Pax-6 and Cdx-2/3 interact to activate glucagon gene expression on the G1 control element.

The promoter element G1, critical for alpha-cell-specific expression of the glucagon gene, contains two AT-rich sequences important for transcriptional activity. Pax-6, a paired homeodomain protein previously shown to be required for normal alpha-cell development and to interact with the enhancer element G3 of the glucagon gene, binds as a monomer to the distal AT-rich site of G1. However, although the paired domain of Pax-6 is sufficient for interaction with the G3 element, the paired domain and the homeodomain are required for high affinity binding to G1. In addition to monomer formation, Pax-6 interacts with Cdx-2/3, a caudal-related homeodomain protein binding to the proximal AT-rich site, to form a heterodimer on G1. Both proteins are capable of directly interacting in the absence of DNA. In BHK-21 cells, Pax-6 activates glucagon gene transcription both through G3 and G1, and heterodimerization with Cdx-2/3 on G1 leads to more than additive transcriptional activation. In glucagon-producing cells, both G1 and G3 are critical for basal transcription, and the Pax-6 and Cdx-2/3 binding sites are required for activation. We conclude that Pax-6 is not only critical for alpha-cell development but also for glucagon gene transcription by its independent interaction with the two DNA control elements, G1 and G3.

The glucagon gene is expressed in the ␣-cells of the pancreatic islets, the L cells of the intestine, and specific areas of the brain (1). The factors controlling glucagon gene expression are still poorly understood. Pancreas-specific expression of the glucagon gene is conferred by the islet-specific enhancer elements G 2 , G 3 , and G 4 (2)(3)(4)(5), and the ␣-cell-specific proximal promoter element G 1 (2,3,(5)(6)(7). G 1 contains two nearly identical 7-bp 1 AT-rich sequences forming a direct repeat that are candidate binding sites for homeodomain transcription factors. At least three protein complexes (B 1 , B 2 , and B 3 ) interact with G 1 , and the integrity of the AT-rich direct repeat is critical for their binding and for transcriptional activity (2,6,7). We and others have recently characterized the transcription factor that binds to the proximal 7-bp site as Cdx-2/3, which is encoded by a caudal-related gene expressed in the endocrine pancreas and the intestine (7,8). Cdx-2/3 is able to bind with high affinity to the 7-bp proximal AT-rich site of G 1 when isolated but binds intact G 1 in glucagon-producing cells preferentially as a multiprotein complex, B 3 (6,7), when both AT-rich sites are present.
We report here that the factor that interacts with Cdx-2/3 to form the B 3 complex is Pax-6, a member of the pax family of vertebrate genes that contain conserved paired and homeo boxes encoding DNA-binding domains (9). Pax-6 has previously been reported to be expressed in the endocrine pancreas (10), to be critical for ␣-cell development (11,12), and to bind to the enhancer element G 3 of the glucagon gene (11). Our results indicate that the paired domain of Pax-6 is sufficient for interaction with the G 3 element, whereas both paired domain and homeodomain are important for interaction with the G 1 element. Furthermore, Pax-6 binds to the promoter element G 1 preferentially as a monomer but also as a heterodimer with Cdx-2/3. When both Pax-6 and Cdx-2/3 are overexpressed in BHK-21 cells, we observe more than additive effects on transcriptional activation of glucagon gene expression, suggesting that Pax-6-Cdx-2/3 interactions have functional consequences on transcription. We conclude that Pax-6 not only is a key regulator of ␣-cell development but is also critical for glucagon gene expression through its independent interaction with the promoter and enhancer elements, G 1 and G 3 , respectively.
CAT and Protein Assays-Cell extracts were prepared 36 -48 h after transfection and analyzed for CAT and alkaline phosphatase activities as described previously (6). Quantification of acetylated and nonacety-lated forms was done with a PhosphorImager (Molecular Dynamics). A minimum of three independent transfections were performed, each of them carried out in duplicate. Protein concentrations were determined with a Bio-Rad protein assay kit.
GST Fusion Proteins and Protein-Protein Interaction in Vitro-For construction of GST fusion proteins, pax-6 paired, homeo, and paired linker homeo boxes were generated from InR1G9 cells by reverse transcription-polymerase chain reaction using the primers Pax-6 PD5Ј, actggatcc-cagcttggtggtgtctttg; Pax-6 PD3Ј, actaagcttgctagccaggttgcgaagaac; Pax-6 HD5Ј, actggatccggctgccagcaacaggaag; and Pax-6 HD3Ј, actaagcttgtgttgctggcctgtcttc and inserted into the BamHI/HindIII sites of pGEX-4T3 (Amersham Pharmacia Biotech). GST fusion proteins were expressed in Escherichia coli and purified according to the manufacturers' recommendations. L-[ 35 S]Methionine-labeled Cdx-2/3 was generated in vitro using the TNT wheat germ extract system (Promega). For protein-protein interaction, 10 g of GST or GST fusion proteins were bound to 25 l of glutathione-Sepharose beads in a total volume of 50 l of incubation buffer containing 12 mM Hepes, pH 7.9, 4 mM Tris/HCl, pH 7.9, 50 mM NaCl, 10 mM KCl, 1 mM EDTA, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride for 20 min at room temperature. Beads were washed three times, resuspended in 20 l of incubation buffer, and incubated with 10 l of L-[ 35 S]methionine-labeled Cdx-2/3 for 40 min on ice. Beads were then washed five times at room temperature with 200 l of washing buffer (20 mM Tris/HCl, pH 8, 100 mM NaCl, 1 mM EDTA, 0.5% Nonidet P-40), recovered in SDS-polyacrylamide gel electrophoresis loading buffer, and bound proteins were analyzed by SDS-polyacrylamide gel electrophoresis followed by autoradiography.

RESULTS
We previously reported that G 1 is a large, 50-bp-long proximal upstream promoter element critical for ␣-cell-specific expression that binds at least three protein complexes, B 1 , B 2 , and B 3 (Ref. 6 and Fig. 1). The more distal enhancer element G 3 can be separated into an A and a B domain; G3 binds four complexes, C 1A and C 1B (A domain), which are islet-specific, and C 2 and C 3 (B domain), representing ubiquitous proteins (Refs. 5 and 20 and Fig. 1). From our previous studies, we concluded that the protein complexes B 1 and C 1A interacting with the G 1 and G 3 elements, respectively, were similar or identical inasmuch as they displayed closely related binding affinities for both G 1 and G 3 and migration characteristics in EMSAs (5,6). Recently, Sander et al. (12) provided evidence that the paired homeodomain transcription factor Pax-6 interacts with the A domain of G 3 and is critical for ␣-cell development. We thus investigated the nature of C 1A and C 1B and their relationship with B 1 . We first performed EMSAs with nuclear extracts from the glucagon-producing hamster cell line InRlG9 and 32 P-labeled oligonucleotides containing either G 1 (G1-56) or G 3 (G3). As shown in Fig. 2A, three specific complexes are detected with G1-56, B 1 , B 2 , and B 3 ; additional complexes are also observed (indicated by asterisks) but have been shown previously to be nonspecific (6). Two major complexes interact with the G 3 oligonucleotide, C 1A and C 1B (Fig. 2B); the ubiquitous complexes C 2 and C 3 are of lower intensity and visible only on longer expositions (5). To more precisely define the relationship between the complexes that bind G 1 and G 3 , we added as competitor cold oligonucleotides containing different portions of the wild-type or mutated element G 1 or the wild-type element G 3 . Like G1-56, G1-33 representing the AT-rich repeat of Schematic representation of oligonucleotides used in EMSAs that correspond to the wild-type or mutated DNA control elements. Two 7-bp AT-rich sequences within G 1 forming an imperfect direct repeat are underlined. In the G 3 sequence, subdomains A and B are underlined. G 1 competed for complexes B 1 , B 2 , and B 3 formed on G1-56 ( Fig. 2A). Mutant G1-33r5 characterized by a mutation of the 5Ј AT-rich site of G 1 only competed for B 3 , whereas the 3Ј AT-rich site mutant (G1-33r3) displayed the same competition characteristics as G1-33. Furthermore, oligonucleotide G1-54 containing the 5Ј AT-rich site of G 1 competed for complexes B 2 and B 3 . These results indicate that B 1 and B 2 interact on overlapping sites within the distal part of G1-33 as previously reported (Ref. 6 and Fig. 1); however, B 2 may interact on a more restricted site compared with B 1 inasmuch as G1-54 efficiently competes for B 2 but not for B 1 . Formation of B 3 , in contrast, involves both the distal and proximal parts of the core element oligonucleotide G1-33. Of note, oligonucleotide G3 competes for both B 1 and B 3 (Ref. 5 and Fig. 2A). We then applied the same competitor oligonucleotides as in Fig. 2A on the complexes formed with InR1G9 nuclear extracts and the 32 P-labeled oligonucleotide G3 (Fig. 2B). Neither formation of the islet-specific complex C 1B nor formation of the ubiquitous complexes C 2 and C 3 are significantly affected by addition of the competing oligonucleotides ( Fig. 2B and data not shown). In contrast, all oligonucleotides competing for complex B 1 on G1-56 (G1-56, G1-33, G1-33r3, and G3), do also compete for C 1A formed on G3, and oligonucleotides that do not interfere with B1 (G1-33r5 and G1-54) do not affect C 1A . These data show that C 1A exhibits similar binding characteristics as B 1 and suggest that C 1A is capable of binding efficiently to the G 1 element.
To assess the nature of complexes B 1 and C 1A , we added in the EMSA incubation reactions antisera raised against the paired or homeodomain of Pax-6 ( Fig. 3A). Anti-paired domain antibodies displaced B 1 and B 3 (G1-56), as well as C 1A and C 1B (G3). Complexes B 1 and C 1A are also supershifted by antibodies raised against the homeodomain of Pax-6. Of note, addition of both anti-Pax-6 antibodies did not completely displace B 1 . The residual complex was, however, of variable intensity in different nuclear extracts of InR1G9 cells and also detected with nuclear extracts from insulin-producing cell lines. Whereas this complex was specifically competed for by G-56, it was less efficiently displaced by competition with G1-33 comprising the core element of G 1 , and not at all with unrelated oligonucleotides (data not shown); it thus may represent a protein present in insulin-and glucagon-producing cells that might play a role in glucagon gene expression via its interaction with G 1 . Incubation of oligonucleotide G3 with extracts from BHK-21 cells transfected with the avian pax-6 cDNA results in a complex that migrates similarly to C 1A . This complex is supershifted with Pax-6 anti-paired and anti-homeodomain antibodies and absent from nuclear extracts from BHK-21 transfected with the vector alone; our data thus indicate that B 1 and C 1A represent Pax-6.
Complex C 1B , which displays binding characteristics that are different compared with B 1 and C 1A (Ref. 5 and Fig. 2), is supershifted with anti-paired domain but not with anti-homeodomain Pax-6 antibodies ( Fig. 3A) and may thus represent a protein antigenically related to Pax-6 or a Pax-6 isoform with a deletion or modification in the homeodomain. Complex B 3 formed on G1-56 has previously been shown by immunological criteria to consist of a protein complex containing Cdx-2/3 (7); its competition by oligonucleotide G3 and its interaction with both anti-Pax-6 antibodies now suggest that B 3 may represent a Pax-6-Cdx-2/3 heterodimer.
To further characterize complexes B 1 and B 3 , we transfected the avian pax-6 or hamster cdx-2/3 cDNA in BHK-21 cells, a cell line that does not express Pax-6 nor Cdx-2/3. As shown in Fig. 3B, incubation of oligonucleotide G1-56 with extracts from BHK-21 cells overexpressing Pax-6 results in a complex that migrates similarly to B 1 and that is absent from nuclear extracts from BHK-21 transfected with the vector alone. Of note, a nonspecific complex with similar migration mobility as Pax-6 and Cdx-2/3 can be observed in vector-transfected BHK-21 cells; this complex does not react with anti-Pax-6 or anti-Cdx-2/3 antibodies (data not shown). When extracts from BHK-21 cells overexpressing Cdx-2/3 are incubated with G1-56, a complex with a slightly lower electrophoretic mobility than B 1 and corresponding to Cdx-2/3 appears (Fig. 3B) as previously reported (7). To test our hypothesis that B 3 represents a Pax-6-Cdx-2/3 heterodimer, we incubated the labeled oligonucleotide G1-56 with a constant concentration of BHK-21 nuclear extracts overexpressing Pax-6 together with increasing amounts of Cdx-2/3 containing extracts or increasing amounts of Pax-6 FIG. 2. Protein complexes from glucagon-producing cells interacting with G 1 and G 3 . EMSAs were performed with 8 g of nuclear extract from glucagon-producing InR1G9 cells and 32 P-labeled G1-56 (A) and G 3 (B) oligonucleotides (see Fig. 1). Competition was performed with 20 and 50 ng of unlabeled wild-type or mutated oligonucleotides as indicated. Specific complexes are indicated on the side with a bold arrowhead, and asterisks correspond to nonspecific complexes.
containing extracts at two different concentrations of Cdx-2/3 containing extracts; in these conditions, an additional complex is observed that migrates like B 3 and supershifted by anti-Pax-6 and anti-Cdx-2/3 antibodies (Fig. 3B). Of note, the intensity of the reconstituted complex did not increase linearly; relatively high concentrations of either Pax-6 or Cdx-2/3 may be necessary for heterodimer versus homodimer formation in this in vitro system. In nuclear extracts from glucagon-producing cells, Cdx-2/3 appears to be present in much lower concentration than Pax-6 because it binds only as the heterodimer B 3 , whereas Pax-6 binds both as a monomer (B 1 ) and a heterodimer (B 3 ). Indeed, when anti-Pax-6 and anti-Cdx-2/3 anti-bodies were added to InRlG9 nuclear extracts incubated with labeled G1-56, anti-Pax-6 antibodies displace B 1 and B 3 , whereas anti-Cdx-2/3 antibodies displace only B 3 (Fig. 4). To test whether the residual complex of low intensity after displacement of B 1 by anti-Pax-6 antibodies corresponds to a Cdx-2/3 monomer, we added both anti-Pax-6 and anti-Cdx-2/3 antibodies to the reaction. However, no further displacement was seen, indicating that in InR1G9 cells, Cdx-2/3 binds G 1 only as the heterodimer B 3 . By contrast to anti-Pax-6 and anti-Cdx-2/3 antibodies, anti-Isl-1 antisera (21) did not affect any of the complexes binding to G1-56.
To analyze the binding characteristics of Pax-6 on the G 1 and G 3 elements, we expressed the Pax-6 paired, homeo, and paired and homeodomains as GST fusion proteins (Fig. 5A). Whereas the Pax-6 homeodomain was unable to bind to G 1 or G 3 , the paired domain interacted with both elements (Fig. 5B); however, the binding of the paired domain to G1-56 was very weak, and the resulting complex was readily competed for by unlabeled G3 oligonucleotides. In contrast, a fusion protein comprising both domains exhibited a high affinity for G 1 and a relatively lower affinity for G 3 ; the complex formed on G 3 was highly competed for by unlabeled G1-56. We therefore conclude that although both the paired and homeodomains of Pax-6 are necessary for maximal interaction with the G 1 and G 3 elements of the glucagon gene, the paired domain alone is sufficient for about half-maximal binding to G 3 , whereas it displays very low affinity for G 1 .
To investigate the functional role of Pax-6, we performed cotransfection experiments in BHK-21 cells using fragments of the 5Ј-flanking sequence of the rat glucagon gene linked to the CAT reporter gene and a SV40-driven expression vector containing the quail pax-6 cDNA. Using the first 350 bp of the glucagon gene promoter (Ϫ350 CAT) containing both Pax-6 binding sites (G 1 and G 3 ), we observed a dose-dependent in-  10 -12 and 13-15, respectively); antibodies directed against Pax-6, its paired domain, its homeodomain, and Cdx-2/3 are indicated by Pax, PD, HD, and Cdx, respectively; bold arrowheads, specific complexes; asterisks, nonspecific complexes; arrows, supershifted complexes.

FIG. 4. Cdx-2/3 binds G 1 only as a heterodimer with Pax-6 (B 3 ).
EMSAs were performed as in Fig. 2. P and Pax indicate preimmune and immune Pax-6 sera, respectively. Cdx and Isl1 indicate Cdx-2/3 and Isl-1 immune antisera, respectively. Bold arrowheads, arrows, and asterisks point to specific, supershifted, and nonspecific complexes, respectively. crease in CAT activity with increasing amounts of expression plasmids containing the pax-6 cDNA. The transcriptional activation observed was similar to that obtained with the transcription factor interacting with the proximal AT-rich site of G 1 , Cdx-2/3 (Figs. 6 and 7A). To characterize the relative functional importance of the proximal (G 1 ) and the distal (G 3 ) Pax-6 binding sites, we investigated the effects of Pax-6 overexpression on promoter constructs containing either one or both Pax-6 binding sites (Fig. 7A); we used the first 138 bp of the glucagon gene promoter either alone (Ϫ138 CAT) or linked to oligonucleotides representing wild-type G 3 (G 3 -138 CAT) or G 3 specifically mutated at the Pax-6 binding site (G 3 M6 -138 CAT). As shown in Fig. 7B, overexpression of Pax-6 resulted in a 98-fold activation of the basal activity obtained with G 3 -138 CAT, whereas only half of this activation was observed for either Ϫ138 CAT or G 3 M6 -138 CAT (37-and 38-fold activation, respectively). These results indicate that both Pax-6 binding sites are necessary for maximal activation of the glucagon gene promoter in BHK-21 cells and that deletion of the G 3 site results in a loss of 50% of the activation potential. We then assessed the consequences of deletions or mutations of the proximal G 1 binding sites by using either the first 75 bp of the glucagon gene promoter linked to wild-type G 3 (G 3 -75 CAT, deletion of the Pax-6 binding site on G 1 ) or the first 350 bp of the promoter with point mutations at nucleotides Ϫ89/Ϫ90 resulting in the loss of Pax-6 binding on G 1 (G 1 M1-350 CAT) (Ref. 6 and Figs. 1 and 8A). A 40-fold stimulation of transcription by Pax-6 was observed for G 3 -75 CAT, which corresponds to half of the activation obtained with G 3 -138 CAT (Fig. 7B). With G 1 M1-350 CAT, transcription increased by 25-fold, which also corresponded to half of the activation seen with the respective control, Ϫ350 CAT. When point mutations were directed to the Cdx-2/3 binding site (nucleotides Ϫ71/72, G 1 M11-350 CAT), Pax-6 induced a 78-fold activation. Our results thus indicate that in BHK-21 cells, each Pax-6 binding site of the rat glucagon gene 5Ј-flanking sequence accounts for about half of the full activation observed when both sites are present. Of note, we always obtained significantly lower stimulation by Pax-6 with the wild-type or mutated 350 first bp of the promoter compared with shorter promoters.
We then analyzed the respective activation potential of Pax-6 and Cdx-2/3 either alone or together on the various promoter mutants described above in BHK-21 cells. 0.25 g of the Pax-6 or Cdx-2/3 expression plasmid was used for transfection because this concentration resulted in submaximal activation and dose-response curves with increasing amounts (from 0.125 to 1 g) of the respective plasmids gave similar activation potentials (Fig. 6). Pax-6 and Cdx-2/3 alone activated Ϫ350 CAT by 21-and 13-fold, respectively; when both pax-6 and cdx-2/3 cDNAs were transfected, we observed an activation reaching 58-fold (Fig. 8A). Similar qualitative results were obtained with Ϫ138 CAT, G 3 -138 CAT and G 3 M6 -138 CAT, which contain both Pax-6 and Cdx-2/3 binding sites on an intact G 1 ; indeed, when tested individually, Pax-6 and Cdx-2/3 increased transcriptional activity from these constructs, but when used together they resulted in a more than additive activation. The weak synergistic effects on transcription observed with Pax-6 and Cdx-2/3 may be explained by the fact that Pax-6 interacts with G 1 primarily as a monomer, B 1 , whereas Cdx-2/3 binds as a heterodimer, B 3 , with Pax-6 (Ref. 7 and Fig. 4).
We then investigated the effects of Pax-6 and Cdx-2/3 on constructs containing point or deletion mutants of G 1 on transcription (Fig. 8A). G 3 -75 CAT, which only contains the distal Pax-6 (G 3 ) and the Cdx-2/3 sites, was activated 26-and 28-fold by Cdx-2/3 and Pax-6, respectively; however, with both Pax-6 and Cdx-2/3, only strictly additive effects were observed. Point mutations of the G 1 Pax-6 binding site that prevent Pax-6 monomer and Pax-6-Cdx-2/3 heterodimer binding (G 1 M1-350 CAT; Ref. 6 and Fig. 8B) decreased the effect of Pax-6 without impairing binding and transactivation by Cdx-2/3; again no synergistic activity was obtained. In contrast, mutation of the major Cdx-2/3 binding site, which decreased Cdx-2/3 monomer binding on G 1 and B 3 formation (G 1 M11-350 CAT; Ref. 6 and Fig. 8B), led to a decreased activation potential of Cdx-2/3 but did not suppress the more than additive effects conferred by both Cdx-2/3 and Pax-6. These binding and transactivation data indicate that the intact Pax-6 binding site may allow some heterodimer formation between Pax-6 and Cdx-2/3 despite a mutated Cdx-2/3 binding site. Indeed, as shown in Fig. 8C, both proteins are able to interact in solution and in the absence of DNA; both the Pax-6 paired and homeodomains participate in this interaction and maximal protein-protein contacts are observed when the two domains are present. However, in the complete absence of binding of Pax-6 as a monomer (G1-56M1), Cdx-2/3 is unable to stabilize the heterodimer B 3 on G 1 ; therefore, heterodimer formation and maximal transcriptional activation may only be possible when some DNA binding of Pax-6 is still present (Fig. 8, A and B).
To investigate the functional importance of the Pax-6 binding sites on G 3 and G 1 and the Cdx-2/3 binding site on G 1 in glucagon-producing cells, we transfected the reporter gene constructs analyzed in Figs. 7 and 8A, in InR1G9 cells. Construct G 3 -138 CAT containing the G 3 element linked directly to the first 138 bp of the glucagon gene promoter conferred a CAT activity identical to the control plasmid Ϫ350 CAT. Mutation (G 3 M6 -138 CAT) or deletion (Ϫ138 CAT) of the enhancer element decreased this activity by about 80% (Fig. 9). Deletion (G 3 -75 CAT) or mutation (G 1 M1-350 CAT) of the Pax-6 binding site on G 1 resulted in 50 -60% loss of activity. Surprisingly, mutation of the Cdx-2/3 binding site on G 1 (G 1 M11-350 CAT), which did not affect the activated transcription conferred by overexpressed Pax-6 and Cdx-2/3 in BHK-21 cells (Fig. 8A), markedly decreased basal CAT activity in InR1G9 cells (80% reduction). This discrepancy may be due to the presence of additional, unidentified factors interacting with Pax-6 and Cdx-2/3 or with Pax6 alone on G 1 in glucagon-producing but not in heterologous cells; alternatively, interactions of the proteins binding to the distal enhancer G 2 or G 3 with those of G 1 that we suggested previously (22) may be impaired by the decreased Pax-6-Cdx-2/3 heterodimer formation.
The same wild-type and mutant reporter gene constructs were transfected along with Pax-6 and/or Cdx-2/3 expression plasmids in InR1G9 cells to observe the consequences of their overexpression on intact or mutated G 1 and G 3 elements; however, we found either no effect or, at the highest amounts of either cDNA transfected, an inhibition of basal activity (data not shown), compatible with a squelching phenomenon. We hypothesize from these results that proteins interacting with Pax-6, Cdx-2/3, or both may be rate-limiting for transcriptional activity and that overexpression of Pax-6 and Cdx-2/3 may decrease the availability of these proteins for transcription. DISCUSSION This study presents evidence that Pax-6 interacts with at least two DNA control elements of the rat glucagon gene promoter, G 3 and G 1 . G 1 is a critical element that confers ␣-cell specificity to the expression of the glucagon gene. Pax-6 is able to bind G 1 either as a monomer or as a heterodimer with Cdx-2/3. When both proteins are overexpressed in BHK-21 cells, they bind preferentially as monomers; formation of heterodimer B 3 is favored by increasing either Pax-6 or Cdx-2/3 concentration but in a nonlinear fashion (Fig. 3B). Binding assays with nuclear extracts from InR1G9 cells suggest that most of Pax-6 binds G 1 as a monomer, as evidenced by the ratio of complexes B 1 /B 3 . However, most if not all of Cdx-2/3 is contained within the heterodimer Pax-6-Cdx-2/3 because we have not been able to detect Cdx-2/3 monomer binding to intact G 1 using InR1G9 nuclear extracts (Ref. 7 and Fig. 4). This is most likely due to the relatively low amounts of Cdx-2/3, as compared with Pax-6, present in InR1G9 cells. Cdx-2/3 monomer binding to its recognition site is only observed when a mutation of the distal AT-rich site on G 1 prevents formation of B 1 (Pax-6), B 2 , and B 3 (Pax-6-Cdx-2/3); the intensity of the monomer complex corresponds to that of the heterodimer in the presence of intact G 1 (7). Despite the relatively low abundance of the heterodimer Pax-6-Cdx-2/3 as compared with the Pax-6 monomer, interactions of Pax-6 with Cdx-2/3 may be functionally important because our results demonstrate that together Pax-6 and Cdx-2/3 activate transcription in a more than additive manner in BHK-21 cells. This effect is entirely lost when Pax-6 binding to G 1 is impaired, such as on the mutant promoters G 1 M1-350 and G 3 -75. In this heterologous system, however, integrity of the Cdx-2/3 site is not required for interactive activation by Pax-6 and Cdx-2/3. Indeed, mutant promoter G 1 M11-350 (6, 7), which decreases Cdx-2/3 binding, does not affect activation by Pax-6 and Cdx-2/3 together in BHK-21 cells. In contrast, mutation of either the Pax-6 or the Cdx-2/3 binding site on G 1 markedly decrease transactivation of the glucagon gene promoter in glucagon-producing cells. This phenomenon may be explained by additional contacts with the proximal AT-rich motif necessary for optimal conformation of the Pax-6-Cdx-2/3 heterodimer and its interaction with yet unidentified proteins in InR1G9 but not in BHK-21 cells. Different expression levels of Cdx-2/3 leading to a variable ratio of Pax-6 monomer/heterodimer formation may serve to modulate transcriptional activity of the glucagon gene during development or in response to physiological stimuli.
Pax-6 has previously been shown to interact with other ho-meodomain proteins such as Engrailed (En-1); heterodimer formation with En-1, however, depends only on the Pax-6 paired domain and down-regulates the DNA binding and transactivation properties of Pax-6 (23). This is in contrast to the Pax-6-Cdx-2/3 heterodimer, whose formation implicates the Pax-6 paired and homeodomains resulting in a transactivation potential that is more than additive compared with both factors taken alone. The role of Pax-6 in endocrine pancreatic development is well established (11,12). Pax-6 is detected in glucagon-producing cells of the mouse embryonic pancreas at day 9.5 and in all pancreatic endocrine cells at later stages of development. In mice homozygous for a targeted null mutation in the Pax-6 FIG. 7. Pax-6 activates glucagon gene expression in BHK-21 cells through both the G 1 and G 3 control elements. A, schematic representation of constructs containing different fragments of the wild-type or mutated rat glucagon gene 5Ј-flanking sequence linked to the CAT reporter gene. Cis-acting DNA control elements G 1 to G 4 , a cAMP response element (CRE), and the TATA box are indicated. Specific mutations in G 1 and G 3 (see Fig. 1 for sequence) are shown by filled circles. A and B correspond to the subdomains of G 3 , and the imperfect direct repeat elements within G 1 are indicated by the letters AT within dotted boxes. Pax-6 monomer and Pax-6-Cdx-2/3 heterodimer binding on G 1 and G 3 is illustrated. B, 1 g of expression vector alone or containing the pax-6 cDNA was cotransfected with 10 g of the indicated glucagon promoter constructs linked to the CAT reporter gene and 1 g of pSV2A pap to monitor transfection efficiency in BHK-21 cells. Data are presented as in Fig. 6. gene, few if any ␣-cells are detected (12); similar findings are reported in the Sey Neu mice, which are characterized by a homozygous mutation of the Pax-6 gene (11). In the latter mice, insulin, somatostatin, and pancreatic polypeptide-producing cells are also markedly decreased, indicating a role for Pax-6 in the development of all islet cells. Mouse genetics has provided direct evidence for the developmental role of tissue-specific transcription factors. Pax-6 interacts with G 1 , a critical determinant of ␣-cell-specific expression of the glucagon gene, and may thus have a key role in its restricted cell-type expression. It has been proposed that Pax-4 and Pax-6 determine the islet cell fate of the pancreas during development; the onset of pax gene expression may then define the lineage of the different endocrine cells (11,12). Cells expressing both Pax-4 and Pax-6 would differentiate into mature ␤, ␦, and ␥ cells, whereas absence of Pax-4 would divert cells to the ␣-cell lineage. It is thus possible that the simultaneous presence of Pax-6 and Cdx-2/3 and the absence of Pax-4 or other ␤ cell-specific factors are the critical determinants for ␣-cell-specific expression of the glucagon gene. Several transcription factors present in ␤ cells or other non-␣-islet cells have been identified (PDX-1 (24), Pax-4 (25), and a recently characterized ␤ cell factor, IB1 (26)), whereas no ␣-cell-specific factor has been identified so far. Interestingly, PDX-1 has recently been shown to directly interact with Cdx-2/3 and to impair its transcriptional activation of the sucrase-isomaltase promoter (27). ␣-cell-specific expression of the glucagon gene might thus occur through a default mechanism whereby ␤ cell-specific factors like Pax-4 or PDX-1 suppress the glucagon promoter in non-␣-islet cells through direct protein-protein interactions with Pax-6-Cdx-2/3, for example.
Pax-6 has previously been shown to interact with the G 3 control element of the rat glucagon gene and to activate tranarrowheads, specific complexes; asterisks, nonspecific complexes; arrows, supershifted complexes. C, GST precipitation assay using 10 g of GST alone or GST-Pax-6 fusion proteins (Fig. 5A) immobilized on Sepharose beads and 35 S-labeled, in vitro synthesized Cdx-2/3 as described under "Experimental Procedures." Input, 1 l of the in vitro translation reaction for Cdx-2/3 using wheat germ lysate.
FIG. 8. Interaction of Pax-6 and Cdx-2/3 on the G 1 control element. A, 0.25 g of expression vector alone or vectors containing either the Pax-6 or Cdx-2/3 cDNAs or both were cotransfected with 10 g of the indicated glucagon promoter constructs linked to the CAT reporter gene and 1 g of pSV2A pap to monitor transfection efficiency in BHK-21 cells. Total DNA was kept constant at 12 g by adding the appropriate amount of empty expression vector. Data are presented as in Fig. 6. B, EMSA performed as in Fig. 2. Nuclear extracts from BHK-21 cells transfected with expression vectors alone or with expression vectors containing the pax-6 and cdx-2/3 cDNAs (Tfd cDNA) or from InR1G9 cells were incubated with 32 P-labeled oligonucleotides corresponding either to the wild-type (G1-56) or mutated (G1-56M1 and G1M11) G 1 DNA control element. Pax, Pax-6; Cdx, Cdx-2/3; bold FIG. 9. Pax-6 and Cdx-2/3 binding sites on the G 1 and G 3 control elements mediate glucagon gene transcription in glucagonproducing cells. InR1G9 cells were transfected with 3 g of the indicated glucagon promoter constructs linked to the CAT reporter gene (Fig. 7A). Data are presented relative to the CAT activity obtained with the control construct Ϫ350 CAT and are the mean Ϯ S.E. of at least four experiments (see "Experimental Procedures"). scription (12). We confirm and extend these findings to the G 1 promoter element. Although the G 3 and G 1 Pax-6 binding sites are clearly homologous in their sequence, they only share 8 of 16 nucleotides (5), implicating functional differences in the DNA-protein interaction. It has indeed been shown that DNA binding induces conformational changes in the Pax-6 protein (28) and that different target sites might have allosteric effects on transcription factors (29). On the glucagon gene 5Ј-flanking sequence, the Pax-6 paired domain alone is sufficient for about half-maximal binding to G 3 , whereas it shows only weak affinity for G 1 . In contrast, a protein comprising the Pax-6 paired and homeodomains exhibits a higher affinity for G 1 than for G 3 (Fig. 5B). These different binding properties might then affect transcriptional potencies and interactions with additional proteins, e.g. Cdx-2/3.
In our heterologous assay system using BHK-21 cells, both Pax-6 binding sites, G 1 and G 3 , appear to be of equivalent functional importance inasmuch as deletions or mutations of either site result in a 50% loss of transcriptional activity. The respective importance of G 1 and G 3 in this heterologous assay does not, however, reflect the basal transcriptional activity of the glucagon gene in glucagon-producing cells. In these cells, deletion of G 3 results in only mild decreases in transcriptional activity when the more proximal enhancer element G 2 is present (2, 6), whereas a point mutation (G 1 M1) of the Pax-6 binding site within G 1 leads to a 60% loss of activity (Fig. 9), indicating that the latter site is crucial for basal transcription. A second difference in the consequences of G 1 mutations is observed between BHK-21 cells overexpressing Pax-6 and Cdx-2/3 and the glucagon-producing InR1G9 cells with a reporter gene construct containing a point mutation in the Cdx-2/3 binding site within G 1 (G 1 M11). Whereas this mutation does not interfere with the more than additive transcriptional activation conferred by Pax-6 and Cdx-2/3, it decreases basal activation in InR1G9 cells by 80% (Figs. 8A and 9). The only change induced by mutation G 1 M11 in the complexes formed with InR1G9 nuclear extracts in EMSA is a reduction of B 3 , and a similar decrease is noted in BHK-21 nuclear extracts containing Pax-6 and Cdx-2/3 (Fig. 8B). Thus, despite the capacity of Pax-6 and Cdx-2/3 to directly interact in the absence of DNA, both binding sites may be important for optimal heterodimer conformation and transcriptional activation in glucagon-producing cells; mutation G 1 M11 may attenuate the formation and change the conformation of the Pax-6-Cdx-2/3 heterodimer and thus impair its interaction with additional factors present in InR1G9 but not in BHK-21 cells. We cannot exclude, however, that the discrepancies in the consequences of G 1 mutations observed between BHK-21 and InR1G9 cells are due to unidentified proteins interacting with Pax-6 and binding to the Cdx-2/3 site.
Pax-6 binds G 3 as a monomer; our results suggest, however, that besides Pax-6, a related protein interacts with G 3 . Indeed, two complexes were previously shown to bind domain A of G 3 , C 1A (Pax-6) and C 1B (5); both complexes share very similar binding characteristics but display differences inasmuch as interference of methylation assays suggest that Pax-6 and C 1B interact with different nucleotides of domain A (5); furthermore, competition for Pax-6 and C 1B by the G 1 binding site in EMSAs affects only Pax-6 interaction (Ref. 5 and Fig. 2B). C 1B , which is recognized by Pax-6 anti-paired but not anti-homeodomain antibodies, displays slightly slower migration than Pax-6. Several isoforms of Pax-6 have been identified in the nervous system and the endocrine pancreas (10, 17, 30); however, functional tests on DNA binding and transactivation with these isoforms suggest that they do not interact with G 3 . 2 Thus, C 1B may represent either an as yet unidentified Pax-6 isoform or a related paired protein. Interestingly, the Pax-6 binding site on G 3 corresponds to the insulin-response element of the glucagon gene and mutations impairing Pax-6 binding and C 1B formation, result in a loss of regulation by insulin (3,5). Formal characterization of C 1B should lead to a better definition of the respective functional role of this Pax-6 related protein and Pax-6 on the insulin regulation of the glucagon gene.
We conclude from our studies that Pax-6 binds independently to at least two DNA control elements of the glucagon gene, G 1 and G 3 . Pax-6 appears functionally important in both ␣-cell development and glucagon gene transcription; it may mediate ␣-cell-specific expression of the glucagon gene as the major transcription factor binding to G 1 and by its interaction with Cdx-2/3 as well as by mediating the inhibitory effects of insulin on glucagon gene transcription by binding to G 3 . Further work will be necessary to appreciate the functional roles of Pax-6 in glucagon gene regulation.