A total of 22 primary tissue samples verified as late-stage pancreatic adenocarcinoma tumor specimens, obtained as autopsy specimens from the Cooperative Human Tissue Network, as previously described (25), were utilized in the current study; their use was compliant with and approved by the University of Chicago institutional review board and Clinical Trials Committees. Slides holding the tissue sections were boiled in 1X citrate buffer for antigen retrieval. Samples were blocked in 1% normal goat serum, followed by overnight incubation at 4°C with primary antibodies against PAX6 (mouse monoclonal; dilution, 1:500; University of Iowa Hybridoma Bank, Iowa City, IA, USA) and CXCR4 (rabbit polyclonal; cat. no. Ab1670; dilution, 1:200; Abcam, Cambridge, MA, USA). Incubation for 1 h at 4°C with goat anti-rabbit fluorescein (cat. no. 31509; dilution, 1:1,000; Thermo Fisher Scientific, Inc., Waltham, MA, USA) or goat anti-mouse dyl647 (cat no. 21235; dilution, 1:1,000; Thermo Fisher Scientific, Inc) secondary antibodies, which enabled the fluorescence-mediated detection of antigens. Samples were scored as ‘positive’ or ‘negative’ for each antigen, with a positive score determined when >25% of the tumor tissue expressed the antigen.

The pancreatic carcinoma cell lines AsPC-1, BxPC-3, Capan-2, HPAF II (HPAF), MIA PaCa-2, PANC-1, CFPAC-1 as well as control cells HEK-293T (human embryonic kidney) and 3T3 (fibroblast) were obtained from American Type Culture Collection (Manassas, VA, USA), the pancreatic carcinoma cell lines COLO-537, CD11 and SW979 were obtained from Dr Ruggeri from Allegeny University (Philadelphia, PA, USA) who was supplied them by Dr Batra from Eppley Institute for Research in Cancer (Omaha, NE, USA) (25), and all the cell lines were cultured as previously described (14,18). Cells were transfected using Effectene Transfection Reagent (Qiagen, Inc., Valencia, CA, USA), according to manufacturer's protocols. In addition to the reporter and/or expression constructs, cells were also transfected with a β-galactosidase-expressing construct (pCMV; Clontech Laboratories, Inc., Mountain View, CA, USA) to serve as an internal control for luciferase assays. The total amount of DNA transfected per cell was maintained at a constant amount by the addition of pBluescript plasmid (Stratagene; Agilent Technologies, Santa Clara, CA, USA). Levels of luciferase and β-galactosidase were measured using assay kits from Promega (Promega Corporation, Madison, WI, USA) and each experiment was performed minimally in triplicate.

The pGL2-CXCR4pm393, pGL2-CXCR4pm393L and pGL2-CXCR4pm393L-ΔISH vectors were constructed as described previously (13). The pGL2-CXCR4pm393 construct contains a sequence of 393 bp 5′-proximal to exon 1 of the CXCR4 gene, as well as the 5′ untranslated region (UTR) from exon 1. The pGL2-CXCR4pm393L vector contains the entire sequence of pGL2-CXCR4pm393 as well as the coding sequence from all of exon 1 and part of exon 2 cloned in-frame with the luciferase reporter gene cassette and the 1,781-bp intron 1. The pGL2-CXCR4pm393L-ΔISH construct is the same as pGL2-CXCR4pm393L with the exception of a deletion of a 267-bp segment from intron 1 containing the island of homology with the PAX binding site.

Whole cell lysates from pancreatic cancer cells were isolated using RIPA buffer, loaded onto 4–15% Tris-Bis gels (50 µg total protein per cell line) for electrophoresis, then transferred onto nitrocellulose membranes. Membranes were probed with an anti-PAX6 antibody and an anti-β-tubulin (cat no. E7) antibody (University of Iowa Hybridoma Bank; dilutions, 1:100 and 1:400, respectively). Immunoreactivity was detected using a WesternBreeze Chemiluminescent kit (cat. no. WB7104; Thermo Fisher Scientific, Inc.), according to the manufacturer's protocol. All incubations were performed at 4°C and for 1 h for primary and secondary antibody incubations and the blocking step. Secondary antibodies utilized were from the WesternBreeze kit. Following secondary antibody incubation the blots were washed at room temperature four times for five min for each wash. Densitometric analyses of resultant western blots were performed with ImageJ software (ImageJ version 1.47 public domain software; National Institutes of Health, Bethesda, MD, USA). For quantification, raw densitometry numbers were recorded for each band and normalized against the β-tubulin readings. The data presented are densitometric readings from three independent western blot analyses. Background readings are arbitrarily set at 0.5 U.

RNA from pancreatic cancer cell lines was isolated using Trizol (Thermo Fisher Scientific, Inc.), DNase-treated with a Ambion DNA-free DNA Removal Kit (Thermo Fisher Scientific, Inc.), and used as a template for RT with the iScript cDNA Synthesis kit (Bio-Rad Laboratories, Inc., Hercules, CA, USA). The mRNA expression levels of CXCR4 transcript variants A and B (CXCR4A and CXCR4B, respectively), PAX6, PAX6(5a) and GAPDH were evaluated using SYBR-Green Master Mix (Bio-Rad Laboratories, Inc.) and the CFX Connect Real-Time System (Bio-Rad Laboratories, Inc.). The template for each RT-qPCR sample was 2.5 µl of a 1:10 dilution of cDNA derived from 1 µg of starting template RNA. Cycling conditions included 40 cycles of 95°C for 10 sec, annealing at 58°C for 10 sec and a 30 sec 72°C extension. The results are presented as the mean and standard error of the mean of three independent experiments, normalized against GAPDH expression and compared to the MIA PaCa-2 cell line for relative expression using the Pfaffl method (26). MIA PaCa-2 was chosen as the cell line of comparison due to this cell line having significant levels of all four transcripts. Primer sequences are presented in Table I.

GraphPad Prism statistical software (version 5.0; GraphPad Software, Inc., La Jolla, CA, USA) was utilized to determine whether the findings were significant. Correlation coefficients between PAX6-expressing and CXCR4-expressing tumor samples were determined with two-tailed Fisher's exact probability tests, with the analysis of nominal data as either positive or negative expression. The significance of the differences between the groups was determined with Student's t-test and χ² analysis, with a confidence interval of 95%. All values stated as significant have P-values of ≤0.05, unless indicated. All experiments were performed in triplicate.

We have previously described PAX6 expression in pancreatic cancer established cell lines and primary tumor samples (14,18). PAX6 is frequently co-expressed with the receptor protein CXCR4 (Fig. 1). In the present study, expression of PAX6 and CXCR4 was measured in 22 primary tumor tissues. Representative immunofluorescent staining results indicating the expression of PAX6 and CXCR4, as well as their co-expression, are shown in Fig. 1A. In the present study, 13/22 samples (59.1%) and 15/22 samples (68.2%) were positive for PAX6 and CXCR4 expression, respectively (Fig. 1B). The expression statuses of PAX6 and CXCR4 were positively correlated when compared using Fisher's exact probability tests (P=0.0066). Of the PAX6-expressing tumor samples, the majority (12/13; 92.3%) also co-expressed CXCR4.

We previously demonstrated that the PAX transcription factor PAX3 promotes CXCR4 expression in melanoma through a highly conserved enhancer in the CXCR4 gene (13). The PAX enhancer element is located within the first intron of the CXCR4 gene, immediately 3′-proximal to exon 1 (Fig. 2A). The element (Fig. 2A, black box) is located in a 52-bp sequence that is highly conserved between mammals [Fig. 2A, grey box (13)]. PAX3 was found to bind to this PAX element. The PAX6 paired domain binds to a sequence with a TT(A/C) ACGC(A/T) core, first identified through in vitro site selection assays (15) and recently supported through in cellulo chromatin immunoprecipitation-sequencing and systematic evolution of ligands by exponential enrichment studies (27,28). The PAX site in the CXCR4 enhancer region closely matches the preferred PAX6 site, and is a more ideal site for PAX6 over PAX3 due to the 5′ TT rather than GT sequence (15,16). To examine whether PAX6 is also capable of driving expression from this element, vectors containing regions of the CXCR4 locus were transfected into 293T cells with or without PAX6 and/or PAX6(5a), an alternative splice version of PAX6. Three constructs were tested: pGL2-CXCR4pm393, pGL2-CXCR4pm393L and pGL2-CXCR4pm393L-ΔISH. The pGL2-CXCR4pm393 vector contains 393 bp of sequence from 5′-proximal to exon 1 and the 5′-UTR. The pGL2-CXCR4pm393L vector contains all the sequence of pGL2-CXCR4pm with the addition of the first intron and part of the coding sequence cloned in-frame with the luciferase gene cassette. The pGL2-CXCR4pm393L-ΔISH vector has the same elements as pGL2-CXCR4pm393L, but with a deletion of the island of sequence homology. These vectors are shown schematically in Fig. 2B. PAX6 or PAX6(5a) did not drive reporter expression from the pGL2-CXCR4pm393 vector (Fig. 2C). However, PAX6 and PAX6(5a) induced significant reporter expression from the pGL2-CXCR4pm393L construct containing an intact intron 1 [2.8±0.69-fold (P=0.011) and 3.8±0.63-fold (P=0.0017) for PAX6 and PAX6(5a), respectively, relative to the reporter vector alone]. The combined addition of PAX6 and PAX6(5a) produced the highest level of reporter expression (8.1±0.25-fold). Deletion of the enhancer element within the intron completely eliminated the ability of either of the PAX6 protein variants to drive luciferase expression.

In order to determine whether the enhancer site containing the PAX element drives expression in pancreatic cancer cells, the three vectors were transfected into PANC-1, MIA PaCa-2, and BxPC-3 cell lines. All cell lines exhibited luciferase activity from the pGL2-CXCR4pm393L vector (set at 100%), and activity was significantly decreased when the intronic enhancer element was removed (pGL2-CXCR4pm393L-ΔISH): 57.09±17.80% in PANC-1 cells (P=0.035), 47.7±23.11% in MIA PaCa-2 cells (P=0.034), and 42.39±19.18% in BxPC-3 cells (P=0.026; Fig. 2D). The expression of luciferase reporter was significantly lower compared with pGL2-CXCR4pm393L, when only the proximal promoter vector was utilized (pGL2-CXCR4pm393) in all cell lines: PANC-1, 16.32±8.48% (P=0.0025); MIA PaCa-2, 11.82±5.25% (P=0.0031); and BxPC-3, 15.53±11.74% (P=0.0059). These data support the hypothesis that PAX6 proteins drive the CXCR4 gene through a conserved intronic element, and that the loss of this site leads to a significant loss of enhancer activity in pancreatic cancer cells.

We have previously demonstrated that PAX6 is expressed in pancreatic carcinoma primary tumors and cell lines (14). Furthermore, unexpectedly, the predominant PAX6 protein expressed in SW979, PANC-1, and MIA PaCa-2 cells was found to be the alternatively spliced 48-kDa version, PAX6(5a), rather than the canonical 46 kDa protein (18). In the present study, PAX6 levels were evaluated in nine independent pancreatic cell lines (Fig. 3A and B). All the cell lines expressed measurable levels of PAX6 protein, except for the low or undetectable levels in HPAF cells. As demonstrated in our aforementioned results, the predominant band was indicated to be ~48 kDa, corresponding to the alternative rather than canonical form of PAX6 (Fig. 3C). Bands at 46 kDa were absent or faint. To determine PAX6 transcript levels, RT-qPCR analysis utilizing transcript-specific primers were performed. The PAX6(5a) transcript was expressed in all eight of the cell lines tested, while 5/8 lines produced the canonical transcript (Fig. 3D); the cell lines AsPC-1, COLO-357, and SW979 did not have measureable levels of PAX6 canonical transcript. No correlation was identified between the levels of transcript and the detected levels of protein. However, HPAF cells expressed significant levels of the two transcripts while producing low or no PAX6 protein.

CXCR4 protein is widely expressed in pancreatic cancer, with significant levels detected in various pancreatic cancer cell lines, including all lines used in the present study (4–6,29,30). Previous reports have identified CXCR4 expression in pancreatic cell lines by RT-PCR, western blot analysis and immunohistochemistry; however, these methods do not differentiate between the two major versions of CXCR4 (canonical CXCR4 protein/CXCR4B transcript and CXCR4-Lo protein/CXCR4A transcript). The human CXCR4 gene comprises two exons and two alternative start codons (Fig. 4A) (22). The two alternatively spliced transcripts (CXCR4B and CXCR4A) differ from one another at the N-terminal ends. CXCR4B (also known as CXCR4 variant 2) encodes the more common CXCR4 protein utilizing codons from exons 1 and 2, while CXCR4A (also referred to as CXCR4 variant 1) produces the longer CXCR4-Lo protein and is encoded entirely from exon 2 (Fig. 4B). The canonical protein encoded from the CXCR4B transcript is expressed in a wide array of tissues, while CXCR4-Lo expression is normally restricted to peripheral blood lymphocytes and spleen cells (22). The proteins differ only in the furthest six or nine (CXCR4 or CXCR4-Lo, respectively) amino acids of the N-terminal ends and, although the functional consequence of this difference is unknown, there is evidence of a differential response to ligand binding (22,24). Our previous study revealed that the canonical CXCR4B transcript was the dominant transcript in melanoma, with expression in all seven lines analyzed (13). However, there was measurable expression of the CXCR4A transcript in three of the seven melanoma lines, albeit to significantly lower levels than those of the canonical transcript. In the present study, pancreatic cancer cell lines were found to express measurable levels of the CXCR4A and CXCR4B transcripts in all eight cell lines measured (Fig. 4C).