FHL3 promotes pancreatic cancer invasion and metastasis through preventing the ubiquitination degradation of EMT associated transcription factors

FHL3 expression correlated with PDAC progression

In our study, we found that there was no significant difference in FHL3 expression in different groups stratified by age (p=0.304, Table 1), gender (p=0.912, Table 1), differentiation grade (p=0.342, Table 1) or M stage (p=0.826, Table 1). However, the quantification of immunohistochemistry (IHC) with Image-Pro^R Plus for 55 paraffin-embedded sections of PDAC showed that a higher expression level of FHL3 correlated with a higher clinical stage in PDAC (Figure 1A). And as Table.1 showed, higher expression level of FHL3 was accompanied with higher T stage (p=0.0165; 7 (26%) T1, 18 (67%) T2, and 2 (7%) T3 tissue sections in the low-FHL3 group; 2 (7%) T1, 16 (57%) T2 and 10 (36%) T3 tissue sections in the high-FHL3 group; Table 1) and N stage (p=0.0437; 22 (81%) N0 and 5 (19%) N1 tissue sections in the low-FHL3 group; 15 (54%) N0 and 13 (46%) N1 tissue sections in the high-FHL3 group; Table 1). In addition, FHL3 was overexpressed in PDAC tissue compared with its expression in adjacent non-tumor tissue in 49 paired paraffin-embedded sections of PDAC samples (p<0.001, Figure 1A and 1B), as same as the outcome of WB in eight matched fresh frozen PDAC samples (p<0.01, Figure 1D). Furthermore, the results of a Kaplan-Meier analysis of 55 PDAC samples grouped by the FHL3 expression measured by IHC indicated that higher expression of FHL3 implied worse prognosis in PDAC (p=0.0169, Figure 1C). In addition, multivariate analysis implied age (p<0.001), tumor site (p<0.025), T stage (p<0.015) and diabetes (p<0.020) made effects on prognosis (Table 2).

Figure 1. FHL3 referred to PDAC progression. (A and B) IHC staining and statistical analysis of FHL3 for sections from 55 matched different lesion stages tissues of PDAC and para-tumor normal area, which showed the higher expression level of FHL3 in PDAC, more than 3-fold, as compared with normal tissue, p<0.001. (C) Survival analysis via Kaplan-Meier analysis in 55 PDAC samples which showed higher expression level of FHL3 was accompanied with worse prognosis, p=0.0168. (D) WB assay of 8 matched fresh frozen PDAC samples, which showed higher expression level of FHL3 in PDAC tissues more than 1.5-fold, p<0.01.

FHL3 expression was upregulated in pancreatic cancer cell lines and was associated with the metastasis ability

Next, our study showed that, comparing with HPDE (normal pancreatic ductal epithelial cells), FHL3 expression was upregulated in four PC cell lines (PANC1, BXPC3, MIAPACA2 and CFPAC1) (Figure 2A–2C), and the highest upregulation of FHL3 was up to 7.07 folds (Figure 2C). As the same time, we found that higher expression levels of FHL3 in four PC cell lines and one normal pancreatic ductal epithelial cell line (HPDEFigure 2D–2F) through linear relationship fitting (r=0.8108, p=0.0371; Figure 2F). In addition, through analyzing the data in TCGA, our study found that expression of FHL3 was negatively correlated to expression level of EMT maker E-cadherin (r=-0.220, p=0.003, Figure 2G₁), and passively correlated to EMT maker vimentin (r=0.627, p<0.001, Figure 2G₂) and EMT associated transcription factors snail1 (r=0.452, p<0.001, Figure 2G₃)and twist1 (r=0.484, p<0.001, Figure 2G₄).

Figure 2. The expression level of FHL3 referred to EMT makers and metastasis in pancreatic cancer cells. (A) IF assay showed the expression level of FHL3 in four pancreatic cell lines and one normal pancreas cell line, and the expression rank was: PANC1>BXPC3>CFPAC1>MIA PACA2>HPDE. (B and C) WB assay showed the same results, and PANC1 and BXPC3 held the highest expression level of FHL3, which were more than 4-fold higher when compared to HPDE, p<0.001. (D and E) transwell assay for evaluating the migration ability in four pancreatic cell lines and one normal pancreas cell line, in which the migration rank was: PANC1>BXPC3>MIA PACA2>CFPAC1> HPDE. (F) linear relationship fitting analysis showed migration ability was correlate with the expression level of FHL3, r=0.8108, p=0.0371. TCGA data analysis showed FHL3 was negatively correlated to expression level of EMT maker E-cadherin (G1) (r=-0.220, p=0.003), and passively correlated to EMT maker vimentin (G2)(r=0.627, p<0.001) and EMT associated transcription factors snail1 (G3) (r=0.452, p<0.001, Fig.2G₃) and twist1 (G4) (r=0.484, p<0.001).

FHL3 knockdown reversed the EMT phenotype and inhibited migration in pancreatic cell lines

For finding out the internal connection between FHL3 and PDAC invasion, our study found the FHL3-knockdown (FHL3-KD) cell lines in PANC1 and BXPC3 (PANC1_KD1 and BXPC3_KD1). As Figure 3A₁–2B₂ showed, whatever the IF or WB, all of the three sequences for FHL3 knockdown were efficient, and the KD-1 cell lines (PANC1 and BXPC3) were the most efficient in FHL3 knockdown, for PANC1 was 92% (p<0.001, Figure 3A_1-2)and for BXPC3 was 87% (p<0.001, Figure 3B₁-₂). So, in the following study, we chose the KD-1 cell lines as our objects. As Figure 3C showed, FHL3 knockdown inhibited the migration of pancreatic cells, in which the migration inhibition rate was 55% in PANC1_KD1 (p<0.01, Figure 3C) and 38% in BXPC3_KD1 (p<0.01, Figure 3C). In another one experiment, named wound healing assay, our study found the lower expression of FHL3 was accompanied with the larger blank area, which mean the weaker ability of metastasis, in PANC1 (P<0.05, Figure 3D) and BXPC3 (p<0.01, Figure 3D) cell lines. We confirmed that the efficient knockdown of FHL3 expression in PANC1_KD1 and BXPC3_KD1 cells resulted in significant upregulation of ZO-1 (PANC1_KD1: about 3 fold, p<0.001; BXPC3_KD1: about 3.5 fold, p<0.001; Figure 3E_1-2) and E-cadherin (PANC1_KD1: about 3.5 fold, p<0.001; BXPC3_KD1: about 2.5 fold, p<0.001; Figure 3E_1-2), and obvious downregulation of MMP2 (PANC1_KD1: about 40%, p<0.05; BXPC3_KD1: about 50% fold, p<0.05; Figure 3E_1-2), snail1 (PANC1_KD1: about 70%, p<0.001; BXPC3_KD1: about 20% fold, p<0.05; Figure 3E_1-2), twist1 (PANC1_KD1: about 60%, p<0.001; BXPC3_KD1: about 40% fold, p<0.05; Figure 3E_1-2). However, the knockdown of FHL3 just made downregulation of vimentin in PANC1_KD1 (about 40%, p<0.05, Figure 3E_1-2), but not in BXPC3_KD1. Meantime, the expression of zeb1 was unchanged after FHL3 knockdown.

Figure 3. FHL3 knockdown restrained migration and reversed EMT phenotype in pancreatic cancer cells. (A1 and B1) IF assay of FHL3 in PANC1_NC/KD_1-3, BXPC3_NC/KD_1-3, which showed KD1 sequence was the most efficient. (A2 and B2) WB assay of FHL3 in PANC1_NC/KD_1-3 and BXPC3_NC/KD_1-3, which showed the expression level of FHL3 was downregulated more than 80% in PANC1_KD1 and BXPC3_KD1 cells, p<0.001. (C) 48h- transwell assay showed more than 40% decreasing of migration ability of PANC1_KD1 and BXPC3_KD1 cells, as compared with PANC1_NC and BXPC3_NC cells, P<0.01. (D) 72h-wound healing assay showed KD1 cell lines held more blank area as compared with NC cell lines, p<0.05. (E1) WB assay showed FHL3 knockdown reversed the expression level of EMT associated proteins and transcriptional factors, except zeb1.

FHL3 regulated EMT process through TGFβ1/Akt/GSK3β/ubiquitin process but not through TGFβ1/smad_2/3/smad4 pathway

Next, our study found that the FHL3 knockdown made significant downregulation of TGFβ1 (>60% in PANC1_KD1 and BXPC3_KD1, p<0.001, Figure 4A_1-2), smad2/3 (>50% in PANC1_KD1 and BXPC3_KD1, p<0.001, Figure 4A_1-2) and smad4 (about 50% in PANC1_KD1 and BXPC3_KD1, p<0.001, Figure 4A_1-2). However, our study found there was almost no influence in the expression of smad_2/3 in nucleus/cytoplasm rate in PANC1 and BXPC3 cell lines, and just not more than 20% decreasing of smad₄, in nucleus/cytoplasm rate, as FHL3 knockdown (Figure 4B_1-2). As previous studies implied, snail1 was regulated by GSK3β-mediated ubiquitin degradation. Therefore, we explored the following research. In the following study, we found FHL3 knockdown changed the TGFβ1/Akt/GSK3β/ubiquitin pathway. As Figure 4C_1-2 showed, FHL3 knockdown not only decreased the absolute level of phosphorylated Akt (ser473-Akt) more than 50%, but also relative level of phosphorylated Akt (ser473-Akt/Akt) more than 50% in PANC1_KD1 and BXPC3_KD1. Meantime, FHL3 knockdown was accompanied with more than 50% downregulation of phosphorylated GSK3β (ser9-GSK3β) and more than 2-fold upregulation of phosphorylated GSK3β (try216-GSK3β), whatever in absolute expression level or relative expression (p-GSK3β/GSK3β) (Figure 4C_1-2). This study implied that FHL3 knockdown decreased the TGFβ1 level, weakened the activity of Akt, result of which further enhanced the activity of GSK3β. In order to verify our study, we use GSK3β inhibitor, 1-Azakenpaullone, for next experiments. On the one hand, as Figure 4D₁ showed, the activity of GSK3β was decreased with the increasing dose of inhibitor both in PANC1_KD1 and BXPC3_KD1. And, the level of snail1 and twist1 were also increased after treatment of GSK3β inhibitor (Figure 4D₁). Finally, EMT maker E-cadherin was significantly downregulated (Figure 4D₁). On the other hand, GSK3β inhibitor absolutely reversed the pancreatic cells migration ability which was blocked by FHL3 knockdown both in PANC1_KD1 and BXPC3_KD1 (Figure 4D₂).

Figure 4. FHL3 regulated EMT mainly by TGFβ1/Akt/GSK3β/ubiquitin pathway. (A1 and A2) WB assay showed FHL3 knockdown made downregulation of TGFβ1, smad_2/3, and smad₄ in PANC1_KD1 and BXPC3_KD1 in total protein level. (B1 and B2) WB assay of nucleus-cytoplasm protein showed that FHL3 knockdown hardly changed the expression level of smad_2/3 and smad₄ in nucleus in PANC1_KD1 and BXPC3_KD1 cells. (C1 and C2) In PANC1_KD1 and BXPC3_KD1 cells, FHL3 knockdown exactly downregulated the absolute and relative expression of phosphorylated AKT (ser473-AKT) more than 50%, p<0.01; and also downregulated the absolute and relative expression of phosphorylated GSK3β (ser9-GSK3β) more than 50%, p<0.01; and upregulated the absolute and relative expression of phosphorylated GSK3β (try216-GSK3β) more than 2-fold, p<0.001. (D1) 0.25μM and 0.50μM GSK3β inhibitor almost eliminated the effect, promoting the TGFβ1/AKT/GSK3β/ubiquitin process, caused by FHL3 knockdown. As treated with GSK3β inhibitor, GSK3β (try216-GSK3β) and E-cadherin were downregulated, snail1 and twist1 were upregulated. (D2) 0.25μM and 0.50μM GSK3β inhibitor reversed the migration ability of PANC1_KD1 and BXPC3_KD1 cells.

FHL3 competitively binded to GSK3β by LIM-3 domain to inhibit ubiquitin process for maintaining the level of EMT associated transcriptional factors

In order to thoroughly determine the roles of FHL3 in the ubiquitin mediated degradation process of EMT associated TFs, we explored the physical interactions between FHL3 and proteins in the ubiquitin process. Coimmunoprecipitation (CO-IP) and mass analysis showed that FHL3 could interact with GSK3β (data not provided) and E3 ligase (RNF146) (data not provided). Based on these results, we performed experiments in HEK293T cells. As our study showed, the expression level of snail1 and twist1 were upregulated about 2-fold with an increased transfection dose of the FHL3-HA plasmid (0μg, 5μg and 15μg, Figure 5A₁). Furthermore, as the transfection dose of the FHL3-HA plasmid increased, more FHL3 bound to GSK3β with less snail1/twist1 bound to GSK3β (decreased more than 50%, Figure 5A₂); few FHL3 molecules bound to the negative control (Figure 4A₂). These results implied that FHL3 was directly involved in the ubiquitin mediated degradation of snail1 and twist1 through competitively binding to GSK3β to weaken the interaction between GSK3β and snail1/twist1. Furthermore, we explored the GSK3β binding ability of the pivotal domain in FHL3. We designed six truncated forms, as shown in Figure 4B, and transfected plasmids into HEK293T cells. Our study showed that only TF-3, TF-4 and TF-5, all of which contained the LIM-3 domain, could bind to GSK3β. In addition, due to LIM-3 domain deletion, TF-1, TF-2 and TF-6 could not bind to GSK3β (Figure 5C). These results implied the LIM-3 domain was required for FHL3 binding to GSK3β.

Figure 5. LIM-3 was the pivotal domain for FHL3 competitively binding to GSK3β. (A1 and A2) Transfection with GSK3β, FHL3 (+:5ug, +++:15ug), snail1 and twist1 in HEK293T for 48h followed by CO-IP assay, which showed the higher FHL3 expression was accompanied with the higher expression of snail1 and twist1, p<0.01; meantime, the higher FHL3 expression made less snail1 and twist1 which binding with GSK3β, p<0.001. (B) Truncated forms form FHL3. (C) Transfection with GSK3β and truncated forms for 48h in HEK293T cells, and the CO-IP assay showed only truncated forms which containing LIM-3 domain could bind with GSK3β.

FHL3 knockdown curbed pancreatic cancer cells growth and metastasis in vivo

Then, we validated the role of FHL3 in growth and migration of pancreatic cancer cells in vivo. As Figure 6A showed, 4 weeks after the orthotopic transplantation or tail intravenous injection of pancreatic cancer cells, the liver and lungs were harvested for tumor detection. During our study in vivo, there was a SCID mice died of unknown cause. In our result, we found orthotopic transplantation tumor of PANC1_KD1 cell was smaller more than 50% as compared with PANC1_NC cell (p<0.01, Figure 6B). And HIC staining of Ki67 for those tumor slices showed that PANC1_KD1 cell tumor held stronger staining signal than PANC1_NC cell tumor (Figure 6C and Supplementary Figure 1). In addition, we also verified the expression level of FHL3 by HIC, which implied PANC1_KD1 cells exactly significantly lose the FHL3 (Figure 6E and Supplementary Figure 1). Next, in the tumor metastasis model experiments, as Figure 6F_{1-a, b} showed, green area were metastatic tumor from circular pancreatic cancer cells, which were from tail vein injection, and lung metastasis occurred in 1 of the 8 SCID mouse in PANC1_KD1 cells (12.5%), which occurred in 6 of 8 SCID mouse in PANC1_NC cells (75%), that mean FHL3 knockdown decreased the lung metastasis from circular tumor cells more than 50%. And we got the same result in BXPC3_KD1 cells (Figure 6F_{1-a, b}). During the experiments of hepatic metastasis model, our study found occurrent rate of hepatic metastasis was 100% (8/8) in PANC1_NC cells group and 87.5% in BXPC3_NC cells group with which was only 37.5% in PANC1_KD1 cells group and 25% in BXPC3_KD1 cells group (Figure 6F_{2-a, b}). Those data showed that FHL3 maintained the invasion and metastasis ability in pancreatic cancer cell lines.

Figure 6. FHL3 knockdown inhibited pancreatic tumor growth and metastasis in vivo. (A) experiments process in vivo. (B) tumor volume of PANC1_KD1 and PANC1_NC, 4 weeks after orthotopic transplantation; and the tumor volume of PANC1_KD1 group was about 1/3 of PANC1_NC group, p<0.01. (C) FHL3 knockdown made the weaker Ki67 staining in tumor sections of PANC1_KD1 group. (D) H&E staining of tumor section. (E) IHC staining of FHL3 of tumor section. (F1a, F1b) lung metastasis from circular pancreatic tumor cells, and FHL3 knockdown made the lower occurrent rate of lung metastasis in PANC1_KD1 and BXPC3_KD1 groups, p<0.05. (F2a, F2b) hepatic metastasis from orthotopic transplantation tumor, and FHL3 knockdown made the lower occurrent rate of hepatic metastasis in PANC1_KD1 and BXPC3_KD1 groups, p<0.05.

Reagents

SB525334, 1-Azakenpaullone, Bortezomib were purchased from Shelleck (USA), and dissolved in DMSO. Recombinant Human Transforming Growth Factor β1 (TGFβ1) was purchased from Proteintech (China), and saved in sterile deionized water within 50% glycerin. Gemcitabine was purchased from Ely Lilly (Bad Homburg, Germany) and dissolved in sterile 0.9% sodiumchloride. Bovine serum albumin (BSA) was purchased from Sigma-Aldrich.

Pancreatic cancer samples preparation

This study was approved by The First Affiliated Hospital of Anhui Medical University Review Board and the ethics committees of Anhui Medical University. 49 matched paraffin-embedded tissue sections, 6 tumor paraffin-embedded tissue sections and 8 paired fresh frozen tissue were collected from tissue bank from January 2011 to January 2018. All patients with pancreatic ductal adenocarcinoma were confirmed by at least two pathologists.

Cell culture

Pancreatic cancer cell lines (PANC1, MIAPACA2, CFPAC1, and BXPC3) and normal pancreas ductal epithelial cell (HPDE) were gained from the cell bank of Chinese academy of science in October 2017 with STR matching analysis. PANC1 and MIAPACA2 were cultured in DMEM (Gibco, USA), CFPAC1 was cultured in IMDM (Gibco, USA), BXPC3 and HPDE was cultured in RPIM-1640 (Gibco, USA). All types of culture media were supplemented with 10% fetal calf serum and 100 units/mL penicillin and streptomycin, but 20% fetal calf serum for CFPAC1.

Cell proliferation and cytotoxicity assays

The cell proliferation was quantified by standard curve (0.1, 0.2, 0.4, 0.8, 1.0, 1.5, 2.0, 3.0×10⁴ cells were detected optical density (OD) via cell counting kit-8 (Japan) after 24h transplanted into 96-wells plates, and then fit linear standard curve between log [cell quantity] and OD), cell cytotoxicity assays was performed via MTT assay, and the detail protocol described in our previous study (PMID29331423 and PMID29800682).

Quantitative real-time PCR (qRT-PCR)

Trizol RNA solation system (Invitrogen, USA) was used for total RNA extraction. The cDNA templates were synthesized through PrimeScript RT Reagent Kit (TaKaRa, China), and qRT-PCR was performed with a 7500 Fast™ System (Applied Biosystems, USA) using the Sensi Mix SYBR Kit (Bio-Rad, USA). The mRNA level was calculated via using (=2^-ΔΔCt), and normalized to GAPDH. All of the sequences of primer were designed by Primer 5 soft, see in Supplementary Table 1.

Western blot analysis

Total protein extraction

Cells were harvested by cytology brush, and lysed with RIPA lysis buffer (Sigma, USA) supplemented with phosphorylase and protease inhibitor mixture (Thermo, USA), quantified by the BCA assay.

Cytoplasmic and nucleus protein extraction

Cells were harvested by Tyrisin (Invitrogen), then cytoplasmic and nucleus protein was extracted by Cytoplasmic and Nucleus Protein Extraction Kit (Thermal Scientific, USA) according to its protocol, quantified by the BCA assay.

The standard detail experimental process of western blot as same as our previous study (PMID29331423 and PMID29800682). Western blot band was quantified through the Image-J software (NIH, USA). Antibodies against GAPDH, FHL3, E-cadherin, MMP2 and GSK3β were purchased from Proteintech (1:1000, Hangzhou, China), antibodies against smad2/3, smad4, twist1, Ubiquitin, Histone 3, ser473-phospho-Akt, try216/279-phospho-GSK3β, ser9-phospho-GSK3β and ser423/425-phospho-smad2/3 were purchased from Huabio (1:1000, Hangzhou, China), antibodies against ZO-1, Akt, vimentin, snail1 and zeb1 were gained from Abcam (1:1000, China).

Small interfering RNA (siRNA) and recombination plasmid (RP)

Small interfering RNA (siRNA) experiments

5 × 10⁵ pancreatic cancer cells were transplanted into 6 wells plates for 24h, and then cells were transfected with three different sequences FHL3 siRNA (GenePharma, Shanghai, China) for 48h, 72h and 96h with Lipofectamine 3000 reagent (Invitrogen, USA) and Opti-MEM (Life Technologies, USA), according to the manufacturer's instructions for gaining the best transfection efficiency. Three siRNA sequences for FHL3 were listed in Supplementary Table 2.

Recombination plasmid experiments

Primers of FHL3, GSK3β, Snail1, Twist1 and RNF146, inserted into plasmid pcDNA 3.1(-) (Addgene), were designed with Primer 5 soft, see in Supplementary Table 3. Briefly, cDNA templates were synthesized through PrimeScript RT Reagent Kit (TaKaRa, China); CDS of genes were amplified with PrimeSTAR® GXL DNA Polymerase (TaKaRa, China); thirdly, products were purified through SanPrep Column DNA Gel Extraction Kit (Sangon Biotech, China); fourthly, the purified products and plasmid were treated with restriction endonuclease (Xho1, EcoR5 and Xba1 were purchased from NEB, USA) respectively; fifthly, recombination of plasmids were performed through homologous recombination with Hieff Clone^TM Plus One Step Cloning Kit (Yeasen Biotech, China). All primers were listed in Supplementary Table 3. 5 × 10⁵ cells were transplanted into 6 wells plates for 24h, and then cells were transfected with RP for 48h, 72h and 96h with Hieff Trans^TM Liposomal Transfection Reagent (Yeasen Biotech, China) for the best transfection efficiency, according to the manufacturer's instructions.

Immunofluorescence analysis

Briefly, 2.5 × 10⁴ pancreatic cancer cells were seeded in 24-well plates for 24h, and then fixed by 4% paraformaldehyde, permeabilized by 0.5% Triton X-100, and blocked with 5% bovine serum albumin (BSA, Sigma) for 1 h at 37 °C. Samples were incubated with primary antibody (FHL3, 1:200, Proteintech) overnight at 4°C. Subsequently, it washed by PBS, incubated with secondary antibodies for 1h in room temperature before being washed again. Finally, nuclei were stained with 15 μl DAPI (Sigma, USA) before detected by fluorescence microscope (Carl Zeiss, Germany).

Immunohistochemistry staining and scoring standard

Experiments procedure of paraffin embedding, tissue section, hematoxylineosin (HE) staining and immunohistochemistry for FHL3 expression level were performed as previously described (PMID: 23200678 and 20571492). What more, the work concentration of antibody against FHL3 (Proteintech, China) was 1:150, and 1:200 for Ki67 proliferation index (Abcam). The protein expression level was assessed by Mean of Integrated Option Density (IOD) with Image-Pro^R Plus. Briefly, all of the Immunohistochemical sections were photographed for three yields in the same standard, and then select Area of Interesting (AOI) and detect IOD to gain Mean of IOD (IOD/AOI, MI), normalized to positive control (vascular smooth muscle cells). Finally, FHL3 expression level was divided into high and low group according to Mean of MI.

Immunoprecipitation

1×10⁷ Cells were harvested by cytology brush, and lysed with RIPA lysis buffer (Yeasen Biotech, 20118ES60) for protein supernatant, followed by adding immune magnetic beads (Anti-Myc, Anti-HA and Anti-Flag, Bimake) for continuous slight mixing in 4°C for 24h. And then gain immune magnetic beads with Magnetic frame (Bimake), followed by TBS washing. Finally, products were boiled before dissolved in 5x SDS (Yeasen) for 5-10 minutes for western blot assay.

Migration ability assay

Migration ability assays contain transwell and wound healing assay. For transwell, 5 × 10⁴ cells, with special treatments or not, were transplanted into transwell plates (24-well, 8.0μm, Corning Incorporated, Corning, NY, USA) with 10% gradient of fetal calf serum for 48h. And the detection procedure was same as our previous study (PMID29331423). Quantification of passed cell area was performed by Image-Pro^R Plus. For wound healing assay, cells were seeded at least 90% fusion in 6-well plates, and scratched by 200ul pipette tip, then washed with PBS to remove shed cells for extra 96h culture (PMID29331423). Scratch area was quantified with Image-Pro^R Plus.

Experimental protocols in vivo

Tumor growth experiments in vivo

Female athymic nude mice (4 weeks), gained from the SLAC (Shanghai, China), were randomly divided into four groups. 1 × 10⁶ cells (PANC1_NC/KD1), in 100ul PBS, were injected into the tail of pancreas. All mouse was sacrificed and the orthotopic pancreatic tumors were harvested for detecting tumor volume (MaA×MiA² / 2; MaA=Major axis, MiA=Minor axis), and followed by being processed into frozen sections for HE staining, immunofluorescence staining and Ki67 staining.

Tumor metastasis experiments in vivo

Female SCID mice (4 weeks), gained from the SLAC (Shanghai, China), were randomly divided into eight groups. 1 × 10⁶ cells (PANC1_NC/KD1, BXPC3_NC/ KD1), in 100ul PBS, were injected into pancreas tail vein for orthotopic-liver-metastasis tumor model, then all mouse were sacrificed and livers were harvested for HE staining and immunofluorescence staining after 4 weeks. For lung metastasis experiments, 1 × 10⁶ cells (PANC1_NC/KD1, BXPC3_NC/KD1), in 100ul PBS, were injected into tail vein, then all mouse was sacrificed and lungs were harvested for HE staining and immunofluorescence staining after 4 weeks.

Statistics

All experimental data were presented as the means ± SD. Statistical Package for the Social Sciences version 21.0 (SPSS Inc., USA) was used for statistical analyses. ANOVA, paired t-test, Chi-square (x²) test and nonparametric test (Mann Whitney U) for statistical analysis of different situations. Statistical significance was considered when p < 0.05 (*p < 0.05; **p < 0.01; ***p < 0.001). All histograms and curves were constructed with GraphPad Prism 6 software (GraphPad Software, La Jolla, CA, USA).