cell-derived factors were involved in CCID formation of pancreatic cancer spheroids, the effects of cancer cell supernatant on CCID formation were investigated. When the cancer cells were subconfluent, the medium was changed to serum-free DMEM and the supernatant was collected after 48 h. Cancer cell culture supernatant (Fig. 4A left panel, from SUIT2 cells; right panel, from MIAPaCa2, SW1990 and SUIT2 cells) was added in LECs for 30 min and then removed. Serum-free DMEM was used in the control group. Next, SUIT2 spheroids were placed on top of the lymphatic endothelial monolayers. The results reveled that CCID area was increased following treatment with cancer cell supernatant (Fig. 4A). Similar results were obtained using the culture supernatants of several lines of pancreatic cancer cells (Fig. 4A). To investigate how gene expression was altered in LECs following treatment with cancer cell supernatant, microarray analysis was performed using LECs with treatment of culture supernatants from three pancreatic cancer cell lines. A heatmap analysis of gene expression profiles revealed common variation in LEC samples treated with culture supernatants from pancreatic cancer cells (Fig. 4B). A total of 133 genes were identified as differentially expressed in LECs following treatment with culture supernatants from pancreatic cancer cells. Out of these 133 genes, S100P was selected for further study. S100P has been reported to be highly expressed in pancreatic cancer (15) and is associated with the cell cytoskeleton (24), migration (24,25), and adhesion (25). In addition, The Cancer Genome Atlas database was explored in regards to S100P expression, and the results revealed that S100P expression was a significant prognostic factor for pancreatic cancer (Fig. 4C), similar to previous reports (16,19).

To confirm whether lymphatic endothelial cells express S100P in human tissues, serial sections from patient pancreatic tumors were used for immunohistochemistry and immunofluorescence staining. Although S100P was highly expressed in tumor cells, it was also expressed in LECs at a site where a tumor cluster was present inside the lymphatic vessel surrounding the primary tumor (Fig. 5A and B). Consistent with the results from the microarray analysis, S100P mRNA expression increased in LECs following treatment with culture supernatant from cancer cells for 48 h (Fig. 5C). S100P expression was also increased at the protein level following treatment with the culture supernatant (Fig. 5D).

A previous report revealed that S100P mRNA expression is increased in prostate cancer cells following treatment with IL-6 (26). It has also been reported that pancreatic cancer cells express higher levels of IL-6 compared with normal human pancreatic ductal epithelium cells (27). Therefore, the changes in S100P expression were investigated in LECs following treatment with IL-6. S100P expression in LECs increased following IL-6 treatment (Fig. 6A). It has been reported that S100P is associated with cell migration (24,25). IL-6 treatment increased migration in LECs (Fig. 6B) and CCID formation (Fig. 6D). To confirm that IL-6-induced LEC migration occurs via S100P, gene knockdown studies were performed. S100P expression was silenced using siRNA (Fig. 6C, left panel). The effect of S100P knockdown on IL-6-induced LEC migration was then investigated. IL-6-induced LEC migration was significantly reduced following S100P silencing in LECs (Fig. 6C, right panel). Extracellular S100P has been demonstrated to activate RAGE (15,28), leading to the increased proliferation, invasion and migration of cancer cells (28,29). Therefore, the migration ability of LECs was evaluated following treatment with culture supernatant from cancer cells, or recombinant human S100P protein with or without RAGE antagonist peptide. Migration was increased by treatment with culture supernatant from cancer cells or recombinant human S100P protein, but this increase was suppressed following treatment with RAGE antagonist peptide (Fig. 6E). CCID area was also increased following treatment with recombinant human S100P protein, which was suppressed by the addition of RAGE antagonist peptide (Fig. 6F).

Therefore, pancreatic cancer cells were cultured in spheroid culture conditions on low-attachment plates or in normal culture conditions on normal plates, and then their S100P mRNA expression levels were compared. Expression of S100P was higher in the cancer cells cultured in low-attachment plates (spheroid culture) compared with cells cultured in normal plates (single cell culture; Fig. 6G). This finding suggested that the expression and secretion of S100P were induced by the spheroid structure, leading to enhanced CCID formation by pancreatic cancer cell spheroids in LECs.