Smad4-dependent suppressor pituitary homeobox 2 promotes PPP2R2A-mediated inhibition of Akt pathway in pancreatic cancer

The importance of Pituitary homeobox 2 (Pitx2) in malignancy remains enigmatic, and Pitx2 has not been previously implicated in pancreatic ductal adenocarcinoma (PDAC). In this study, we performed gene expression profiling of human PDAC tissues and identified Pitx2 as a promising candidate. Pitx2 expression was decreased from 2.6- to 19-fold in human PDAC tissues from microarray units. Immunochemistry staining showed that Pitx2 expression was moderate to intense in normal pancreatic and pancreatic intraepithelial neoplastic lesions, whereas low in human PDAC tissues. The Pitx2 levels correlated with overall patient survival post-operatively in PDAC. Induction of Pitx2 expression partly inhibited the malignant phenotype of PDAC cells. Interestingly, low Pitx2 expression was correlated with Smad4 mutant inactivation, but not with Pitx2 DNA-methylation. Furthermore, Smad4 protein bound to Pitx2 promoter and stimulated Pitx2 expression in PDAC. In addition, Pitx2 protein bound to the promoter of the protein phosphatase 2A regulatory subunit B55α (PPP2R2A) and upregulated PPP2R2A expression, which may activate dephosphorylation of Akt in PDAC. These findings provide new mechanistic insights into Pitx2 as a tumor suppressor in the downstream of Smad4. And Pitx2 protein promotes PPP2R2A expression which may inhibit Akt pathway. Therefore, we propose that the Smad4-Pitx2-PPP2R2A axis, a new signaling pathway, suppresses the pancreatic carcinogenesis.


INTRODUCTION
Microarray has been used to further understand the molecular pathogenesis of carcinogenesis and explore differentially expressed genes which may be potential therapeutic targets [1,2]. In this paper, microarray analysis was performed to identify potential markers correlated with the known suppressors Smad4, p16, Tp53 and BRCA2, which are most frequently inactivated in pancreatic ductal adenocarcinoma (PDAC). Using gene microarray, we identified pituitary homeobox 2 (Pitx2) as a promising candidate marker.
Previous studies showed that Pitx2 was a pairedlike homeodomain transcription factor that was identified as one of the genes responsible for Axenfeld Rieger syndrome [3], and was induced by the Wnt-β-catenin pathway and TGF-β1-FGF9 pathway in cell proliferation during development [4,5].
Recently, low Pitx2 expression was reported in patients with breast cancer [6,7], prostate cancer [8,9] and colon cancer [10], and it was considered as a potential predictor of poor clinical outcomes, which were correlated with its hypermethylation. Hirose et al. also reported that Pitx2 levels were inversely correlated with in vitro colon cancer cell growth and invasion [10]. According to these observations, Pitx2 may suppress carcinogenesis. However, interestingly, several studies showed that Pitx2 may promote tumor progression, and it was considered to be a potential oncogene in thyroid cancer [11,12], prostate cancer [13] and ovarian cancer [14,15]. These studies indicate the diverse and complex functions of Pitx2 in carcinogenesis.
The regulatory mechanism of Pitx2 as an oncogene had also been partly discovered in ovarian cancer, which was associated with the Wnt pathway [14,15]. However, the regulatory mechanism of Pitx2 as a tumor suppressor is still unclear, and Pitx2 has not been previously implicated in PDAC.
In this study, using microarray analysis, we found that Pitx2 expression was significantly downregulated in PDAC tissues and decreased Pitx2 expression was correlated with Smad4 inactivation, but not with Pitx2 DNA-methylation. Furthermore, Pitx2 may act as a downstream suppressor in the TGF-β1-Smad4 pathway, and promote the protein phosphatase 2A regulatory subunit B55α (PPP2R2A) expression which may inhibit the Akt pathway in pancreatic carcinogenesis.

Gene expression profiling for identification of the potential markers correlated with known suppressors Smad4, p16, Tp53 and BRCA2 in human PDAC
To identify genes with potential roles in the progression of PDAC, we performed gene expression profiling of the microdissected tissues of human PDAC using 4 × 44K Agilent microarrays. To investigate mRNAs with statistically significant differences in expression, heatmap and volcano plot filtering on eight human PDAC tissues and paraneoplastic tissues were performed (Supplementary Figure S1A and S1B), and thirty-five upregulated genes and twenty-nine downregulated genes (≥2-fold, p-value< 0.05) were identified (Supplementary Table S1).
We aimed to identify potential markers correlated with the known suppressors Smad4, p16, Tp53 or BRCA2 with the most frequently inactivated mutation in PDAC, among the sixty-four differentially expressed genes. Interestingly, according to the microarray results, Pitx2 expression was strongly correlated with Smad4, p16 and Tp53 expression using Pearson's correlation test (Supplementary Figure S1C), and it significantly decreased from 2.65-to 19.62-fold in six of the eight PDAC tissues compared to the control ( Figure 1A). We observed that in the PDAC and paraneoplastic tissues, the Pitx2 levels were significantly correlated with the Smad4 levels (PDAC: p = 0.003; paraneoplastic tissues: p = 0.002), p16 levels (PDAC: p = 0.049; paraneoplastic tissues: p = 0.046), and Tp53 levels (PDAC: p = 0.045; paraneoplastic tissues: p = 0.038). However, the Pitx2 levels did not show a correlation with the BRCA2 levels in PDAC (p = 0.896) or paraneoplastic tissues (p = 0.573) ( Figure 1A).
Real-time PCR analysis of Pitx2, Smad4, p16, Tp53 and BRCA2 expression showed that in six of the eight PDAC tissues, Pitx2, Smad4, p16, and Tp53 levels were significantly decreased from 1.71-to 19.20-fold, 1.55to 10.83-fold, 2.01-to 6.54-fold, and 1.75-to 6.75-fold, respectively. However, BRCA2 levels were decreased from 1.48-to 4.32-fold in two tissues, compared to the control. The PCR results were consistent with the microarray data ( Figure 1B).
To further examine whether Pitx2 expression had a significant correlation with Smad4, p16 or Tp53 expression, Pearson's correlation test was performed on the PCR data. As shown in Figure 1B, Pitx2 expression was significantly correlated with Smad4 expression in PDAC (p = 0.004) and paraneoplastic tissues (p = 0.001). However, Pitx2 expression was not correlated with p16 expression in PDAC (p = 0.394) or paraneoplastic tissues (p = 0.144) and with Tp53 expression in PDAC tissues (p = 0.119) or paraneoplastic tissues (p = 0.135). We then evaluated the Pitx2 and Smad4 levels in an increased sample size of thirty-six PDAC and paraneoplastic tissues by real-time PCR. Pearson's correlation analysis revealed that Pitx2 expression had a strong correlation with Smad4 expression in the thirty-six PDAC (p = 0.008) and paraneoplastic tissues (p = 0.017) ( Figure 1C).
These data demonstrate that the significant decrease in the Pitx2 levels may be correlated with decrease in the Smad4 levels in PDAC.

Dynamic alteration and clinical significance of Pitx2 expression in human PDAC tissues
To determine whether the differential expression of Pitx2 was clinically significant, immunohistochemistry staining was performed to assess the Pitx2 levels in 256 PDAC tissues (including 162 pancreatic intraepithelial neoplastic lesions[PanIN]) and 256 paraneoplastic tissues. The percentage of stained cells was scored from grade 0 to 3. Weak, moderate, and intense Pitx2 nuclear staining was observed in 83, 75 and 28 of the 256 paraneoplastic pancreatic tissue samples, respectively. Weak, moderate, and intense Pitx2 nuclear staining was detected in 104 of the 162 PanIN lesions. In contrast, no Pitx2 nuclear staining was observed in 59 of the 92 welldifferentiated adenocarcinoma (WD-PDAC) tissues and in 131 of the 164 moderately-or poorly-differentiated adenocarcinoma (MD-PDAC or PD-PDAC) tissue samples ( Figure 1D and 1E). The Pitx2 levels did not show significant downregulation in PanIN lesions compared to paraneoplastic pancreatic tissues (p = 0.076), however, were sharply decreased in WD-PDAC tissues (p = 0.001) and MD or PD-PDAC tissues (p < 0.001) compared to paraneoplastic pancreatic tissues ( Figure 1E).
Furthermore, the PDAC samples were divided into three groups for clinicopathological evaluation according to the Pitx2 levels: strong positive Pitx2 www.impactjournals.com/oncotarget staining (grade 2 or 3), weak positive Pitx2 staining (grade 1) and negative Pitx2 staining (grade 0). Kaplan-Meier analysis showed that the postoperative median survival was 24.97 months for patients with strong positive Pitx2 staining, 17.72 months for patients with weak positive Pitx2 staining and 13.1 months for patients with negative Pitx2 staining (p = 0.029) ( Figure 1F). These data demonstrate that Pitx2 expression was sharply decreased in PDAC tissues, which is probably correlated with the poor prognosis in PDAC patients.
These data suggest that Pitx2 acted as a potential tumor suppressor in PDAC cells.

Pitx2 is a target gene of Smad4 in PDAC cells
Next, we aimed to explore the underlying mechanism of low Pitx2 expression in PDAC. To determine whether Pitx2 DNA-hypermethylation caused low Pitx2 expression in PDAC, the Pitx2 DNAmethylation levels and Pitx2 mRNA levels were detected in human PDAC cell lines HPAF-II, Miacapa-2, Panc-1, BxPC-3, Capan-1, AsPC-1 and CFPAC-1. Human breast cancer cell line MCF-7 with Pitx2 DNA-hypermethylation was used as the positive control [6]. The results showed that relative Pitx2 DNA-methylation levels were extremely low, between 0.14% and 6.67%, in PDAC cells compared to MCF-7 cells (58.1%), and Pitx2 mRNA levels in PDAC cell lines were low, from 0.03% to 7.16 % (Supplementary Figure S3A). We also compared the Pitx2 DNA-methylation levels in thirty-six fresh human PDAC and paraneoplastic tissues. There was no significant difference of Pitx2 DNA-methylation between the two groups (Supplementary Figure S3B). Importantly, we found that the Pitx2 DNA-methylation levels did not correlate with the Pitx2 mRNA levels in thirty-six human PDAC tissues using Pearson's correlation test (p = 0.275) (Supplementary Figure S3C). These results reveal that factors other than Pitx2 DNA-hypermethylation lead to low Pitx2 expression in PDAC.
To confirm the correlation of Pitx2 with Smad4, Tp53, p16 and BRCA2 in human PDAC and paraneoplastic tissues, lentivirus-mediated Smad4, p16, BRCA2, Pitx2-shRNA vector or the control were stably transduced into SHPAN cells or DTPCa cells [16,17]. The knockdown effects of Smad4, p16 and BRCA2 were shown in Supplementary Figure S4 and the pairs of shRNA with the best knockdown effect were used, respectively. The genotype of the cell lines was shown in Figure 3A. We found that the Pitx2 levels were significantly decreased in SHPAN-lent-Smad4i cells compared to the control, whereas knockdown of Pitx2 did not alter the protein levels of Smad4, p16, Tp53, or BRCA2 in SHPAN cells ( Figure 3A). The results indicate that Pitx2 expression might be correlated with Smad4 expression in mouse cells.
To identify whether Smad4 protein binds to Pitx2 promoter, we constructed a series of reporter plasmids that cover various lengths of the Pitx2 promoter from -2023 to +16. Luciferase promoter-reporter assays showed that the effect of the Smad4 expression vector in Panc-1 and Miacapa-2 cells was maintained in the -1012/+16, -557/+16, and -268/+16 truncated constructs, but lost in the -73/+16 construct ( Figure 3D). These results indicate that the regulatory region primarily involved in Smad4mediated Pitx2 expression was located in the -268/+16 upstream region of the promoter, which contained three putative Smad4-binding sites, namely M1, M2, and M3 ( Figure 3E). To validate whether M1, M2 or M3 was www.impactjournals.com/oncotarget  Smad4-binding site, the three ones were deleted to abolish Smad4 binding, respectively. The results demonstrated that the ΔM1 and ΔM3 constructs did not significantly inhibit Smad4 binding, but the ΔM2 mutated construct significantly inhibited Smad4 binding compared to the control ( Figure 3E). The protein levels of Pitx2 were shown in Supplementary Figure S5A. These results suggest that the putative M2 site in the Pitx2 promoter may be essential for Smad4-mediated activation. To confirm whether Smad4 directly binds to Pitx2, ChIP analysis was conducted. Primers were designed to amplify the -268/+16 site in the Pitx2 promoter. ChIP analysis showed that Pitx2 was markedly detected from the Smad4immunoprecipitated Panc-1 and Miacapa-2 chromatins, but absent from chromatin immunoprecipitated by the control rabbit immunoglobulin G ( Figure 3F).
Taken together, these data suggest that Smad4 protein directly binds to the promoter of Pitx2 and that ectopic expression of Smad4 results in an increase in the Pitx2 levels in PDAC.

Pitx2 is a downstream tumor suppressor in the TGF-β-Smad4 pathway in PDAC
To clarify whether Pitx2 is a tumor suppressor in the downstream of TGF-β-Smad4 pathway in PDAC, we firstly detected the Pitx2 levels with or without inhibitors against ERK1/2 (PD98059), p38 (SB202190), TGFβ1-Smads (SB431542), and JNK (SP600125). As shown in Figure 4A, only SB431542 resulted in a decrease of Pitx2 expression in Panc-1 and Miacapa-2 cells with or without TGF-β1 treatment, whereas treatment with PD98059, SB202190, or SP600125 had no effect. Then, we examined whether TGF-β1 would promote Pitx2 expression. Lentivirus-mediated Smad4-shRNA vector or the control was stably transduced into Panc-1 cells, and the Pitx2 levels in the cells with or without TGF-β1 treatment were determined. The results showed that TGF-β1 may result in an increase in the Pitx2 levels in the presence of Smad4; however, TGF-β1 may not alter the Pitx2 levels in the absence of Smad4 ( Figure 4B). In addition, lentivirus-mediated Smad4-overexpression vector or the control was stably transduced into Panc-1 cells, and immunofluorescence staining and immunoblotting analysis were performed to measure the levels and locations of Pitx2 and Smad4 with or without TGF-β1 treatment in the cells. Nuclear and cytoplasm Pitx2 and Smad4 staining was sharply increased with overexpression of Smad4, and treated with TGF-β1 further increased the Pitx2 and Smad4 levels and induced Pitx2 and Smad4 nuclear import ( Figure 4C). These data show that Pitx2 expression may be altered by TGF-β1 in the presence of Smad4.
These data suggest that Pitx2 acted as a potential suppressor downstream of the TGF-β1-Smad4 pathway in PDAC cells.

Pitx2 may promote Akt dephosphorylation via upregulation of PPP2R2A in PDAC
We next identified the potential downstream markers of Pitx2. Lentivirus-mediated Pitx2-overexpresion vector or the control was stably transduced into in Panc-1 cells and lentivirus-mediated Pitx2-shRNA vector or control was stably transduced into in HPDE6-C7 cells. Microarray was performed on cell lines in both groups, and genes were considered differentially expressed if expression values changed by at least 2-fold. Microarray analysis identified 107 differentially expressed genes in Panc-1 cells with Pitx2-overexpression and 128 differentially expressed genes in HPDE6-C7 cells with Pitx2-shRNA, compared to the control ( Figure 5A, Supplementary  Table S2 and Supplementary Table S3). The thirteen differentially expressed genes that overlapped between the two cell groups were chosen for validation by realtime PCR. According to the PCR data, the protein phosphatase 2A regulatory subunit B55α (PPP2R2A) was the most significantly differentially expressed gene among the thirteen genes in both groups of cells, which were consistent with the microarray data. Induction of Pitx2 upregulated B55α (PPP2R2A) levels in Panc-1 and Miacapa-2 cells, while knockdown of Pitx2 downregulated B55α levels in HPDE6-C7 cells ( Figure 5B). To testify whether Pitx2 was correlated with PPP2R2A, Pitx2 and PPP2R2A levels were detected in thirty-six fresh PDAC www.impactjournals.com/oncotarget  or paraneoplastic tissue samples. Pearson's correlation analysis showed that Pitx2 expression was significantly correlated with PPP2R2A expression in thirty-six PDAC or paraneoplastic tissues ( Figure 5C). To identify whether Pitx2 protein binds to PPP2R2A promoter, a series of reporter plasmids that cover various lengths of the PPP2R2A promoter from -1975 to +46 were constructed. Luciferase promoter-reporter assays showed that the effect of the Pitx2 expression vector in Panc-1 and Miacapa-2 cells was maintained in the -1962/+46, -412/+46, and -198/+46 truncated constructs, but lost in the -61/+46 construct, which demonstrated that the regulatory region primarily involved in Pitx2-mediated PPP2R2A expression was located in the -198/+46 upstream region of the promoter ( Figure 5D). The -198/+46 region may contain three putative Pitx2-binding sites, namely M1, M2, and M3 ( Figure 5E). To testify whether M1, M2 or M3 was Pitx2-binding site in Panc-1 and Miacapa-2 cells, the three ones were deleted to abolish Pitx2 binding, respectively. Expression of the ΔM1 and ΔM2 constructs did not significantly alter Pitx2 binding, but expression of the ΔM3 construct resulted in a significant decrease in Pitx2 binding ( Figure 5E). The protein levels of Pitx2 were shown in Supplementary Figure S5B. The results indicate that the putative M3 site in the PPP2R2A promoter may be essential for Pitx2-mediated activation. To confirm whether Pitx2 protein directly binds to PPP2R2A promoter, ChIP analysis was conducted. Primers were designed to amplify the -198/+46 site in the PPP2R2A promoter. ChIP analysis showed that PPP2R2A was markedly detected from the Pitx2-immunoprecipitated Panc-1 and Miacapa-2 chromatins, but absent from chromatin immunoprecipitated by the control rabbit immunoglobulin G ( Figure 5F). Taken together, these data suggest that Pitx2 protein directly binds to the promoter of PPP2R2A and that ectopic expression of Pitx2 results in an increase of PPP2R2A levels in PDAC.
To determine whether PPP2R2A expression inhibited Akt in PDAC, lentivirus-mediated PPP2R2Aoverexpression vector or the control was stably transduced into Panc-1 and Miacapa-2 cells. Immunoblotting analysis indicated that B55α (PPP2R2A) promoted dephosphorylation at Thr-308 of Akt in Panc-1 and Miacapa-2 cells ( Figure 6A). As shown in Figure 5B that Pitx2 overexpression upregulated B55α levels, we then determined whether Pitx2 promoted B55α expression, which may inhibit Akt pathway in PDAC cells. We immunoprecipitated endogenous Akt by an anti-Akt antibody from the Panc-1 and Miacapa-2 cells and examined the association of the protein phosphatase 2A (PP2A) complexes in the immunoprecipitated Akt. The interaction between Akt and B55α, PP2A-C, or PP2A-Aα subunit was markedly increased in the cells with Pitx2 overexpression, which suggests the possible effects of Pitx2-induced B55α on Akt phosphorylation. In addition, Pitx2-induced B55α mainly binds to the PP2A holoenzyme to catalyze dephosphorylation at Thr-308 of Akt but not at Ser-473 in Panc-1 and Miacapa-2 cells ( Figure 6B).
To confirm whether inhibition of the Akt pathway by Pitx2-induced B55α was correlated with cell invasion, Akt activators IGF-1 and KGF/FGF-7 were used. The results showed that IGF-1 and KGF/FGF-7 may stimulate the phosphorylation at Thr-308 of Akt in Panc-1 and Miacapa-2 cells with PPP2R2A or Pitx2-overexpression, and that IGF-1 had more significant effect ( Figure 6C). Moreover, the invasion of Panc-1 and Miacapa-2 cells with Pitx2-overexpression was significantly activated by IGF-1 ( Figure 6D).
Taken together, these results demonstrate that B55α (PPP2R2A) induced by Pitx2 inhibited the Akt pathway, which may be responsible for suppressing PDAC cell invasion. A simplified schematic diagram of the genetic pathway was summarized in Figure 6E.

DISCUSSION
Recently, Pitx2 was found to play diverse and complex roles in carcinogenesis. Because of its hypermethylation, low Pitx2 expression had been reported in patients with breast cancer [6,7], prostate cancer [8,9], and colon cancer [10], which was considered to be a potential predictor of poor clinical outcomes. Moreover, Hirose et al. reported that Pitx2 expression was inversely correlated with colon cancer cell growth and invasion in vitro [10], which suggests that Pitx2 may act as a tumor suppressor in colon cancer. Interestingly, Huang et al. investigated that knockdown of Pitx2 in human thyroid cancer cells significantly reduced cell proliferation [11]. Vela et al reported that Pitx2 specific shRNA inhibited prostate cancer cell migration toward bone cell derived chemoattractant [13]. These studies suggest that Pitx2 may act as an oncogene in the cancer types.
Using microarray analysis, we tried to identify the potential markers correlated with Smad4, p16, Tp53 and BRCA2 in human PDAC tissues. We found that Pitx2 expression was sharply decreased in PDAC tissues, and was significantly correlated with Smad4 expression. As Pitx2 has not been previously implicated in PDAC, we were interested in exploring the expression and function of Pitx2 in PDAC. We found that low Pitx2 expression was correlated with poor prognosis in PDAC patients, and Pitx2 may act as a tumor suppressor in PDAC.
Low Pitx2 expression had been reported correlated with its hypermethylation status [6,7]. To examine whether low Pitx2 levels in PDAC tissues were correlated with its DNA-hypermethylation, we examined the Pitx2 DNA-methylation levels in PDAC. Interestingly, we identified that the Pitx2 DNA-methylation levels in PDAC were extremely low.
Previous reports had indicated the role of Pitx2 as an oncogene that was associated with the Wnt pathway [14,15]. However, the regulatory mechanism of Pitx2 as a tumor suppressor is still unclear. Therefore, we tried to www.impactjournals.com/oncotarget explore the underlying mechanism in PDAC. Remarkably, we found Pitx2 levels were significantly correlated with Smad4 levels in PDAC tissues according to microarray units and following verification. Luciferase and ChIP analysis indicated that Smad4 protein bound to Pitx2 promoter and stimulated Pitx2 expression. Moreover, TGF-β1 may stimulate Pitx2 expression and induces Pitx2 nuclear import in the presence of Smad4. Taken together, these findings suggest that Pitx2 may be a downstream suppressor in the TGFβ1-Smad4 pathway.
Furthermore, we tried to elucidate how Pitx2 regulated downstream targets during pancreatic carcinogenesis. CyclinD2 and CyclinA1 had been reported to be downstream targets of Pitx2 that promote thyroid carcinogenesis [11,12]. In our study, we identified that Pitx2 expression stimulated B55α (PPP2R2A) expression. Previous studies demonstrated that B55α was a regulatory subunit of PP2A, and PP2A may inhibit Akt pathway [18,19]. We previously showed that B55α may cause dephosphorylation both at Thr-308 and Ser-473 of Akt in mouse cells [20]. Consistent with previous reports [21], in this study, we found that B55α bound to the PP2A holoenzyme to catalyze dephosphorylation at Thr-308 of Akt but not at Ser-473 in human PDAC cells, which may be responsible for the inhibition of PDAC cell invasion.
In summary, our findings show that Smad4 protein may stimulate the transcription of Pitx2 and Pitx2 may promote B55α (PPP2R2A) expression which may inhibit Akt pathway in PDAC. We propose that the Smad4-Pitx2-PPP2R2A axis, a new signaling pathway, suppresses the pancreatic carcinogenesis.

Patients and tissue samples
Two hundred and fifty-six PDAC and paraneoplastic tissues were obtained from Ruijin Hospital (Shanghai, China). None of the patients underwent radiotherapy or chemotherapy before surgery. The tissues were embedded in paraffin wax for analysis. Histological diagnoses were performed by two independent senior pathologists. Thirtysix fresh PDAC tissues and paraneoplastic tissues obtained after surgery were frozen for the following experiments.

Reagents
The MEK inhibitor PD98059 and TGF-β1-Smad4 inhibitor SB431542 were purchased from Calbiochem. A selective inhibitor of p38 mitogen-activated protein kinase SB202190 and JNK inhibitor SP600125 were purchased from Biomol. Human TGF-β1 and Akt activators KGF/FGF-7 and IGF-1 were purchased from R&D Systems Inc.

Gene expression analysis and quantitative real-time PCR
Total RNA was isolated by Trizol, and RNA quality was checked using denaturing agarose gels. For microarray, the cRNA was purified and then hybridized to the 4×44K oligo microarray manufactured by Agilent Technologies. Microarray data were analyzed using an Agilent G2565BA Microarray Scanner. Comparative analysis was done using GeneSpring GX 10 (Agilent Technologies) [2,[25][26]. The experimental data were compared to baseline data. Genes were considered differentially expressed if expression changed by at least two-fold.

Methylation-specific real-time PCR and quantitative real-time PCR
Genomic DNA from seven human pancreatic cancer cell lines and 36 human PDAC tissues were extracted with the DNeasy Tissue kit (Qiagen). Methylation-specific PCR (MSP) was carried out using the MethylDetector kit (Active Motive). Purified genomic DNA (1μg) was converted with bisulfite, and purified according to the manufacturer's instructions. The methylation status of the upstream transcription start site (P1 region) of the Pitx2 gene was analyzed by real-time PCR. Cycle threshold values obtained from each probe (Ctm and Ctu represent the methylated and unmethylated status, respectively) were used to calculate the methylation rate (=100/ [1 + 2 {Ctm-Ctu} ]) [6]. The reverse transcription and quantitative real-time PCR were done using the ABI PRISM 7300 Sequence Detection System (Applied Biosystems). Relative quantification of gene expression was determined using the comparative CT method. Gene expression levels in A relative to B were calculated using the following formulas: ΔΔCT=ΔCTA-CTB, fold change = 2 −ΔΔCT . Primers were shown in Supplementary Table S4.

luciferase reporter assay and chromatin immunoprecipitation
All transfections were performed using Lipofectamine 2000 (Invitrogen). The luciferase assay was performed with the Dual Luciferase Assay System kit (Promega Corp.). Relative luciferase activity was reported as the fold induction after normalization for transfection efficiency. ChIP was performed using the EZ ChIP™ Chromatin Immunoprecipitation Kit (Millipore). Chromatin was immunoprecipitated using the mouse monoclonal anti-Smad4 antibody (ab3219, abcam) or anti-Pitx2 antibody (ab192495, abcam). Normal polyclonal rabbit anti-mouse IgG was used as negative control. Primer sequences were listed in Supplementary Table S4.

Cell proliferation, soft agar assay and invasion assay
A detailed description was previously described [25].

Nude mouse subcutaneous xenograft and metastasis model
The BALB/c nude mice were divided into experimental and control groups. The cells (5×10 6 per mouse) were injected into the subcutis of nude mice in groups of five of experimental or control group, and 25 days later, tumor volume (V) was estimated using the formula V = LW 2 /6 (L: length of tumor; W: width of tumor). Also, the cells (1 ×10 6 per mouse) were injected into the tail vein of nude mice in groups of five of experimental or control group, and 60 days later, lung of each mouse was removed and lung metastases were determined. The micrometastases in the lung tissues had been quantified by two independent pathologists (doubleblinded). The BALB/c nude mice were maintained under specific pathogen-free conditions at the Shanghai Experimental Animals Centre of Chinese Academy of Sciences, and they were cared for in accordance with the institutional guidelines.

Histology, immunohistochemistry, and immunofluorescence
Four-micrometer-thick sections were stained with hematoxylin and eosin for histological verification. Primary antibody against Pitx2 was purchased from Santa Cruz (sc-8748). To evaluate the immunohistochemistry results, the percentage of positive cells was scored from grade 0 to grade 3: Grade 0 (negative), <1% or 0% of the cells were stained; Grade 1 (weak), 1% to 49% of the cells were stained; Grade 2 (moderate), 50% to 70% of the cells were stained; and Grade 3 (intense), >70% of the cells were stained [26,28]. For immunofluorescence staining, the antibodies used included anti-Smad4 (ab110175, abcam) and anti-Pitx2 (ab55599, abcam). Rhodamine-conjugated goat anti-rabbit IgG and FITC-conjugated goat anti-mouse IgG were used as secondary antibodies. The stained cells were mounted on glass slides and examined by confocal microscopy.

Statistical analysis
Statistical analyses were carried out using SPSS version 17.0. Each experiment was repeated at least three times. The results were presented as mean ± SD. Significant differences were assessed using Student's ttest or Kruskal-Wallis test. Postoperative survival was evaluated using the Kaplan-Meier method and log-rank test. Pearson's correlation test was used for correlation analysis. p values less than 0.05 were considered statistically significant.