Autoregulation and Maturity Onset Diabetes of the Young Transcription Factors Control the Human PAX4 Promoter*

During pancreatic development, the paired homeodomain transcription factor PAX4 is required for the differentiation of the insulin-producing beta cells and somatostatin-producing delta cells. To establish the position of PAX4 in the hierarchy of factors controlling islet cell development, we examined the control of the human PAX4 gene promoter. In both cell lines and transgenic animals, a 4.9-kilobase pair region directly upstream of the human PAX4 gene transcriptional start site acts as a potent pancreas-specific promoter. Deletion mapping experiments demonstrate that a 118-base pair region lying approximately 1.9 kilobase pairs upstream of the transcription start site is both necessary and sufficient to direct pancreas-specific expression. Serial deletions through this region reveal the presence of positive elements that bind several pancreatic transcription factors as follows: the POU homeodomain factor HNF1α, the orphan nuclear receptor HNF4α, the homeodomain factor PDX1, and a heterodimer composed of two basic helix-loop-helix factors. Interestingly, mutations in the genes encoding four of these factors cause a dominantly inherited form of human diabetes called Maturity Onset Diabetes of the Young. In addition, PAX4 itself has at least two high affinity binding sites within the promoter through which it exerts a strong negative autoregulatory effect. Together, these results suggest a model in which PAX4 expression is activated during pancreatic development by a combination of pancreas-specific factors but is then switched off once PAX4 protein reaches sufficient levels.

During pancreatic development, the paired homeodomain transcription factor PAX4 is required for the differentiation of the insulin-producing beta cells and somatostatin-producing delta cells. To establish the position of PAX4 in the hierarchy of factors controlling islet cell development, we examined the control of the human PAX4 gene promoter. In both cell lines and transgenic animals, a 4.9-kilobase pair region directly upstream of the human PAX4 gene transcriptional start site acts as a potent pancreas-specific promoter. Deletion mapping experiments demonstrate that a 118-base pair region lying approximately 1.9 kilobase pairs upstream of the transcription start site is both necessary and sufficient to direct pancreas-specific expression. Serial deletions through this region reveal the presence of positive elements that bind several pancreatic transcription factors as follows: the POU homeodomain factor HNF1␣, the orphan nuclear receptor HNF4␣, the homeodomain factor PDX1, and a heterodimer composed of two basic helix-loop-helix factors. Interestingly, mutations in the genes encoding four of these factors cause a dominantly inherited form of human diabetes called Maturity Onset Diabetes of the Young. In addition, PAX4 itself has at least two high affinity binding sites within the promoter through which it exerts a strong negative autoregulatory effect. Together, these results suggest a model in which PAX4 expression is activated during pancreatic development by a combination of pancreas-specific factors but is then switched off once PAX4 protein reaches sufficient levels.
Organogenesis of the mammalian pancreas and its differentiation into discrete populations of endocrine and exocrine cells depends on the tightly regulated temporal and spatial expression of an array of transcription factors (for reviews see Refs. [1][2][3]. Some of these factors, such as HNF3␤ (4,5), PDX1 (6,7), and HLXb9 (8,9), are involved in the formation and outgrowth of the dorsal and ventral pancreatic buds. Within the forming pancreas, the basic helix-loop-helix transcription factor neuro-genin3 functions as a pro-endocrine gene, initiating the program of endocrine differentiation in selected cells (10,11). The complete differentiation and maturation of the endocrine cells requires additional factors, including neuroD1/BETA2 (12), NKX2.2 (13), ISL1 (14), and PAX6 (2,15).
Once neurogenin3 expression sets a progenitor cell on a course of endocrine differentiation, additional factors are necessary to determine which of the four endocrine cell types it will become. One of these islet cell-type determination factors is the paired homeodomain transcription factor PAX4. PAX4 functions as a potent transcriptional repressor and is expressed only transiently in the fetal pancreas, peaking during the period of beta and delta cell differentiation (16). Mice with a targeted disruption of the pax4 gene have a marked decrease in beta and delta cells with a commensurate increase in alpha cells (17), suggesting that PAX4 functions by repressing the alpha cell differentiation program and thereby deviates cells from a default alpha cell fate to an alternate beta or delta cell fate. In this model, once beta/delta cell type fate has been set, PAX4 expression is switched off.
To understand the mechanisms that control the expression of PAX4 and thereby impact endocrine cell type fate decisions, we mapped and characterized the regulatory regions of the human PAX4 gene. We found that a small promoter element lying approximately 1.9 kb 1 upstream from the transcription start site of the PAX4 gene is both necessary and sufficient to drive pancreas-specific expression. Remarkably, this sequence contains at least four functionally important binding sites for transcription factors, including four factors, HNF4␣, HNF1␣, pdx1, and neuroD1, that have been implicated in human beta cell dysfunction and early onset diabetes. Simultaneous exogenous expression of the above factors can stimulate the activity of the defined promoter element by 9-fold in a fibroblast cell line. In addition, the PAX4 gene promoter contains several binding sites for PAX4 itself, suggesting a model in which PAX4 inhibits its own expression, providing transient, selflimited expression of the PAX4 gene during pancreatic development.

5Ј-Rapid Amplification of cDNA Ends (RACE)-
The 5Ј ends of human and murine PAX4 cDNA were identified by 5Ј-RACE, using a modification of the protocol from the 5Ј-RACE System 2.0 (Life Technologies, Inc.). 2.5 pmol of either human PAX4-specific primer (5Јggacactcacccttcagctg-3Ј) or murine PAX4 primer (5Ј-tcacacagccattagata-3Ј) were annealed to 1 g of human islet RNA or murine E15.5 fetal pancreas RNA. Reverse transcription was carried out using Superscript II reverse transcriptase (Life Technologies, Inc.) according to the manufacturer's instructions. After first strand synthesis, the mRNA template was digested with RNase, and a homopolymeric tail added to the 3Ј end using terminal deoxynucleotidyltransferase. 35 cycles of primary PCR were used to amplify fragments, followed by 35 cycles of nested PCR according to the manufacturer's instructions, using Universal Amplification primers (Life Technologies, Inc.) in conjunction with PAX4-specific primers: human primer 1 (5Ј-ggcaattcgagccaccacag-3Ј) and primer 2 (5Ј-gcggatccaatgccctttggctccaag-3Ј); murine primer 1 (5Ј-ggctccgtgaaatgtcacag-3Ј) and primer 2 (5Ј-ttgctagctgcacaatctgc-3Ј). PCR products were subcloned into BluescriptII (Stratagene) and sequenced.
Cell Culture and Transfection-NIH3T3, ␣TC1.6, and ␤TC3 cells were cultured as described previously (18) and transiently transfected using TRANSFAST lipid reagent (Promega) according to the manufacturer's instructions. 2 g of DNA were transfected per million cells. Where expression vectors were cotransfected, 0.2 g were transfected per million cells. Cells were harvested and extracts assayed for luciferase activity as described previously (19) using 2 g of cellular protein extract.
The human PAX4 gene was obtained from human cosmid clone g1572c264, and the sequences between Ϫ4958 bp and ϩ46 relative to the transcription start site were subcloned upstream of the firefly luciferase gene in the pFOXLuc1 plasmid (20). Truncated forms of the PAX4 promoter and a full-length promoter with a deletion between Ϫ1963 and Ϫ1845 bp were amplified by PCR from the full-length promoter and confirmed by sequencing. The construction of the plasmid expressing luciferase under the control of the Rous sarcoma virus-long terminal repeat has been described previously (21).
Construction of pBAT12.mPAX4, the DNA plasmid expressing mouse PAX4 under the control of the human cytomegalovirus (CMV) immediate-early gene promoter, has been described previously (16). Mutations in the paired domain (S43P and R44P) and the homeodomain (N220P and R222P) in the mouse PAX4 cDNA were produced in the pBAT12.mPAX4 plasmid and confirmed by sequencing. Each mutation has previously been shown to disrupt DNA binding through the paired domain and homeodomain, respectively (22,23).
Two complementary oligonucleotides encompassing the pancreasspecific domain of the human PAX4 promoter (bp Ϫ1960 to Ϫ1909, 5Ј-gatctggcaagacctttgagttaatgtataattgtgagcagatggcggggg-3)Ј were ligated into BglII/BamHI sites upstream of the herpes simplex virus thymidine kinase (HSV-TK Ϫ109 to ϩ 51bp) promoter driving luciferase expression in the pFOXluc1 vector and confirmed by sequencing.
In Vitro Protein Expression-The construction of the in vitro expression vector containing the mouse PAX4 cDNA has been described previously (16). We used a carboxyl-terminally truncated PAX4 cDNA terminating at codon 274 because the synthesis of full-length PAX4 is less efficient but gives the same binding specificity (16); all results with truncated PAX4 were confirmed with full-length PAX4. The mouse PAX6 expression vector (pmPAX6) (24) was a gift from M. Busslinger, Research Institute of Molecular Pathology, Vienna, Austria. PAX4 and PAX6 proteins were produced by in vitro transcription/translation reaction, using the TNT-coupled reticulocyte lysate system (Promega, Madison, WI) according to the manufacturer's instructions.
Electromobility Shift Assays (EMSA)-Single-stranded oligonucleotide probes were 5Ј-end-labeled with [␥-32 P]ATP using T4 polynucleotide kinase. Labeled oligonucleotides were column-purified and annealed to an excess of the complementary strand. EMSA buffers and electrophoresis conditions were as described previously (18) using 1.5 g of poly(dI-dC)⅐poly(dI-dC) per lane. For PAX4 and PAX6 protein, 1 l of the 50-l in vitro translation reaction mix was added. Where antisera were used, 1 l was added to the binding reaction 10 min prior to probe addition. HNF4␣ and HNF1␣ antisera were purchased from Santa Cruz Biotechnology; PDX1 antisera were raised against bacterially expressed PDX1. Blocking peptides were added at a concentration of 1 g per 10-l binding reaction 10 min prior to the addition of probe. Unlabeled competing oligonucleotides were added at a 50-fold excess.
RT-PCR-RT-PCR was performed as described previously (16) using cDNA from mPAC. ␣TC1.6, ␤TC3, and NIH3T3 cell lines, adult rat liver and purified islets, and dated mouse embryonic pancreas. The primers used to amplify all cDNAs were separated by an intron to distinguish contaminating genomic DNA from cDNA. Primer sequences for HNF4␣ amplification were primer 1 (5Ј-tgccctctcacctcagcaatg-3Ј) and primer 2 (5Ј-cccctcagcacacggttttg-3Ј). Products were analyzed by agarose gel electrophoresis and ethidium bromide staining. The identity of PCR products was confirmed by cloning into the TA vector (Invitrogen) and sequencing.
Transgenic Mice-Both the wild-type full-length PAX4 promoter and the promoter containing a deletion between Ϫ1963 to Ϫ1845 were cloned into KpnI/XhoI sites upstream of the ␤-galactosidase gene in the plasmid p␤gal-Enhancer (CLONTECH). The plasmid was linearized and microinjected (1.5 ng/l) into murine pronuclei. Injected embryos were transferred to pseudopregnant females, and the fetal pancreata were harvested at E14.5. Tissues genotyping positive by PCR for the transgene were fixed for 30 min at 4°C in 4% paraformaldehyde and then incubated overnight in phosphate-buffered saline supplemented with 5-bromo-4-chloro-3-indolyl ␤-D-galactopyranoside (X-gal) (400 g/ ml) at 37°C. Tissues were then fixed again in 4% paraformaldehyde for 30 min, paraffin-embedded, sectioned at 5 m, and used for immunohistochemistry as described previously (2). Commercially available primary antibodies were used for staining of glucagon (dilution 1:10000) and insulin (1:8000) (Linco Research Immunoassay) according to the manufacturer's instructions. The rabbit antisera raised against mouse neurogenin3 has been described previously (25). Secondary antibodies were linked to alkaline phosphatase (insulin and glucagon) or peroxidase (neurogenin3) used at a dilution of 1:5000. Vector Red (Vector Laboratories) was used as a substrate for alkaline phosphatase staining.

RESULTS
Mapping the PAX4 Promoter-By using RNA purified from adult human islets and from embryonic day 15.5 (E15.5) fetal mouse pancreas, the pancreatic transcription start sites of the human and murine PAX4 genes were determined by 5Ј-RACE. Although not identical, the 5Ј-RACE products identify tightly clustered start sites and the same first intron in both genes (Fig. 1A). Interestingly the pancreatic transcripts contain an additional two exons compared with the transcript previously characterized from human placenta (26).
To establish whether the 5Ј-flanking sequences can act as a cell-specific promoter, approximately 5 kb of flanking sequence was amplified by PCR from human cosmid clone g1572c264. The full-length promoter (Ϫ4958 bp) and four 5Ј-truncated promoters were placed in front of a luciferase reporter gene and transfected into the pancreatic beta cell line ␤TC3, the pancreatic alpha cell line ␣TC1.6, and non-islet NIH3T3 cells (Fig. 1B).
Removal of the region between Ϫ4958 and Ϫ2153 bp increases reporter activity in all cell types, and deletion of sequences between Ϫ2153 and Ϫ1614 bp markedly decreases promoter activity in the pancreatic cell lines. An additional (although smaller) decrease in pancreatic promoter activity results from the deletion of sequence from Ϫ1614 to Ϫ1078 bp. Addition of 5Ј genomic sequences between ϩ46 bp and the translation start site at ϩ2543 bp gives no increase in promoter activity (data not shown).
Identification of a Pancreas-specific Promoter Element-Inspection of the promoter sequence between Ϫ2153 and Ϫ1614 bp reveals several potential binding sites for pancreatic transcription factors, all clustered between Ϫ1960 and Ϫ1861 bp ( Fig. 2A). These include a potential HNF4␣-binding site (27), four TAAT sequences that are commonly found in homeodomain protein-binding sites, and an E box, a recognition sequence for the binding of basic helix-loop-helix protein (bHLH).
To test the importance of these potential binding sites, a series of small deletions were made between Ϫ1960 and Ϫ1866 bp in the PAX4 promoter (Fig. 2B). In the ␤TC3 cells, removal of each of these sites in series, the potential HNF4␣ site (deletion Ϫ1947), TAAT1 and -2 (deletion Ϫ1930), the E box (deletion Ϫ1909), and TAAT3 and -4 (deletion 1866), produces a stepwise reduction in promoter activity. By contrast, in ␣TC1.6 cells, only the last of these deletions (TAAT3 and -4, Ϫ1866) significantly reduces promoter activity.
We also deleted the sequences between Ϫ1963 and Ϫ1845 bp in the intact PAX4 promoter to test the importance of the pancreas-specific element in the context of the full-length promoter. In the ␤TC3 cells, this deletion completely disables the promoter (Fig. 3A). In addition, a single copy of the 5Ј portion of the pancreas-specific element (Ϫ1960 to Ϫ1909 bp) can function as a pancreas-specific enhancer when linked to a heterol-ogous promoter, the TK promoter from the herpes simplex virus (Fig. 3B). Although the enhancement given by the pancreas-specific element is small (2-fold), this represents a significant increase above the high basal activity of the TK promoter. With two copies of the (Ϫ1960 to Ϫ1909) region ligated 5Ј of the TK promoter, reporter gene activity increases to 12-fold above the basal level in ␤TC3 but not 3T3 cells (Fig. 3B).
Activity of the Pancreas-specific Element in Vivo-Whereas the pancreas-specific element may be essential for the function of the PAX4 promoter in ␤TC3 and ␣TC1.6 cells, these tumor cells may not be representative of the cells in the developing pancreas where PAX4 is predominantly expressed. Therefore, we produced mice carrying a transgene with the Ϫ4958-bp FIG. 1. Mapping the PAX4 promoter. A, the human PAX4 gene sequence is shown with the transcriptional start sites for murine fetal pancreas (q), human adult pancreatic islets (f), and human placenta (Ϫ). The translational start site is indicated in bold, and the intron sequence is shown in lowercase, and the exon sequence is shown in uppercase. B, the activity of various lengths of the human PAX4 promoter designated by their 5Ј ends and driving a luciferase reporter gene were transfected into the indicated cell types. Resulting luciferase activity is expressed relative to the activity of the promoterless expression vector. The mean Ϯ S.E. is shown for transfections performed in duplicate on at least three separate occasions. human PAX4 promoter driving the bacterial gene encoding ␤-galactosidase. Initial analysis of mice harvested at E12.5 reveals that ␤-galactosidase expression is restricted to the pancreas (Fig. 4).
Additional founder mice were harvested at embryonic day 14.5, 24 h after the peak of PAX4 mRNA expression in the fetal pancreas (16) for further analysis (Fig. 5). Out of 12 independent transgenic founders, strong ␤-galactosidase activity was detected in the pancreas of 8. In contrast, in a parallel experiment, no ␤-galactosidase activity was detectable in 9 founders using the Ϫ4958-bp promoter with the sequences between Ϫ1963 and Ϫ1845 bp (the pancreas-specific element) deleted.
In the transgenic mice carrying the wild-type promoter construct, we used immunohistochemistry to identify the cells expressing ␤-galactosidase. The ␤-galactosidase expressing cells are located predominantly in the central area of the pancreas where the main ducts lie and endocrine cell genesis normally occurs. Most of the insulin-expressing cells at this stage also express ␤-galactosidase, but there are also many non-insulin-expressing cells that express ␤-galactosidase. Oc-casional cells co-express glucagon and ␤-galactosidase; but these cells are uncommon; and most alpha cells do not express ␤-galactosidase. Interestingly, not all ␤-galactosidase expressing cells express pdx1, and a subset express neurogenin3, a basic helix-loop-helix transcription factor and a marker for islet cell precursors (11,25). Together these results suggest that the PAX4 promoter is active in beta cell precursors and possibly occasionally in alpha cell precursors as well.
Multiple Factors Bind to the Pancreas-specific Element-To identify nuclear factors that bind to the pancreas-specific element in the PAX4 promoter, we synthesized a series of oligonucleotides spanning the transcriptionally active region and tested them for binding to nuclear proteins by electromobility shift assay (EMSA).
An oligonucleotide (H4) encompassing the putative HNF4␣binding site binds to a low mobility complex present in liver and ␤TC3 cell nuclear extracts. The mobility of this complex from both liver and ␤TC3 cells is retarded (supershifted) by the addition of antisera directed to the carboxyl terminus of mouse HNF4␣, an effect that is prevented by the addition of excess HNF4␣ carboxyl-terminal peptide (Fig. 6A). A similar complex and supershifted complex is not seen in nuclear extracts from ␣TC1.6 cells (data not shown). In addition, in vitro produced HNF4␣ protein readily binds to the H4 oligonucleotide (data not shown).
These results demonstrate that HNF4␣ is present in the ␤TC3 cell line and binds to the H4 site. HNF4␣ expression has previously been demonstrated in the pancreas but with unspecified cellular distribution (28,29). To assess the expression of HNF4␣ in the fetal pancreas when the PAX4 gene is actively transcribed, we gauged HNF4␣ mRNA expression by RT-PCR and found that HNF4␣ mRNA peaks in the pancreas at E13.5 (Fig. 6B), in parallel with the peak of PAX4 mRNA (16). Unlike PAX4, however, HNF4␣ mRNA persists in the adult islet at detectable levels. Two HNF4␣-derived PCR products of very similar size were detected in the products depicted in Fig. 6B; these were sequenced and represent two previously reported alternate splice forms of HNF4␣ mRNA (30). Interestingly, no HNF4␣ mRNA was detected in ␣TC1.6 cells consistent with the observation that removal of the H4 site had no effect on PAX4 promoter activity in these cells.
An oligonucleotide (oligonucleotide TAAT 1/2) containing the 5Ј two TAAT sequences binds three major complexes (Fig. 7A). The lowest mobility complex present in pancreatic cell types is shifted by HNF1␣ antiserum (Fig. 7A). The highest mobility complex in ␤TC3 extracts is recognized by pdx1 antiserum (Fig.  7B). The ␣TC1.6 nuclear extract produces a complex with the same mobility but less intensity than the PDX1 band in ␤TC3 extracts, but this complex is not recognized by the pdx1 antiserum (data not shown).
The E box in the pancreas-specific element from the PAX4 promoter is identical to the conserved 8-base pair sequence of the E boxes present in the human and rat insulin promoters. As shown in Fig. 7C, identical slow migrating complexes present in both ␣TC1.6 and ␤TC3 extracts bind to both the human insulin promoter E1 element and the PAX4 promoter E element (oligonucleotide E). This complex has previously been shown to consist of a heterodimer containing ubiquitously expressed bHLH proteins such as E2A gene products (31-34) and neuroD1/BETA2 (35), a protein of limited tissue distribution.
Finally, an oligonucleotide (oligonucleotide TAAT3/4) containing the two 3Ј TAAT sequences binds three complexes recognized by pdx1 antiserum in ␤TC3 extracts (Fig. 7D). Whereas one of the two more slowly migrating complexes most likely results from PDX1 binding to both TAAT sites, the other band probably is produced by interactions with an additional protein, such as PBX1 (36).
Activation of the Ϫ1960 to Ϫ1909 Promoter Region in a Non-pancreatic Cell Line-To test the ability of the identified transcription factors to activate the pancreas-specific element FIG. 3. Elements between ؊1963 and ؊1845 are necessary and sufficient to activate transcription. A, the full-length PAX4 promoter, a similar construct containing a 118-bp deletion (⌬-1963 to Ϫ1845 bp) and a truncated promoter fragment were transfected into ␤TC3 cells. Resulting luciferase activity is expressed relative to the activity of the promoterless expression vector. B, PAX4 promoter sequences from Ϫ1960 to Ϫ1909 were ligated upstream of the HSV-TK gene promoter driving the luciferase reporter gene. The resulting construct and the parent vector lacking the PAX4 promoter sequences were transfected into the indicated cell line. Resulting luciferase activity is expressed relative to the activity of the promoterless expression vector. The mean Ϯ S.E. is shown for transfections performed in duplicate on at least three separate occasions. in a non-pancreatic cell line, we expressed each of the factors in 3T3 fibroblast cells along with a luciferase reporter construct driven by two copies of pancreas-specific element upstream of the minimal TK promoter. Although no single cDNA activates the pancreas-specific element (Fig. 8A), a combination of the transcription factors can activate the construct up to 9-fold (Fig. 8B). We also tested the ability of neurogenin3 to replace neuroD1 in this combination of factors because neurogenin3 is expressed abundantly in some progenitor cells that express PAX4 in vivo (Fig. 5). Neurogenin3 and neuroD1 are closely related, and when dimerized with Pan1/E47 both bind to the same E box consensus sequence (37). 2 Although the complex in ␤TC3 nuclear extract that binds to the E box in the pancreasspecific element does not include neurogenin3 (data not shown), neurogenin3 is expressed at much lower levels in ␤TC3 cells than in the embryonic pancreas ( Fig. 5 and data not shown). Expressing neurogenin3 in place of neuroD1 in the transcription factor combination results in a further in-crease in the activity of the pancreas-specific element in 3T3 cells (Fig. 8B).
Autoregulation of the PAX4 Promoter-Inspection of the PAX4 promoter sequence reveals a number of sequence elements that match or closely resemble the consensus DNA binding sequence for the PAX4 protein (38,39), including both paired domain and homeodomain binding motifs. To test whether the PAX4 promoter is autoregulated, the PAX4 promoter reporter plasmid was co-transfected with a plasmid vector expressing the PAX4 protein (Fig. 9A). PAX4 protein represses the activity of the full-length promoter, and this effect is most remarkable in ␣TC cells, consistent with previous evidence that PAX4 repression is most fully realized in ␣-cell lines (16). Removal of 5Ј sequences from the promoter reduces the repression by PAX4. This repression depends on an intact homeodomain and to a lesser extent on the paired domain (Fig. 9B). FIG. 6. HNF4␣ binds to the pancreas-specific element. A, EMSA using the H4 DNA probe is shown. Probe was incubated with 2 g of nuclear extract from rat liver or ␤TC3 cells. 1 l of antiserum and/or 5 l of blocking peptide were preincubated with the samples as indicated prior to probe addition and electrophoresis. B, RT-PCR was performed using 250 ng of RNA derived from rat liver, cell lines, or mouse embryonic pancreas. HNF4␣ or ␤-actin cDNA were amplified for 30 cycles. PCR products were analyzed by agarose gel electrophoresis and ethidium bromide staining. FIG. 7. Multiple factors bind to the pancreas-specific element. A and B, EMSA using the TAAT1/2 probe is shown. Probe was incubated with 2 g of nuclear extract from the indicated cell lines. Samples were preincubated with 1 l of HNF1␣ antiserum and/or 5 l of blocking peptide (A) or with 1 l of PDX1 antiserum or preimmune serum (B) as indicated prior to probe addition and electrophoresis. C, EMSA using the E probe and the E2 element of the rat insulin I promoter is shown. Probes were incubated with 2 g of nuclear extract from the indicated cell line.
Two oligonucleotides (oligonucleotides P4.1 and P4.2) corresponding to PAX4 sites located at bases Ϫ4611 and Ϫ4148 of the PAX4 promoter bind PAX4 (Fig. 9C), with an affinity comparable to the C2 element of the insulin promoter. In contrast to the C2 element (16,39,40), the PAX4-binding sites of the PAX4 promoter show a significantly higher affinity for PAX4 than for PAX6. DISCUSSION In the present study, we have shown that the human and mouse PAX4 genes have adjacent transcription start sites and use the same first two exons in the developing and mature pancreas. It should be noted, however, that a previous study using human placental mRNA identified a more 3Ј transcription start site without the first two exons and introns described here (26). These differences most likely result from the use of alternate promoters in distinct tissues. A previous study of PAX4 mRNA in the rat pancreas identified alternate splice forms, but none of these differences involved the 5Ј end (41).
The experiments using the 5Ј-flanking sequence from the PAX4 gene confirm the presence of an active, pancreas-specific promoter driving expression from the mapped start sites. Although this promoter lacks a consensus TATAA element, there Competition was performed with a 50-fold excess of the indicated unlabeled oligonucleotide. D, EMSA using the TAAT3/4 probe is shown. Probe was incubated with 2 g of nuclear extract from the indicated cell line. Samples were preincubated with 1 l of HNF1␣ antiserum and/or 5 l of blocking peptide as indicated prior to probe addition and electrophoresis.
FIG. 8. A combination of transcription factors activates the PAX4 promoter region ؊1960 to ؊1909 in the NIH3T3 fibroblast cell line. A, NIH3T3 cells were transfected with a reporter construct containing 2 copies of the (Ϫ1960 to Ϫ1909)) PAX4 promoter region located 5Ј to the HSV-TK promoter, upstream of the firefly luciferase (LUC) gene (2 g per million cells). Additionally the cDNA for the indicated transcription factor (ev is an expression vector with no cDNA) was cotransfected under the control of the CMV promoter (0.1 g per million cells). Relative luciferase activity is shown as a multiple of that from the promoterless reporter construct. B, transfections were performed as in A, except a combination of transcription factors were co-transfected together (HNF4␣2, HNF1␣, PDX1, and PAN1) with either NeuroD1 or neu-rogenin3 as indicated.
are two CCAAT sequences adjacent to the transcription start sites, in keeping with observations of other TATAA-less promoters (42). A more distal, pancreas-specific element, however, contributes an essential component to the overall activity of the promoter.
Interestingly, the pancreas-specific element has remarkable similarities to the E2-A3/4 element (previously called the Far-FLAT or FF minienhancer) present in the rat insulin I promoter (18,43). Like the E2-A3/4 element in the insulin promoter, the pancreas-specific element is required for full activity of the PAX4 promoter and is also sufficient to drive pancreasspecific transcription when linked to a heterologous promoter. But it is the similarity in DNA-binding sites that is most remarkable. The two elements contain identical sequences that can function as binding sites for bHLH proteins. In ␤TC3 cells, these sites bind a heterodimer of two bHLH proteins, a ubiquitous class A bHLH protein such as E47 (31)(32)(33)(34)44) and the neuroendocrine-specific class B bHLH protein neuroD1/BETA2 (35). In the developing pancreas, neurogenin3 may replace neuroD1/BETA2 in the heterodimer binding to this sequence. 3 Adjacent to the bHLH-binding sites in both elements lie nearly identical AT-rich sequences that function as binding sites for both the pou-homeodomain protein HNF1␣ (45) and the pancreatic/duodenal homeodomain protein PDX1 (46 -48).
But if the same factors control insulin and PAX4 gene expression, why do the two genes show marked differences in their temporal expression pattern? In the developing pancreas, PAX4 mRNA levels peak at E13.5, when insulin mRNA levels are just starting to rise, and PAX4 expression is shut off by the time insulin mRNA reaches its peak (16). This pattern is confirmed in individual cells in the transgenic mice with the PAX4 promoter driving ␤-gal, since ␤-gal is expressed in many cells that do not express insulin and is co-expressed in neurogenin3 expressing cells that represent progenitor cells in which insulin is not yet expressed (25). This pattern of PAX4 activation in non-hormone-expressing progenitor cells is consistent with a role for PAX4 in beta/delta-cell type determination.
The differences in insulin and pax4 gene expression suggest that additional, distinct factors play essential roles in the expression of the two genes. HNF4␣ could be one such factor, since the pancreas-specific element from the PAX4 promoter contains an HNF4␣-binding site that is not present in the rat insulin I promoter, and HNF4␣ mRNA shows an expression pattern that is similar to PAX4 mRNA. In addition, the deletion experiments demonstrate that sequences immediately downstream from the bHLH-binding site contribute to PAX4 promoter activity in both ␣TC and ␤TC cells. Although there are at least two PDX1-binding sites in this region, it seems unlikely that PDX1 alone can explain the activity, since the ␣TC cells do not express PDX1. Other factors, possibly including other homeodomain factors, must explain this activity in ␣TC cells and likely in progenitor cells in vivo as well. Finally, although the pancreas-specific element is critical for the overall activity of the PAX4 promoter, and it can function on its own, 3 V. Schwitzgebel, H. Watada, and M. German, unpublished data.
transfected into ␣TC1.6 cells and co-transfected with pBAT12, pBAT12 expressing wild-type PAX4 cDNA, or pBAT12 expressing PAX4 cDNA carrying a mutation in the homeodomain (N220P and R222P) or paired domain (S43P and R44P). Resulting luciferase activity is expressed relative to the activity of the promoterless expression vector. The mean Ϯ S.E. is shown for transfections performed in duplicate on at least three separate occasions. C, EMSA using the rat insulin I promoter C2 site and the human PAX4 P4.1 and P4.2 probes is shown. Probes were incubated with 1 l of in vitro produced PAX4, PAX6, or luciferase (control).
FIG. 9. The PAX4 promoter is subject to autoregulation by a mechanism requiring DNA binding. A, luciferase reporter vectors driven by the indicated PAX4 or Rous sarcoma virus promoters were transfected into the cell line shown. Cells were co-transfected with either the PAX4 expression vector pBAT12.mPAX4 which produces full-length PAX4 protein under the control of the CMV promoter or the parent vector pBAT12 without cDNA insert. Relative luciferase activities were calculated with the activity of cells transfected with the parent pBAT12 vector without cDNA insert set at 1. In order to highlight differences in repression, data are shown as fold repression, the inverse of relative luciferase activity. B, the long PAX4 promoter (Ϫ4958 bp) upstream of the luciferase reporter gene in pFOXLuc1 was it may require cooperative interactions with other sequences binding to distinct factors in order to realize its full activity.
One of the most interesting aspects of the PAX4 gene is its capacity for auto-repression. If PAX4 functions solely to determine islet cell type fate during the differentiation of the endocrine pancreas, then persistent expression of PAX4 after this decision may be detrimental, possibly by repressing insulin gene expression (16,40). As has been shown for the Drosophila homeobox gene goosecoid, auto-repression is one mechanism to prevent persistent expression (49). Such a model allows for the rapid activation and subsequent extinction of PAX4 gene expression observed during pancreas development. Of course other mechanisms would be required to keep the gene off, and our data demonstrate that portions of the promoter can repress transcription in the absence of PAX4 protein.
But if PAX4 limits its own expression, why does it persist in the beta-cell tumor line ␤TC3? Tumor lines do not reflect completely normal beta cell biology, and in addition we have not consistently found PAX4 in other beta cell tumor lines (data not shown). The unique inability of ␤TC3 cells to extinguish PAX4 gene expression could result from differences in expression or function of co-repressors that normally cooperate with the PAX4 protein in affecting transcriptional repression. Evidence that PAX4 functions poorly as a transcriptional repressor in ␤TC3 cells supports this explanation (16). Alternatively, persistent low level PAX4 gene expression may allow for a balance of PAX4 gene expression and feedback repression by PAX4 protein. This pattern is quite different from the rapid accumulation of PAX4 protein and subsequent repression of PAX4 gene expression that normally occurs in the developing pancreas.
Finally, the presence of HNF4␣, HNF1␣, PDX1, and neu-roD1 among the proteins binding to the active region of the PAX4 promoter is intriguing given the role of these proteins in human diabetes. Mutations in the genes encoding any of these four transcription factors can cause a form of autosomal dominant diabetes in humans called maturity onset diabetes of the young (50,51). Although studies in mice carrying targeted mutations in the genes encoding HNF4␣ and HNF1␣ suggest that secondary defects in the expression of genes involved in glucose sensing may cause the insulin secretory defects observed in patients with MODY1 (52) and MODY3 (53), defects in beta cell development could also contribute to the eventual beta cell deficiencies seen in the human syndromes. By understanding the hierarchy of gene regulation events that control the development of the pancreas and the differentiation of the insulin-producing beta cells eventually may provide new insights into the inherited defects that cause insulin deficiency and diabetes.