Microlumens in the Pdx1^−/− bud expand to form an epithelial cyst

Since defects were evident early in Pdx1^−/− pancreatic epithelium, we characterized microlumen formation and coalescence. Analysis of whole-mount immunofluorescent pancreata stained for E-cad and Mucin1 (Muc1) showed that microlumens in control buds rapidly increased in number, but remained of constant size (Fig. 1A-H). Pdx1^+/− epithelium had 270±22 microlumens at E10.5, with a maximum number at E11.0 of 334±49 (Fig. 1I,K,M). As the bud developed, these microlumens connected into a fine plexus of narrow lumens and individual microlumens could not be counted thereafter. By E11.5, 150±30 microlumens and at E12.5, 111±10 microlumens were found in Pdx1^+/− buds (Fig. 1A-D,M). By contrast, in Pdx1^−/− buds, the number of microlumens decreased over time after E10.5, with an average of 263±34, 115±5 at E11.0, 35±2 at E11.5 and by E12.5, it was further reduced to 9±2 (Fig. 1E-H,M). Imaris surface reconstructions of whole-mount pancreata showed that microlumens initially formed and continued to emerge until E10.5, in both Pdx1^+/− and Pdx1^−/− buds, with rosette formation and microlumens in equal numbers and similar size (Fig. 1I,K). However, by E11.5, mutant lumens significantly expanded (Fig. 1J,L,N and Movies 1 and 2). We note that Pdx1^−/− bud defects were variable, although images shown are representative and quantifications reflect observed phenotypes.

Open in a separate window

Fig. 1.

In the absence of Pdx1, the pancreatic ductal plexus transforms into a bag-like cyst. (A-H) Z-stack images of whole-mount tissues stained with Muc1 to visualize lumens, reconstructed using 3D projection (ImageJ). (I-L) Imaris surface reconstruction of lumens (Muc1, red; E-cad, green). Note that whereas E10.5 microlumen size is similar in Pdx1^+/− and Pdx1^−/− buds (I,K), lumen sizes become expanded in Pdx1^−/− by E11.5 (L) when compared with Pdx1^+/− (J). Representative microlumens included for comparison (insets, panels J,L). (M) Quantification of number of microlumens at different stages. (N) Comparison of microlumen diameter as lumens start to fuse. ns, not statistically significant, *P<0.05, **P<0.01, #P<0.001 (Student's t-test). Scale bars: 50 μm in A-H; 25 μm in I-L.

We also performed analyses on sections to further understand the Pdx1^−/− luminal phenotype. Similar to whole mounts, small microlumens of comparable sizes and shapes were found at E10.5 in Pdx1^+/− and Pdx1^−/− buds (Fig. 2A,E). However, by E11.5-E12.5, instead of a complex luminal network of epithelial tubes as observed in Pdx1^+/−, Pdx1^−/− microlumens enlarged and formed a progressively simpler plexus consisting of fewer connections, which were later eliminated. By E13.5, Pdx1^−/− buds displayed a single large cystic lumen (Fig. 2B-D,F-H). The average lumen diameter in Pdx1^+/− at E13.5 was 9.184±1.1 μm at the widest point, while that of Pdx1^−/− buds was 48.99±2.2 μm (Fig. 1N and Fig. 2D,H). Maintenance of narrow lumen diameters thus failed in the Pdx1^−/− epithelium.

Open in a separate window

Fig. 2.

Pdx1^−/− pancreatic buds display epithelial lumens with progressively expanded diameters after E11.5. Sections through pancreatic buds show epithelial lumen diameters at different stages in (A-D) Pdx1^+/− and (E-H) Pdx1^−/− embryos. Insets are close-up images of representative lumens, and yellow lines indicate lumens measured at widest diameter (Muc1, red) visible. Note normal microlumens at E10.5 (A,E), and cystic lumen in Pdx1^−/− at E13.5 (H), compared with Pdx1^+/− (D). Scale bars: 20 μm.

Pdx1^−/− pancreatic buds are smaller because of decreased proliferation, not cell death

To assess whether the decrease in size over time in Pdx1^−/− buds was due to cell death, we examined cleaved caspase 3 and TUNEL immunostaining on cryosectioned E10.5-E13.5 pancreata (Fig. S5A-F and data not shown). Quantification showed that there are low levels of cell death at these stages in normal pancreas. Similarly, Pdx1^−/− pancreata displayed few cells stained for cell death markers. Using cleaved caspase 3 antibody staining, at E10.5 we observed 1.4±0.4% positive cells out of total Pdx1^+/− epithelial cells and 1.6±0.2% in Pdx1^−/−, at E11.5 we observed 1.6±0.2% apoptotic cells in Pdx1^+/− and 1.6±0.3% in Pdx1^−/−, and at E12.5 we observed 1.2±0.1% positive cells in Pdx1^+/− and 1.5±0.2% in Pdx1^−/− (Fig. S5G). Loss of Pdx1 therefore does not result in epithelial cell death.

We then investigated whether proliferative capacity was decreased in Pdx1^−/− buds by staining for phospho-histone H3 (pHH3). We found no difference in proliferation between heterozygotes and nulls at E10.5 (Fig. 3A,D,G) (13.6±0.5% pHH3⁺ cells in Pdx1^+/− and 12.8±0.4% pHH3⁺ cells in Pdx1^−/− epithelium). However, fewer pHH3⁺ cells were observed in Pdx1^−/− buds at E11.5 and E12.5 (Fig. 3B,C,E-G). The difference in proliferation at these stages, between Pdx1^+/− and Pdx1^−/− buds, is largely because fewer internal cells undergo mitosis, whereas Pdx1^−/− peripheral cap cells exhibited similar low pHH3⁺ positivity. At E11.5, Pdx1^+/− buds had 8.9±1.2% pHH3⁺ cap and 12.6±1.0% pHH3⁺ internal cells, whereas Pdx1^−/− had 7.5±1.0% pHH3⁺ cap and 5.3±0.9% pHH3⁺ internal cells. By E12.5, proliferation had significantly decreased in both newly forming peripheral cap/tip domains and internal/trunk cells in Pdx1^−/− buds, with the difference more striking in trunk cells (2.9±0.3% Pdx1^−/− compared with 7.1±0.4% Pdx1^+/− in tips and 3.8±0.6% Pdx1^−/− compared with 12.0±0.9% Pdx1^+/− in trunks).

Open in a separate window

Fig. 3.

The Pdx1^−/− epithelium has decreased proliferative capacity over time. (A-F) Images of sections through pancreatic buds at E10.5 and E11.5 (pHH3, red; E-cad, green; DAPI, blue). Scale bars: 20 μm. (G) Quantification of pHH3⁺ cells in the E10.5 epithelium shows no difference between Pdx1^+/− and Pdx1^−/−. Comparison of pHH3⁺ cells at E11.5 shows decreased proliferation in internal cells, but not cap cells, of the Pdx1^−/− bud. By E12.5, the Pdx1^−/− epithelium exhibits decreased proliferation in both cap and internal cells. ns, not statistically significant; ^#P<0.001 (Student's t-test).

These findings are interesting because previous reports (Zhou et al., 2007) show that proliferating cells fueling pancreas growth primarily reside in tip cell domains after E12.5. We subsequently verified that at early stages in the pancreas, more proliferation occurred within internal body cells with pHH3 stains on wild-type (WT) pancreas in 0.5 day increments from E10.5-E13.5 (Fig. S6A-G). At E10.5-E12.5, mitosis was indeed more frequent in the internal/trunk cell domain (6.8-9.4% compared with 2.5-6.1% in cap/tip domain). By E13.5, however, we observed a distinct shift in proliferative regionalization, as tip cells displayed increased pHH3 positivity (5.1% tip compared with 1.8% in trunk cells). Our findings show that early growth primarily occurs in the heart of the stratified epithelium and that the shift in proliferation from internal to external cells was correlated with epithelial resolution and branching initiation.

Multipotent progenitor cells (MPCs) are lost in the Pdx1^−/− pancreatic bud

To determine whether progenitors were affected in Pdx1^−/− pancreata, we examined MPC marker expression. We performed immunostaining for CPA1, Ptf1a and Nr5a2 on E11.5 and E12.5 bud sections (Fig. 4). Although CPA1 staining was evident in heterozygote littermates at E11.5 and E12.5, it was absent from Pdx1^−/− epithelium at both stages (Fig. 4A,D and G,J). Similarly, although the MPC marker Ptf1a was present in E11.5 Pdx1^+/− and Pdx1^−/− buds (Fig. 4B,E), its expression had ceased in Pdx1^−/− buds by E12.5 (Fig. 4H,K). By contrast, Nr5a2 continued to be expressed in Pdx1^−/− buds at both stages (Fig. 4C and F,I and L). We confirmed these results through both semi-quantitative PCR and qPCR and observed that in Pdx1^−/− buds, Cpa1 transcripts were reduced, whereas Ptf1a transcripts were absent, compared with controls (Fig. 4M,N).

Open in a separate window

Fig. 4.

Progenitor pool is impaired early in the Pdx1^−/− pancreas. (A-L) MPC marker expression analyzed in the absence of Pdx1. CPA1 (red) is absent at both E11.5 and E12.5 (compare D,J with A,G), whereas Ptf1a (red) is reduced by E12.5 (K compared with H) in the Pdx1^−/− epithelium. Nr5a2 is present at both stages (F,L) in Pdx1-null epithelium. Scale bars: 20 μm. (M) Semiquantitative PCR was performed on cDNA isolated from single E12.5 pancreatic buds for Cpa1 and Ptf1a. Cpa1 transcript levels are decreased in the Pdx1^−/− bud, and Ptf1a is completely absent by this stage. First lane for each individual set of samples per genotype is a water blank stopped after 36 cycles, and second to fourth lanes correspond to cDNA containing reactions stopped at 30, 33 and 36 cycles, respectively. (N) qPCR analysis performed on cDNA isolated from single E11.5 buds (n=5 for each genotype, in triplicate) show that Cpa1 and Ptf1a transcripts are significantly lower in the Pdx1^−/− pancreas, compared with Pdx1^+/−. ^#P<0.001 (Student's t-test).

Similarly, other markers of pancreatic progenitors were reduced in the absence of Pdx1. In WT pancreatic buds, SRY-box 9 (Sox9) is expressed broadly in the epithelium, albeit at different intensities in different cells, with both Sox9^hi and Sox9^lo cells observed. Sox9^hi cells are believed to be bipotential cells that give rise to endocrine and ductal cells (Kopp et al., 2011; Lynn et al., 2007). We found that whereas Sox9 staining is present throughout E11.5 Pdx1^−/− pancreatic buds, Sox9^hi cells were reduced by 30% (Fig. S7A-B′). Analysis of overall intensity of Sox9 staining at E12.5 (Fig. S7C,D) confirmed decreased levels of Sox9 in Pdx1^−/− buds, although Sox9 transcript levels were unchanged. This finding suggests that the bipotent Sox9^hi progenitor population is reduced in the absence of Pdx1.

E-cadherin and β-catenin are downregulated in Pdx1^−/− pancreatic epithelium

Given that epithelial morphology was significantly disrupted in Pdx1^−/− buds, we visualized the epithelium using the classical adhesion components E-cad and β-catenin (β-cat). While the Pdx1^−/− epithelium generally remained cohesive, we nonetheless observed a marked reduction of E-cad (Fig. 5A,C), β-cat (Fig. 5B,D) and α-catenin (α-cat, data not shown) expression by immunofluorescence at E11.5. However, staining in the esophagus (where Pdx1 is not normally expressed) was unaffected (Fig. 5E-H). In E10.5 Pdx1^−/− buds, levels appeared relatively normal, suggesting that E-cad expression was lost at the onset of destratification. Quantification of pixel intensity of E-cad and β-cat determined that levels of adhesion molecules in the absence of Pdx1 were approximately half that obtained in Pdx1^+/− (Fig. 5I).

Open in a separate window

Fig. 5.

Expression of E-cadherin and β-catenin is decreased in the Pdx1^−/− pancreatic epithelium. (A-D) Pancreas sections stained for E-cad and β-cat (in red) and DAPI (in blue). (E-H) Esophagus sections from the same slides show that E-cad and β-cat staining is not affected in tissues not expressing Pdx1. Scale bars: 25 μm. (I) Pixel intensity quantifications of E-cad and β-cat in pancreas of Pdx1^+/− and Pdx1^−/−. ^#P<0.001 (Student's t-test).

Pdx1 regulates expression of E-cadherin, but not β-catenin

We therefore asked whether Pdx1 might play a direct role in the regulation of E-cad and β-cat. Using the Evolutionary Conservation of Genomes (ECR) browser (Ovcharenko et al., 2004) and performing sequence alignments with consensus Pdx1 binding sequence CATYAS, we identified two conserved predicted binding sites for Pdx1 in the E-cad promoter and five conserved predicted binding sites in the β-cat (Ctnnb1) promoter (Fig. 6B,D,F). Pdx1 was previously observed to bind the promoters of E-cad and β-cat in adult β-cells, via a chromatin immunoprecipitation using anti-Pdx1 in both mouse and human adult pancreas (Khoo et al., 2012).

Open in a separate window

Fig. 6.

Pdx1 binds to and activates the E-cad promoter. (A) Luciferase assays show that Pdx1 transcriptionally activates both a 2.8 kb and a 300 bp fragment of the WT E-cad promoter (Prom) fused to a luciferase reporter gene. (B) Schematic showing two predicted Pdx1 binding sites (consensus sequence CATYAS) in the E-cad WT 2.8 kb and 300 bp promoter fragments. (C) When predicted binding sites in the 300 bp E-cad promoter fragment are mutated, reporter activity decreases. (D) Schematic showing promoter fragments analyzed in C, with the WT E-cad 300 bp promoter fragment with and without Pdx1 binding site point mutations. (E) Addition of Pdx1 protein to the 2.3 kb β-cat promoter did not increase luciferase reporter activity. (F) Schematic comparing fragments used for luciferase assays in E; 2.8 kb E-cad promoter fragment has two predicted Pdx1 binding sites and the 2.3 kb β-cat promoter has five predicted Pdx1 binding sites. (G) E-cad and β-cat transcript levels from individual E12.5 pancreatic buds were reduced using semi-quantitative PCR in the Pdx1^−/− pancreas (three different primer pairs, n=3/genotype, in triplicate); β-cat levels remained unchanged. Pdx1 and aldolaseA (AldoA) primer sets were used as controls. (H) qPCR analysis shows that E-cad transcripts are reduced in Pdx1^−/− buds at E11.5, whereas β-cat transcripts were unaffected (n=3/genotype, in triplicate, 6-12 repeats). ns, not statistically significant; **P<0.01 (Student's t-test).

To determine whether Pdx1 was capable of activating transcription of E-cad and β-cat, we performed luciferase assays in HEK293T cells (which normally do not express Pdx1). We found that Pdx1 bound to the E-cad promoter and activated expression of E-cad, using both 2.8 kb and 300 bp promoter fragments (both containing predicted Pdx1 binding sites). Increasing concentrations of Pdx1 protein resulted in a corresponding increase in luciferase reporter activity (Fig. 6A,B). However, mutation of binding sites in the 300 bp E-cad promoter fragment, either individually or combined, hampered luciferase reporter activity, with the highest luminescence signal corresponding to less than half that seen with the WT promoter fragment (Fig. 6C,D). We also assayed a 2.3 kb fragment of the β-cat promoter that encompassed five predicted Pdx1 binding sites, but found that increasing Pdx1 concentration did not activate luciferase (Fig. 6E,F). These results suggested that Pdx1 directly regulates E-cad in the developing pancreas.

To further examine this possibility, we carried out semi-quantitative RT-PCRs and qPCRs on Pdx1^+/+, Pdx1^+/− and Pdx1^−/− pancreatic epithelium to assess E-cad transcripts. In the absence of Pdx1, we found a significant decrease in E-cad mRNA, but not β-cat mRNA (Fig. 6G,H, Fig. S8). Conversely, we used HEK293T cells to test whether increased levels of Pdx1 might promote an increase in the E-cad protein level. In this context, we found a modest but measurable increase in E-cad protein in transfected cells (Fig. S9), further suggesting a positive, regulatory relationship between Pdx1 and E-cad.

Epithelial reacquisition of apical polarity during microlumen formation is not affected

Given the dramatic differences in epithelial organization in Pdx1^+/− versus Pdx1^−/− buds, and our finding that Pdx1 binds and activates the E-cad promoter, we examined the cellular events that might result in loss of lumen diameter maintenance. We previously showed that epithelial cells acquire apicobasal polarity following stratification, undergoing apical constriction and rosette formation and resulting in microlumen formation (Villasenor et al., 2010). Given the defects in maintenance of microlumen diameter in Pdx1^−/− buds, we assessed cell polarity determinants at stages when the epithelium is stratified and forming microlumens. Immunostaining for F-actin (phalloidin), ZO-1, GM-130, atypical protein kinase C (aPKC), Par3 and ezrin showed no evident defects in apical polarity when comparing Pdx1^+/− and Pdx1^−/− buds, either early at E10.5 or during resolution at E12.5 (Figs S10, S11 and data not shown).

Pdx1^−/− epithelial cells fail to maintain apical constriction

Although apical polarization of Pdx1^−/− epithelial cells was grossly normal, we assessed cellular morphology. We speculated that inability to maintain cell shape might impact maintenance of luminal diameter. By E10.0, stratified pancreatic epithelial cells undergo something akin to apical constriction, changing their morphology from cuboidal to ‘bottle’ shaped (Villasenor et al., 2010). Groups of cells coordinately adopt this shape to generate rosettes that open microlumens at their centers, as described in forming kidney tubules (Lienkamp et al., 2012) and zebrafish gut epithelium (Horne-Badovinac et al., 2001).

We found that Pdx1^+/− pancreatic epithelial cells acquired this ‘bottle shape’ and formed normal microlumens by E10.5 (Fig. 7A-A″). However, although Pdx1^−/− cells developed apically constricted ends around E10.5, they did not maintain this bottle morphology over time (Fig. 7B-B″). Pdx1^−/− cells suffered rapid loss of apical constriction and abnormal cell morphology. We used ImageJ to quantify cell shape parameters, including area, circularity and the apico-basal surface ratio. We found that Pdx1^−/− cells were generally larger (67.51±2.38 µm² in Pdx1^+/− compared with 95.30±3.35 µm² in Pdx1^−/−) (Fig. 7C) and adopted a columnar, rather than bottle-shaped, morphology (as per ImageJ circularity assessment, where values equal to 1.0 indicate an ideal circular shape). Pdx1^+/− epithelial cells exhibited an average circularity value of 0.69±0.01, whereas Pdx1^−/− measurements indicated linear morphology with a value of 0.56±0.01 (Fig. 7D). Measurements of the apico-basal surface ratio showed that Pdx1^+/− cells were significantly more constricted at their apical surface (area less than half that of basal surface, ratio=0.48±0.03) (Fig. 7E). However, Pdx1^−/− epithelial cells showed altered cell shape at E11.5 and E12.5, with cells almost equal in their apical and basal surface lengths (ratio=0.88±0.05).

Open in a separate window

Fig. 7.

Cell shape is altered in the Pdx1^−/− pancreas by E12.5. (A-B″) E12.5 pancreas sections stained for epithelial marker β-cat (in red), apical polarity determinant aPKC (in green) and DAPI (in blue). Squares in A and B indicate regions shown at high magnification in A′,B′, and those in A′ and B′ indicate regions shown in A″,B″. White outlines indicate cells used for morphometric analysis. (Contrast is enhanced in the red channel to allow for visualization of cell shape in Pdx1^−/−.) Only cells with defined apical and basolateral domains were used for analysis. Scale bars: 20 μm. (C-E) Cell morphology parameters assessed indicate that Pdx1^−/− cells have a larger area, are less circular and have a higher apical:basal surface ratio, demonstrating that these cells do not apically constrict. ^#P<0.001 (Student’s t-test).

β-galactosidase reaction

Pancreata were fixed using glutaraldehyde for 15 min, rinsed in PBS and stained for β-galactosidase overnight as previously described (Villasenor et al., 2008). Images were taken with a NeoLumar stereomicroscope (Zeiss) using a DP-70 camera (Olympus).

Immunofluorescence on sections

Fixed pancreata were washed in PBS and cryoprotected in 30% sucrose overnight. Tissues were embedded in Tissue-Tek OCT compound and sectioned at 10 µm on a cryostat. Sections were washed in 1× PBS and blocked for 1 h at room temperature (RT) in 5% serum. Primary antibody incubations were done at 4°C overnight (for antibodies and dilutions, see Table S1). Slides were washed in PBS, incubated in secondary antibody (Jackson ImmunoResearch or Invitrogen) for 1 h at RT and subsequently incubated in DAPI. Slides were rinsed in PBS and mounted using Prolong Gold Mounting Medium (Invitrogen). Images were obtained using either an LSM510 or LSM710 Meta Zeiss confocal. In cases where signal amplification was required, TSA kit 12 (Invitrogen) was used according to the manufacturer's directions.

Whole-mount immunofluorescence

Fixed embryonic guts were dehydrated in methanol and incubated in Dent's bleach (methanol:DMSO:30% H₂O₂, 4:1:1). Tissues were then rehydrated to PBS and blocked using TNB reagent (PerkinElmer). Embryonic guts were incubated in primary antibodies overnight at 4°C. Tissues were washed in PBS and incubated in secondary antibodies overnight at 4°C and then dehydrated in methanol. Tissues were visualized after clearing in BABB (benzyl alcohol, benzyl benzoate) using an LSM710 Meta Zeiss confocal to take optical sections every 2 µm in tissues from E10.5-E11.5 and every 5 µm in tissues older than E12.5. Quantification of microlumens was performed in at least three pancreata per genotype.

Quantification of cell shape parameters

ImageJ was used to quantify cell morphology in the Pdx1^+/− and Pdx1^−/− strains at E12.5. The parameters assessed were the circularity, area and the ratio of apical to basal surface. At least three pancreata were analyzed per genotype, with 50 cells analyzed per pancreas. Only cells with defined apical (aPKC⁺) and basolateral (bud periphery) domains were used for analysis. Average values were calculated and mean differences were tested for statistical significance using the Student's t-test.

E-cadherin and β-catenin promoter cloning and mutagenesis

E-cad (2.8 kb and 300 bp) and β-cat (2.3 kb) promoter fragments encompassing predicted Pdx1 binding sites were amplified from mouse genomic DNA (Clontech) with Accuprime Pfx Supermix polymerase (Life Technologies), using primers (IDT) indicated in Table S2. Promoter fragments were inserted into the SmaI site of the pGL3 Luciferase vector (Promega). Full-length Pdx1 was amplified from plasmid DNA (ThermoScientific, 40045664) using primers indicated in Table S2 and inserted into the XhoI-XbaI sites of the CS2-eGFP vector. Mutations in the 300 bp E-cad promoter fragment were generated by stitching PCR using transition substitution for nucleotides.

Luciferase assays

HEK293T cells were seeded (4×10⁴ cells/well) in DMEM/10% fetal bovine serum (FBS) in 24-well tissue culture plates and transfected with 400 ng plasmid DNA using Fugene6 (Promega) after 24 h. Cells were lysed in passive lysis buffer (Promega) 24 h after transfection and subjected to freeze-thaw at −80°C (Anderson et al., 2009). Luciferase activity was measured by reacting 20 µl lysate with 50 µl luciferase assay buffer (Promega) using a Fluostar Optima microplate reader (BMG Labtech), with normalization to β-galactosidase activity using the FluoReporter lacZ/galactosidase quantitation kit (Life Technologies). Experiments were performed in triplicate and repeated five times. Further details can be found in supplementary Materials and Methods.

Real-time and semiquantitative PCR

Briefly, total RNA (250 ng) from mouse pancreata was isolated (RNeasy Micro Kit, Qiagen) and treated with DNAse. For RT-PCR, cDNA was generated (iScript cDNA synthesis kit, Bio-Rad) and 1 µl (1/100th bud) of the cDNA and 1 µl MMTaq was used with 0.5 µl primer (20 nM) for analysis. Individual reactions were stopped at 30, 33 and 36 cycles, to assess levels within the linear range (water blanks stopped at 36th cycle). For qPCR, 1 μl cDNA was generated (SuperScript II, Invitrogen) and used for analysis (Power SybrGreen Master Mix, Applied Biosystems on CFX96, Bio-Rad, using 95°C, 30 s; 60°C, 30 s; 72°C, 30 s; 35 cycles). Fluorescence was measured at 72°C. Transcript levels were assessed by threshold cycle (C_t) calibrated to a standard curve generated for each assay using a five-step 1:5 dilution curve of WT mouse E11.5 pancreas and lung cDNA, and were normalized to cyclophilin (Das et al., 2013). All primers listed in Table S3. Data for RT-PCR were collected from three individual pancreatic buds per genotype, in triplicate, and data for qPCR from five individual isolated buds per genotype, in triplicate (6-12 repeats for each gene analyzed). For further details of methods used, see supplementary Materials and Methods.

Transfection for immunofluorescent staining

HEK293T cells were seeded at 1.6×10⁵ cells/well in DMEM/10% FBS onto coverslips coated with fibronectin in six-well tissue-culture plates, then transfected with 1600 ng plasmid DNA (CS2GFP or CS2GFP-Pdx1). After 24 h, medium was removed, cells were fixed in 4% PFA/PBS, washed in 0.1% NP40/PBS (PBSN), blocked in 5% serum for 30 min, incubated for 1 h with primary antibodies (mouse anti-E-cadherin and chicken anti-GFP, Table S1). Cells were rinsed in PBSN, incubated with secondary antibodies (Invitrogen Alexa Fluor 555 donkey anti-mouse and Alexa Fluor 488 goat anti-chicken) for 1 h, washed in PBSN, processed and visualized as above. E-cadherin levels were quantified using the mean gray value function in ImageJ, on 150 individual transfected cells, obtained from three separate transfection and staining assays in which three replicate wells were used for each transfection type. Further details can be found in supplementary Materials and Methods.

Thanks to the Johnson and MacDonald labs at UT Southwestern Medical Center for aliquots of Ptf1a antibody, as well as the Olson lab for CS2-GFP and pGL3 vectors and HEK293T cells. We are grateful to the Johnson, MacDonald, Olson, Cleaver and Carroll labs for invaluable discussions and assistance.