Transcriptional Regulation of the Human Carboxyl Ester Lipase Gene in Exocrine Pancreas

The human carboxyl ester lipase (CEL) is an important enzyme for the intestinal absorption of dietary lipids. The gene is highly expressed in exocrine pancreas and in the mammary gland during pregnancy and lactation. In this paper, we have focused on its transcriptional regulation in exocrine pancreas. Reporter gene analysis in cell cultures reveals that a high level of tissue-specific expression is established by the proximal 839 base pairs of the 5′-flanking region. This is due to a strong enhancer, located at −672 to −637. Transfections in mammary gland-derived cells reveal that the enhancer is pancreas-specific and does not contribute to the mammary gland expression. This indicates that the expression of the CEL gene in the mammary gland and pancreas, respectively, is due to two different regulatory systems. Further characterizations of the enhancer reveal that it is composed of two closely located cis-elements. The proximal element mediates a positive effect, whereas the distal element exerts a silencing effect on the positive proximal element. The functional enhancer complex is composed of ubiquitously expressed factors, since similar interactions are achieved with nuclear extracts from cells derived from other tissues. However, no enhancer activity is achieved in such cells. Hence, the net enhancer activity is the result of a tissue-specific balance between factors interacting with the two elements. Since none of the described cis-elements show any clear homology to known cis-elements, we propose that the interacting complex is composed of yet unidentified transcription factors.

The need for fat is of special importance for infants, since malabsorption would influence both growth and development (1). To secure the requirement, several different lipolytical enzymes, secreted from the pancreas, contribute to the digestion of dietary fat. One of the important enzymes is carboxyl ester lipase (CEL), 1 since it, because of its broad substrate specificity, is believed to be the main enzyme for the hydrolysis of fat-soluble vitamin esters, monoglycerides, and cholesterol esters (2,3). CEL is synthesized in the acinar cells of exocrine pancreas, secreted with the pancreatic juice, and activated in the intestinal lumen by bile salts (for reviews, see Refs. 4 and 5). CEL seems to be a very old enzyme. A pancreatic expression has been found in all vertebrates analyzed so far (6,7). In fact, in the pancreas of leopard sharks, CEL is the only lipolytic enzyme for hydrolyzing dietary lipids (8), which suggests that CEL, because of its broad substrate specificity, might even have been one of the first lipolytic enzymes expressed in the pancreas. The importance of CEL for intestinal absorption of dietary lipids is strongly supported by the fact that in mammals it is also synthesized in lactating mammary gland, secreted into the milk, and activated in the intestine (9,10). Since the capacity of human infants to digest milk lipids is poor during the first months, due to poorly developed pancreatic function (11), the presence of CEL in milk seems to secure fat absorption. After cloning of the CEL cDNA and gene, respectively, we and other groups could show that there is only one gene responsible for the expression in the two tissues (12)(13)(14)(15). Neither the enzyme nor the gene shows any homology to the classical lipases or their corresponding genes; instead, homologies are found to esterases, such as acetylcholine esterase (12), which might indicate that the CEL gene has evolved as a separate lipase/esterase gene.
Besides the predominant CEL gene expression in mammary gland and pancreas of humans it has recently been shown that small but significant levels of expression could be detected in other tissues, such as macrophages and liver and endothelial cells. The function of this expression is still unknown, but it has been suggested that CEL interacts with cholesterol esters and oxidized lipoproteins and thereby modulates the progression of atherosclerosis (16,17). Furthermore, there seem to be differences with respect to the tissue-specific expression among different species (18). This might be explained by differences that have taken place during the evolution of the CEL gene locus. For example, in humans, the CEL gene has been duplicated, and the original CEL gene, the mouse homologue, has become a pseudogene (19).
The digestive enzymes have been in focus for a long time. As proposed by Pavlov in the early 1900s (20) and first shown by Grossman et al. (21), the variations in levels between different secreted digestive enzymes are controlled by changes in the levels of circulating hormones in response to food composition (22). This response is fast and dependent on translational regulation and increased secretion of stored enzymes (23,24). Changes in mRNA levels can only be induced following prolonged stimulation by nutrients (22). The regulation of the CEL gene activity in pancreas has been studied by Huang and Hui (25), who have shown that the pancreatic cell line AR4-2J constitutively expresses CEL and that this protein expression can be induced by cholecystokinin and secretin without any changes in mRNA level. Further investigations have indicated that Ca 2ϩ and protein kinase C are part of the signal mechanism by which the acinar cells are induced to secret CEL (26). The same group was also able to show that a high fat/high cholesterol diet exhibited increased CEL mRNA levels in rats (27) and that cationizied low density lipoprotein can induce a 2-fold increase in the CEL mRNA level in AR4-2J cells (28). A similar induction by prolonged cholesterol treatment has recently also been shown to occur in rabbits (29). The ability to respond to cholesterol has been shown for several other genes such as the low density lipoprotein receptor gene (30) and the cholesteryl ester transfer protein gene (31). However, no response element has yet been reported to mediate cholesterol induction of gene activation. Hence, studies of the CEL gene may also provide important insight into cholesterol-induced gene activation.
During the last decade, the tissue-specific regulation of genes expressed in exocrine pancreas has been extensively studied, and the transcriptional regulation has been characterized to some extent. Most work has concerned genes encoding proteases and amylases, in which the pancreatic transcription factor 1 (PTF-1) seems to play a major role (32,33). Almost nothing is known about the transcriptional regulation of the corresponding lipase genes. Recently, some reports have indicated that the amount and type of fat can modulate the rate of transcription of pancreatic lipase (34,35). Furthermore, PTF-1 seems to be involved in the regulation of the pancreatic colipase gene (36,37).
The high expression level in two different tissues, constitutively in pancreas and induced during lactation in the mammary gland, makes the CEL gene an interesting system for studying tissue-specific regulation. In this first report, we have, by using reporter gene analysis of promoter constructs in the rat pancreatic cell line AR4-2J, analyzed the basic structure of the transcriptional regulatory machinery responsible for the expression of the human CEL gene.

EXPERIMENTAL PROCEDURES
Construction of Reporter Plasmids-The proximal part of the human CEL gene promoter was amplified by polymerase chain reaction from the pTZ19R subclone b (15) with the use of two 25-mer oligonucleotides. The distal oligonucleotide includes the SphI site at nucleotide (nt) Ϫ384, and the other introduces a restriction enzyme (RE) site just in front of the translation start at nt ϩ13. After digestion, the fragment was subcloned into pTZ19R and sequenced. The promoter region in the subclone was then further elongated to nt Ϫ4740 by stepwise insertion of RE fragments from the BSSL1 clone (15). The polylinker in the luciferase reporter plasmid pGL-2 basic (Promega) was changed, and an upstream mouse sequence (UMS) (38) was inserted immediately upstream from the new polylinker (the UMS-Luc construct). In this vector, the subcloned CEL promoter fragment was inserted (the CEL4740Luc construct). All other constructs were made by deletions of this construct, either by restriction enzymes or ExoIII nuclease digestion. Point mutations were made by the Stratagene Quikchange site-directed mutagenesis kit. The number in the name of each construct indicates the absolute content of CEL gene 5Ј-flanking sequence present in the construct.
Cell Culture and Transfection-The rat pancreatoma cell line AR4-2J and the RAT2 cell line, respectively, were obtained from ATCC. The mouse mammary epithelial cell line HC11 was kindly provided by Dr. R. Ball, Friedrich Miescher-Institut, Basel. The AR4-2J cells were cultured at 37°C in a 5% CO 2 , 95% air atmosphere in DulbeccoЈs modified EagleЈs medium containing 2 mM glutamine and 4.5 g/liter glucose and supplemented with 10% fetal calf serum and 1% penicillin/ streptavidin. Cell media were changed every second day, and the cultures were split 1:3 every fourth day. The AR4-2J cells form and grow in clumps and therefore never reach confluence. This forced us to modify the standard transfection techniques. The AR4-2J cells were seeded into 6-cm cell culture dishes, and instead of waiting with the transfection until the next day, which would have resulted in clumps, the cells were transfected after 5-7 h, when they had just attached to the dishes. This also allowed us to use a high density of cells. The cells were washed once with serum-free medium and transfected with a mixture of 30 l of Lipofectin and 8 g of promoter construct in serumfree medium. To correct for variations in transfection efficiency, 2 g of pCMV-CAT reporter plasmid was co-transfected as an internal control. After incubation for 12-15 h, fresh medium was added, and after an additional 25-h incubation, the cells were harvested. The RAT2 cells were maintained in the same way as the AR4-2J cells with the exception that they were split 1:6. The HC11 cells were maintained as the RAT2 cells with the exception that the medium was RPMI 1640 and the culture flasks were collagen-coated. The RAT2 and HC11 cells, respectively, were transfected in the same way as the AR4-2J cells. For stably transfecting the HC11 cells, approximately 0.5 ϫ 10 6 cells were seeded into 35-mm culture dishes and transfected with 30 l of Lipofectin, 10 g of reporter plasmid, and 1 g of pTX-1 neo expression vector. Two days after transfection, 350 g/ml G418 was added and selection was performed. Resistant colonies were pooled and expanded. Hormone induction was performed as described previously (39).
Cell Extract and Reporter Gene Assays-The cells were washed once with ice-cold 1ϫ Tris-buffered saline. After that, 200 l of ice-cold 0.5ϫ Tris-buffered saline was added, and the cells were scraped together and lysed by three freeze/thaw cycles. A luciferase assay was performed using the Promega kit with 50 l of extract and assayed in a luminometer (Berthold). After removal of endogenous deacetylation activities by heat inactivation, 50 l of extract were assayed for CAT activities using the diffusion assay and an automatic scintillation counter (Beckman) as described previously (40). CAT activity was calculated as cpm/h. The luciferase activities were normalized to equivalent CAT activities to compensate for variations in transfection efficiency. Each construct was transfected on duplicate dishes, and average luciferase activities were calculated and represented as mean Ϯ S.E. based on a minimum of three independent transfections. For the stably transfected HC11 cells, the luciferase activity was normalized to equal protein concentration of each extract.
Nuclear Protein Preparation-Protease inhibitors (Complete ® , Pepstatine, L-1-tosylamido-2-phenylethyl chloromethyl ketone, 1-chloro-3tosylamido-7-amino-2-heptanone, phenylmethylsulfonyl fluoride, benzamidine, aprotinin, chymostatin, and soy trypsin inhibitor, all from Boehringer Mannheim) were included in the buffers. Nuclear protein extracts from the cell lines were prepared according to current protocols (41) with some modifications. Briefly, cells were homogenized in H buffer (10 mM Hepes, pH 7.9, 10 mM KCl, 0.75 mM spermidine, 0.15 mM spermine, 0.1 mM EDTA, 0.1 mM EGTA, 2 mM dithiothreitol), nuclei were extracted in E buffer (20 mM Hepes, 0.4 M KCl, 0.75 mM spermidine, 0.15 mM spermine, 0.2 mM EDTA, 0.2 mM EGTA, 20% glycerol, 2 mM dithiothreitol), and the extracts were dialyzed in D buffer (20 mM Hepes, 100 mM KCl, 0.2 mM EDTA, 0.2 mM EGTA, 20% glycerol, 2 mM dithiothreitol). Preparation of nuclear extracts from rat tissue was performed as above with some modifications. Briefly, the pancreas was rapidly frozen in liquid nitrogen and pulverized by a mortar. The pulverized tissue was then dredged in 10ϫ volumes of H buffer and homogenized in a glass-Teflon Yamato homogenizer (1500 rpm, 10 -15 strokes). The homogenate was filtered through gauze, and the crude nuclei were pelleted and resuspended in 10 package nuclei volumes of H buffer plus 0.3 M sucrose and purified over a 1.7 M sucrose bed. The nuclei were then resuspended in 5 package nuclei volumes of H buffer and extracted in the same way as for cell cultures. Protein concentrations were determined using the Bradford assay (Bio-Rad).
EMSA and DNase I Footprinting-Oligonucleotides for EMSA were made on an oligonucleotide synthesizer (Beckman) as trityle-on and purified on a fast protein liquid chromatography column (pepRPC column, reverse phase) (Amersham Pharmacia Biotech). After annealing, 5 pmol of double-stranded oligonucleotide was labeled with ␣-32 P by the use of Klenow fill in and purified on a 0.3-mm, 10% polyacrylamide gel. For each binding reaction, 3 g of poly(dI-dC), 1.5 g of bovine serum albumin, 3 l of 5ϫ binding buffer (0.1 M Hepes, pH 7.9, 250 mM KCl, 10 mM MgCl 2 , 5 mM EDTA, 50% glycerol, 2.5 mM dithiothreitol), 4 g of nuclear extract, and competitor was mixed in a volume of 15 l on ice. An approximately 15,000 cpm probe was added, and the reaction was incubated at room temperature for 30 min. After incubation, 1 l of loading buffer was added, and the reactions were loaded on a 1-mm, 4% polyacrylamide gel (Tris/glycine, 5% glycerol) and electrophoresed at 250 V in a cold room.
For the footprint assay, subcloned fragments were digested and end-labeled by ␣-32 P using Klenow fill in. After a second digestion, the labeled fragments were purified on a 2% agarose gel using DEAE-membranes. Footprint reactions were performed as described by Jones et al. (42), using 20 g of nuclear extract and 15,000 cpm of fragment for each reaction.

RESULTS
The Proximal 839 bp of the 5Ј-Flanking Region Is Sufficient for Tissue-specific Expression-To localize promoter regions required for tissue-specific transcriptional activity of the human CEL gene in exocrine pancreas, a reporter gene analysis was performed. 5Ј-Deletion constructs were generated by progressive deletions of the 4740-bp flanking region, including the first 11 bp of exon 1 (relative the transcription initiator site), and fused to a promoterless luciferase reporter vector. The resulting 5Ј-deletion series were transiently transfected into the rat pancreatic cell line AR4-2J and the rat fibroblast cell line RAT2, and promoter activities were deduced from a luciferase assay. To normalize for differences in transfection efficiency, co-transfections with a cytomegalovirus-CAT reporter plasmid were performed, and the relative promoter activity was calculated as luciferase activity/CAT activity.
The AR4-2J cell line is of exocrine origin and has previously been reported to express CEL (25) and was therefore expected to contain adequate levels of trans-acting factors to support the activation of the chimerical CEL promoter-reporter gene constructs. The rat fibroblast cell line RAT2 was used as a negative control, since no CEL gene expression could be detected in these cells. 2 Data revealed that a stepwise increase in expression level is correlated with increasing promoter length (Fig. 1). A minimal increase in activity was observed with constructs increasing in length up to Ϫ653. Extension of the promoter up to Ϫ839 results in a highly significant increase in transcriptional activity, approximately 18 times. This suggests that an important positive cis-regulatory element is present within this region. Further extension of the promoter, up to Ϫ1356 and Ϫ3200 increases the luciferase expression approximately 2 and 4 times, respectively. No further increase in activity was seen by elongation of the promoter up to Ϫ4740. Compared with the construct CEL62Luc, which contains the first 62 bp of the 5Ј-flanking region, including the TATA box, the longest construct, CEL4740Luc, gives an approximately 700 times higher activity. This transcriptional activity of the human CEL gene promoter seems to be tissue-specific, since almost no promoter activity was achieved from transfection of any of the constructs in the RAT2 cells.
The human CEL gene is also highly expressed in the lactating mammary gland. To see if the same regulatory elements are responsible for the mammary gland expression, some of the 5Ј-deletion constructs were transfected into the mouse mammary gland-derived cell line HC11. The HC11 cells have to be hormonally induced for several days to show some of the characteristic properties of lactation i.e. expression of milk protein genes (39), and a similar treatment is also needed for CEL gene expression in these cells. 3 This excluded the use of transient transfection assays, and instead the constructs were stably transfected into the HC11 cells. After selection, the transfectants were subjected to hormonal treatment as described previously (39) and then assayed for luciferase activity. As can be seen in Fig. 1, the deletion series exhibit a distinctly different pattern compared with that seen in cells of pancreatic origin. A high level of expression is achieved by the Ϫ150 construct, which is further increased by extending the promoter to Ϫ650. Extending the promoter up to Ϫ839 or higher does not affect the transcriptional activity. Hence, the strong positive effect achieved in pancreatic cells by the Ϫ839 to Ϫ653 region is not present in cells of mammary gland origin. This clearly demonstrates that two different sets of promoter elements are responsible for the CEL gene activity in mammary gland and pancreas, respectively.
The Ϫ317 to Ϫ156 Region Contains Elements That Are Needed for Proper Promoter Function-Promoters usually involve cooperative interactions between several regulatory elements. These elements can be located at great distance from the core promoter, but in general most of them are located within the first few hundred base pairs. To identify regions 2 U. Lidberg, unpublished observation. 3 M. Kannius-Janson, unpublished observation.

FIG. 1. Reporter gene analysis of a 5-deletion construct series of the human CEL gene promoter.
A series of 5Ј-deletion constructs of the human CEL gene promoter, extending from ϩ11 to Ϫ4740 relative to the transcription initiation site, was transiently transfected into AR4-2J and RAT2 cells, respectively, and luciferase activity was measured in cell lysates after 40 h. The luciferase activities were normalized to equivalent CAT activities as described under "Experimental Procedures." Some of the constructs were also stably transfected into HC11 cells. Bars represents luciferase activities, which are expressed in arbitrary units relative to the activity produced by the UMS-Luc construct, which was adjusted to the value of 1. Data represent mean and S.D. of at least three independent transfections. The x axis indicates the nucleotide position of the promoter sequence relative the transcription initiation site, which is indicated by an arrow.
containing elements of importance for proper promoter function, we generated internal deletion mutations of the CEL1640Luc construct. Fragments of a size of approximately 100 -200 bp were cut out by use of different restriction enzymes, blunted, and religated, hence creating a series of overlapping deletion mutant constructs ( Fig. 2A). (It should be noted that due to the type and location of usable restriction enzyme sites, sticky ends etc., the end points do not exactly correspond to the end points in the 5Ј-deletion series shown above.) These constructs were transfected into the AR4-2J cells, and luciferase activities were measured ( Fig. 2A). The results revealed that construct CEL1640⌬66/ϩ11Luc, in which the entire transcription start has been deleted, is inactive. This indicates that our constructs are under the control of a correct transcription initiation site. Deletions in the proximal part of the promoter, constructs CEL1640⌬156/63Luc and CEL1640⌬402/153Luc, reduce the activity down to 74 and 42% respectively, but the most dramatic effects are found in the region Ϫ836/Ϫ628. Deletion of this region abolishes the promoter activity. When the same deletion was introduced into the longest construct, CEL4740⌬836/628Luc, the activity was reduced down to 7%. Hence, this region, which overlaps with the region identified by the 5Ј-deletion series, not only contains elements required for the tissue specificity of the promoter but is also absolutely necessary for the promoter activity in exocrine pancreas. The same series of constructs were analyzed in RAT2 cells, but no significant change in activity was observed (data not shown).
The first 300 bp of the CEL gene promoter is well conserved between human and rat, and consequently it might contain elements of importance for promoter function (15). Although this conservation might contain elements of importance for the mammary gland expression, since a strong increase in promoter activity was achieved by this region in the HC11 cells, The reporter gene constructs CEL4740Luc and CEL1640Luc were used to generate two series of internal deletion constructs by the use of different restriction enzymes. These series were analyzed by transient transfection into AR4-2J cells, and luciferase activity was measured in cell lysates after 40 h. The luciferase activities were normalized to equivalent CAT activities as described under "Experimental Procedures." A, transfection of a series of internal deletion mutant constructs. Data are presented as percentage of the activity of the CEL1640Luc construct and represent the mean and S.D. of at least three independent transfections. In the construct schematic drawings, the dark boxes represent the extent of promoter sequence remaining in each construct. Black boxes represents the 5Ј-untranslated region, and deletions are indicated by thin lines. Nucleotide positions with respect to the transcription initiation site are indicated above. Note the different scales for the largest constructs. Names of the constructs are written to the left and contain numbers representing the promoter length of each construct relative the transcription initiator site. Deletions are marked by ⌬ followed by the numbers of the nucleotide position of the deleted segment. If downstream from the transcription initiation site, a deletion is indicated by ϩ before the number. B, transfection of a 3Ј-deletion construct series. The 3Ј-end point in this deletion series is the SacI site at Ϫ63 except for the CEL1640⌬631/321Luc construct. In the construct schematic drawings, the hatched box represent an unrelated DNA fragment. Other details are as in A.
the deletion construct CEL1640⌬402/153Luc results in a significantly reduced promoter activity in the AR4-2J cells. We therefore constructed a series of 3Ј-deletion mutants of the construct CEL1640Luc (Fig. 2B). The 3Ј-end point in this deletion series is the SacI site at Ϫ63, which is 33 bp upstream from the TATA box. As shown above, deletion of the region Ϫ156/Ϫ63 reduces the activity down to 74%, while an extended deletion up to Ϫ317 abolishes the promoter activity. If the deletion is further extended up to Ϫ631, the activity is almost restored (80%), while an extension of the deletion, past the important region identified above, up to Ϫ1154 completely abolishes promoter activity. Hence, the results imply that the conserved proximal region may contain elements of importance for normal promoter activity in pancreas. The restoration of promoter activity following the extension of the deletion from Ϫ317 to Ϫ631 suggests that this region encompasses a negative acting cis-element. To evaluate this possibility, a new deletion construct was made, CEL1640⌬631/321Luc, in which this region was deleted and analyzed in the AR4-2J cells (Fig. 2B). Surprisingly, instead of the expected increase, the promoter activity was reduced to approximately 50%. This result might indicate the presence of positional effects (i.e. activity is dependent on distance between the main positive element and the basal promoter). Moreover, insertion of an unrelated sequence, of identical size as the Ϫ631/321 region, from within the CEL gene, into the CEL1640⌬631/63luc construct (i.e. exchanging the Ϫ631/321 region in the CEL1640⌬317/63Luc) results in a similar decrease in activity (Fig. 2B, CEL1640⌬631/63ϩunLuc and CEL1640⌬631/63Luc versus CEL1640⌬317/53Luc). The same series of constructs was analyzed in RAT2 cells, but no significant change in activity was observed (data not shown).
Based on these results, we conclude that the human CEL gene promoter contains several regions of importance for the proper function in exocrine pancreas. Distal positive elements are located between Ϫ3200 and Ϫ1640. Another positive element is located between Ϫ1350 and Ϫ1194. The region from the CelII site at Ϫ839 to the ApaI site at Ϫ632 seems to be most important, since it contains positive elements absolutely necessary for both the activity and the tissue specificity of the CEL gene promoter. Besides the TATA box region, the proximal region, Ϫ317 to Ϫ63, is necessary for normal promoter activity and is an absolute requirement, provided the Ϫ631 to Ϫ318 region is present.
The Ϫ839/Ϫ632 Region Contains a Tissue-specific Enhancer Element-From what has been described above, the CEL839Luc is the shortest 5Ј-deletion construct, and the CEL1640⌬631/63Luc construct is the largest 3Ј-deletion that exhibits promoter activity. Therefore, it is clear that the Ϫ839 to Ϫ632 region in the CEL gene promoter contains positive elements that are absolutely necessary for the activity of the CEL gene promoter. To test for its ability to enhance transcription from an unrelated promoter, the CelII to ApaI restriction fragment containing this region was inserted in front of the SV40 promoter in the pGL-2 promoter vector (Promega) in both forward and reverse orientations, respectively (Fig. 3A). These two chimeric constructs, pEnhLuc and pEnhRLuc, were transfected into several cell lines. When transfected into the AR4-2J cells, these constructs were able to enhance transcription 26 and 15 times, respectively, compared with the pGL-2 promoter vector (Fig. 3B). Furthermore, to test for the ability to enhance transcription from longer distances, the fragment was inserted in reverse orientation approximately 5 kilobase pairs upstream from the SV40 promoter, the pEnhDRLuc construct. Transfection revealed that this construct was able to enhance the activity 13 times i.e. similar activity as for the pEnhRLuc construct. Hence, this indicates that the Ϫ839/Ϫ632 fragment of the CEL gene promoter has the characteristic properties of an enhancer. To exclude the possibility that any DNA fragment could enhance the activity, a construct was made with an unrelated fragment of identical length as the Ϫ839/Ϫ632 fragment in front of the SV40 promoter, the pUnLuc construct. Transfection revealed no enhanced activity from this construct. The same constructs were also transfected into the RAT2 cells and the HC11 cells, respectively. Almost no induction of activity was seen compared with the pGL-2 promoter vector for any of these constructs, which indicates that the enhancer is tissue-specific.
The Enhancer Is Composed of both a Positive and a Negative Factor with Specific DNA Binding Properties-Sequence analysis of the enhancer fragment reveals several putative ciselements. To localize the element responsible for the enhancement of the activity, the region was first analyzed with the DNase I footprinting assay with crude nuclear extracts from the AR4-2J cells in an attempt to identify regions of protein-DNA interactions. Two protected regions were detected (data not shown), the first at Ϫ754, which covers a STAT5 consensus element, TTCNNNGAA (where N represents any nucleotide), and the second at Ϫ696, which covers a sequence with homology to an AP-2 element (43). To determine the effect of these elements, new 5Ј-deletion constructs were made and transfected into the AR4-2J and RAT2 cells, respectively (Fig. 4). Surprisingly, all new constructs maintained the high expression level except for the CEL663Luc construct, which shows an increased activity, approximately 3-fold. This indicated that none of the protein-DNA interactions found by the DNase I footprinting assay are mediating the enhancer effect. Instead, the enhancer seems to be composed of both a negative element located within the region Ϫ681 to Ϫ664 and a positive element located in close proximity. For this region, no footprint could be detected. To map the 3Ј border of the positive element, a new 3Ј-deletion construct, CEL839⌬651/63Luc was made. As can be seen in Fig. 4, by elongation of the 3Ј-deletion from Ϫ631 to Ϫ651, the promoter activity is abolished. Together with the results from the CEL663Luc construct, this indicates that the positive element is located within the region Ϫ663 to Ϫ632.
Since the new 5Ј-deletion series closely narrowed the location of the enhancer, we could use the more sensitive electrophoretic mobility shift assay (EMSA) for further analysis of the enhancer structure. Two double-stranded, partly overlapping, oligonucleotides, GO22 and GO13, which together span the location of the enhancer, were synthesized (Fig. 5A). GO22 covers the 5Ј-half of the enhancer, which should contain the negative acting cis-element, whereas GO13 covers the 3Ј-half, which should contain the cis-element mediating the enhancer activity. After radiolabeling, the two oligonucleotides were used in binding assays with nuclear extract from AR4-2J cells. The nuclear protein interaction with GO22 reveals two strong protein-DNA complexes, whereas interaction with GO13 re-veals a weaker protein-DNA complex (Fig. 5B). The specificity of the interactions is confirmed by the fact that they can be efficiently competed by an excess of unlabeled oligonucleotides but not by an excess of an unrelated unlabeled oligonucleotide. GO13 cannot compete with the GO22 interaction; nor can GO22 compete with the GO13 interaction. This indicates that the enhancer is a complex composed of both a positively and a negatively acting transcription factor that interacts with two closely located but distinct cis-elements. To analyze whether the two trans-acting factors could interact with their respective cis-elements simultaneously, an oligonucleotide, GO26, covering the entire enhancer, was synthesized. As can be seen in Fig.  5B, several shifts were obtained. One interaction, indicated by an arrow in Fig. 5B, seems to be specific, since it can be competed by unlabeled GO28 but not by an unrelated oligonucleotide. Also, this shift seems to be the result of an interaction of both factors simultaneously, since competition by either cold GO13 or GO22, respectively, was able to disrupt this complex. Several nonspecific DNA complexes that could not be eliminated by competition with a large excess of unlabeled oligonu- cleotide were reproducibly seen.
The Negatively Acting Element Binds a Ubiquitously Expressed Factor-As shown above, the interaction to GO22 reveals two complexes. The same band pattern is also achieved with nuclear extract from rat pancreas (Fig. 6A), which verifies that the AR4-2J cell line to some extent reflects the conditions of the pancreas in vivo. Surprisingly, similar shifts are also obtained with nuclear extract from RAT2 cells and HC11 cells, respectively (Fig. 6A). A closer analysis of the band shift pattern shows that there is a tissue-dependent difference in the ratio between the higher and lower band. In pancreatic extracts, the lower band is much stronger compared with the higher band, whereas in RAT2 cells the opposite pattern is seen. This might be the result of two different isoforms of a protein or two different proteins, or it might be two differently composed complexes. To further elucidate the interaction, we competed the complex with an increasing amount of unlabeled GO22. As shown in Fig. 6B, the two band shifts seem to have similar affinity, since they are competed equally well, suggesting that the band pattern is due to different isoforms.
In an attempt of identify the sequence necessary for the protein interaction, a series of mutated GO22 oligonucleotides were constructed and used as unlabeled competitors to the radiolabeled GO22 (Fig. 6C). EMSA with AR4-2J nuclear extract revealed that the formation of the complexes could be competed by an excess of oligonucleotide Mut7, Mut6, or Mut5, respectively, and to some extent by the Mut4 oligonucleotide but not by the Mut1 or Mut3 oligonucleotide, respectively (Fig.  6C). This indicates that the region Ϫ668 to Ϫ654 contains the cis-element responsible for the interaction with the negative acting trans-factor.
Comparison of the binding sequence with known cis-elements did not reveal any clear homology, but analysis of the promoters for other lipolytic or pancreatic genes reveals several sequences with a high degree of similarity to the binding sequence. Based on these similarities, a set of oligonucleotides were made and used as unlabeled competitors to GO22. As can be seen in Fig. 6D, two of them, one from the mouse CEL gene promoter 3 and one from the dog co-lipase gene promoter (36), were able to compete the formation of the complexes. Sequence comparison of the interacting oligonucleotides reveals the consensus sequence for the cis-element to be TcTGTCAcTCTG FIG. 6. Analysis of the protein-DNA interaction at the negative element. A, EMSA analysis of the GO22 interactions with 4 g of nuclear extract from different cell lines or tissues. B, the specificity of the two shifts achieved with the GO22 oligonucleotide was analyzed by competition using increasing amounts of cold oligonucleotide. The molar excess of cold oligonucleotide in each line is shown. Eight g of nuclear extract from the AR4-2J cells were used. C, competition of the GO22 interaction by block-mutated GO22 oligonucleotides. The upper part schematically presents the block-mutated GO22 oligonucleotides used in the competition assay. The lower part shows the results from the competition assay with 4 g of nuclear extract from the AR4-2J cell line. Competitor oligonucleotides are in 100-fold molar excess. D, competition analysis of the GO22 interaction with oligonucleotides from homologous sequences found in the promoter of other lipolytical genes. The competitor labeled un represents an unrelated oligonucleotide. GO29, GO30, and GO31 are from the mouse CEL gene promoter nt Ϫ1782/Ϫ1756, Ϫ851/Ϫ823, and Ϫ810/Ϫ782, respectively 3 ; GO32 is from the human lipoprotein lipase gene promoter nt Ϫ648/Ϫ620; GO33 and GO34 are from the human pancreatic lipase gene promoter nt Ϫ575/Ϫ549 and Ϫ294/Ϫ266, respectively; GO35 is from the dog colipase gene promoter nt Ϫ565/Ϫ588; and GO36 is from the human colipase gene promoter nt Ϫ99/Ϫ74 (GenBank TM accession numbers X68111, L11242, M63427, and M95529, respectively). 4 g of AR4-2J nuclear extract was used. Competitor oligonucleotide is in 100-fold molar excess. The lower part shows a comparison of the competing oligonucleotides. The identical nucleotides are shown in boldface type. The locations of the promoter sequence for respective oligonucleotides are indicated by numbers. hCEL, human CEL gene promoter; mCEL, mouse CEL gene promoter; dCoLip, dog co-lipase gene promoter. E, site-specific mutations were introduced in the wild type construct CEL839Luc as described under "Results" and analyzed by transient transfection in AR4-2J and RAT2 cells, respectively. The luciferase activities were normalized to equivalent CAT activities as described under "Experimental Procedures." Data are presented as percentage of the activity of the CEL839Luc construct and represent the mean and S.D. of at least three independent transfections.
(where lowercase represents 2 identical of 3) (Fig. 6D). To verify this sequence as the cis-element of the negative factor, we introduced a mutation, TGTCA 3 GAGAG in the CEL839Luc construct. This construct was transfected in to the AR4-2J and RAT2 cells, respectively, and luciferase activities were determined (Fig. 6E). In AR4-2J cells, the CEL839 m1Luc construct shows a 3-fold increase in activity compared with the CEL839Luc construct. A similar increase in activity is also seen in the RAT2 cells. Hence, these data show that the mutated base pairs are critical for the function of the negatively acting complex.
The sequence TGTCA, which seems to be most conserved in the consensus sequence, is identical to the estrogen response element (ERE) half-site, identified in the rat prolactin gene promoter (44). However, an oligonucleotide containing the ERE could not compete in EMSA; nor could estrogen induce any change in transcriptional activity in a reporter gene assay (data not shown).
The Response Element Mediating the Enhancer Activity Is Composed of Two Separate Binding Sequences-The interaction with GO13 seen with nuclear extract from AR4-2J cells is also obtained with extract from rat pancreas (Fig. 7A). Again, and surprisingly, a complex migrating at a similar rate is also achieved with nuclear extracts from the RAT2 cells and HC11 cells, respectively (Fig. 7A), although much weaker, which indicates that also this factor is ubiquitously expressed. To identify the cis-element, a series of mutated GO13 oligonucleotides were made and used as unlabeled competitors to the radiolabeled GO13 (Fig. 7B). EMSA with AR4-2J nuclear extract revealed that the formation of the complex could be efficiently competed by an excess of oligonucleotide Mut19 and Mut16, respectively, but not by an excess of the oligonucleotide Mut18, Mut11, or Mut12, respectively. This suggests that the positive response element is composed of two separated subelements located in the region Ϫ661 to Ϫ637. To verify this region as the positive element, we introduced a mutation in the 5Ј-part, CTGCCTGG 3 AATGACTT in the CEL839Luc construct. This construct was transfected into the AR4-2J and RAT2 cells, respectively, and luciferase activities were determined (Fig.  7C). In AR4-2J cells, this mutation almost abolished the promoter activity, which indicates the critical function of these base pairs for the enhancer function. The spacing between the two subelements seems to be important, since a deletion construct, CEL839⌬652/649Luc, in which 4 bp are deleted in the region that could be mutated without affecting the DNA interaction (i.e. Mut16) decreases the promoter activity down to 30% (Fig. 7C). In RAT2 cells, no change in activity was observed for any of the new constructs.
In an attempt to identify the factor(s) interacting with this cis-element, several tests were done. The sequence that, based on the EMSA analysis, constitutes the positively acting ciselement contains a perfect and spaced palindrome motif, ctgC-CTGGGcatgcCCCAGGgcc (where boldface and uppercase indicates the motif). Furthermore, each repeat is centered in two sequences that show homology to the AP-2 consensus sequence, GCCN 2-4 GGC. Also, there is homology to the binding sequence for the p64 subunit of the PTF-1, TGGGA. However, an oligonucleotide containing the SV40 high affinity AP-2 site (43) or the Amy2-IV oligonucleotide (33) containing the PTF-1 ciselement from the amylase 2 promoter, respectively, was unable to compete the GO13 interaction in EMSA (data not shown). Nor could phorbol esters induce any change in transcriptional activity in a reporter gene assay (data not shown). Sequence comparison to the promoter for other lipolytic genes did not reveal any clear homology. But when the sequence was compared with the promoter sequence of the gene for the p48 subunit of the PTF-1 transcription factor, we found a 79% homology (15 bp out of 19) to the sequence that defines the footprint domain III (45). However, oligonucleotide III, containing the protected region, was unable to compete the GO 13 interaction in EMSA (data not shown). DISCUSSION The elements responsible for the tissue-specific expression of genes expressed in exocrine pancreas are mainly located within the first few hundred bp of the 5-flanking region (33,46). The expression of such genes is the effect of a cooperative interaction between several elements as described previously and exemplified by the elastase I gene enhancer (32). For this gene, it has been shown that PTF-1, which binds to the A element, is the critical key in directing the expression to the acinar cells (47). The activity is further enhanced by other elements such as the B and C elements, respectively.
Although expression of the CEL gene in the pancreas is highly restricted to exocrine tissue, it seems to be regulated in a different manner. The main response element in the CEL gene promoter is located approximately 650 bp upstream from the transcription initiation site. This element is an absolute requirement for activity as well as for maintenance of the tissue specificity of the promoter. Although the high activity of the largest construct, CEL4740Luc, is the combined effect of several regions, as shown in Fig. 1 by the stepwise increase in activity, its activity is almost abolished if the main element is deleted ( Fig. 2A). Despite the fact that PTF-1 does not seem to be involved in this activation, this element is acting as a pancreas-specific enhancer as demonstrated by transfection of the chimeric CEL/SV40 construct (Fig. 3). In contrast, and as shown by transfection of HC11 cells, the enhancer does not contribute to the high CEL expression seen in lactating mammary gland. Hence, the dominating dual expression pattern of the CEL gene, mammary gland, and pancreas, respectively, is due to two different regulatory systems.
Tissue-or cell-specific regulation of transcription, especially of genes with highly restricted expression, often involves cooperative interactions between several regulatory elements (32, 48 -50). Such interactions between enhancer elements and their cognate promoters have also been described (51,52). Sequence analysis has shown that the proximal part of the CEL gene promoter has been conserved during evolution and is highly similar between human and rat (15). Although the CEL enhancer itself seems to be able to establish tissue-specific expression, the abolished promoter activity following the deletion of the Ϫ317 to Ϫ63 region, as shown in Fig. 2B, indicates the presence of other important elements. As shown in Fig. 1, this region is not in itself sufficient to activate the expression of the human CEL gene in AR4-2J cells. Hence, the results obtained in this study may be consistent with described models of cell-specific gene regulation involving cooperative interactions. However, a further deletion of most of the proximal promoter region, Ϫ631 to Ϫ63 (Fig. 2B), resulted in a restoration of the promoter activity (which again indicates the importance of the element located at Ϫ650), suggesting that a negative regulatory element is located in the region Ϫ631 to Ϫ318. As a consequence, the deletion construct CEL1640⌬631/322Luc, in which this region has been deleted, should then give rise to an increased activity; instead, the observed activity is reduced. Moreover, the insertion of an unrelated DNA fragment of identical length as the fragment deleted above into the CEL1640⌬631/63Luc results in a similar reduction of the promoter activity. Taken together, a more likely explanation is that the restored promoter activity observed by the elongation of the deletion Ϫ63/Ϫ317 up to Ϫ631, is the result of moving the enhancer closer to the core promoter. This might also indicate that the effect of the enhancer in the CEL gene promoter is not absolutely position-independent at short distances to the core promoter.
The enhancer seems to be built up by two subelements, enhansons: a distal negative element and a proximal positive element. As demonstrated in Fig. 5B, each trans-acting factor is capable of binding DNA by itself. Enhancers composed of both positive and negative elements often mechanistically involve competition for DNA binding. Overlapping response elements will prevent simultaneous DNA interaction, as seen in the ␦1-crystallin gene enhancer (53). Alternatively, too close location of the response elements might result in steric hindrance, as seen for YY1 and the MGF site in the rat ␤-casein gene promoter (54). As shown in Fig. 5B, this seems not to be the case for the CEL gene enhancer, since EMSA with GO26, containing both the response elements, results in a large complex. This complex seems to be composed of both factors, since it can be competed by oligonucleotides containing a single positive or negative element, respectively. A cooperative interaction like that seen for the sterol regulatory element-binding protein 1 factors in the low density lipoprotein receptor gene promoter (49) cannot be excluded, although it seems less likely, since a similar amount of cold oligonucleotide was required for competition as for the single elements. However, as has been shown in Figs. 4 and 6E, respectively, deletion or mutation of the negative element results in a highly increased promoter activity. On the other hand, the negative factor fails to influence the basal promoter activity when the positive element is deleted or mutated. This indicates that the negatively acting factor seems to influence the promoter activity by modulating the activity of the positively acting factor, which is similar to the hierarchical model suggested for the main positive element in the apolipoprotein B gene promoter (40).
As shown in Fig. 6A and 7A, the factors that bind to the enhancer seem to be ubiquitously expressed, since all studied cell types contain nuclear proteins capable of binding to the unique enhancer elements in the CEL gene promoter. Furthermore, GO26, covering the entire enhancer, revealed similar but weaker band patterns with nuclear extract from other cells (data not shown). The ubiquity of factors giving rise to mobility shifts, but not of enhancer activity, may be explained in several ways. The binding to these specific DNA sequences, respectively, could represent the interaction of several distinct nuclear proteins or heteropolymers of nuclear proteins. The cellspecific differences in the subunit composition of a multimeric binding protein may then account for the difference in activity, as is seen in the case of the NF-B family (55,56). An alternative could be the existence of cell-specific differences in the phosphorylation state, as is seen in interferon-regulated genes and lactation associated genes (57,58). Also small differences in the balance between a positive and a negative factor may strongly influence the activation potential as seen for the CCAAT/enhancer-binding protein ␤ isoforms ratio in the mammary gland (59). The difficulties in comparing the amount of a transcription factor in different nuclear extracts are obvious. Nevertheless, it seems to us that the factor binding to the positive element is significantly more highly expressed in cells derived from exocrine pancreas. This suggests that the net enhancer activity might be the result of a tissue-specific balance between the positively and negatively acting factors.
Sequence analysis of the functional enhancer reveals no homology to the A, B, or C elements, respectively, in the elastase I gene promoter. Nor does a comparison to other known transcription factor binding sites show any clear homology; however, a comparison of respective subelements to previously reported cis-elements reveals some homologies, which mainly FIG. 7. Analysis of the protein-DNA interaction at the positive element. A, EMSA analysis of the GO13 interaction with 4 g of nuclear extract from different cell lines or tissues. B, competition of the GO13 interaction by cold block-mutated GO13 oligonucleotides. The upper part schematically presents the block mutated GO13 oligonucleotides used in the competition assay. Arrows indicate the spaced palindrome motif. The lower part shows the results from the competition assay with 4 g of nuclear extract from the AR4-2J cell line. Competitor oligonucleotides are in 100-fold molar excess. C, site-specific mutations were introduced in the wild type construct CEL839Luc as described under "Results" and analyzed by transient transfection in AR4-2J and RAT2 cells, respectively. The luciferase activities were normalized to equivalent CAT activities as described under "Experimental Procedures." Data are presented as percentage of the activity of the CEL839Luc construct and represent the mean and S.D. of at least three independent transfections. constitute half-sites. However, none of the oligonucleotides containing the corresponding cis-element tested in EMSA (ERE, AP-2, PTF-1) were able to compete with the interactions achieved by GO22 (negative subelement) or GO13 (positive subelement), respectively. The subelement mediating the positive effect contains a spaced palindrome motif that might indicate the binding of a homodimer. When the sequences of the subelements were compared with other promoter sequences, it was found that the subelement mediating the negative effect is present in the promoters of several other lipolytical genes, but none of them is yet connected to any functional activity. Hence, it is possible that the functional enhancer complex might be composed of not yet described transcription factors with new unique DNA binding specificities. Alternatively, it might be composed of common subfactors organized in a unique tissue-specific enhancer structure, as suggested by Tjian and Maniatis (60).
An attractive hypothesis is that the enhancer element in the CEL gene promoter in some way might be connected with the observed cholesterol induction. The lack of sequence homology for the enhancer element to any known response elements might support this hypothesis. Also, the inactivity of the pancreatic enhancer element in mammary gland is in agreement with this hypothesis, since other factors/hormones are believed to induce milk protein gene expression. However, the reported up-regulation of the CEL gene by cholesterol is a process both too slow and too limited to be studied by transient transfection in the highly CEL expressing AR4-2J cells and will await stably transfected cell lines or more truly transgenes. Hence, analysis of the CEL gene regulation might contribute to the understanding of cholesterol-induced gene activation.
In conclusion, the data presented here demonstrate that the expression of the human CEL gene in exocrine pancreas, but not in the lactating mammary gland, is due to a strong pancreas-specific enhancer. This suggests that the tissue-specific expression of the CEL gene is due to different regulatory systems. Also, in contrast to other studied genes expressed in exocrine pancreas, the PTF-1 does not seem to play a major role in the regulation of the CEL gene. Instead, yet unidentified factors interact with the unique enhancer elements. Although these factors are ubiquitously expressed, the functional enhancer complex seems to be organized in a tissue-specific structure. Hence, further characterization of the enhancer complex will provide additional insights into the regulation of exocrine pancreas-specific genes.