GAGA Factor Down-regulates Its Own Promoter*

GAGA factor is involved in many nuclear transactions, notably in transcription as an activator inDrosophila. The genomic region corresponding to theTrl promoter has been obtained, and a minimal version of a fully active Trl promoter has been defined using transient transfection assays in S2 cells. DNase I footprinting analysis has shown that this region contains multiple GAGA binding sites, suggesting a potential regulatory role of GAGA on its own promoter. The study shows that GAGA down-regulates Trl expression. The repression does not depend on the GAGA isoform, but binding to DNA is absolutely required. A fragment of the Trl promoter can mediate repression to a heterologous promoter only upon GAGA overexpression in transiently transfected S2 cells. Chromatin immunoprecipitation analysis of S2 cells confirmed that GAGA factors are bound to the Trl promoter over a region of 1.4 kbp. Using a double-stranded RNA interference approach, we show that endogenous GAGA factors limit Trl expression in S2 cells. Our results open the possibility of observing similar GAGA repressive effects on other promoters.

GAGA factor is involved in many nuclear transactions, notably in transcription as an activator in Drosophila. The genomic region corresponding to the Trl promoter has been obtained, and a minimal version of a fully active Trl promoter has been defined using transient transfection assays in S2 cells. DNase I footprinting analysis has shown that this region contains multiple GAGA binding sites, suggesting a potential regulatory role of GAGA on its own promoter. The study shows that GAGA down-regulates Trl expression. The repression does not depend on the GAGA isoform, but binding to DNA is absolutely required. A fragment of the Trl promoter can mediate repression to a heterologous promoter only upon GAGA overexpression in transiently transfected S2 cells. Chromatin immunoprecipitation analysis of S2 cells confirmed that GAGA factors are bound to the Trl promoter over a region of 1.4 kbp. Using a double-stranded RNA interference approach, we show that endogenous GAGA factors limit Trl expression in S2 cells. Our results open the possibility of observing similar GAGA repressive effects on other promoters.
GAGA is a Drosophila nuclear factor that is encoded by the Trithorax-like gene (Trl) and is of maternal effect (1). At least two isoforms, GAGA 519 and GAGA 581 , are known to be produced by alternative splicing (2), and both the sequence of the isoforms and their splicing patterns are highly conserved between Drosophila melanogaster and the distant species Drosophila virilis (3). The GAGA factors present an overall modular structure that is formed by an N-terminal BTB/POZ domain involved in protein oligomerization (4,5); a DNA binding domain, composed of a single zinc-finger and three adjacent basic regions, that confers sequence-specific DNA binding (6 -8); and a C-terminal domain (Q-domain) that in both isoforms is glutamine-rich, although the amino acid sequences differ, and is the activation domain (9 -11). 1 There is some evidence that the Q-domain may also be involved in GAGA aggregation (12,13). Between the POZ/BTB domain and the DNA-binding domain there is a fourth domain, that will be referred to here as the X-domain, about 200 amino acids long and of unknown structure and properties. To date, both isoforms have been found to be rather equivalent, and have largely overlapping functions in transgenic flies although some differences have been noticed (11).
GAGA binds to DNA sequences with a consensus (GAGAG) (7), but neither a strict consensus nor a single site is efficient enough for GAGA binding. In fact, a GAG or CTC trinucleotide sequence has been described as being sufficient for specific binding (8). However, as noted by others (14), GAGA binding sites on natural promoters show a clear tendency to cluster and give rise to strong composite GAGA binding sites. This is in agreement with the finding that GAGA is oligomeric and can interact with specific sequences that are variably spaced by other irrelevant DNA sequences that are neither recognized nor bound (4,5). The need for the clustering of GAGA binding sites apparently arises from the fact that GAGA activation is very weak (if not negligible) when only single or double sites are present (4).
Since its discovery, GAGA was counted as a transcription activation factor because, upon binding to (GA) n -rich sequences, it stimulated transcription of genes such as Ultrabithorax (Ubx) and engrailed (en) (15,16). This role has frequently been reported for many other promoters, either in vitro and/or in vivo, including actin 5C, krü ppel, fushi-tarazu (ftz), etc. (recapitulated in Ref. 17). Nevertheless, several early observations suggested that GAGA was a complex nuclear factor. For instance, unexpectedly, during mitosis GAGA remains bound to chromatin, whereas transcription factors are usually displaced (18,19). Also, at metaphase GAGA is redirected from its interphasic locations on the promoters (20 -22) to heterochromatic GA-rich regions located in the vicinity of Drosophila centromeres (18,19). These features appear to be common to both GAGA isoforms (2,11). This mitotic location might explain the phenotypes observed for the hypomorphic mutant (Trl 13C ), which include serious defects on the synchrony in cleavage cycles, failure in chromosome condensation, abnormal chromosome segregation, and chromosome fragmentation (23). Other explanations invoke the evidence that GAGA co-operates with chromatin remodeling machines like NURF 2 to open several promoters (22,24). Also GAGA and ISWI, the catalytic subunit shared by several chromatin remodeling complexes, present some limited co-localization at some loci on polytenic chromosomes but do not co-localize at all on mitotic chromosomes (25). Recently, a direct interaction has been observed between GAGA and the largest NURF subunit, NURF301. The interaction involves two regions in NURF301 and a GAGA region around the DNA binding domain. On the other hand, the chromatin remodeling activity and the stimulatory activity of GAGA have been shown to be independent in transgenic flies (26).
The existence of GAGA binding sites on several Polycombresponsive elements (PRE) and its co-localization with some proteins of the Polycomb group also indicates a connection of GAGA with gene silencing (27,28). The interaction of GAGA with SAP18, a Sin3-associated polypeptide, and the presence of GAGA in a complex containing the RPD3 deacetylase and the Polycomb group factors, polyhomeotic and posterior sex combs, indirectly suggested a connection with repression too (21,29). A functional requirement for both GAGA and pleiohomeotic has been established to account for the silencing activity of the iab-7 PRE and the MCP silencer of Abdominal-B (30,31).
Here, we have analyzed the Trl promoter and provide direct evidence that both GAGA isoforms can mediate repression of its own promoter. Repression is promoter-specific and requires GAGA DNA binding activity, whereas transactivation and oligomerization domains are dispensable.

EXPERIMENTAL PROCEDURES
DNA Constructs-The sequences corresponding to the 5Ј upstream region to the Trl coding sequence and released in Flybase before the whole Drosophila genome was available were used to isolate the sequences putatively corresponding to the Trl promoter. An ϳ4.3 kbp fragment was PCR-amplified from P1 clone ds01921 (kindly provided by the European Drosophila Genome Project) using 5Ј-TAC ATG GAT AAG ATT CTG ACG G-3Ј and 5Ј-GGG AGG CGG CGC ACA GG-3Ј as primers and cloned into pGEM-T (Promega). This fragment comprised about 3.5 kbp of putative promoter region, the three 5Ј end sequences of the cDNAs reported previously (2), and the first 168 bp of the GAGA coding sequence. Automated sequencing exactly confirmed the genomic sequence deposited later on by Celera Genomics. For transfection assays, subclones were generated using PCR in a way that coding sequences were omitted while leaving intact a long 5Ј-UTR. Promoter deletions were introduced either by restriction at unique sites whenever possible or by PCR with specific primers. All of them were inserted between MluI and HindIII sites in pGL3 vector (Promega).
The eve-luciferase reporter was obtained by inserting a 1.8-kbp fragment of the eve promoter that directs the embryonic expression of stripe 2 in the embryo (from position Ϫ1759 to ϩ102; kindly provided by Dr. M. Levine, University of California, Berkeley) into pGL3 vector (Promega).
Constructs for expressing GAGA isoforms and mutants in S2 cells were all prepared using Act5PPA vector. Constructs for GAGA 519 , GAGA⌬Q, GAGA⌬Q-VP16, and GAL4BD have already been described (9). The expression construct for GAGA 581 was generated by PCR cloning the C terminus of GAGA 581 and replacing the C-terminal region of GAGA 519 construct. The expression construct for GAGA⌬POZ was generated by subcloning the previously prepared ⌬POZ 122 construct into Act5PPA vector (4). The construct expressing the mutated DNA binding domain (GAGA 519 H361P) was generated by PCR across the DNA binding domain using appropriate primers that introduced a His-3613 Pro change at the zinc-finger. This mutant was initially prepared in a bacterial expression vector (pET14b) and then subcloned into Act5PPA for expression in S2 cells.
Constructs containing the CMV promoter and a luciferase reporter gene were fused to fragments FI, FII, FIIϩFIII, FIII, FIV, and FV derived from the Trl promoter using either an MluI site, located upstream of the CMV promoter, for the FI to FIII fragments or a HindIII site, located downstream of the CMV promoter, for fragments FIV and FV, in an attempt to mimic as much as possible the relative location of the Trl-derived fragments with respect to the transcription start sites.
Transient Cell Transfections-S2 cells were grown and transfected as described (9). Each transfection included 3 g of CMV-␤-galactosidase, variable amounts of reporter constructs fused to the luciferase gene and expression vectors driven by the actin 5C promoter of Drosophila, and 7 g of pGL3 vector (Promega). The final amount of DNA was adjusted to 20 g by the addition of Act5PPA empty vector. After 48 h of incubation with the DNA, cell lysates were prepared according to manufacturer's instructions. Luciferase and ␤-galactosidase activities to correct for transfection efficiency were determined using commercial kits (Promega and Roche Molecular Biochemicals, respectively). Results for each condition represent the average of three to five independent experiments, each performed in duplicate.
DNase I Footprinting Assays-The portion of the Trl promoter to be studied was divided into five different regions and subcloned into pBS(Ϫ) (Stratagene). These five regions contained sequences between positions Ϫ679/Ϫ462, Ϫ462/Ϫ204, Ϫ204/Ϫ49, Ϫ49/ϩ347, and ϩ347/ ϩ737, respectively, according to the numbering system indicated in Fig.  3A. These fragments were end-labeled and purified on native polyacrylamide gels following standard protocols. Recombinant GAGA factor used in these experiments corresponded to GAGA 519 and was expressed and purified in Escherichia coli BL21(DE3) as described (32). DNase I footprinting reactions were carried out as described (33).
RNase Protection Assays-An RNase protection analysis was performed using the Direct Protect Lysate RPA kit following the manufacturer's instructions (Ambion). Total RNA was isolated from 4 ϫ 10 7 Drosophila S2 cells, and for analysis of Trl transcripts, 25-65 g of total RNA was mixed with 1 ϫ 10 4 cpm of 32 P-labeled antisense riboprobes. Protected RNAs were analyzed on denaturing 5% polyacrylamide gels containing 8 M urea. Gels were dried, exposed at Ϫ80°C for 24 h with intensifier screens, and developed. Uniformly radiolabeled RNA probes R1 (from position Ϫ49 to ϩ347) and R2 (from position ϩ347 to ϩ737) were generated by in vitro transcription using either T3 (for R1) or T7 (for R2) RNA polymerases from the respective plasmids bearing these sequences in pBS(Ϫ). Riboprobes were purified on polyacrylamide gels.
Chromatin Immunoprecipitation Analysis-Drosophila Schneider S2 cells were cross-linked and chromatin-immunoprecipitated as described (34) with the modification in the PCR amplification protocol subsequently described (35). Polyclonal antibody raised in rabbits against bacterial expressed GAGA 519 isoform was affinity-purified using standard protocols. DNA amplified from immunoprecipitations was analyzed by slot-blot hybridization. Radiolabeled DNA probes corresponding to regions Ϫ679/Ϫ462 (FI), Ϫ462/Ϫ204 (FII), Ϫ204/Ϫ49 (FIII), Ϫ49/ϩ347 (FIV), and ϩ347/ϩ737 (FV) were prepared using the Ready-to-Go labeling kit (Amersham Biosciences). Hybridizations were carried out following standard protocols. Blots were exposed at Ϫ80°C, and films were quantified using a laser microdensitometer (Amersham Biosciences). All of the results were verified to be in the linear response range as indicated by comparison with corresponding sets of dilution input standards (not shown).
RNA Interference Experiments-To generate the dsRNA an EcoRI-BamHI fragment of GAGA (718 bp long) encoding a protein sequence common to both isoforms of GAGA and including most of the POZ/BTB domain and part of the adjacent sequence was subcloned into pBS(Ϫ). The RNAs were prepared by in vitro transcription as described above and resuspended in water. To prepare dsRNAs equimolar amounts of sense and antisense RNAs were mixed, heated for 5 min at 90°C, cooled down slowly to room temperature, and stored at Ϫ20°C until use. For transient transfection assays, dsRNA (0.5 g to 5 g) was added to the transfection mix and the assays proceeded as described. Luciferase and ␤-galactosidase activities were assayed after 48 or 72 h as described above. A control dsRNA for the ␤-galactosidase mRNA was similarly prepared from a lacZ SstI-PvuII fragment about 0.7 kbp long cloned into pBS(Ϫ).

RESULTS
Isolation and Analysis of the Trl Promoter-A 4.2-kbp DNA fragment was obtained by PCR amplification from P1 clone ds01921 (obtained from the European Drosophila Genome Project) using primers specific for the Trl gene. The complete sequence was fully consistent with the sequence deposited by Celera Genomics in the Berkeley Drosophila Genome Project Database. The data available suggested the existence of at least three different transcription start sites according to the different 5Ј ends of the cDNAs isolated so far (2). Here, we assigned position ϩ1 to the hypothetical start site the most upstream from the coding sequence. Our clone obtained contained about 0.73 kbp of the 5Ј-UTR sequence of the Trl gene and ϳ3.4 kbp of the upstream sequence.
The study of the Trl promoter elements included transient transfection in Schneider's S2 cell line. As a first step, Trl transcription start sites used in S2 cells were mapped. RNase protection assays (RPA) with total RNA from S2 cells revealed at least three start sites (Fig. 1B). From their positions with respect to the coding sequence, the start sites were named, distal, intermediate, and proximal. These three sites reasonably correspond to the three 5Ј cDNA ends already described (2). Other start sites may exist, because an intense band appeared in the RPA analysis a few nucleotides upstream of the position of the proximal start site (*1, Fig. 1, A and B, left panel). Also, another band appeared some 50 nucleotides upstream of the distal site (*2, Fig. 1, A and B, right panel), which is clearly different from the undigested R1 probe. From the previously reported results, the proximal start site accounts for most of the GAGA 519 isoform and the distal start site for the GAGA 581 isoform in embryos (2). In S2 cells, because the RPA technique is quantitative, we observed that the proximal start site is also the most frequently used, as indicated by the high intensity of the protected band.
The Trl promoter fragment described above was linked to a luciferase reporter gene and studied in S2 cells. Transcriptional activity from this construct reached values above 2600fold with respect to empty pGL3 vector and was taken as 100% of Trl promoter activity in our analysis (not shown). A deletion analysis of the Trl promoter showed that nearly full activity could be observed in serial 5Ј deletions from position Ϫ3470 down to position Ϫ345 (Fig. 2). Deletion to position Ϫ270 resulted in a drop of about 50% of the promoter activity. Deletion to position Ϫ204 could only support about 20% of total activity, and activity was lost completely by deletion to position Ϫ49. Deletion of 400 bp at the 3Ј end of the fragment resulted in about 50% of total activity. Because this deletion removed the proximal start site, it is reasonable to assume that the two other sites were functional and accounting for reasonable levels of mRNAs. We concluded that the Trl promoter in S2 cells is small and extends to about position Ϫ345.
Computer analysis of the promoter sequence highlighted several potential GAGA factor sites and some potential binding sites for other regulatory proteins (not shown). Because GAGA binds to sequences showing a considerable degree of heterogeneity and a good binding site has been shown to have composite nature rather than a single linear sequence (4,5,8), a DNase I footprinting analysis using recombinant GAGA 519 factor was performed. Because full promoter activity in S2 cells was confined to position Ϫ345, we limited our assay to a slightly larger fragment (up to position Ϫ680) and included the 737 bp of the 5Ј-UTR. Because the region to be analyzed was rather long (ϳ1.4 kbp), it was subdivided by convenience into five fragments (denoted as FI to FV, Fig. 3A) covering the entire length of the region studied. Fig. 3B shows that a large number of GAGA sites were present on the Trl promoter (see Fig. 3A for a general scheme and summary of results). Especially rich in GAGA binding sites were regions FII, FIV, and FV with 5, 8, and 8 sites, whereas regions FI and FIII showed only 2 and 3 sites, respectively. The relative affinities of the binding sites were not identical, and some showed a high affinity (e.g. FIII no. 1, FIV nos. 6 and 7, and FV nos. 6, 7, and 8), whereas others were of low affinity (e.g. FII no.1, FIV nos. 1 and 2, FV nos. 1 and 3, and results not shown). These differences suggest that the relative occupancies of the sites could be different in vivo. Some hypersensitive sites that might suggest a stressed DNA conformation upon GAGA binding were also noted (indicated by an asterisk in Fig. 3B).
GAGA Represses the Trl Promoter-GAGA factor has always been regarded as an activator because in the fly it stimulates the transcriptional activity of genes under its control (e.g. ftz, en, Ubx, etc.) (1,11,17,23). GAGA also stimulates transcription of many reporter genes when assayed using transient transfection in Drosophila cell lines (2,9,14) and in transgenic flies (26). From our results, an appealing possibility was that GAGA could regulate the expression of its own promoter. This prospect was studied by performing transient transfection experiments in which GAGA factors were overexpressed and their activity on the Trl promoter analyzed. Unexpectedly, overexpression of either GAGA 519 or GAGA 581 resulted in a strong repression of the Trl promoter. Repression was efficient and dose-dependent (Fig. 4A), reaching a minimal level of activity at about 15% with respect to controls without GAGA overexpression. Overexpression of larger amounts of GAGA factor did not result in further repression (results not shown). On the contrary, expression of increasing amounts of GAGA 519 stimulated transcription of the eve stripe 2 promoter up to ϳ15-fold in a dose-dependent manner (Fig. 4B). Overexpression of an unrelated factor, the yeast GAL4BD, showed no effect on the activity of either promoter at any dose ( Fig. 5 and results not shown).
RNA interference assays were performed to assess that the stimulation of eve and the repression of Trl were due to GAGA overexpression. Both repression of Trl (Fig. 3, light gray columns) as well as stimulation of eve (dark gray columns) were efficiently abolished by co-transfection of constructs and specific RNAi GAGA at low doses (Fig. 4C). At the highest dose, Trl expression was stimulated ϳ2-fold above the starting level, clearly suggesting a GAGA-mediated down-regulation of Trl promoter even in the absence of expressed GAGA (Fig. 4C, compare columns at 0 and 5 g of RNAi GAGA ; see also Fig. 7B). As expected, ␤-galactosidase activity was not affected by cotransfection of RNAi GAGA , and vice versa, co-transfection of the RNAi ␤-gal did not affect luciferase levels (results not shown).
Although the BTB/POZ Domain and Q-domain Are Dispensable for GAGA-mediated Repression of Trl, the DNA-binding Domain Is Required-In order to define the domain(s) of GAGA required for the repression of the Trl promoter, the two GAGA isoforms, GAGA 519 and GAGA 581 , which are identical in sequence except for the glutamine-rich C-terminal domain (Qdomain) were assayed initially. Both isoforms similarly repressed Trl, much the same as GAGA⌬Q did, confirming that Q-domains were not required for repression (Fig. 5). On the contrary, both isoforms stimulated transcription of eve stripe 2 promoter ( Fig. 4B and results not shown).The POZ/BTB domain was not required either, because its deletion did not prevent repression (Fig. 5, GAGA 519 ⌬POZ). However, a single point mutation at His-361 (GAGA 519 (H361P)), which disrupted the unique zinc-finger domain and rendered GAGA completely unable to recognize GAGA sites (as shown by DNase I footprinting experiments; results not shown), significantly abolished repression of Trl. Replacement in GAGA of the Q-domain by the VP16 activation domain (GAGA⌬QVP16) reverted the situation and could even stimulate Trl transcription ϳ2.5-fold. Expression of an irrelevant factor (GAL4BD) had no effect at all on Trl expression. In all cases, analysis by Western blot indicated that expression of all of the constructs resulted in proteins of the expected sizes that were produced in similar amounts (results not shown).
A Region of the Trl Promoter Is Required to Mediate GAGA Repression-Because the DNA binding activity of GAGA was required, it was of interest to identify the DNA sequence of the Trl promoter that mediated repression. As shown in Fig. 3, there were so many GAGA binding sites on this promoter that assignment of the sites involved in repression and discrimination of those that might not take part became rather complex, even more so because GAGA does not simply operate on a single-site basis but binds clusters of relatively close binding sites (2, 4, 5, 14, 15, 17). Therefore, we decided to identify regions rather than sites that could confer GAGA-mediated repression. The region required for GAGA to repress Trl was delimited, using the promoter deletion constructs described before, to position Ϫ345 in S2 cells because deletions from position Ϫ3470 showed no effect on repression (Fig. 6A). Deletion to position Ϫ270 reduced the repression levels slightly. Further deletions could not be assayed because of the very low transcriptional activity (see Fig. 2). These results indicated that the promoter element involved in GAGA-mediated repression had to be located between positions Ϫ345 and ϩ737. To better define and assay which region was required, we considered that repression might be transferred to heterologous promoters. To perform such an assay, a prerequisite was that the recipient promoter had to be insensitive to GAGA overexpression. CMV was found to be insensitive (Fig. 6C) and then, fragments FI to FV, which spanned positions Ϫ679 to ϩ 737, were inserted into the CMV-luciferase reporter. Fragments FI to FIII were inserted upstream of the core CMV promoter, and fragments FIV and FV were inserted downstream in an attempt to mimic as much as possible the relative situation of the Trl promoter regions with respect to the transcription start site of the core CMV promoter. Transfection of these constructs in S2 cells resulted in transcription levels of the luciferase reporter gene higher than observed for the parental CMV construction, suggesting that they contained some stimulatory sequence elements (Fig. 6B). The sole exception was FI (spanning positions Ϫ679 to Ϫ462), which did not show any effect, likely because positions upstream of Ϫ345 were not required for full activity of the Trl promoter (Fig. 2). Upon GAGA 519 overexpression, fragments FII and FIII could mediate partial repression of the CMV promoter with respect to their corresponding controls (Fig. 6C, columns FII-CMV and FIII-CMV). This repression, however, still left about 70% of activity, whereas the values observed with the complete Trl promoter left only 15-20%. The very weak repression activity observed with fragment FI was not considered significant. On the other hand, GAGA overexpression clearly had no effect on CMV-FIV construct, and transcription from CMV-FV construct was stimulated ϳ4-fold (Fig. 6C, columns CMV-FIV and CM-FV). Because these results indicated that a region (perhaps two) of the Trl promoter could direct partial GAGA-mediated repression when transported to an unrelated heterologous promoter and the Trl regions were defined and inserted in the CMV promoter for convenience, it was possible that the partial repression observed could be due to disruption of a genuine repressive element of the Trl promoter. Thus, insertion of a fragment encompassing the complete FII and FIII regions upstream of the CMV promoter resulted in higher levels of repression only upon GAGA overexpression (50% of activity left, Fig. 6C, FIIϩFIII-CMV). In the absence of GAGA overexpression, insertion of this region resulted in a weak stimulation of CMV transcription as observed for the individual fragments (Fig. 6B,  FIIϩFIII-CMV).
In S2 Cells Endogenous GAGA Is Bound to Trl Promoter and Down-regulates Its Transcription-To gain some insight into the state of the genomic Trl promoter in intact cells, we performed chromatin immunoprecipitation analysis. Using as probes the same fragments described for the footprinting experiments (FI to FV, Fig. 3), the presence of GAGA across the minimal Trl promoter region presenting full activity was revealed (Fig. 7A). GAGA could be detected bound to all of them, although a clear accumulation of GAGA factors on fragments FII and FIV was detected. Accumulation of GAGA factors was especially intense at region FII (100-fold), one of the regions able to mediate repression to a heterologous CMV promoter. Fragment FIV, in which binding of GAGA was shown to have no effect, in the same assay was highly enriched in GAGA factors as well (65-fold). Enrichment of GAGA was clearly similar but lower for the other three fragments (FI, FIII, and FV, 8-, 14-, and 10-fold, respectively) (Fig. 7A, see lower panel for quantification). Overall the data supported the results obtained by DNase I footprinting in vitro and suggested that GAGA could have a regulatory role on its own promoter. Accordingly, the repressive effect on the Trl promoter could also be observed in S2 cells by depletion of the endogenous GAGA content using RNA interference assays. Co-transfection of the Trl luciferase reporter construct and increasing amounts of RNAi directed against GAGA mRNA (RNAi GAGA ) resulted in a dose-dependent stimulation of transcription that reached an ϳ4-fold increase at the highest dose tested (Fig. 7B). Because in this experiment GAGA was not overexpressed but RNAi knocked down the endogenous content of GAGA factors, we conclude that normal physiological levels of GAGA factors limit Trl transcription in S2 cells so that a reduction in the GAGA content results in a derepression of Trl promoter. DISCUSSION A functional promoter of the Trl gene has been isolated and the regions relevant for expression in S2 cells defined. A region of about 0.35 kbp upstream of the distal transcription start site was sufficient to support full transcription in S2 cells. Transcription dropped to a minimum when only 0.2 kbp were left. and completely disappeared when only 49 bp were left (Fig. 2). Preliminary results in transgenic flies confirm these results and indicate that both the Ϫ3470 to ϩ737 promoter and the minimal Ϫ345 to ϩ737 promoter appear to be functional and to direct similar embryonic expression of a reporter gene, whereas the Ϫ49 to ϩ737 cannot (results not shown).
Three transcription start sites of the Trl gene have been mapped in S2 cells that closely correlate with the start sites of the three different cDNA clones from a Drosophila ovarian cDNA library reported previously (2). In addition, the possible existence of other two start sites has been described. Notably, the most distal site observed (*2 in Fig. 1) matches reasonably well with one of the start sites predicted by GadFly for Trl gene (Fly Base entry number CG9343). The possible existence of other start sites cannot be discarded at this point, especially one slightly upstream to the proximal one (*1 in Fig. 1).
Inspection of the Trl promoter sequence immediately suggested several potential GAGA factor binding sites that were confirmed by DNase I footprinting analysis. In fact, as many as 27 sites were identified in an ϳ1.5 kbp region around the transcription start sites, and there may likely be some others that could not be resolved in our analysis. It was not our aim to make an extensive cartography of the GAGA binding sites on this promoter, and in fact, this high abundance of sites was not expected. The analysis of the sites showed that 18 of them presented GAGA sequences of high affinity (7). For the remaining sites, eight presented at least a GAG/CTC trinucleotide sequence that in principle would suffice to bind GAGA factor  (8), and one did not present any sequence motif reminiscent of a GAGA site. From all of the sites examined, no correlation could be established with the relative affinity estimated from the gels because, whereas some look weak, others look much stronger irrespective of the motif recognized in each case. These results suggest that, depending on the context and likely due to the oligomeric nature of GAGA factor, some potentially weak sites may become of higher occupancy than expected and be clearly observable if surrounded by high affinity sites, and vice versa, some sites showing weak binding may contain a high affinity GAGA binding site, which per se would not be sufficient, as has already been observed (4,36). Our results are in agreement with a distribution of GAGA footprints centered around 14 to 15 bp (14). Some of the sites appear to be double, as they often contain two high affinity GAGA binding sites, and in general, protections larger than the average tend to be observed on these sites suggesting that both sequences are similarly bound. Similar results showing an extensive GAGA binding have been reported for other natural promoters (5,15,37).
These results suggested a potential autoregulation of Trl promoter by GAGA factors. So far, GAGA factors have never been reported to repress transcription of any promoter but to stimulate transcription much like we have shown for the eve ) or using anti-GAGA 519 antibody (middle row (␣GAGA antibody)) or an unrelated antibody (bottom row (unrelated antibody)). Each sample was analyzed for the five regions studied (columns labeled I to V). The relative enrichment for GAGA is indicated below the blot for each fragment and was calculated as the ratio of signal in antibody to mock immunoprecipitations. In the bottom panel, relative enrichment is plotted versus promoter position to give a visual representation of GAGA promoter occupancy in vivo. B, transcription of Trl promoter encompassing positions Ϫ3470 to ϩ737 is stimulated by depletion of endogenous GAGA factors in S2 cells. Stimulatory effect of increasing RNAi GAGA dosage on the expression of Trl promoter in the absence of GAGA overexpression. Luciferase and ␤-galactosidase activities were measured at 72 h post-transfection. stripe 2 promoter. The presence of several GAGA binding sites on the eve promoter was previously shown and therefore was suspected to be under GAGA regulation (5,38). Unexpectedly however, co-transfection of either GAGA 519 or GAGA 581 led to a dose-dependent down-regulation of Trl that reduced transcription levels to 15%. The results indicated that whereas neither the Q-domain nor POZ/BTB domain was required for repression, a single amino acid change affecting the zinc-finger DNA binding domain largely abolished repression. As in the absence of the POZ/BTB domain GAGA binds as a monomer and cannot impose any three-dimensional structure on the promoter (4, 5), architectural arguments cannot explain GAGA repressive effect. On the other hand, the fact that this repression can be changed to a weak activation by replacing the activation domain may be indicative that steric hindrance to other factors may not be the only event, unless GAGA is unable to stimulate Trl transcription because of some basal promoter selectivity. Neither is it a simple switch of the Q-domain from activation to repression because the Q-domain is dispensable. On the other hand, the weak stimulation observed with GAGA⌬QVP16 cannot be explained by the basal promoter selectivity that VP16 presents (39) because a similar weak activation (9-fold) was previously observed for this mutant using a basal promoter that responds robustly to activation by Gal4VP16 (more than 1000-fold; not shown). Thus, the basis of its weak activation potential is likely to be the particular features of the chimeric construct itself (9). Our results suggest that the GAGA DNA binding domain on its own or in combination with the X-domain may be responsible for the repression of the Trl promoter. In addition, a double deletion mutant has shown that the POZ/BTB domain and Q-domains do not act as redundant, repressive domains that might separately be sufficient for repression in combination with the DNA binding domain (results not shown).
The negative regulation of the Trl gene by its products GAGA 519 and GAGA 581 suggests that intranuclear concentration of GAGA is presumably kept at a constant level in S2 cells. In fact, endogenous GAGA depletion experiments indicate that Trl gene is actively kept at a submaximal level of transcription. Nevertheless, repression is not complete; intact S2 cells express normal levels of GAGA factors, whereas a high amount of GAGA can be detected on the Trl promoter, especially on the region that has been shown to mediate repression. The occupancy of a low affinity GAGA site, which would be sensitive to changes in GAGA concentration, might be a mechanism to explain the repression. Nevertheless, the oligomeric nature of GAGA factor and its simultaneous binding to several sites question this simple mechanism. In any case, it is very unlikely that GAGA is acting alone but rather in combination with other factor(s), which should be Trl-specific. Two models that are not mutually exclusive and involve the presence of at least an additional factor can be envisaged.
In the first model, upon interacting with some factor(s) GAGA would trigger the repressive effect. Using synthetic promoters in which bona fide GAGA binding sites were inserted near a TATA box, or in minimal natural promoters in which this situation is reproduced, GAGA only stimulated transcription in vitro and in vivo (2,4,9,14,26,33). From these results GAGA should essentially be considered as a transcriptional activator. On the other hand, the hypothetical factor that drives GAGA-mediated repression of the Trl promoter must be specific for this promoter and requires GAGA binding to DNA. This hypothesis implicitly requires the existence of a DNA binding site for the unknown factor(s) in the vicinity of the GAGA sites because the silencing observed can be transported to the CMV promoter by the FIIϩFIII region, whereas it is not observed with other promoters (for instance eve, which was assayed here). The observation that each of these two Trl promoter regions can still mediate GAGA silencing on its own, albeit partially, might suggest either the presence of one binding site, or more, for the unknown factor in each region or the existence of two or more factors, which co-operate and bind to either one or the other region. In this scheme, GAGA would bind to both regions and would act as a necessary factor to elicit repression. In agreement with this proposal, other fragments from the Trl promoter, despite having in some cases several good GAGA binding sites, could not support repression, suggesting that although GAGA factors (either GAGA 519 or GAGA 581 ) are necessary, they are not sufficient. This model has a precedent in Dorsal; although it is an activator, its binding to a silencer region of the zerknü llt gene results in recruitment of the co-repressor Groucho. This does not take place with Dorsal sites alone but requires the presence of an AT2 DNA element that, by binding cut and dead ringer factors, collaborates with Dorsal in the recruitment of Groucho and results in its conversion to a repressor (40,41).
The second model suggests that GAGA displaces an activating factor from the Trl promoter, in such a way that transcription is not actually repressed but deactivated. The model also requires another factor, in this case an activator different to GAGA, bound close enough to some GAGA sites in the FIIϩFIII region so that it can gradually be displaced by increasing amounts of GAGA. To explain why GAGA binding does not result in some activation, we have to consider that the unknown activator is largely more active than GAGA itself and/or that Trl basal promoter is not responsive to GAGA. This model has a precedent in SV40 virus large T-antigen, which initially stimulates transcription from the late promoter of the virus. As the viral cycle progresses, large T-antigen accumulates, resulting in its binding to the early promoter with a displacement of the stimulatory Sp1 factor from this promoter, which leads to a repression of its own transcription unit (42,43).
The existence of a GAGA site on a certain promoter cannot be correlated directly to an activation of transcription. Whether GAGA binding to a promoter results in activation or repression, it must be encoded in the specific DNA sequence of the promoter itself and must depend on factor(s) other than GAGA. Although peculiar, in the sense that GAGA factors usually are activators, this activation-to-repression conversion is not unprecedented.